Aurora kinase A drives non-canonical YAP1/TAZ crosstalk to sustain primary resistance to anti-EGFR therapies in colorectal cancer

Abstract

Anti-epidermal growth factor receptor (EGFR) therapies are the most recommended first-line treatment for RAS/RAF wild-type unresectable metastatic colorectal cancer (CRC) according to the European Society for Medical Oncology guidelines. However, primary resistance renders this treatment ineffective for almost 40% of patients. Our previous work identified Aurora kinase A (AURKA) as a key resistance driver through non-canonical, Hippo-independent Yes-associated protein 1 (YAP1) activation. However, the role of the other main Hippo coactivator, transcriptional coactivator with PDZ-binding motif (TAZ), in this resistance mechanism remains unexplored. By integrating preclinical in vitro and in vivo models, including cell lines and patient-derived xenografts, with RNA sequencing, we investigated the impact of TAZ overexpression in cetuximab resistance driven by the AURKA/YAP1 axis. Our findings reveal that TAZ overexpression sustains YAP1-mediated resistance and stemness. Even under YAP1 suppression, TAZ-overexpressing cells remain unresponsive to anti-EGFR therapies, whereas dual YAP1/TAZ silencing restores sensitivity. Treatment with alisertib, a phase III AURKA inhibitor, simultaneously destabilizes YAP1 and TAZ, restoring anti-EGFR efficacy by suppressing stemness. Transcriptomic analyses further show that AURKA inhibition and dual YAP1/TAZ suppression disrupt stem-like traits and reveal transcriptional deregulations affecting nucleotide metabolism. These findings demonstrate that AURKA orchestrates YAP1/TAZ crosstalk, which is crucial for driving stemness and resistance to anti-EGFR therapies, highlighting AURKA inhibitors as a promising strategy to enhance anti-EGFR therapies in metastatic CRC.

Graphical abstract

The AURKA/YAP1/TAZ axis drives resistance and stemness in metastatic colorectal cancer. Targeting AURKA with the phase III inhibitor alisertib suppresses resistance by destabilizing YAP1 and TAZ, reversing anti-EGFR therapy failure in a YAP1/TAZ-dependent manner, highlighting a promising therapeutic approach to expand the benefits of cetuximab to a broader population.

Introduction

While only 25% of colorectal cancer (CRC) patients are diagnosed in metastatic stages,¹ they account for 85% of deaths attributed to this disease due to the difficulty of undergoing surgical resection and the limited efficacy of existing treatments.² For patients with RAS/RAF wild-type unresectable metastatic CRC (mCRC), the European Society for Medical Oncology recommends anti-epidermal growth factor receptor (EGFR) therapies, such as cetuximab. It is a monoclonal antibody that blocks ligand-induced activation of the EGFR, thereby inhibiting downstream signaling pathways critical for tumor cell proliferation and survival, like the mitogen-activated protein kinase and phosphatidylinositol 3-kinase/protein kinase B (PI3K/AKT) pathways.³ Despite improved clinical outcomes since its approval, nearly two out of every five individuals fail to benefit from this treatment, suffering adverse effects that not only endanger their survival but also reduce their quality of life.⁴

Cancer stem cells (CSCs) play a crucial role in mediating drug resistance across various cancer types, including CRC, owing to their phenotypic adaptability and heightened quiescence, enabling them to effectively withstand stress induced by cancer treatments and facilitate disease progression.^5,6,7 A key regulator of CSC traits is the Hippo signaling pathway, a highly conserved tumor suppressor cascade that limits cell proliferation and promotes apoptosis in normal tissues. The core Hippo kinase module—MST1/2 and LATS1/2—phosphorylates and inactivates the transcriptional co-activators Yes-associated protein 1 (YAP1) and transcriptional coactivator with PDZ-binding motif (TAZ), retaining them in the cytoplasm and targeting them for degradation.⁸ In cancer contexts, this tumor suppressor pathway is often deregulated, leading to the overactivation of YAP1/TAZ and favoring sustained proliferation, stemness, and resistance to both conventional^9,10 and targeted therapies.^11,12 However, the absence of specific inhibitors for YAP1 and TAZ beyond phase 1 clinical trials renders them challenging to target pharmacologically, posing obstacles to their translation into clinically applicable interventions.

Our recent findings have revealed that the phosphorylation of YAP1 by Aurora kinase A (AURKA), a master mitotic regulator, serves as a mediator of resistance to anti-EGFR therapies by augmenting CSC characteristics.¹³ These results underscore alisertib—a selective AURKA inhibitor currently in phase 3 clinical trials—as a promising therapeutic strategy to indirectly inhibit YAP1, for which direct inhibitors are not yet available beyond phase 1 trials. By targeting AURKA, alisertib enhanced the efficacy of cetuximab, potentially expanding treatment benefits to a larger subset of patients with cetuximab-resistant CRC.

However, while the role of YAP1 in therapeutic resistance in cancer is increasingly well understood,¹⁴ the functions of TAZ—the other major effector of the Hippo pathway that shares some redundant functions with YAP1—remain unclear. This prompted us to investigate whether TAZ not only contributes to cetuximab resistance but also is involved within the AURKA/YAP1 axis. Our findings indicate that TAZ is upregulated in CRC cell lines and patients refractory to cetuximab, where it enhances YAP1-mediated stemness and contributes to cetuximab resistance in an AURKA-dependent fashion. Knocking down YAP1 alone proves ineffective in overcoming cetuximab resistance and stemness when TAZ is overexpressed. However, simultaneous silencing of both YAP1 and TAZ effectively restores the anti-EGFR efficacy and suppresses stemness. Furthermore, AURKA inhibition not only prevents YAP1 activation but also destabilizes TAZ, suggesting its potential as a therapeutic approach for all cetuximab-resistant tumors exhibiting YAP1/TAZ overactivation. RNA sequencing (RNA-seq) analysis reveals shared transcriptional deregulations between AURKA inhibition and YAP1/TAZ silencing, highlighting nucleotide metabolism as a potential mediator of this resistant and stem-like phenotype.

Results

Transcriptomic analyses reveal TAZ upregulation during cetuximab primary resistance

Deregulation of the Hippo pathway stands out as one of the most extensively studied mechanisms contributing to therapeutic resistance in cancer. Our previous investigations have established that YAP1 plays a crucial role in driving cetuximab resistance in mCRC through an AURKA-dependent mechanism. However, the role of TAZ in cetuximab resistance has not yet been investigated. To address this gap, we conducted a comprehensive bioinformatic analysis of publicly available transcriptomic datasets containing microarray gene expression data from CRC cell lines (GEO: GSE59857) annotated for cetuximab responsiveness and RAS/RAF status. Samples with activating mutations in RAS or RAF were excluded from the analysis. Then, for every dataset, we evaluated the expression levels of TAZ, encoded by the WW domain-containing transcription regulator 1 (WWTR1) gene, in both cetuximab-sensitive and -resistant CRC samples. Notably, individuals responsive to cetuximab exhibited markedly lower levels of WWTR1 gene expression, whereas resistant individuals displayed a spectrum ranging from a pattern akin to sensitivity to notable overexpression. We identified a subgroup of high-TAZ individuals among the resistant cases, revealing a significant upregulation of these genes compared with their sensitive counterparts (Figure 1A).

Figure 1.

TAZ is upregulated in cetuximab-resistant CRC cell lines and patients

(A) WWTR1 expression in CRC cell lines sensitive and resistant to anti-EGFR therapies included in the GSE59857 dataset. (B) Scores for gene signatures previously related to cetuximab sensitivity and resistance calculated for each patient included in the COAD cohort of the TCGA. Patients were classified into WWTR1-high or low using the 75th percentile as the cutoff value. Differences between groups were evaluated using a two-tailed t test. (C) Kaplan-Meier curve evaluating the overall survival of CRC patients contained in the COAD cohort, obtained from GEPIA2¹⁵ (http://gepia2.cancer-pku.cn/#index). The 35th percentile was used as the cutoff for stratifying samples according to WWTR1 gene expression. Statistical differences between groups were calculated using a log rank test. (D) Kaplan-Meier curve evaluating the disease-free survival of CRC patients contained in the COAD cohort, obtained from GEPIA2¹⁵ (http://gepia2.cancer-pku.cn/#index). The 35th percentile was used as the cutoff for stratifying samples according to WWTR1 gene expression. Statistical differences between groups were calculated using a log rank test. ∗∗p < 0.01, ∗∗∗p < 0.001.

To further confirm these findings, we analyzed transcriptomic data from The Cancer Genome Atlas (TCGA)-COAD cohort, including information from 512 CRC patients. Using two established gene signatures for predicting cetuximab sensitivity and resistance,¹⁶ we assigned a score to each patient and each signature using single sample Gene Set Enrichment Analysis (ssGSEA) guidelines (see supplemental information). Subsequently, we stratified the samples into two groups based on WWTR1 gene expression levels and compared, for each signature, the average of the scores of all patients in each group. As expected, patients with elevated WWTR1 expression levels exhibited significantly higher resistance-signature scores compared with those with low expression of WWTR1. Conversely, patients with higher sensitivity scores were predominantly found in the low-WWTR1 expression group. Notably, the difference observed in the sensitivity signature was relatively smaller, aligning with the existence of a subgroup of resistant patients exhibiting similar low WWTR1 expression patterns as the sensitive group (Figure 1B). This suggests a potential role for TAZ overexpression in the absence of cetuximab response, while its low expression may not necessarily predict sensitivity. These findings underscore the complex nature of cetuximab response mechanisms and highlight the need for a comprehensive understanding of TAZ involvement in resistance.

To further explore the prognostic implications of TAZ expression in CRC patients, we analyzed the prognostic impact of TAZ gene expression in CRC patients of the COAD-TCGA cohort. As expected, higher levels of TAZ expression may predict not only lower overall survival but also an increased relapse risk, supporting its importance in CRC (Figures 1C and 1D). These findings justify additional research into the involvement of TAZ in CRC and its possible function in facilitating resistance to anti-EGFR therapies.

TAZ enhances YAP1-induced cetuximab resistance in CRC cell lines

To assess the role of TAZ as a driver of primary resistance to anti-EGFRs, we selected two cetuximab-sensitive CRC cell lines alongside two resistant ones. We examined the levels of YAP1, YAP phosphorylation at Ser397, and TAZ proteins within the four cell lines. As we have previously described, YAP1 exhibited hyperphosphorylation in the resistant CRC cell lines compared with the sensitive ones. Interestingly, TAZ overexpression was observed only in one of the resistant CRC cell lines (Figure 2A), prompting further investigation into the individual contribution of TAZ to this primary resistance phenomenon.

Figure 2.

TAZ overexpression sustains YAP1-dependent cetuximab resistance

(A) Western blot illustrating the levels of TAZ, YAP1, and YAP1-activating phosphorylation at Ser397 (n = 3). (B) Western blot illustrating the levels of MET (n = 3), TAZ (n = 3), YAP1 (n = 3), and p-YAP1(Ser397) (n = 4) after YAP1 knockdown in the SW48 cell line. Results are plotted as the mean ± SD. Differences between groups were studied using a two-tailed t test. (C) Western blot illustrating protein expression of MET (n = 3), p-MET (n = 4), TAZ (n = 3), YAP1 (n = 3), and p-YAP(Ser397) (n = 4) after silencing YAP1 or TAZ in the C10 cell line. Results are plotted as the mean ± SD. Differences between groups were studied using an one-way ANOVA test followed by Tukey’s test for multiple comparisons. (D) Cell proliferation analysis of SW48 cell line after silencing YAP1 (n = 3). Results are plotted as the mean ± SD. Differences between groups were studied using a two-tailed t test. ∗∗p < 0.01. (E) Cell proliferation analysis of C10 cell line after silencing YAP1 or TAZ (n = 3). Results are plotted as the mean ± SD. Differences between groups were studied using a two-tailed t test. ∗p < 0.05.

To address this question, we conducted short hairpin RNA (shRNA)-based knockdowns of YAP1 or TAZ in the resistant C10 and SW48 cell lines and evaluated their response to cetuximab treatment. In the TAZ-deficient SW48 cell line, silencing of YAP1 alone was sufficient to restore cetuximab sensitivity (Figures 2B and 2D), whereas resistance persisted in the C10 cell line. Silencing of TAZ alone was also ineffective in reinstating cetuximab efficacy in the C10 cell line. However, simultaneous silencing of both genes resulted in a more than 35% reduction in cell proliferation for the C10 cell line after exposure to cetuximab (Figures 2C and 2E).

YAP1-driven resistance often occurs in a c-MET-dependent manner.^8,13 Therefore, we assessed its protein levels following YAP or TAZ knockdown. As expected, in the TAZ-deficient SW48 cell line, silencing of YAP1 alone led to a reduction in c-MET expression by more than 50% (Figure 2B). In the C10 cell line, both YAP1 and TAZ knockdown resulted in decreased c-MET expression, comparable between both single knock-downs and the double YAP1/TAZ silencing (reductions of 57%, 71%, and 59% compared with the control; p < 0.01). However, the double knockdown displayed an additive effect compared with the single silencing in the phosphorylated active form of this protein (70% reduction compared with control [p < 0.01], compared with 42% and 53% under YAP1 and TAZ knockdown, respectively) (Figure 2C). These findings suggest that TAZ can sustain YAP1-mediated resistance and compensate for its inhibition, emphasizing the importance of dual inhibition strategies to enhance cetuximab efficacy in a greater number of resistant CRC patients.

TAZ sustains YAP-1-mediated stem-like phenotypes in cellular CRC models of cetuximab resistance

Given the well-established role of YAP1 in inducing stemness in CRC, we sought to investigate whether TAZ influences this phenotype through similar pathways. To gain a comprehensive understanding of TAZ’s involvement in the CSC phenotype, we assessed the correlation between WWTR1 expression and each of the gene signatures using TCGA data. By using 26 publicly available gene signatures associated with CSC traits, we calculated a stemness score for each individual sample. Subsequently, we assessed the correlation between WWTR1 expression and each of the gene signatures. Of the 26 signatures, 21 displayed a positive correlation with WWTR1 expression, with 8 of them showing a particularly strong correlation (adjusted R² > 0.25, p < 0.01) (Figure S1).

To validate the individual contributions of YAP1 and TAZ to the CSC phenotype, we examined the effect of silencing these genes on stemness, including colony formation ability, the levels of well-established CRC stemness markers like CD133, CD44, and OCT-4 (coded by the POU5F1 gene), together with aldehyde dehydrogenase-1 (ALDH1) activity, and colonosphere formation. As expected, silencing of YAP1 alone effectively reduced colony formation ability (Figure 3A) and ALDH1 activity (Figure 3B) in the TAZ-deficient SW48 cell line. However, in the C10 cell line, while YAP1 silencing impaired colony formation ability, it had no significant effect on either ALDH1 activity or spheroid formation ability. A similar trend was observed with the WWTR1 knockdown, with no notable changes observed in any of these three markers compared with the cell line carrying the empty vector. Notably, simultaneous knockdown of both YAP1 and TAZ replicated the effects observed with YAP1 silencing alone in the SW48 cell line, resulting in a 75% reduction in the number of colonies and spheroid size, along with a more than 40% reduction in ALDH1 activity (Figures 3A–3C). To further validate these findings, we assessed the expression of the stemness markers CD133 and CD44 by western blot, and POU5F1 by RT-qPCR, in C10 cells following YAP1, TAZ, or dual silencing, compared with cells transduced with the empty vector. Consistent with previous results, the most pronounced reductions across all markers were observed upon simultaneous knockdown of YAP1 and TAZ. While single shYAP1 know-down led to a statistically significant decrease in CD44 (50% vs. control; p < 0.05), it was numerically higher after concomitantly silencing both Hippo pathway effects (57% vs. control; p < 0.01). For the other two stemness markers, only statistically significant differences were achieved when both genes were silenced (64% and 59% reductions in CD133 and POUF1 vs. control, respectively; p < 0.05) (Figures 3D and 3E), confirming that TAZ can sustain the stemness phenotype driven by YAP1, and the combined suppression of both effectors is required to efficiently mitigate stemness characteristics.

Figure 3.

TAZ sustains YAP1-driven acquisition of stemness traits in CRC cell lines

(A) Colony-formation assay after silencing of YAP1 or TAZ in the cetuximab-resistant SW48 and C10 cell lines. (B) ALDH1 activity assay performed in SW48 and C10 cell lines with knockdown of YAP1 or TAZ compared with the empty vector controls. (C) Spheroid formation assay of C10 after YAP1 or TAZ silencing. Spheroid volume and number were measured. (D) Western blot illustrating the levels of the stemness markers CD133 (n = 4) and CD44 (n = 3) in the C10 cell line after knocking-down YAP1, TAZ, or both, compared with cell line transduced with an empty vector. (E) Gene expression levels of the stemness marker POU5F1 in the C10 cell line in the C10 cell line after knocking-down YAP1, TAZ, or both compared with cell line transduced with an empty vector (n = 4). Results are plotted as the mean ± SD. Differences between groups were studied using a two-tailed t test in SW48 cell line. A one-way ANOVA and Tukey's multiple comparisons test were performed to analyze data obtained from the C10 cell line. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

AURKA inhibition promotes TAZ destabilization and suppresses stemness both in vivo and in vitro

Our previous findings demonstrated the efficacy of AURKA inhibition with alisertib in preventing YAP1 activation, thus contributing to overcoming cetuximab resistance. This led us to investigate whether AURKA inhibition could also affect TAZ stability, potentially impeding compensatory resistance mechanisms. Subsequently, the C10 cell line was treated with alisertib for 48 h, and TAZ levels were assessed via western blot analysis. As anticipated, both AURKA inhibition alone and in combination with cetuximab resulted in a reduction of TAZ levels by up to 50% (Figure 4A). To validate these findings in vivo, we examined TAZ levels in our previously established cetuximab-resistant PDX model treated with alisertib alone and in combination with cetuximab. As expected, tumors treated with the AURKA inhibitor exhibited decreased TAZ levels compared with the control and cetuximab monotherapy groups (Figure 4B). These results confirm that AURKA inhibition not only suppresses YAP1-driven cetuximab resistance but also mitigates compensatory mechanisms involving the overactivation of its ortholog TAZ.

Figure 4.

AURKA inhibition compromises TAZ stability and suppress stem-related transcriptional reprograming

(A) Western blot illustrating TAZ expression levels after treatment with the AURKA inhibitor alisertib alone or in combination with cetuximab in the C10 cell line (n = 4). Results are plotted as the mean ± SD. Differences between groups were studied using an one-way ANOVA test followed by Tukey’s test for multiple comparisons. Cells were stimulated with 40 ng/mL EGF after both treatments to EGFR activation. (B) TAZ protein levels after alisertib or cetuximab treatment in the cetuximab-resistant PDX STF557. The selected image is representative of two biological replicates for each condition (n = 4). Differences between groups were assessed using a one-way ANOVA test followed by Tukey’s test for multiple comparisons. (C) Volcano plot illustrating 10 of the most DEGs detected in the C10 cell line after alisertib treatment. (D) GSEA analysis of the 10 most significantly deregulated stemness signatures after AURKA inhibition in the C10 cell line. (E) GSEA plot of the three deregulated stemness signatures with the best correlation with WWTR1 expression in the COAD TCGA cohort. (F) Protein levels of CD133 in the C10 cell line after treatment with alisertib or cetuximab. Results are plotted as the mean ± SD. Differences between groups were studied using a one-way ANOVA test followed by Tukey’s test for multiple comparisons. (G) Protein levels of CD133 in the C10 cell line after treatment with alisertib or cetuximab. Results are plotted as the mean ± SD. Differences between groups were studied using a one-way ANOVA test followed by Tukey’s test for multiple comparisons. ALS, alisertib; COM, combined; CTR, control; CTX, cetuximab; EGF, epidermal growth factor. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

In our previous work, we demonstrated that AURKA inhibition suppresses stemness reprogramming, thereby overcoming cetuximab resistance. To further investigate the transcriptional changes underlying this effect, we performed RNA-seq on the C10 cell line following 24 h of treatment with either alisertib or DMSO (control). AURKA inhibition revealed 815 differentially expressed genes (DEGs), with 432 upregulated and 383 downregulated. Notably, some significantly downregulated genes, such as MYB,¹⁷ PROM1,¹⁸ and TP73¹⁹ (Figure 4C), have been extensively implicated in the acquisition of CSC traits in various cancers, suggesting their potential role as drivers of stemness in CRC in an AURKA-dependent manner. To corroborate the suppression of stemness phenotypes, we performed GSEA analysis using 26 stemness-related signatures defined in http://stemchecker.sysbiolab.eu. Ten of these signatures exhibited significantly different expression between the alisertib-treated and control groups. In line with our hypothesis and prior research, nine of them were significantly downregulated following alisertib exposure (Figure 4D), supporting the potential of AURKA inhibition in targeting CSCs and rendering them more susceptible to therapeutic agents. Notably, three of the downregulated signatures demonstrated a robust correlation with WWTR1 expression in TCGA data (Figures 4E and S1), suggesting that YAP1 and TAZ may be involved on this mechanism.

As noted, PROM1, one of the most downregulated stemness-associated genes following AURKA inhibition in our RNA-seq analysis, was also significantly reduced upon YAP1/TAZ silencing (Figure 3D). To validate that CD133 (encoded by PROM1) is regulated by AURKA in a YAP1/TAZ-dependent manner, we examined CD133 protein levels after alisertib treatment. As expected, AURKA inhibition resulted in a greater than 50% reduction in CD133 expression in both C10 (Figure 4F) and SW48 (Figure 4G) cell lines. These results reinforce the notion that AURKA regulates the YAP1/TAZ balance to induce stemness-related transcriptional programs and confer cetuximab resistance in CRC.

AURKA inhibition modulates gene signatures related to prognosis after cetuximab treatment

Given that AURKA inhibition restores cetuximab efficacy in a YAP1/TAZ-dependent manner, we performed a GSEA analysis to study the expression changes in biological process terms in Gene Ontology. Sunakawa et al. demonstrated in 2017²⁰ that certain gene sets predict better or worse outcomes in patients treated with cetuximab (Figure 5A). Hence, we aimed to determine whether AURKA inhibition could modulate these pathways in the C10 cell line following alisertib treatment. Four of the gene sets associated with worse outcomes after anti-EGFR treatment were downregulated following AURKA inhibition, including processes related to DNA metabolism and repair, biosynthesis of nucleosides, and cell-cycle phase transition (Figure 5B). Conversely, gene sets associated with a good prognosis in patients treated with cetuximab, such as leukocyte migration, vasculature development, and Notch1-related signaling pathways, were upregulated after AURKA inhibition (Figure 5C). Altogether, these results suggest that AURKA inhibition contributes to gene expression reprogramming, rendering cells more sensitive to anti-EGFR therapies in mCRC.

Figure 5.

Alisertib treatment reduces the expression of gene signatures related to poor prognosis after cetuximab treatment and potentiates the good-prognostic ones

(A) Schematic figure illustrating the good and poor prognostic signatures previously described by Sunakawa and colleagues.²⁰ (B) Summary of the GSEA results for the prognostic-related signatures after cetuximab treatment. (C) GSEA plots of the four poor-prognostic signatures after alisertib treatment in the C10 cell line. (D) GSEA plots of the four good-prognostic signatures after alisertib treatment in the C10 cell line.

Shared gene expression changes led by AURKA and YAP1/TAZ point to nucleotide metabolism as a potential driver of stemness and cetuximab resistance

To validate the transcriptomic changes associated with an alisertib-induced sensitive-like state, particularly dependent on YAP1/TAZ, we analyzed the RNA-seq data featuring the HCT116 cell line data with simultaneous YAP1 and TAZ silencing (GEO: GSE176475). Despite its KRAS-mutant status, this cell line demonstrates substantial dependence on the Hippo pathway,²¹ and the suppression of the cetuximab resistance by concomitant YAP1/TAZ silencing in this cell line suggests a shared resistance mechanism with our RAS wild-type C10 cell line.²² DEGs resulting from YAP1/TAZ silencing confirmed the downregulation of the two Hippo pathway effectors (Figure S2A).

GSEA was conducted to examine whether the suppression of stemness signatures induced by alisertib correlated with alterations resulting from YAP1/TAZ knockdown. Six out of the eight stemness signatures downregulated by AURKA inhibition exhibited a similar reduction after YAP1/TAZ silencing, as shown in Figure 6A. Further exploration of changes in gene signatures associated with prognosis following cetuximab treatment revealed that all the signatures linked to unfavorable outcomes post-anti-EGFR treatment experienced a decrease (Figure 6B). This suggests that inhibiting YAP1/TAZ activity effectively attenuates both stemness and cetuximab resistance, mirroring the effects observed with AURKA inhibition.

Figure 6.

Shared transcriptomic features between YAP1/TAZ silencing and AURKA inhibition

(A) GSEA results of the most deregulated stemness signatures after YAP1/TAZ knockdown in the HCT116 CRC cell line. (B) GSEA results of the cetuximab prognostic signatures described by Sunakawa and colleagues²⁰ after YAP1/TAZ knockdown in the HCT116 CRC cell line. (C) Venn diagram illustrating the common differentially negatively expressed genes (|logFC| > 0.329, FDR < 0.05) between YAP1/TAZ silencing and alisertib treatment. (D) A PPI network of the common downregulated genes obtained in the Venn diagram. Darker red indicates a more negative logFC, while increased size indicates a higher degree of the node. (E) Western blot illustrating the protein levels of RRM2 in the C10 cell line after being transduced with lentiviral vectors for silencing YAP1, TAZ, or both compared with the cell line transduced with the empty vector (n = 4). Differences between groups were assessed using a one-way ANOVA test followed by Tukey’s test for multiple comparisons. (F) Gene over-representation analysis of the Biological Processes Gene Ontology (GO) term of genes contained in the PPI network. The bar graph represents the 10 most significant terms deregulated according to their p value. (G) Gene over-representation analysis of the Molecular Function GO term of genes contained in the PPI network. The bar graph represents the 10 most significant terms deregulated according to their p value. (H) Gene over-representation analysis of the Cellular Component GO term of genes contained in the PPI network. The bar graph represents the 10 most significant terms deregulated according to their p value. (I) Analysis of RRM2 expression in cetuximab-sensitive and resistant CRC cell lines contained in the GSE59857 dataset. Differences between groups were evaluated using a two-tailed t test. ∗p < 0.05. Distribution of cetuximab-sensitive and resistant CRC cell lines included in the GSE59857 dataset according to their RRM2 expression. The median value was used as the threshold to classify samples according to RRM2 expression. Differential distribution of cetuximab response according to RRM2 expression category was evaluated using a Fisher’s exact test. The p value is plotted below the graph.

To comprehend the processes influenced by YAP1/TAZ activation through AURKA, we extracted the most frequently deregulated genes shared between AURKA inhibition and YAP/TAZ suppression. A set of 50 commonly downregulated genes was identified and visualized as a protein-protein interaction (PPI) network using String and Cytoscape. Within this set, 31 genes formed a highly interconnected network (Figures 6C and 6D). In contrast, no overlap was observed among common genes between YAP1/TAZ-upregulated genes and those downregulated by alisertib (Figure S2B), indicating that the shared downregulation specifically contributes to cellular behavioral alterations. Notably, this common downregulated gene set shows clear upregulation in CRC tumors compared with healthy tissues (Figure S2C), highlighting its potential as a therapeutic target.

An over-representation analysis of this network using BinGO revealed that these downregulated genes are predominantly involved in DNA replication and nucleotide metabolism, consistent with the poor prognosis associated with nucleotide synthesis-related signatures (Figure 6F). Additionally, the primary altered molecular functions relate to DNA binding and various nucleotide metabolism activities, corroborating the biological process findings (Figure 6G). In line with the previous, the most significantly affected cellular components were nuclear (Figure 6H).

To explore the association between the PPI network genes and CSCs, we examined correlations between the combined expression of these genes—calculated as a gene signature score using ssGSEA—and the 26 CSC signatures from TCGA COAD cohort data. This gene set showed positive correlations with many CSC signatures, several with strong significance (adjusted R² > 0.5; p < 0.05) (Figure S3). Furthermore, this gene set positively correlated with cetuximab resistance signatures and negatively with the cetuximab-sensitivity ones (Figure S4).

Among the key genes included within this PPI network, we identified ribonucleotide reductase regulatory subunit M2 (RRM2), a critical mediator of ribonucleotide metabolism, as one of the most downregulated and highly interconnected genes in our PPI network (Figure 6D). Given its reported association with YAP1 and its role in CSC transformation and drug resistance,^23,24 we studied whether RRM2 expression is regulated by both YAP1 and TAZ in the C10 cell line (Figure 6E). Consistent with transcriptomic analyses in HCT116 cells, RRM2 protein levels decreased by 57% in the C10 cell line following YAP1/TAZ silencing compared with control cells. Interestingly, YAP1 silencing alone had no significant effect on RRM2 levels, whereas TAZ knockdown reduced RRM2 by 35% (p < 0.05), supporting a cooperative role between these two effectors in driving specific transcriptional programs.

Having established that RRM2 reduction is at least partly driven by the AURKA/YAP1/TAZ axis, we evaluated RRM2 expression as a predictor of cetuximab response in CRC cell lines with known sensitivity status from the GSE59857 dataset. RRM2 expression was significantly higher in resistant cell lines compared with the sensitive ones. Nearly all the sensitive lines exhibited low RRM2 levels, while resistant lines showed heterogeneous expression (Figure 6I, left). Using the median expression as a cutoff, we classified samples as RRM2-high or -low and analyzed their distribution across sensitive and resistant groups. Fisher’s exact test confirmed significant enrichment of RRM2-high lines in the resistant group, while the sensitive group was enriched for RRM2-low lines (Figure 6I, right).

These findings highlight nucleotide metabolism as a key mechanism regulated by the AURKA/YAP1/TAZ axis that contributes to stemness and cetuximab resistance in mCRC. Further studies are warranted to clarify RRM2’s role as a driver of stemness in CRC and its potential as a surrogate biomarker to identify cetuximab-resistant patients with YAP1/TAZ-overactivated tumors who may benefit from adding AURKA inhibitors to their targeted therapy regimens.

Discussion

The Hippo pathway effectors, YAP1 and TAZ, are recognized as key oncogenic drivers in CRC.^25,26 Their overactivation leads to more aggressive tumor phenotypes,²⁷ increased metastatic potential,²⁸ and diminished responses to both chemotherapeutic agents⁹ and targeted therapies.¹² Our previous work established that YAP1 activation by AURKA-mediated phosphorylation at Ser397 orchestrates transcriptional reprogramming, which enhances stemness features and drives cetuximab resistance.¹³ However, the role of TAZ in this process remains unexplored. Here, we demonstrate that AURKA also regulates TAZ stability and that TAZ overexpression cooperates with YAP1 to sustain cetuximab resistance and stemness in cetuximab-resistant mCRC models.

Through a comprehensive bioinformatic analysis of publicly available transcriptomic data from CRC cell lines and patients, we identified a TAZ upregulation in a subset of cetuximab-refractory individuals (Figure 1). This observation prompted experimental validation, revealing that TAZ significantly contributes to sustaining YAP-1-mediated cetuximab resistance. Specifically, while YAP1 silencing restored cetuximab sensitivity in the TAZ-deficient SW48 cell line, it failed to do so in C10 cells, in which TAZ is overexpressed. Notably, this sensitivity was only restored upon concurrent YAP1/TAZ knockdown. These results align with evidence of functional redundancy between YAP1 and TAZ,¹⁴ underscoring the necessity of co-targeting both to enhance responses to tyrosine kinase inhibitors, such as sorafenib in hepatocellular carcinoma²⁹ and tamoxifen in breast cancer.³⁰ Furthermore, Plouffe et al.¹⁴ demonstrated in 2018 that YAP1 exerts a stronger influence on Hippo-mediated processes than TAZ, reinforcing YAP1’s dominant role in cetuximab resistance, with TAZ amplifying this effect.

CSCs are pivotal mediators of therapeutic resistance across multiple cancers, including CRC, due to their phenotypic plasticity and self-renewal capacity.^5,6,7 In cetuximab-resistant CRC models, the acquisition of stemness traits is mediated by the AURKA-YAP1 axis, prompting us to investigate whether TAZ acts independently of this axis or in an AURKA-dependent manner. Prior studies explored the potential interaction between AURKA and TAZ: Chang et al.³¹ demonstrated that TAZ’s promoter-binding activity is enhanced by AURKA but without elucidating their functional consequences. Additionally, recent findings by Marugán et al.³² demonstrate that the overexpression of TPX2, a bona fide activator of AURKA, increases TAZ expression. In line with these observations, we demonstrate that AURKA inhibition not only prevents YAP1 activation but also reduces TAZ stability, leading to a 40% reduction in TAZ levels following treatment with the AURKA inhibitor alisertib both in vitro and in vivo. This leads to the transcriptional reprogramming that suppresses stemness traits. Notably, AURKA inhibition downregulated CD133 expression at both the RNA and protein levels in C10 and SW48 CRC cell lines. CD133, a well-established marker of CSCs in CRC, is implicated in a bi-directional relationship with the Hippo pathway. While YAP1 promotes CD133 expression,^33,34 CD133 has been shown to stabilize TAZ,³⁵ enhancing its transcriptional activity. In our model, the reduction in CD133 expression was observed only when YAP1 and TAZ were silenced simultaneously, but not with individual knockdowns, indicating that both effectors act cooperatively in an AURKA-dependent manner to promote stemness traits and cetuximab resistance.

To gain further insights into the specific functions regulated by the AURKA/YAP1/TAZ axis, we performed an analysis of publicly available data from the HCT116 cetuximab-resistant CRC cell line after YAP1 and TAZ knockdown. Our analysis revealed that most of the CSC-related signatures downregulated by AURKA inhibition were also downregulated upon knockdown of these Hippo pathway effectors. To investigate potential drivers of this phenotype, we identified commonly downregulated genes from these two experiments and visualized them using a PPI network. This analysis revealed a highly interconnected network of genes primarily associated with nucleotide metabolism and DNA replication, with RRM2 emerging as one of the most significant genes. RRM2 is a central effector of the ribonucleotide metabolism that has been shown to be cooperatively regulated by YAP1 and TAZ, driving the acquisition of the CSC phenotype in breast cancer.²³ Additionally, TAZ silencing has been shown to reduce RRM2 expression, while its overexpression enhances RRM2 levels in renal cancer and lung adenocarcinoma cell lines in an MESH1-dependent manner.³⁶ Moreover, RRM2 has been identified as a transcriptional target of YAP1 in pancreatic ductal adenocarcinoma models.³⁷ In line with previously published evidence, our study demonstrates that this YAP1/TAZ-dependent regulation also occurs in CRC in an AURKA-dependent fashion, leading to the acquisition of stem cell properties and acting as a potential driver of primary resistance to anti-EGFR resistance in mCRC.

Overexpressed in 30 tumor types, including CRC, RRM2 is notably upregulated in cetuximab-resistant CRC cell lines (Figure 6I), underscoring its role as a potential predictive biomarker of cetuximab resistance, through the AURKA/YAP1/TAZ axis, and as a promising therapeutic target to overcome this resistance. RRM2 has been implicated in resistance to both chemotherapeutic agents³⁷ and immune checkpoint inhibitors, sustaining stemness properties in a PI3K/AKT-dependent manner.^38,39 Furthermore, RRM2 has been shown to be directly regulated by both YAP1 and TAZ, leading to cell growth and oncogenic senescence,²³ and also has been observed to mediate aggressiveness and stemness in multiple myeloma through the Wnt pathway.⁴⁰ Our analyses demonstrate that the genes within our network positively correlate with the majority of the stemness signatures (Figure S3), while patients with high expression of this signature also display higher scores for cetuximab-resistant signatures (Figure S4). These results align with pre-clinical studies performed in lung cancer, in which osalmid, an investigational compound that specifically targets RRM2, restores sensitivity to EGFR tyrosine kinase inhibitors in lung cancer models,⁴¹ laying the groundwork for future research into the role of this protein not only as a predictive biomarker but also as a therapeutic target.

Our research has unveiled the pivotal role of AURKA in orchestrating the collaborative actions of YAP1 and TAZ, driving the acquisition of a CSC phenotype and fostering resistance to cetuximab. Notably, the use of AURKA inhibitors has proven effective in suppressing YAP1/TAZ-mediated stemness and resistance to cetuximab. Mechanistically, the AURKA/YAP1/TAZ axis induces RRM2 expression and leads to alterations in nucleotide metabolism, contributing to these refractory stem-like states. These findings pave the way for further investigations into RRM2 as a potential biomarker for cetuximab unresponsiveness and as a therapeutic target to enhance the efficacy of anti-EGFR therapies.

Materials and methods

Bioinformatic analysis of publicly available transcriptomic datasets

Gene expression data from 157 CRC cell lines were obtained from the GEO database under the identifier GSE59857. To closely reflect the clinical context of anti-EGFR treatments, samples harboring activating mutations in RAS and BRAF were excluded from the analysis.

The cell lines COLO320 and its derivates, COLO320HSR and COLO320DM (GSM1448152, GSM1448173, and GSM1448182), were excluded from the analysis due to their neuroendocrine features. Additionally, the cell line HuTu80 (GSM1448180) was also excluded as it originates from the small intestine, leading to a distinct expression profile compared with colorectal-derived cell lines. The intensity data of the microarray experiments available in GEO were obtained using the “getGEO” function of the GEOquery package.

TCGA data from the COAD cohort was obtained from https://xenabrowser.net/datapages. Count matrix was normalized using EdgeR. A gene signature score for 25 CSC-related signatures (publicly available at http://stemchecker.sysbiolab.eu/) and two signatures related to cetuximab resistance and responsiveness published by Schutte and colleagues¹⁶ was calculated for each sample using the ssGSEA function contained in the ssGSEA package. This score was then correlated with WWTR1 gene expression. AURKA/YAP/TAZ downregulation score was constructed using the 31 commonly downregulated genes represented in the PPI network.

Cell lines

Human CRC cell line HCA46 was obtained from European Collection of Authenticated Cell Cultures. SW48 and NCIH-508 were obtained from the American Type Culture Collection. Meanwhile, C10 and HEK293T cell lines were kindly provided by Prof. Alberto Bardelli (Torino University) and Dr. Pedro Mateos (University of Alcalá), respectively. HCA46, C10, and HEK293T cells were cultured with DMEM (HyClone), while SW48 and NCIH-508 cells were cultured with RPMI medium. In both cases, the media were supplemented with 25 mM HEPES (Gibco; ref. 22400_097), 10% fetal bovine serum, and 1% penicillin/streptomycin. All cells were maintained in cell culture flasks incubated at 37°C with 5% CO₂ and controlled humidity. Routine screening for the presence of Mycoplasma occurred bimonthly. All cell lines used in this work are RAS/RAF wildtype and were authenticated by short tandem repeat profiling.

Cell engineering

Cells expressing shRNAs for silencing YAP1 and/or TAZ (WWTR1) were generated. For YAP1 knockdown, the ready-to-use vector pLV[shRNA]-Hygro-U6>hYAP1[shRNA#1] (Vector Builder, VB900139-1590efz) was used. For silencing WWTR1, shRNA was constructed into a pLKO plasmid by using the sequences described in Table S1.

Lentivirus production and cell infection

HEK293T cells were cultured until they reached 90% confluence. Subsequently, they were transfected with 10 μg of lentiviral expression vector along with 10.5 μg of viral packaging plasmids (comprising 3 μg of pVSVG, 5 μg of RRE, and 2.5 μg of RRV) using polyethyleneimine at a concentration of 1 mg/mL, with a ratio of 3:1 relative to the quantity of DNA. The media were replaced 8 h post transfection. After 48 h, the media were collected, filtered using 0.45-nm filters (Fisher brand ref. 15216869), and supplemented with 10 ng/μL of polybrene (Merck H9268-5G). The viral solution was then added to SW48 and C10 cell lines for infection, with media renewal every two hours for a total of four times. Blasticidin (20 μg/mL, Sigma-Aldrich, ref. SBR00022-1ML) was introduced after 24 h to select the transduced cell lines, and the selection process continued for a total of 120 h.

Western blot

Proteins were extracted from the cells using RIPA buffer, resolved by SDS-PAGE, and then transferred to nitrocellulose membranes. The membranes were probed with anti-human antibodies against TAZ (Cell Signaling, #4883), CD133 (Proteintech, 18470-1-AP), YAP1 (Cell Signaling, #14074), p-YAP1 (Cell Signaling, #13619) (S397), β-actin (Cell Signaling; #3700), c-MET (Proteintech, 25869-1-AP), p-c-MET (Abcam, ab5662), CD-44 (Proteintech, 15675-1-AP), CD-133 (Proteintech, 18470-1-AP), and RRM2 (Cell Signaling, #65939) overnight at 4°C. Horseradish peroxidase-linked anti-rabbit immunoglobulin (Ig)G or anti-mouse IgG was used and the antigen-antibody reaction was visualized by enhanced chemiluminescence assay. Densitometry was performed using Image QuanTLsoftware. Beta-actin run on the same blot was used as the loading control.

RNA extraction, complementary DNA synthesis, and qPCR

The detailed protocol of RNA extraction and RT-qPCR was reported previously.⁴² TaqMan probes (Thermo Fisher Scientific) used for evaluating gene expression are the following: POU5F1 (Hs04260367_gH) and RPLP0 (Hs99999902_m1). The expression ratio was calculated by the ΔΔCt method as described by Livak and Schmittgen,⁴³ using RPLP0 as the housekeeping gene for data normalization. Each sample was analyzed in duplicate or triplicate.

Cell viability assays

To assess the influence of YAP1 and TAZ on cetuximab responsiveness, shRNAs targeting YAP1 or TAZ were transduced into cetuximab-resistant CRC cell lines SW48 and C10. Each cell line was then seeded with 5,000 cells in 96-well plates and incubated for 24 h at 37°C in a humidified atmosphere with 5% CO₂. Following media replacement, cells were exposed to cetuximab (10 ng/μL for C10 and 1 ng/μL for SW48) for 5 days, after which 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) was added at a concentration of 1 mg/mL and incubated at 37°C for 2–4 h. Subsequently, the MTT-containing medium was aspirated, and the formazan crystals were dissolved in 100 μL of DMSO. Optical absorbance was then measured at 570 nm using a microplate reader.

Colony formation and colonosphere assay

The C10 and SW48 cell lines with empty or shYAP/TAZ vectors were plated in 100-mm dishes and allowed to reach 70% confluence within 24 h. Then, cells were stained with Trypan blue and counted.

For the colony formation assay, 3,000 viable cells were seeded in 100-mm plates. After 10 days, colonies were stained with methyl violet dye (Thermo Fischer Scientific) and counted manually.

In the colonosphere formation assay, 1,000 viable cells were cultured in specialized polystyrene plates (Corning) designed to support the growth of cells in ultra-low adherence conditions. These cells were maintained in a mixture of DMEM/F12 supplemented with insulin (20 mg/mL, Sigma), EGF (20 ng/mL, R&D Systems), basic fibroblast growth factor (10 ng/mL, R&D Systems), glucose (3 mg/mL, Sigma), and antibiotics (1% penicillin-streptomycin, PanBiotech). The plates were then incubated at 37°C with 5% CO₂ for 10 days, and photographs of the spheres were captured. The size of the spheres was quantified using ImageJ (National Institutes of Health).

Aldehyde dehydrogenase assay

ALDH1 activity detection kit (Cat # MAK082, Sigma) was used to determine the ALDH1 activity according to the recommended procedure from manufacturers.

Patient-derived xenograft

Tumor samples used in these experiments were derived from the in vivo model outlined in our prior work,¹³ which evaluated the efficacy of AURKA inhibitors in PDX models with high YAP1 phosphorylation. TAZ expression was analyzed to confirm its high levels of TAZ, indicating its suitability for this study alongside YAP1. As stated in the original article, all animal procedures were approved by the Ethical Animal Research Committee at IIS-Fundación Jiménez Díaz and conducted following institutional and government regulations (Reference n°: PROEXP 142-17).

RNA-seq and bioinformatic analysis

The C10 cell line underwent a 24-h treatment with 350 nm alisertib. Total RNA was extracted from six independent biological replicates using the commercial NucleoSpin RNA Isolation Kit (MACHEREY-NAGEL), and sequencing libraries were generated on the Illumina NovaSeq 6000 platform (Novogene). Sample alignment to the Ensembl GRCh38 genome was conducted using the STAR program (v. 2.7.10b). Matrix counts were obtained using htseq-count. DEGs were obtained using DeSEQ2⁴⁴ applying DESeq normalization. GSEA⁴⁵ was performed in R with the fgsea package. DEGs were considered significant with an absolute fold change of 25% or higher (|log2FC| > 0.3219) and a false discovery rate (FDR) of less than 0.05.

Significant DEGs were used to construct a PPI network utilizing the STRING database⁴⁶ (using default edges features) and visualized in Cytoscape (v3.9.1).⁴⁷ Gene enrichment analysis of the PPI network was carried out using the BinGO app⁴⁸ within Cytoscape.

Statistical analysis

The experimental data provided are presented as the mean ± SD. Each experiment was independently repeated three times unless the opposite was stated in the figure captions. p values were calculated using GraphPad Prism 5 software. Two-sample comparisons utilized the two-tailed Student’s t test. For comparisons involving three or more conditions, an initial analysis of variance (one-way ANOVA) was performed, followed by Tukey’s multiple comparisons test. Statistical significance was determined when p values were less than 0.05. In GSEA analyses, a signature was considered significantly enriched when the FDR was less than 0.25 and adjusted p value was less than 0.05, as recommended by the guidelines (https://www.gsea-msigdb.org/gsea/doc/GSEAUserGuideFrame.html).

Additional methodological details can be found in the supplemental information.

Data availability

All RNA-seq data generated in this study have been deposited at the GEO Database under the accession code GSE261294.

Acknowledgments

We thank Professor Alberto Bardelli (Torino University, Italy) for kindly providing the C10 cell line.

This work was supported by grant PI19/01231 funded by Agencia Española de Investigación (AEI) and the European Union. A.R.-V.’s contract was funded by the program “CONTRATOS PREDOCTORALES DE FORMACIÓN EN INVESTIGACIÓN EN SALUD (PFIS),” grant FI20/00213 from the Instituto de Salud Carlos III (ISC-III), and co-funded by the European Regional Development Fund/European Social Fund (European Union) (ERDF/ESF, “A way to make Europe”/”Investing in your future”) associated with the project PI19/01231.

Author contributions

Conceptualization: A.R.V., A.C., and J.G.F.; data curation: A.R.V., M.S., L.G.H., A.G., and A.C.C.; formal analysis: A.R. and A.C.; funding acquisition: J.G.F. and A.C.; methodology: A.R.V., A.C., L.P.N., P.M., P.A.M.G., N.B., and O.A.; project administration: L.G., A.C., and J.G.F.; resources: N.B., A.C., L.G., and J.G.F.; software: A.R.V. and P.M.; supervision: A.C., P.M., and J.G.F.; validation: A.C.; visualization: A.R.V.; writing – original draft: A.R.V.; writing – review & editing: A.C., J.G.F., and A.R.V.

Declaration of interests

The authors declare no competing interests.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work the authors used ChatGPT in order to improve the grammar and language quality of this work. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.