Fig 2: Experimental confirmation of the Aurora B and ZAK interactions in MDA-MB-231. a Validation of the AURKB–ZAK interactions using two Qiagen siRNAs (siA1 and siA2) and two Ambion siRNAs (siA3 and siA4) for AURKB, and similarly for ZAK (siZ1–siZ4). For each siRNA, 16 nM of final concentrations were used for both single siRNAs and double siRNAs (i.e., an 8+8 nM combination in double siRNAs). The highest single agent (HSA) synergy scores were calculated as the difference between the siRNA double knockdown effects minus the maximal effects of the single knockdowns in cell viability inhibition (CellTiter-Glo) and toxicity (CellTox Green) assays, respectively (see Methods for details). Standard error of means was calculated over three replicates. b Knockdown effect of AURKB and ZAK by each of the individual Qiagen siRNAs and their combinations using Western blot assays. Standard error of means was calculated over three replicates. c HSA synergy scores for AURKB and ZAK double knock-out using combinatorial sgRNAs (sgA1 for AURKB and sgZ1, sgZ2 for ZAK) in CRISPR/Cas9 system. Standard error of means was calculated over eight replicates. d Cell inhibition and toxicity effects were measured for Aurora B inhibitors combined with the two ZAK siRNAs. *p < 0.05; **p < 0.01; ***p < 0.001 (Wilcoxon rank sum test, two-sided). The labels on the x-axis indicate the different siRNA combinations

Fig 3: Experimental validation of ZAK function in cancer.(a) Immunoblots of ZAK TV1 and TV2 expressions show that protein level of TV1 is higher in gastric tumours and cell lines, compared with normal stomach tissues. ZAK TV1 was detected with Bethyl α-ZAK antibody and TV2 with Sigma α-ZAK antibody (see Methods). (b) ZAK TV1, but not TV2, can stimulate multiple transcriptional programs related to cancer pathways. Transcription reporter assay in 293 cells transfected with empty vector, TV1 or TV2 along with the indicated firefly luciferase reporter construct (AP1, NFkB and TCF/LEF). Activity is normalized to cell number using CellTiter-Glo. Immunoblot shows relative ZAK isoform expression from 293 cells transfected with the indicated construct. ZAK was detected with Sigma α-ZAK antibody. (c) Depletion of ZAK from gastric cancer cell lines inhibits cell growth. In cell lines where ZAK knockdown led to reduced viability, there was consistently high TV1 expression, while TV2 expression was marginal and variable (for example, IM-95m cell line, see for example, panel a). Cell viability analysis was carried out 6 days after infection of gastric cancer cell lines with independent ZAK shRNAs. Cell number is normalized to shNTC-infected cells. Immunoblot indicates the level of ZAK-TV1 depletion 4 days after infection. ZAK was detected with Bethyl α-ZAK antibody.

Fig 4: Differential ZAK isoform usage between normal and tumour samples.(a) ZAK gene model and protein domain structure. Thick bars: coding exons; thin bars: UTR. TV2 lacks the last nine exons and has a long terminal coding and non-coding exon. Blue and red indicate unique TV1 and TV2 sequences. Transcript variant 1 (TV1) encodes a longer protein product with a sterile alpha motif domain. (b) ZAK TV1 fraction is significantly higher in gastric tumours (left) and colon tumours (right), compared with normal adjacent tissues. The fraction of TV1 was measured by the ratio between the number of reads uniquely assignable to TV1 and the number of reads mapped to the entire ZAK gene. To account for smooth muscle contamination in normal tissues, we fit a linear model with smoothelin expression as predictor and the log isoform fraction as response, and used the residuals of the model as the ‘adjusted isoform fraction’. Dots represent samples. Grey lines connect matched tumour and normal samples. The boxes in the box-and-whisker plots represent the interquartile range between the first and third quartiles; the dashed lines (whiskers) extend to the most extreme data points, which is no more than 1.5 times the interquartile range from the box. (c) ZAK isoform fractions derived from RNAseq data correlate with quantitative PCR (qPCR) measurements. For nine gastric cancer cell lines in our study, we quantified the ratio between ZAK total expression and ZAK TV1 expression using qPCR, and compared the measurements with the isoform fraction we derived from the RNAseq data. The two measurements have significant correlation (Pearson’s correlation coefficient r=0.91, P-value=0.00077). (d) ZAK isoform expression in six TCGA data sets where there are >10 normal samples. Normal samples are represented by blue dots and tumour samples by red dots. ZAK TV1 fraction is significantly higher (adjusted P-value <0.001 and fold change >2) in breast and bladder cancer data (marked by the green asterisks).

Fig 5: Dynamic modeling of MDA-MB-231 signaling network supports the context-specific combination effects. a Signaling network based on selected interaction partners of Aurora B, ZAK, and CSF1R. The node colors indicate the log2 mRNA expression levels of the genes. Arrow-heads represent activation and bar-headed edges represent inhibition of the target proteins, retrieved from OmniPath.23 Red-circled area highlights the p38 pathway that may be activated by ZAK and blue-circled area suggested the role of TGF-β pathway that involves CSF1R. b Simulated cell viability in response to single and double gene knock-downs. The fraction of viable cells decreased further when both AURKB and ZAK were silenced, while a simultaneous knock-down of AURKB and CSF1R increased the cell proliferation compared to the knock-down of AURKB alone. Standard error of means was calculated over ten replicates. c Left panel: the steady state expression levels of genes inferred by SGNS2 and COPASI. Right panel: the average influence of the Kd (degradation rate) and Kp (production rate) parameters of each gene on the expression level of all the genes in the signaling network. d Left panel: the overall survival curves for breast cancer patients with higher ZAK gene expression (z-score > 1.5 in RNA-Seq data, n = 59) versus the others (n = 1036). Right panel: the AURKB gene expression and TP53 mutation frequency differences between TNBC and non-TNBC patients. Error bars represent standard errors