Chromosome 1q21.3 amplification is a trackable biomarker and actionable target for breast cancer recurrence

Journal name:
Nature Medicine
Year published:
DOI:
doi:10.1038/nm.4405
Received
Accepted
Published online

Abstract

Tumor recurrence remains the main reason for breast cancer–associated mortality, and there are unmet clinical demands for the discovery of new biomarkers and development of treatment solutions to benefit patients with breast cancer at high risk of recurrence. Here we report the identification of chromosomal copy-number amplification at 1q21.3 that is enriched in subpopulations of breast cancer cells bearing characteristics of tumor-initiating cells (TICs) and that strongly associates with breast cancer recurrence. Amplification is present in ~10–30% of primary tumors but in more than 70% of recurrent tumors, regardless of breast cancer subtype. Detection of amplification in cell-free DNA (cfDNA) from blood is strongly associated with early relapse in patients with breast cancer and could also be used to track the emergence of tumor resistance to chemotherapy. We further show that 1q21.3-encoded S100 calcium-binding protein (S100A) family members, mainly S100A7, S100A8, and S100A9 (S100A7/8/9), and IL-1 receptor–associated kinase 1 (IRAK1) establish a reciprocal feedback loop driving tumorsphere growth. Notably, this functional circuitry can be disrupted by the small-molecule kinase inhibitor pacritinib, leading to preferential impairment of the growth of 1q21.3-amplified breast tumors. Our study uncovers the 1q21.3-directed S100A7/8/9–IRAK1 feedback loop as a crucial component of breast cancer recurrence, serving as both a trackable biomarker and an actionable therapeutic target for breast cancer.

At a glance

Figures

  1. Genomic interrogation of TICs identifies 1q21.3 amplification in breast cancer.
    Figure 1: Genomic interrogation of TICs identifies 1q21.3 amplification in breast cancer.

    (a) Schematic overview of sample preparation for RNA-seq analysis. The heat map displays 1,401 genes upregulated in 12 patient-derived tumorsphere samples compared to matching bulk tumor cells as determined by RNA-seq. Green, downregulated; red, upregulated; black, unchanged. (b) Copy-number variation (CNV) of 1,401 genes upregulated in tumorspheres (TS) in the breast cancer TCGA data set (breast invasive carcinoma)8 on different chromosomes (subjects with CNV profile, n = 816). Red, gene amplification; blue, gene deletion. (c) Location of the 17 upregulated genes on chromosome 1q21.3. (d) Percentage of subjects with 1q21.3 amplification in different breast cancer subtypes in the TCGA data set (invasive ductal cancer, n = 481 subjects). (e) Correlation of 1q21.3 or 8q amplification and corresponding gene expression in the TCGA breast cancer data set. Normalized gene expression values were downloaded from the TCGA Pan-Can RNA-seq data set. Mean gene expression for 1q21.3 (17 genes) and 8q (36 genes) was determined for each subject. The number of subjects per group is indicated below the graph. P values were calculated with the Kruskal–Wallis test. Dunn's post hoc test was applied to correct for multiple comparisons. (f) Time-to-death analysis of subjects with 1q21.3 or 8q amplification in the TCGA breast cancer data set. The number of subjects per group is indicated below the graph. P values were calculated with the Mann–Whitney test. *P < 0.05; n.s., not significant. (g) Kaplan–Meier relapse-free survival (RFS) analysis of the combined gene signature (17 genes on 1q21.3) in the indicated subjects with breast cancer. The number of subjects (n) per group is indicated in the graph. For box plots, the median value (center line), the minimum and maximum (whiskers), and the 25th and 75th percentiles (box perimeters) are presented.

  2. 1q21.3 amplification is enriched in TICs and is associated with tumor recurrence.
    Figure 2: 1q21.3 amplification is enriched in TICs and is associated with tumor recurrence.

    (a) Representative DNA FISH images of primary mammary epithelial cells and patient-derived tumorsphere cells hybridized with DNA probes to detect 3 (red) and 1p32.3 (green). The percentage of cells with different copy numbers of 3 is shown in the pie charts below the corresponding representative images. Scale bar, 5 μm. (b) ddPCR analysis of 3 amplification in primary tumors and matched tumorsphere cells. RPP30 was used as a reference gene for normalization. Data are expressed as means ± s.d. of four technical replicates. (c) ddPCR analysis of primary tumors from subjects without recurrence within 5 years of diagnosis. Shown are the averaged ratios of the three target genes, TUFT1, S100A8, and S100A7, relative to reference gene RPP30. Data were generated from the means of two technical replicates. The number of subjects (n) per group is indicated. TN, triple negative. (d) ddPCR analysis of primary tumors and matched recurrent tumors from subjects who had recurrence within 5 years of diagnosis. Shown are the averaged ratios of the three target genes, TUFT1, S100A8, and S100A7, relative to reference gene RPP30. Data were generated from the means of two technical replicates. The number of subjects (n) per group is indicated. (e) Scatterplot summarizing the results from c and d. P values were calculated with the McNemar test. **P < 0.01, ***P < 0.001.

  3. cfDNA detection of 1q21.3 amplification is associated with recurrence and poor patient outcome.
    Figure 3: cfDNA detection of 1q21.3 amplification is associated with recurrence and poor patient outcome.

    (a) Correlation analysis between tumor DNA and cfDNA in both development and validation cohorts (n = 54). Dashed lines denote the cutoff threshold for each axis. (b) ROC curves using individual genes (TUFT1, S100A8, and S100A7) or these three genes in combination for detection of 1q21.3 amplification in 54 subjects with breast cancer (development and validation cohorts). The AUC value for the three-gene combination is 0.997 (95% CI, 0.927 to 1.00; P < 0.0001). The P value was calculated with the nonparametric t-test. The dotted diagonal line denotes an AUC of 0.50. (c) Scatterplot showing 1q21.3 amplification in the cfDNA of healthy control subjects as well as subjects with breast cancer at the time of diagnosis and recurrence. The number of subjects (n) per group is indicated. The P value was calculated with the Mann–Whitney test. ***P < 0.001. (d) Kaplan–Meier progression-free survival analysis of Denmark OUH patients with early-stage breast cancer with or without 1q21.3 amplification in cfDNA at the time of diagnosis. The number of patients (n) per group is indicated. (e) Kaplan–Meier progression-free survival analysis of the Singapore NUHS neoadjuvant cohort with or without 1q21.3 amplification in cfDNA. The number of subjects (n) per group is indicated. (f) Kaplan–Meier overall survival analysis of the Singapore NUHS neoadjuvant cohort with or without 1q21.3 amplification in cfDNA. The number of patients (n) per group is indicated.

  4. 1q21.3-encoded S100A7/8/9 forms a functional feedback loop with IRAK1 to drive tumorsphere growth.
    Figure 4: 1q21.3-encoded S100A7/8/9 forms a functional feedback loop with IRAK1 to drive tumorsphere growth.

    (a) Microarray gene expression analysis of S100A family members in tumors versus corresponding derived tumorspheres (n = 4; left) and 3-amplified (MB436, MB468, MCF-7, and T47D) versus non-amplified (BT-549, BT-474, MB361, and MB231) breast cancer cell lines (right). Green, top three upregulated genes; red, downregulated; blue, upregulated. (b) Representative images (n = 6) and quantification of the number of tumorspheres after individual knockdown of S100A7, S100A8, or S100A9 in MB436, MB468, and MB231 cells. shLUC was used as a non-targeting control. Data are expressed as means ± s.e.m. of three independent knockdown experiments. Scale bar, 100 μm. (c) Representative western blot (n = 2) showing a reduction in phosphorylated IRAK1 (pIRAK1) in MB436 and MB468 cells upon individual shRNA-mediated knockdown of S100A7, S100A8, or S100A9. Actin was used as a loading control. (d) Quantification of the number of tumorspheres after individual knockdown of S100A7, S100A8, or S100A9 in two different PDX-derived tumorsphere lines. Data are expressed as means ± s.e.m. of three technical replicates. (e) Representative western blot (n = 2) showing induction of IRAK1 phosphorylation upon individual treatment with recombinant S100A7, S100A8, or S100A9 protein in MB231 cells at the indicated concentrations. Actin was used as a loading control. (f) Real-time PCR analyses of S100A7, S100A8, and S100A9 gene expression after inducible knockdown of IRAK1 in MB436 tumorsphere cells. Dox, doxycycline. Data are expressed as means ± s.d. of three technical replicates. (g) Real-time PCR analysis of S100A7, S100A8, and S100A9 gene expression after overexpression of IRAK1-WT or IRAK1-K239S in MB436 tumorsphere cells. Data are expressed as means ± s.d. of three technical replicates. (h) Quantification of the number of tumorspheres (TS) after overexpression of IRAK1-WT or IRAK1-K239S in MB436, MB468, and MB231 cells. Recombinant S100A8 and S100A9 proteins were added separately to IRAK1-K239S-expressing cells to rescue tumorsphere formation. Data are expressed as means ± s.e.m. of three independent overexpression experiments. (i) Representative IHC images (n = 6) of phosphorylated IRAK1 (Ser376) and S100A8 in matched primary and recurrent breast tumor samples. Scale bars, 100 μm. (j) Quantification of phosphorylated IRAK1 and S100A8 levels in IHC analyses of 18 paired primary and recurrent breast tumor samples. Recurrent tumors showed an increased 3 copy-number ratio.P values were calculated with paired two-tailed t-tests. (k) Correlation analysis of S100A8 and phosphorylated IRAK1 IHC staining in 18 paired tumor samples showing an increased 3 copy-number ratio in recurrent tumors. Linear regression was performed using GraphPad Prism. The linear regression Pearson's correlation coefficient (R2) and its P value are given. All P values were calculated with two-tailed t-tests. *P < 0.05, **P < 0.01, ***P < 0.001.

  5. Pacritinib effectively disrupts the S100A7/8/9-IRAK1 feedback loop to inhibit tumorsphere growth.
    Figure 5: Pacritinib effectively disrupts the S100A7/8/9–IRAK1 feedback loop to inhibit tumorsphere growth.

    (a) Representative western blot (n = 2) showing inhibition of phosphorylated IRAK1 and phosphorylated JAK2 (pJAK2) within 6 h of pacritinib treatment in MB468 and MB231 cells. Actin was used as a loading control. (b) Quantification of the number of tumorspheres after pacritinib treatment for MB468 and MB231 cells. Data are expressed as means ± s.e.m. of three independent drug treatment experiments. (c) Real-time PCR analysis of S100A7, S100A8, and S100A9 gene expression after 24 h of pacritinib treatment in MB468 cells. Data are expressed as means ± s.d. of three technical replicates. (d) Representative western blot (n = 2) showing prolonged inhibition of phosphorylated IRAK1 24 h following pacritinib (2.5 μM) treatment of various PDX-derived tumorspheres with and without IL-1β. Actin was used as a loading control. (e) Real-time PCR analysis of S100A7, S100A8, and S100A9 transcript levels 24 h following pacritinib (Pac; 2.5 μM) treatment. Data are expressed as means ± s.d. of three technical replicates. (f) Quantification of the number of tumorspheres after pacritinib treatment for four different PDX-derived tumorspheres. Data are expressed as means ± s.e.m. of three independent experiments. (g) Representative images (n = 6) and quantification of the number of tumorspheres in rescue assays. Treatment with recombinant S100A8 or S100A9 protein (10 ng/ml) was able to rescue pacritinib-treated (2.5 μM) tumorspheres from various cell lines. Data are expressed as means ± s.d. of three independent experiments. Scale bar, 100 μm. All P values were calculated with two-tailed t-tests. **P < 0.01, ***P < 0.001.

  6. Amplification status of 1q21.3 correlates with the efficacy of pacritinib in vitro and in vivo.
    Figure 6: Amplification status of 1q21.3 correlates with the efficacy of pacritinib in vitro and in vivo.

    (a) ddPCR analysis of genomic DNA from various breast cancer cell lines to determine amplification status of 1q21.3. RPP30 was used as a reference gene for normalization. Data are expressed as means ± s.d. of four technical replicates. (b) Representative western blot (n = 2) showing phosphorylated IRAK1 and phosphorylated JAK2 levels in tumorspheres from the breast cancer cell lines in a. Actin was used as a loading control. Red labels, cell lines with 1q21.3 amplification. (c) Growth curve of tumorspheres from ER-positive and ER-negative breast cancer cell lines treated with increasing doses of pacritinib. Red, amplified; black, non-amplified. Data are expressed as means ± s.e.m. of three independent experiments. (d) NOD-SCID mice bearing MB231 and HCC70 tumors were treated with 50, 100, or 150 mg per kg body weight (mg/kg) pacritinib by oral gavage (n = 10 mice per group). (e) Western blot analysis (n = 2) showing the level of phosphorylated IRAK1 in HCC70 xenograft mice (n = 4) treated with pacritinib. Actin was used as a loading control. (f) NOD-SCID mice bearing PDX3 tumors were treated with 20 mg per kg body weight paclitaxel by intravenous tail vein injection every other day for 14 d to induce tumor regression. The residual tumors were surgically removed, and mice were rested for 14 d before treatment with 20 mg per kg body weight paclitaxel, 150 mg per kg body weight pacritinib, or both drugs for 14 d. The sample size (n) per group is indicated. (g) Schematic representation of the proposed mechanisms of an S100A8/9–IRAK1 feedback loop in 1q21.3-amplified tumors, as well as the application of 1q21.3 amplification as a biomarker for use in companion diagnostics and tracking of tumor response. (h) Genomic alterations of S100A8, S100A9 (located on 1q21.3), and genes with targeted therapy in various TCGA cancer data sets. In d and f, P values were calculated with two-tailed unpaired t-tests. **P < 0.01, ***P < 0.001.

Introduction

Despite advances in the treatment of breast cancer, numerous patients develop recurrence and die from metastasis years after treatment. In spite of intensive research, the molecular driver events leading to tumor recurrence remain elusive. Identifying the key molecular events associated with tumor recurrence will allow for the development of new diagnostic and treatment solutions, which are essential for improving patient survival. One of the reasons for cancer relapse is intratumor heterogeneity arising from distinct populations of cancer cells, known as TICs or cancer stem cells, that are rare, aggressive, and resistant to conventional cytotoxic treatments1, 2. Therefore, identifying molecular features associated with this subpopulation of TICs, which often survive cancer therapy, is expected to deliver more sensitive diagnostic biomarkers and unveil new therapeutic targets in recurrent tumors.

This study was undertaken by utilizing integrative genomic analyses to identify TIC-associated genetic changes in breast cancer. We identified chromosomal 1q21.3 amplification as highly enriched in TICs and recurrent tumors. Using droplet digital PCR (ddPCR), we developed a blood-based molecular assay to detect 1q21.3 amplification in cfDNA and investigated whether amplification can serve as a circulating biomarker for the prediction of early relapse and monitoring of tumor response to chemotherapy. We identified a molecular mechanism through which 1q21.3 amplification is associated with breast cancer recurrence, including a functional link between 1q21.3-encoded S100A7/8/9 and IRAK1. Notably, we also demonstrate that 1q21.3 amplification is an actionable target, with therapeutic implications spanning different breast cancer subtypes.

Results

Integrative genomic analyses of TICs identify 1q21.3 amplification in breast cancer

Cancer cells growing as tumorspheres in serum-free medium are enriched for TICs and are highly tumorigenic1, 3. Using tumorsphere cells derived from 12 subjects with breast cancer, we performed RNA-seq analysis to compare their gene expression profiles with those of matched bulk tumor cells. This analysis identified 1,401 genes whose expression was commonly upregulated in the tumorspheres as compared to the matched bulk tumors (fold change > 2; false discovery rate (FDR) < 0.05; Fig. 1a). In accordance with other reports4, 5, 6, several members of the ALDH gene family were significantly upregulated in the patient-derived tumorspheres (Supplementary Fig. 1a). These patient-derived tumorsphere cells displayed much higher aldehyde dehydrogenase (ALDH) activity than the matched bulk tumor cells (Supplementary Fig. 1b) and were able to initiate tumor formation when engrafted at low cell numbers into immunodeficient NOD-SCID gamma (NSG) mice (Supplementary Fig. 1c).

Figure 1: Genomic interrogation of TICs identifies 1q21.3 amplification in breast cancer.
Genomic interrogation of TICs identifies 1q21.3 amplification in breast cancer.

(a) Schematic overview of sample preparation for RNA-seq analysis. The heat map displays 1,401 genes upregulated in 12 patient-derived tumorsphere samples compared to matching bulk tumor cells as determined by RNA-seq. Green, downregulated; red, upregulated; black, unchanged. (b) Copy-number variation (CNV) of 1,401 genes upregulated in tumorspheres (TS) in the breast cancer TCGA data set (breast invasive carcinoma)8 on different chromosomes (subjects with CNV profile, n = 816). Red, gene amplification; blue, gene deletion. (c) Location of the 17 upregulated genes on chromosome 1q21.3. (d) Percentage of subjects with 1q21.3 amplification in different breast cancer subtypes in the TCGA data set (invasive ductal cancer, n = 481 subjects). (e) Correlation of 1q21.3 or 8q amplification and corresponding gene expression in the TCGA breast cancer data set. Normalized gene expression values were downloaded from the TCGA Pan-Can RNA-seq data set. Mean gene expression for 1q21.3 (17 genes) and 8q (36 genes) was determined for each subject. The number of subjects per group is indicated below the graph. P values were calculated with the Kruskal–Wallis test. Dunn's post hoc test was applied to correct for multiple comparisons. (f) Time-to-death analysis of subjects with 1q21.3 or 8q amplification in the TCGA breast cancer data set. The number of subjects per group is indicated below the graph. P values were calculated with the Mann–Whitney test. *P < 0.05; n.s., not significant. (g) Kaplan–Meier relapse-free survival (RFS) analysis of the combined gene signature (17 genes on 1q21.3) in the indicated subjects with breast cancer. The number of subjects (n) per group is indicated in the graph. For box plots, the median value (center line), the minimum and maximum (whiskers), and the 25th and 75th percentiles (box perimeters) are presented.

We hypothesized that the identification of copy-number alterations associated with TICs might lead to the discovery of biomarkers associated with tumor recurrence. To identify genes in our RNA-seq analysis whose upregulation might potentially be associated with copy-number gain, we profiled the copy-number status of the 1,401 genes upregulated in TICs according to chromosome location using The Cancer Genome Atlas (TCGA) breast cancer genomic data set7, 8, 9. This analysis revealed two subgroups of genes located on chromosome 1 (86 genes) and chromosome 8 (38 genes) displaying amplification in a substantial number of subjects with breast cancer (~12% and ~16%, respectively; Fig. 1b). Further analysis revealed that 17 of the 86 genes on chromosome 1 are clustered at 1q21.3 (Fig. 1c), indicating gain of 1q21.3. Detailed stratification of the subjects into the four different molecular subtypes of breast cancer revealed that amplification of 17 genes at 1q21.3 could be found in all breast cancer subtypes, but in a higher percentage of subjects with basal-like breast tumors (31% of basal-like tumors versus 12% of human epidermal growth factor receptor 2 (HER2)+ tumors and 10% of luminal tumors; Fig. 1d).

Further analyses in the TCGA data set showed the correlation of both 1q21.3 (17 genes) and 8q (38 genes) copy-number amplification with respective gene expression (Fig. 1e). However, only subjects with 1q21.3 amplification had a significantly shorter time to death from initial pathologic diagnosis (P = 0.0248, Mann–Whitney test; Fig. 1f), indicating a potential role of 1q21.3 amplification in breast cancer progression. In accordance with this hypothesis, meta-analysis of survival for subjects with breast cancer in a different data set (see URLs)10 showed that the 17-gene expression signature for 1q21.3 identified in our RNA-seq analysis was associated with poor relapse-free survival in subjects with breast cancer regardless of estrogen receptor (ER) status (Fig. 1g), demonstrating the potential prognostic value of 1q21.3 amplification in breast cancer. Similar observations were noted in a larger breast cancer cohort (Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) database11, 12; Supplementary Fig. 2a–c).

To validate this genomic finding, we analyzed a retrospective cohort of 67 breast cancer specimens (Singapore Tan Tock Seng Hospital (TTSH) discovery cohort; Supplementary Table 1). Using genomic DNA for real-time PCR of the S100A8 gene as a readout of 1q21.3 copy number, we confirmed that 11 of the 67 primary tumors (16.4%) had amplification of the S100A8 gene (Supplementary Fig. 2d). Furthermore, subjects with breast cancer who were positive for S100A8 amplification had significantly lower overall survival than subjects without amplification (Supplementary Fig. 2e; hazard ratio (HR), 5.279; 95% confidence interval (CI), 1.868 to 14.916; log-rank P = 0.002). In particular, the group of subjects with S100A8 amplification had a significantly higher death rate within 10 years of clinical follow-up (5 of 11 subjects, 45.5%) than patients without S100A8 amplification (10 of 56 patients, 17.9%; P = 0.0447). Multivariate analysis confirmed S100A8 amplification as an independent predictor of poor survival (HR, 7.134; 95% CI, 2.226 to 22.869; log-rank P = 0.001), which outperformed other tumor characteristics, including ER or HER2 status, node status, and tumor size (Supplementary Table 2). These findings collectively support a prognostic value for 1q21.3 amplification in breast cancer.

1q21.3 amplification is enriched in TICs and associates with tumor recurrence

To validate the relevance of 1q21.3 amplification to TICs, we performed DNA fluorescence in situ hybridization (FISH) on patient-derived tumorsphere cells using a molecular probe to detect 1q21.3 and another probe to detect 1p32.3, which served as an adjacent-genomic-region control. DNA FISH analysis revealed multiple copies of 1q21.3 in examined tumorspheres derived from both ER-positive and ER-negative tumors (Fig. 2a). Of note, the tumorsphere cells were highly heterogeneous, with anywhere from 3 to 12 copies of 1q21.3 present (Supplementary Table 3). Also, the majority of the cells in which copy number was quantified had more copies of 1q21.3 than 1p32.3, indicating that the 1q21.3 gain seen in TICs is not simply a consequence of chromosome aneuploidy (Fig. 2a).

Figure 2: 1q21.3 amplification is enriched in TICs and is associated with tumor recurrence.
1q21.3 amplification is enriched in TICs and is associated with tumor recurrence.

(a) Representative DNA FISH images of primary mammary epithelial cells and patient-derived tumorsphere cells hybridized with DNA probes to detect 3 (red) and 1p32.3 (green). The percentage of cells with different copy numbers of 3 is shown in the pie charts below the corresponding representative images. Scale bar, 5 μm. (b) ddPCR analysis of 3 amplification in primary tumors and matched tumorsphere cells. RPP30 was used as a reference gene for normalization. Data are expressed as means ± s.d. of four technical replicates. (c) ddPCR analysis of primary tumors from subjects without recurrence within 5 years of diagnosis. Shown are the averaged ratios of the three target genes, TUFT1, S100A8, and S100A7, relative to reference gene RPP30. Data were generated from the means of two technical replicates. The number of subjects (n) per group is indicated. TN, triple negative. (d) ddPCR analysis of primary tumors and matched recurrent tumors from subjects who had recurrence within 5 years of diagnosis. Shown are the averaged ratios of the three target genes, TUFT1, S100A8, and S100A7, relative to reference gene RPP30. Data were generated from the means of two technical replicates. The number of subjects (n) per group is indicated. (e) Scatterplot summarizing the results from c and d. P values were calculated with the McNemar test. **P < 0.01, ***P < 0.001.

To develop a robust assay for detection of 1q21.3 amplification, we used the highly sensitive and quantitative ddPCR assay and designed three PCR probes targeting TUFT1, S100A8, and S100A7 as proxies covering 1q21.3 and a probe for a reference gene, RPP30. We validated the specificity of this newly developed assay (Online Methods and Supplementary Fig. 3a,b) using three 1q21.3-amplified breast tumors (Supplementary Fig. 2d) and matched adjacent normal tissues (Supplementary Fig. 4a).

Using this ddPCR assay, we examined 1q21.3 amplification in tumorspheres and matched bulk tumors. In all seven paired samples, of which some were ER positive and some were ER negative, tumorspheres displayed an enrichment of 1q21.3 amplification as compared to corresponding bulk tumor cells (Fig. 2b). Moreover, FACS-sorted tumorsphere cells with high ALDH activity showed further enrichment of 1q21.3 amplification as compared to tumorsphere cells with low ALDH activity (Supplementary Fig. 4b). In contrast to genes at 1q21.3, those located at other bands of the 1q arm, including NECTIN4 (PVRL4) at 1q23.3 and LAMB3 at 1q32.2, did not exhibit a similar extent of copy-number gain (Supplementary Fig. 4c), supporting regional amplification of 1q21.3 in TICs.

Next, we sought to investigate whether 1q21.3 amplification is associated with disease recurrence. To do this, we procured primary tumors from subjects who did not develop recurrence and subjects who did develop recurrence within 5 years of diagnosis, as well as matched recurrent tumors (Supplementary Table 4). An averaged copy-number ratio for the three target genes (TUFT1, S100A8, and S100A7) relative to the reference gene, RPP30, was calculated for each sample, and a cutoff value of ≥3 s.d. above the mean, as determined from 30 normal breast tissues adjacent to breast cancer tumors, was used to define samples positive for 1q21.3 amplification (Supplementary Fig. 5a).

Using this criterion, we detected that 2 of the 28 primary tumors were positive for 1q21.3 amplification among the subjects without recurrence (7.1%; 95% CI, 1.3% to 25.0%; Fig. 2c and Supplementary Table 5), whereas 12 of the 34 primary tumors were positive among the subjects with matched clinical characteristics who later developed recurrence (35.3%; 95% CI, 20.3% to 52.1%; Fig. 2d and Supplementary Table 5). Intriguingly, of the 34 paired recurrent tumors, 26 were positive (76.5%; 95% CI, 58.4% to 88.6%; Fig. 2d and Supplementary Table 5). Of note, 14 of the 34 patients with recurrent disease (41.2%) were negative for 1q21.3 amplification in their primary tumors but had acquired it upon recurrence (McNemar test, P < 0.001). A similar result was observed in another validation cohort obtained from Odense University Hospital (OUH) in Denmark (Fig. 2e, Supplementary Fig. 5b, and Supplementary Table 5). Enrichment of 1q21.3 amplification in tumor recurrence was seen in tumors from different subtypes, suggesting broad coverage of this biomarker in subjects with breast cancer. Statistical analysis with Pearson's chi-squared test of independence between the primary tumors of subjects with and without recurrence within 5 years of diagnosis showed a significant correlation of 1q21.3 amplification with early tumor recurrence (P < 0.008; odds ratio, 7.091).

Moreover, we used a patient-derived xenograft (PDX) mouse model for triple-negative breast cancer to test whether 1q21.3 amplification is associated with tumor recurrence following chemotherapy. In this model, mice orthotopically injected with PDX tumors were treated with paclitaxel to induce tumor regression, and tumors were subsequently surgically removed before monitoring the mice for tumor recurrence (Supplementary Fig. 5c). Indeed, ddPCR analysis revealed that both residual tumors following paclitaxel treatment and recurrent tumors showed marked enrichment of 1q21.3 amplification as compared to treatment-naive PDX tumors (Supplementary Fig. 5d).

cfDNA detection of 1q21.3 amplification in blood is associated with disease progression and poor outcomes

ddPCR has previously been successfully used to detect gene amplification in cfDNA13, 14, 15, 16, 17, 18. We thus evaluated whether our ddPCR assay could be translated to a liquid biopsy test for detection of 1q21.3 amplification in cfDNA from a patient's blood. Using 2 ng of cfDNA for ddPCR (Online Methods and Supplementary Fig. 6a), we first determined the background and cutoff value for 1q21.3 amplification in plasma using plasma from 30 healthy women (Supplementary Fig. 6b and Supplementary Table 6). Next, we proceeded to test the assay in two separate cohorts of plasma samples from subjects with breast cancer and corresponding tumor samples (Supplementary Fig. 7a). In the first cohort of 21 subjects (development cohort), we detected 8 primary tumor samples that were positive for 1q21.3 amplification; 7 of these had matched plasma samples that were also positive for 1q21.3 amplification (Fig. 3a, Supplementary Fig. 7b, and Supplementary Table 7). Overall, we obtained only 1 false positive and 1 false negative in the 21 plasma samples tested. In the second cohort of 33 subjects with breast cancer (validation cohort) that we analyzed for verification, we detected 7 subjects who were positive for amplification in both the primary tumor and corresponding plasma sample from blood, with 100% tumor–plasma concordance (Fig. 3a, Supplementary Fig. 7b, and Supplementary Table 7). Therefore, the ddPCR assay using plasma cfDNA was able to detect 1q21.3 amplification with a sensitivity of 93.3% and a specificity of 97.5%. Analysis of the combined cohorts by receiver operating characteristic (ROC) curve showed an increased area under the curve (AUC) of 0.997 when using the three-gene average (95% CI, 0.927 to 1.00; P < 0.0001) as compared to ratios for individual genes (Fig. 3b).

Figure 3: cfDNA detection of 1q21.3 amplification is associated with recurrence and poor patient outcome.
cfDNA detection of 1q21.3 amplification is associated with recurrence and poor patient outcome.

(a) Correlation analysis between tumor DNA and cfDNA in both development and validation cohorts (n = 54). Dashed lines denote the cutoff threshold for each axis. (b) ROC curves using individual genes (TUFT1, S100A8, and S100A7) or these three genes in combination for detection of 1q21.3 amplification in 54 subjects with breast cancer (development and validation cohorts). The AUC value for the three-gene combination is 0.997 (95% CI, 0.927 to 1.00; P < 0.0001). The P value was calculated with the nonparametric t-test. The dotted diagonal line denotes an AUC of 0.50. (c) Scatterplot showing 1q21.3 amplification in the cfDNA of healthy control subjects as well as subjects with breast cancer at the time of diagnosis and recurrence. The number of subjects (n) per group is indicated. The P value was calculated with the Mann–Whitney test. ***P < 0.001. (d) Kaplan–Meier progression-free survival analysis of Denmark OUH patients with early-stage breast cancer with or without 1q21.3 amplification in cfDNA at the time of diagnosis. The number of patients (n) per group is indicated. (e) Kaplan–Meier progression-free survival analysis of the Singapore NUHS neoadjuvant cohort with or without 1q21.3 amplification in cfDNA. The number of subjects (n) per group is indicated. (f) Kaplan–Meier overall survival analysis of the Singapore NUHS neoadjuvant cohort with or without 1q21.3 amplification in cfDNA. The number of patients (n) per group is indicated.

We next used the cfDNA assay to assess the role of 1q21.3 amplification in metastatic progression. Of 51 plasma samples collected upon tumor recurrence after chemotherapy, 37 were positive for 1q21.3 amplification (72.5%; 95% CI, 58.0% to 83.7%), which is much higher than the proportion collected at initial diagnosis before any treatment (P < 0.0001, Mann–Whitney test). Moreover, among treatment-naive samples, more positive samples were collected from subjects with metastatic cancer than subjects with non-metastatic cancer (47.4% versus 27.1%; Fig. 3c and Supplementary Table 8). Together, these findings establish the association of 1q21.3 amplification with disease progression and highlight the value of 1q21.3 as a potential circulating biomarker for clinical metastasis and relapse of breast cancer.

Next, we tested the ability of our cfDNA assay to assess the clinical outcomes of patients with breast cancer. We first sought to investigate whether our assay is sensitive enough in patients with early-stage breast cancer, as the abundance of cell-free tumor DNA (ctDNA) is much lower in patients with early-stage cancer than in those with advanced cancer19. In a cohort of 80 patients with newly diagnosed stage 1 and 2 tumors from Denmark (OUH early-stage breast cancer cohort), from whom serum samples were collected before surgery (Supplementary Fig. 8a and Supplementary Table 9), 4 patients were positive for 1q21.3 amplification in their cfDNA. Remarkably, disease relapse occurred in all four of these patients within 5 years (HR, 37.05; 95% CI, 7.660 to 233.336; log-rank P < 0.001; Fig. 3d and Supplementary Table 10).

To determine whether our cfDNA assay can similarly assess the outcome of patients with locally advanced breast cancer, we procured serum samples collected from a cohort of 52 patients with newly diagnosed stage 2 and 3 cancers before treatment with neoadjuvant chemotherapy (Singapore NUHS neoadjuvant breast cancer cohort; Supplementary Fig. 8b and Supplementary Table 11). Analysis of these cfDNA samples identified eight patients who were positive for 1q21.3 amplification; all of them developed relapse (HR, 7.578; 95% CI, 2.935 to 19.569; log-rank P < 0.001; Fig. 3e and Supplementary Table 12) and six died (HR, 11.798; 95% CI, 2.898 to 48.028; log-rank P < 0.001) within 3 years of initial diagnosis (Fig. 3f and Supplementary Table 12). Further multivariate analysis demonstrated that cfDNA detection of 1q21.3 amplification was a strong and independent prognostic factor for survival outcome in this neoadjuvant cohort (HR, 16.351; 95% CI, 3.089 to 86.554; P = 0.001; Supplementary Table 13). These results indicate that cfDNA detection of 1q21.3 amplification in blood is a potential liquid biopsy for identification of patients with high-risk early-stage cancer who might suffer an early relapse. The assay also identifies a group of patients with particularly aggressive disease, for whom current chemotherapy regimens appear inadequate. These patients may be candidates for additional and/or alternative treatments.

cfDNA detection of 1q21.3 amplification monitors dynamic tumor response to chemotherapy

To determine the potential utility of the cfDNA assay for detection of 1q21.3 amplification in monitoring tumor response to treatment, we took advantage of a retrospective cohort of patients with refractory metastatic breast cancer in a phase 2 study20. In this study, patients were administered gemcitabine and carboplatin for a maximum of six cycles, and blood samples were collected before the start of every new cycle and on days 1, 8, and 15 of the first cycle (Supplementary Fig. 9a). Patients stopped treatment if they were deemed to have disease progression, as evaluated through radiological computed tomography (CT) scans performed every 6 weeks. Among the 29 patients in this study, 22 were found to be positive for 1q21.3 amplification at baseline (before gemcitabine and carboplatin treatment), and 10 of these patients had serial serum samples available from the start of chemotherapy to disease progression (Supplementary Fig. 9b). All of the patients, except one (C-S29), were positive for 1q21.3 amplification at baseline and became negative by day 8 or 15 of cycle 1 of treatment, indicating a decrease in tumor burden and an initial drug response. However, the copy-number ratio of cfDNA marker rebounded in most of these patients by cycle 2 or 3, although CT scans for some patients (C-S05, C-S20, C-S21, and C-S29) showed a partial radiological response of the tumors. On the other hand, the three patients (C-S26, C-S37, and C-S30) with CT scan evaluation indicating stable disease at the end of cycle 2 showed no change or a decrease in the copy-number ratio of cfDNA marker, whereas all patients who had progressive disease despite receiving chemotherapy showed either persistent or rebounded copy-number ratio of cfDNA marker during the later cycles of treatment, which preceded radiological evidence of tumor progression by 2.2 months on average (range, 0.6–5.7 months). These results suggest that cfDNA detection of 1q21.3 amplification could be used to predict the early emergence of chemoresistance and to detect radiologically occult disease. Taken together, our results demonstrate the potential application of our cfDNA assay to track and monitor tumor response to treatments earlier than is possible with conventional radiological scans.

1q21.3-encoded S100A7/8/9 and IRAK1 form a functional regulatory circuit to drive tumorsphere growth

To identify therapeutic solutions to target 1q21.3-amplified tumors, we next sought to interrogate the functionality of 1q21.3 amplification in breast cancer. 1q21.3 harbors up to 17 members of the S100A gene family, and several of these have been previously implicated in breast cancer progression21, 22, 23, 24. Moreover, in other contexts, some S100A proteins have been shown to act upstream of Toll-like receptor (TLR) signaling to induce activation of IRAK–nuclear factor (NF)-κB signaling22, 25. We have recently shown that increased IRAK1 phosphorylation is associated with breast cancer recurrence and that IRAK1-directed NF-κB signaling plays an important role in breast cancer metastasis, chemoresistance, and tumor recurrence26. These findings prompted us to test the functional relevance of 1q21.3 amplification in IRAK1 activation and tumorsphere growth.

In a series of ex vivo PDX-derived tumorsphere and breast cancer cell line models, S100A7, S100A8, and S100A9, but not other genes encoding S100A family members, consistently showed upregulation in 1q21.3-amplified tumorspheres and breast cancer cell lines as compared to corresponding bulk tumor cells and 1q21.3-non-amplified cancer cell lines, respectively (Fig. 4a). This observation suggests a close correlation of S100A7, S100A8, and S100A9 expression with 1q21.3 amplification. Efficient knockdown of S100A8 or S100A9 in 1q21.3-amplified MDA-MB-436 (MB436) and MDA-MB-468 (MB468) cells (Supplementary Fig. 10a) was sufficient for impairment of tumorsphere growth (Fig. 4b) as well as IRAK1 phosphorylation (Fig. 4c), although S100A7 knockdown had only a modest effect on MB436 cells owing to the low level of S100A7 expression in this cell line. In contrast, these knockdowns did not seem to affect MDA-MB-231 (MB231) cells, which do not harbor 1q21.3 amplification (Fig. 4b). Similarly, knockdown of S100A7, S100A8, or S100A9 effectively abolished the growth of PDX-derived tumorspheres (Fig. 4d). Notably, the knockdown described above did not affect the proliferation of these cancer cell lines cultured as a monolayer regardless of 1q21.3 status (Supplementary Fig. 10b), suggesting that the impairment of tumorsphere growth upon knockdown of S100A7, S100A8, or S100A9 was not a consequence of reduced cell proliferation. These findings demonstrate a preferential role of S100A7/8/9 in the growth of TICs with 1q21.3 amplification. In addition, treatment of MB231 cells with as little as 1 ng/ml recombinant S100A7, S100A8, or S100A9 protein was able to induce IRAK1 phosphorylation (Fig. 4e). Moreover, when we performed long-term treatment of MCF10A cells with recombinant S100A7, S100A8, or S100A9 individually for 10 d, we observed increased IRAK1 phosphorylation and mammosphere formation in cells treated with S100A8 or S100A9 (Supplementary Fig. 10c,d). No obvious effect was observed in the growth of MCF10A cells in monolayer (Supplementary Fig. 10e).

Figure 4: 1q21.3-encoded S100A7/8/9 forms a functional feedback loop with IRAK1 to drive tumorsphere growth.
1q21.3-encoded S100A7/8/9 forms a functional feedback loop with IRAK1 to drive tumorsphere growth.

(a) Microarray gene expression analysis of S100A family members in tumors versus corresponding derived tumorspheres (n = 4; left) and 3-amplified (MB436, MB468, MCF-7, and T47D) versus non-amplified (BT-549, BT-474, MB361, and MB231) breast cancer cell lines (right). Green, top three upregulated genes; red, downregulated; blue, upregulated. (b) Representative images (n = 6) and quantification of the number of tumorspheres after individual knockdown of S100A7, S100A8, or S100A9 in MB436, MB468, and MB231 cells. shLUC was used as a non-targeting control. Data are expressed as means ± s.e.m. of three independent knockdown experiments. Scale bar, 100 μm. (c) Representative western blot (n = 2) showing a reduction in phosphorylated IRAK1 (pIRAK1) in MB436 and MB468 cells upon individual shRNA-mediated knockdown of S100A7, S100A8, or S100A9. Actin was used as a loading control. (d) Quantification of the number of tumorspheres after individual knockdown of S100A7, S100A8, or S100A9 in two different PDX-derived tumorsphere lines. Data are expressed as means ± s.e.m. of three technical replicates. (e) Representative western blot (n = 2) showing induction of IRAK1 phosphorylation upon individual treatment with recombinant S100A7, S100A8, or S100A9 protein in MB231 cells at the indicated concentrations. Actin was used as a loading control. (f) Real-time PCR analyses of S100A7, S100A8, and S100A9 gene expression after inducible knockdown of IRAK1 in MB436 tumorsphere cells. Dox, doxycycline. Data are expressed as means ± s.d. of three technical replicates. (g) Real-time PCR analysis of S100A7, S100A8, and S100A9 gene expression after overexpression of IRAK1-WT or IRAK1-K239S in MB436 tumorsphere cells. Data are expressed as means ± s.d. of three technical replicates. (h) Quantification of the number of tumorspheres (TS) after overexpression of IRAK1-WT or IRAK1-K239S in MB436, MB468, and MB231 cells. Recombinant S100A8 and S100A9 proteins were added separately to IRAK1-K239S-expressing cells to rescue tumorsphere formation. Data are expressed as means ± s.e.m. of three independent overexpression experiments. (i) Representative IHC images (n = 6) of phosphorylated IRAK1 (Ser376) and S100A8 in matched primary and recurrent breast tumor samples. Scale bars, 100 μm. (j) Quantification of phosphorylated IRAK1 and S100A8 levels in IHC analyses of 18 paired primary and recurrent breast tumor samples. Recurrent tumors showed an increased 3 copy-number ratio.P values were calculated with paired two-tailed t-tests. (k) Correlation analysis of S100A8 and phosphorylated IRAK1 IHC staining in 18 paired tumor samples showing an increased 3 copy-number ratio in recurrent tumors. Linear regression was performed using GraphPad Prism. The linear regression Pearson's correlation coefficient (R2) and its P value are given. All P values were calculated with two-tailed t-tests. *P < 0.05, **P < 0.01, ***P < 0.001.

Conversely, exploration of the functional activity of IRAK1 with respect to the regulation of S100A7/8/9 revealed that IRAK1 might also regulate the expression of S100A7, S100A8, and S100A9. As shown below, inducible IRAK1 knockdown in MB436 tumorspheres resulted in downregulation of S100A8 and S100A9 expression, although S100A7 expression was not affected (Fig. 4f). We further demonstrated that the kinase activity of IRAK1 is required to regulate the expression of S100A7/8/9, as ectopic overexpression of a kinase-dead IRAK1 (IRAK1-K239S) in MB436 cells resulted in downregulation of S100A8 and S100A9, whereas overexpression of wild-type IRAK1 (IRAK1-WT) led to induction of their expression (Fig. 4g). Accordingly, as compared to IRAK1-WT, IRAK1-K239S inhibited tumorsphere growth in 1q21.3-amplified MB436 and MB468 cells but not in MB231 cells (Fig. 4h). Addition of recombinant S100A8 or S100A9 to the culture medium of IRAK1-K239S-expressing cells was able to rescue tumorsphere growth (Fig. 4h). Moreover, overexpression of IRAK1-WT, but not IRAK1-K239S, in the MCF10A noncancerous breast epithelial cell line led to increased tumorsphere formation and upregulation of S100A7, S100A8, and S100A9 expression, which was more evident in the presence of IL-1β treatment (Supplementary Fig. 10f,g). Through both loss- and gain-of-function experiments in various cellular models, these studies have established a functional regulatory circuit involving S100A7/8/9 and IRAK1 signaling with a role in driving tumorsphere growth. Of note, the results also suggest that S100A8 and S100A9 promote tumorsphere growth both through IRAK1 (via feedback regulation) and in an IRAK1-independent manner. Furthermore, in 18 of 25 sets of paired primary and recurrent tumors with enrichment of 1q21.3 amplification (Supplementary Table 14), immunohistochemistry (IHC) analysis showed corresponding induction of IRAK1 phosphorylation and S100A8 expression in recurrent tumors as compared to corresponding primary tumors (Fig. 4i,j) and that increased IRAK1 phosphorylation is significantly correlated with S100A8 expression (Fig. 4k). These results demonstrate the clinical relevance of IRAK1 phosphorylation and S100A8 expression in tumor recurrence.

Pacritinib treatment disrupts S100A7/8/9–IRAK1-mediated tumorsphere growth

We next explored the utility of the S100A7/8/9–IRAK1 circuitry as an actionable drug target in breast cancer. Pacritinib, a potent small-molecule inhibitor of Janus kinase 2 (JAK2) currently being evaluated in multiple phase 2 and 3 clinical trials for the treatment of myelofibrosis27, 28, has been recently reported to be a potent IRAK1 inhibitor29. We thus tested the ability of pacritinib to inhibit S100A7/8/9–IRAK1 signaling in breast cancer cells. Pacritinib was able to effectively block the phosphorylation of both IRAK1 and JAK2 in a dose-dependent manner in MB468 cells as early as 6 h following the start of treatment, although it also inhibited phosphorylation of JAK2 in MB231 cells (Fig. 5a). Remarkably, pacritinib treatment resulted in a strong inhibitory effect on tumorspheres in the MB468 cell line but had little effect in the MB231 cell line (Fig. 5b), suggesting that JAK2 inhibition alone was insufficient to lead to growth inhibition in MB231 cells. In accordance with IRAK1 knockdown or mutant IRAK1 overexpression, pacritinib treatment also reduced the expression of S100A7, S100A8, and S100A9 in MB468 cells in a dose-dependent manner (Fig. 5c). In addition, pacritinib treatment resulted in sustained inhibition of phosphorylated IRAK1 (Fig. 5d) and reduced expression of S100A7, S100A8, and S100A9 (Fig. 5e), concomitantly with strong inhibition of tumorsphere growth (Fig. 5f), in patient-derived tumorsphere cells treated for 24 h. Notably, IL-1β treatment induced JAK2 phosphorylation but did not have an obvious effect on IRAK1 phosphorylation, suggesting that IRAK1 is constitutively phosphorylated in these patient-derived tumorspheres (Fig. 5d). Of note, although expression of S100A8 and S100A9 was consistently reduced by pacritinib, the effect of pacritinib on S100A7 expression seemed to be cell line specific, indicating a more consistent role for S100A8 and S100A9 in regulating tumorsphere growth. Crucially, addition of recombinant S100A8 or S100A9 protein to the culture medium effectively rescued pacritinib-induced growth inhibition in multiple cellular models of tumorspheres from both cell lines and patient-derived samples (Fig. 5g). These observations reveal a robust function for S100A8 and S100A9 in driving tumorsphere growth and demonstrate that inhibiting IRAK1-mediated S100A8 and S100A9 expression contributes substantially to the effect of pacritinib.

Figure 5: Pacritinib effectively disrupts the S100A7/8/9–IRAK1 feedback loop to inhibit tumorsphere growth.
Pacritinib effectively disrupts the S100A7/8/9-IRAK1 feedback loop to inhibit tumorsphere growth.

(a) Representative western blot (n = 2) showing inhibition of phosphorylated IRAK1 and phosphorylated JAK2 (pJAK2) within 6 h of pacritinib treatment in MB468 and MB231 cells. Actin was used as a loading control. (b) Quantification of the number of tumorspheres after pacritinib treatment for MB468 and MB231 cells. Data are expressed as means ± s.e.m. of three independent drug treatment experiments. (c) Real-time PCR analysis of S100A7, S100A8, and S100A9 gene expression after 24 h of pacritinib treatment in MB468 cells. Data are expressed as means ± s.d. of three technical replicates. (d) Representative western blot (n = 2) showing prolonged inhibition of phosphorylated IRAK1 24 h following pacritinib (2.5 μM) treatment of various PDX-derived tumorspheres with and without IL-1β. Actin was used as a loading control. (e) Real-time PCR analysis of S100A7, S100A8, and S100A9 transcript levels 24 h following pacritinib (Pac; 2.5 μM) treatment. Data are expressed as means ± s.d. of three technical replicates. (f) Quantification of the number of tumorspheres after pacritinib treatment for four different PDX-derived tumorspheres. Data are expressed as means ± s.e.m. of three independent experiments. (g) Representative images (n = 6) and quantification of the number of tumorspheres in rescue assays. Treatment with recombinant S100A8 or S100A9 protein (10 ng/ml) was able to rescue pacritinib-treated (2.5 μM) tumorspheres from various cell lines. Data are expressed as means ± s.d. of three independent experiments. Scale bar, 100 μm. All P values were calculated with two-tailed t-tests. **P < 0.01, ***P < 0.001.

1q21.3 amplification is associated with efficacy of pacritinib both in vitro and in vivo

We next explored whether 1q21.3 amplification correlates with pacritinib sensitivity in breast cancer cells. For this purpose, we profiled 1q21.3 gene amplification in a series of breast cancer cell lines using ddPCR (Fig. 6a). Of note, 1q21.3 status in these cell lines was generally correlated with phosphorylation of IRAK1 and expression of S100A8 and S100A9 (Fig. 6b and Supplementary Fig. 11a) but not with the phosphorylation status of JAK2 (Fig. 6b). Accordingly, pacritinib treatment was more effective in impairment of tumorsphere growth in cancer cell lines positive for 1q21.3 amplification than in cell lines without amplification (Fig. 6c). Of note, HCC1937 cells, despite being negative for 1q21.3 amplification, had high levels of both phosphorylated IRAK1 and phosphorylated JAK2 and thus were sensitive to pacritinib treatment. These observations suggest that pacritinib response is more associated with 1q21.3 amplification and IRAK1 than with JAK2; these could be potential biomarkers for stratification of patients to obtain optimal drug response.

Figure 6: Amplification status of 1q21.3 correlates with the efficacy of pacritinib in vitro and in vivo.
Amplification status of 1q21.3 correlates with the efficacy of pacritinib in vitro and in vivo.

(a) ddPCR analysis of genomic DNA from various breast cancer cell lines to determine amplification status of 1q21.3. RPP30 was used as a reference gene for normalization. Data are expressed as means ± s.d. of four technical replicates. (b) Representative western blot (n = 2) showing phosphorylated IRAK1 and phosphorylated JAK2 levels in tumorspheres from the breast cancer cell lines in a. Actin was used as a loading control. Red labels, cell lines with 1q21.3 amplification. (c) Growth curve of tumorspheres from ER-positive and ER-negative breast cancer cell lines treated with increasing doses of pacritinib. Red, amplified; black, non-amplified. Data are expressed as means ± s.e.m. of three independent experiments. (d) NOD-SCID mice bearing MB231 and HCC70 tumors were treated with 50, 100, or 150 mg per kg body weight (mg/kg) pacritinib by oral gavage (n = 10 mice per group). (e) Western blot analysis (n = 2) showing the level of phosphorylated IRAK1 in HCC70 xenograft mice (n = 4) treated with pacritinib. Actin was used as a loading control. (f) NOD-SCID mice bearing PDX3 tumors were treated with 20 mg per kg body weight paclitaxel by intravenous tail vein injection every other day for 14 d to induce tumor regression. The residual tumors were surgically removed, and mice were rested for 14 d before treatment with 20 mg per kg body weight paclitaxel, 150 mg per kg body weight pacritinib, or both drugs for 14 d. The sample size (n) per group is indicated. (g) Schematic representation of the proposed mechanisms of an S100A8/9–IRAK1 feedback loop in 1q21.3-amplified tumors, as well as the application of 1q21.3 amplification as a biomarker for use in companion diagnostics and tracking of tumor response. (h) Genomic alterations of S100A8, S100A9 (located on 1q21.3), and genes with targeted therapy in various TCGA cancer data sets. In d and f, P values were calculated with two-tailed unpaired t-tests. **P < 0.01, ***P < 0.001.

We next tested the efficacy of pacritinib in vivo using mice orthotopically engrafted with 1q21.3-amplified HCC70 cells, which express a high level of phosphorylated IRAK1 but a low level of phosphorylated JAK2, as well as 1q21.3-non-amplified MB231 cells, which express a low level of phosphorylated IRAK1 but a high level of phosphorylated JAK2 (Fig. 6b). Pacritinib treatment in different doses resulted in substantial inhibition of tumor growth for HCC70 xenograft tumors but had no obvious effect on the growth of MB231 xenografts (Fig. 6d). The efficacy of pacritinib on HCC70 tumor growth was confirmed to be associated with downregulation of phosphorylated IRAK1 (Fig. 6e) and S100A7, S100A8, and S100A9 expression but not with JAK2 (Supplementary Fig. 11b). The efficacy of pacritinib on 1q21.3-amplified tumors was further confirmed in three additional xenograft tumor models (Supplementary Fig. 11c).

To assess the potency of pacritinib in mitigating tumor recurrence in a mouse model, we explored a clinically relevant 'neoadjuvant' setting, in which a PDX mouse model of triple-negative breast cancer was treated with paclitaxel to induce tumor regression (first-line treatment) before surgical removal of the remaining residual tumor as described in Supplementary Figure 5c. Mice were rested for 2 weeks to recover from surgery before further treatment with paclitaxel, pacritinib, or both for 2 weeks (second-line treatment). Although tumor regrowth was slowed by paclitaxel or pacritinib treatment, tumors showed facilitated regrowth upon termination of treatment. In contrast, the combination of paclitaxel and pacritinib resulted in tumor regression, which lasted for up to 2 months after termination of treatment (Fig. 6f). Of note, pacritinib both alone and in combination with paclitaxel was well tolerated by the mice during the course of treatment, and there was no obvious toxicity or body-weight loss (Supplementary Fig. 11d,e). These in vivo findings demonstrate the potential use of pacritinib in combination with the standard chemotherapy regimen in management of 1q21.3-amplified breast tumors.

Discussion

Our study identified 1q21.3 amplification and the associated S1007/8/9–IRAK1 signaling feedback loop as an important driver event in breast cancer progression. We demonstrated the enrichment of 1q21.3 amplification in breast cancer TICs and in more than 70% of recurrent breast cancers; detection of amplification in blood using cfDNA could serve as a potential circulating biomarker for prediction of disease outcome and for monitoring of treatment response. Notably, we established 1q21.3-directed molecular events as an actionable target that has considerable implications for the clinical management of patients with breast cancer.

The widespread occurrence of 1q21.3 amplification in recurrent breast cancers indicates that this biomarker might be useful in covering a large proportion of patients with breast cancer across different subtypes. We show that, in patients with early-stage breast cancer as well as those with locally advanced breast cancer, this blood-based assay is able to identify a tendency to recur early (within the first 5 years) at the time of diagnosis. Although ER- or HER2-positive tumors tend to develop recurrence late owing to improved treatments, a sizable proportion of these patients still experience early recurrence, and thus identifying these patients would potentially allow for early consideration of alternative treatments.

The use of circulating ctDNA as a liquid biopsy is becoming a reality in cancer diagnosis and monitoring30, 31. With the rapid advancement of next-generation sequencing (NGS) technology, it is now possible to track and anticipate tumor recurrence from the ctDNA of patients with various solid cancers32, 33, 34, 35. However, owing to intertumor heterogeneity in breast cancer36, 37, 38 and a lack of prevalent common mutations spanning different subtypes, current efforts in breast cancer research have focused on NGS approaches to identify somatic mutations in the primary tumor before development of personalized ctDNA assays to track tumor recurrence39, 40. In addition, the NGS-based cfDNA assays currently under development lack prognostic value at the time of diagnosis, although they could predict outcomes of patients on the basis of the amount of ctDNA detected after surgery35, 41.

We propose that, for early-stage patients, detection of 1q21.3 amplification could identify a subgroup of women with particularly aggressive tumors; this has important implications for prognostication as well as treatment recommendations. It can also be used to monitor tumor response in the metastatic setting. Available serum tumor markers for breast cancer (CA15-3, CEA, and CA27.29) are often limited by reduced sensitivity and specificity42, 43, 44, and radiologic imaging approaches (CT scans, magnetic resonance imaging (MRI), or positron emission topography (PET)–CT scans), although noninvasive, introduce radiation exposure and potential complications of hypersensitivity, contrast extravasation, and renal impairment to patients. Our study suggests that cfDNA may provide an early indication of treatment failure in metastatic breast cancer and can potentially replace CT scans as the primary modality for monitoring. CT scans can be reserved for investigation of any rise in cfDNA. An early indication of treatment failure could allow a prompt switch in treatment regimens.

We further demonstrated the functional relevance of 1q21.3 amplification to IRAK1 activation and unraveled a functional circuitry involving S100A7/8/9 and IRAK1 in the enhancement of tumorsphere growth (Fig. 6g). S100A8, secreted by myeloid cells found in the tumor microenvironment, has been previously shown to promote breast cancer chemoresistance and metastasis21. Our findings, however, demonstrated a cell-autonomous role, resulting from 1q21.3 amplification, in cancer for S100A8 and S100A9 in disease progression. We have previously shown that IRAK1 activity is important in metastasis and chemoresistance for triple-negative breast cancer26. Here we showed elevated IRAK1 phosphorylation and S100A8 protein level in up to 70–80% of both ER-positive and ER-negative recurrent breast cancer tumors, in accordance with the high proportion of recurrent breast cancer tumors showing 1q21.3 amplification.

Notably, we further demonstrated that 1q21.3 amplification and associated IRAK1 activation could be targeted by an existing small-molecule agent, pacritinib, as demonstrated in in vitro and in vivo models. Intriguingly, pacritinib-induced inhibition of tumor growth was effectively rescued by treatment with recombinant S100A8 or S100A9 protein, indicating a crucial role for S100A8 and S100A9 in mediating the effects of pacritinib. Moreover, we show that breast cancer cells harboring 1q21.3 amplification are more sensitive to pacritinib treatment than cancer cells without 1q21.3 amplification, suggesting the potential value of 1q21.3 amplification as a biomarker for the identification of patients who could benefit from pacritinib treatment. Pacritinib can be repurposed for breast cancer treatment to reduce tumor recurrence. Of note, 1q21.3 amplification is also found in other cancers, where it does not seem to coexist with other druggable driver mutations (Fig. 6h). Therefore, targeting 1q21.3-amplified tumors with pacritinib may provide benefit to patients without actionable mutations and encourage the design of 'basket' clinical trials to test pacritinib in patients carrying 1q21.3 amplification.

Hence, the widespread 1q21.3 amplification and S100A7/8/9–IRAK1 functional loop present in recurrent tumors across distinct breast cancer subtypes highlight the clinical relevance of 1q21.3 amplification as a common feature in breast cancer progression. This genomic aberration may represent a necessary subclonal event in driving tumor recurrence and chemoresistance that can be therapeutically mitigated using existing clinical inhibitors, such as pacritinib.

Methods

Breast tumor dissociation and patient-derived tumorsphere culture.

Surgically resected tumor specimens were obtained from consenting patients. Tumors were first washed with PBS supplemented with antibiotic-antimycotic (Invitrogen, cat. no. 15240-062) and then mechanically disaggregated followed by enzymatic digestion at 37 °C for 2 h in DMEM/F12 medium with 1 mg/ml collagenase type IV (Sigma, St. Louis, MO). After incubation, the cell suspensions were triturated and passed through a 40-μm-pore strainer (BD Falcon, San Jose, CA), and single cells were seeded onto an ultra-low-attachment plate (Corning, Kennebunk, ME) at 50,000 cells/ml in serum-free DMEM/F12 supplemented with N2 and B27 (Gibco, Grand Island, NY), 20 ng/ml epidermal growth factor (MACS, Auburn, CA), 20 ng/ml basic fibroblast growth factor (MACS, Auburn, CA), 0.2 μM thiazovivin (STEMCELL Technologies, Vancouver, BC), and penicillin-streptomycin (Gibco, Carlsbad, CA). After 12–15 d, tumorspheres were passaged with 0.05% trypsin digestion followed by replating in the same manner as the previous generation. All cells were maintained at 37 °C in a humidified atmosphere at 5% CO2.

Cell lines.

All cell lines were obtained, authenticated, and cultured according to American Type Culture Collection (ATCC, Manassas, VA) instructions, unless otherwise stated. All cell lines used for functional studies were tested and found to be free of mycoplasma contamination. BT-474, BT-549, MDA-MB-231, MDA-MB-361, MDA-MB-436, MDA-MB-468, MCF-7, and T-47D breast cancer cell lines and HEK293T and Platinum-A (Plat-A) cell lines were grown in DMEM (Invitrogen) supplemented with 10% FBS. The Plat-A cell line was obtained from Cell Biolabs. HCC70 and HCC1937 cells were maintained in RPMI (Invitrogen) supplemented with 10% FBS. SUM159PT cells obtained from Asterand Bioscience were maintained in Ham's F-12 (Invitrogen) supplemented with 5% FBS, 5 μg/ml insulin (Invitrogen), and 1 μg/ml hydrocortisone (Invitrogen). All media were supplemented with 5,000 U/ml penicillin-streptomycin (Invitrogen). All cell lines were maintained at 37 °C in a humidified atmosphere at 5% CO2.

Reagents.

Recombinant IL-1β (cat. no. 200-01B) was purchased from Peprotech (Rocky Hill, NJ). Recombinant S100A7 (cat. no. pro-149), S100A8 (cat. no. pro-800), and S100A9 (cat. no. pro-814) were purchased from ProSpec (East Brunswick, NJ). Cells were treated with recombinant proteins at the stated concentration for 20 min before harvesting for western blot analysis. Paclitaxel (cat. no. P-9600) was purchased from LC Lab (Woburn, MA). Pacritinib (cat. no. HY-16379) was purchased from MedChem Express (Princeton, NJ).

Mammosphere formation assay.

PDX-derived tumorsphere cells were trypsinized and passed through a 40-μm cell strainer to achieve single-cell suspensions. 3 × 104 cells per well were seeded in six-well ultra-low-attachment plates (Corning, NY; CLS3471) in serum-free DMEM/F12 supplemented with N2 and B27 (Gibco, Grand Island, NY), 20 ng/ml epidermal growth factor (MACS, Auburn, CA), 20 ng/ml basic fibroblast growth factor (MACS, Auburn, CA), 0.2 μM thiazovivin (STEMCELL Technologies, Vancouver, BC), and penicillin-streptomycin (Gibco, Carlsbad, CA). For breast cancer cell lines and the MCF10A cell line, 3 × 104 cells per well were seeded in six-well ultra-low-attachment plates in MammoCult Medium (STEMCELL Technologies, Vancouver, BC) supplemented with fresh hydrocortisone (0.5 μg/ml) and heparin (1:500). Mammospheres were maintained at 37 °C with 5% CO2 and topped up with medium every 3 d. After 7–9 d, mammospheres were stained with 2-(4-iodophenyl)-3-(4-nitrophenyl)-5-phenyl-2H-tetrazolium chloride (INT) (Sigma-Aldrich) and quantified. Imaging and quantification were done using the GelCount apparatus and associated software (Oxford Optronix, Abingdon, UK). For pacritinib treatment, cells were seeded in mammosphere medium, treated at indicated concentrations, and supplemented with fresh drug at the respective dosage every 3 d until quantification. For S100A8 and S100A9 recombinant-protein rescue experiments, tumorspheres were pretreated with 10 ng/ml of the respective recombinant S100A8 or S100A9 protein for 1 h before treatment with 1 μM pacritinib. Every 3 d, 0.5 ml of fresh growth medium, pacritinib, and the respective recombinant protein were added to the corresponding initial tumorsphere treatment condition.

Cell proliferation assay.

For the cell proliferation assay, optimal cell seeding was first determined empirically for all cell lines by examining the growth of a wide range of seeding densities in a 96-well format to identify conditions that permitted proliferation for 7 d. Cells were then plated at the optimal seeding density in triplicate. Plates were incubated for 7 d at 37 °C in 5% CO2. Cells were then lysed with CellTiter-Glo (CTG, Promega, Madison, WI), and the chemiluminescent signal was detected with a microplate reader on days 0, 1, 3, 5, and 7. Luminescence signal values obtained during the 7 d were plotted against time.

shRNAs and ectopic overexpression.

Constructs for inducible IRAK1 shRNA as well as ectopic IRAK1-WT and IRAK1-K239S kinase-dead mutant were generated as described previously26. To generate S100A7-, S100A8-, and S100A9-knockdown cell lines, shRNA oligonucleotides were subcloned into the pLV-H1-EF1a-RFP-Puro (cat. no. SORT-B31) vector according to the manufacturer's instructions (Biosettia, San Diego, CA). The shRNA sequence was confirmed with DNA sequencing of the plasmid. Stable knockdown cell lines were generated via lentiviral infection using HEK293T cells to package lentivirus. Briefly, 2 μg of shRNA plasmid construct together with 1.5 μg of psPAX2 plasmid (Addgene) and 0.5 μg of pMD2.G plasmid (Addgene) was transfected into HEK293T cells using Lipofectamine 2000 (Invitrogen). Media containing the transfection reagent were removed 8 h post-transfection and replaced with 5% FBS–containing DMEM. Lentivirus-containing media were collected 24 h later and passed through a 0.45-μm filter to remove detached HEK293T cells. Adherent monolayer cells were infected with lentivirus harboring shRNA plasmids at a multiplicity of infection (MOI) of 3 for 24 h in the presence of 8 μg/ml polybrene and then maintained with puromycin (MB436, 1.5 μg/ml; MB468, 0.5 μg/ml). Cells selected for 7 d were immediately seeded for tumorsphere assays and cell proliferation assays concurrently.

For PDX-derived tumorsphere cell lines, lentivirus-containing media were first concentrated with Amicon (Millipore, Ireland) ultra-centrifugal filter columns (cat. no. UFC910096) before infection of tumorsphere cells at an MOI of 3 for 24 h in the presence of 8 μg/ml polybrene. Positive cells were selected with puromycin (1 μg/ml) before performing downstream assays. The specific shRNA oligonucleotides used for cloning are summarized in Supplementary Table 15.

RNA-seq.

Twelve primary human breast tumors and corresponding tumorsphere cells were used to generate RNA-seq data. RNA-seq libraries were generated by using the cDNA amplification kit SMARTer Ultra Low RNA Kit (cat. no. 634935, Clontech Laboratories, Mountain View, CA) for small amounts of RNA or <200 cells, according to the manufacturer's manual, followed by DNA library construction using the NEBNext DNA Library Prep Master Mix Set for Illumina kit (cat. no. E6040S, New England BioLabs, Ipswich, MA). In brief, approximately 50 ng of total RNA or 200 tumor cells were first lysed in reverse-transcription buffer, and the reaction was initiated with oligo(dT) primer. Complete first-strand synthesis was followed by template switching and the incorporation of SMARTer oligonucleotide. Full-length cDNAs were amplified using PCR to obtain DNA. Fragmentation and adaptor introduction were performed using acoustic shearing to obtain a length of ~200–500 bp and the NEBNext DNA Library Prep kit for incorporation of multiplex index primers. A pooled multiplexed library, consisting of equal amounts of six individual libraries, was sequenced on the HiSeq 2500 system by the Genome Institute of Singapore (GIS) core facility.

For data processing and analyses, Illumina 100-bp paired-end sequenced reads were aligned to reference human genome hg19. All further analyses were performed using R statistical programming. Genes were ranked according to significance of differential expression between primary tumors and derived tumorspheres using DESeq2 (Bioconductor). Genes showing >2-fold alteration in expression and FDR < 0.05 (the cutoff value) were considered as having significantly altered expression. The heat map of upregulated genes was generated using the gplots package.

Expression microarray analyses.

To identify the S100 family members important in breast cancer stem cells, two sets of expression microarray analysis were performed. The first microarray set consisted of RNA extracted from four PDX tumors and the corresponding ex vivo tumorsphere cultures. The second microarray set consisted of four 1q21.3-amplified cell lines (MDA-MB-436, MDA-MB-468, MCF-7, and T-47D) and four cell lines not amplified at 1q21.3 (BT-549, BT-474, MDA-MB-361, and MDA-MB-231). RNA extracted from cells was used for expression microarray analysis with the Illumina Gene Expression Sentrix BeadChip HumanHT-12_V4 (Illumina, San Diego, CA) according to the manufacturer's recommended protocol. In brief, 500 ng of total RNA was converted to single-stranded cDNA using a T7 oligo(dT) primer and subsequently converted to double-stranded DNA (dsDNA) template for transcription. The dsDNA was then amplified and labeled with biotin to generate biotinylated cRNA. The labeled cRNAs were hybridized to Illumina BeadChip HumanHT-12_V4 microarray slides for 20 h. After extensive washing, Cy3-conjugated streptavidin (Cy3-SA) was introduced to bind to the analytical probes that had been hybridized to the BeadChip. The microarrays were then scanned with the Illumina BeadArray Reader. The Illumina BeadArray Reader used a laser to excite the fluorescent signals of the hybridized single-stranded product on the beads of the BeadChip sections. Light emissions from these fluorescent signals were then recorded in high-resolution images of the BeadChip sections. Data from these images were analyzed using Illumina's GenomeStudio Gene Expression Module. The raw intensity data for the gene expression profile were further analyzed using GeneSpring GX software (Agilent Technologies, Santa Clara, CA). The differentially expressed S100 genes (fold change ≥2) were identified by comparison of tumorspheres versus PDX tumors and amplified versus non-amplified breast cancer cell lines.

TCGA and METABRIC data set analyses.

Copy number variation (CNV) of upregulated genes was analyzed with the web-based cBioPortal (http://www.cbioportal.org/) using the TCGA breast invasive carcinoma data set8. Briefly, gene sets of interest were submitted through cBioPortal, and putative copy-number alterations from Genomic Identification of Significant Targets in Cancer (GISTIC)45 for each gene were retrieved and presented in graphical OncoPrints showing putative homozygous (deep) genetic deletion and genetic amplification (gain of two or more copies).

To study CNV in relation to gene expression, the TCGA data set with normalized gene expression for breast invasive carcinoma8, along with corresponding copy-number data (with GISTIC annotations), was downloaded from the UCSC Cancer Genomics Browser. METABRIC breast cancer gene expression and copy-number data (with GISTIC annotations)11, 12 were downloaded from cBioPortal. Patients were stratified according to their copy-number status ('Deletion', −1; 'Neutral', 0; 'Gain', +1; 'Amp', +2). The mean mRNA expression of the 17 genes identified on 1q21.3 was calculated for each individual and plotted against corresponding copy-number status. P values were calculated with Kruskal–Wallis tests.

To correlate CNV with clinical prognosis, clinical information (time to death from initial diagnosis) and copy-number data (with GISTIC annotations) for the TCGA breast invasive carcinoma data set8 were downloaded from the UCSC Cancer Genomics Browser. Copy-number status at 1q21.3 and 8q was analyzed. Patients were grouped into two groups (Neutral versus Gain + Amp), and their 'Time to Death' clinical feature was analyzed. Clinical information and copy-number data (with GISTIC annotations) for the METABRIC breast cancer data set were downloaded from cBioPortal. Copy-number status at 1q21.3 and Time to Death were analyzed according to respective GISTIC annotations (Neutral, Gain, Amp). P values were calculated with Mann–Whitney tests.

Breast cancer survival analyses of a public data set.

Kaplan–Meier (KM) survival analyses for relapse-free survival (RFS) were performed using the online database (http://www.kmplot.com/). The following 17 probes were used to generate the KM plots: TUFT1 (205807_s_at), S100A10 (200872_at), S100A11 (200660_at), SPRR1A (213796_at), SPRR1B (205064_at), S100A9 (203535_at), S100A8 (202917_s_at), S100A7 (205916_at), S100A6 (217728_at), S100A2 (204268_at), S100A16 (227998_at), S100A14 (218677_at), SNAPIN (223066_at), JTB (200048_s_at), RAB13 (202252_at), UBE2Q1 (222480_at), and EFNA (210132_at). Mean expression of the 17 probes was used to generate survival curves. The ER status of each subject was derived from the gene expression data set. The percentiles of the subjects between the upper and lower quartiles were autoselected on the basis of the computed best-performing thresholds used as cutoffs. All other parameters were left at default settings unless otherwise stated. For the METABRIC breast cancer data set, subjects were stratified into two groups ('High', 'Low') according to the mean expression of the 17 genes identified in RNA-seq analysis using the Z-score and best-performing thresholds as cutoffs. KM plots were generated for the respective groups using GraphPad Prism version 6.0.

Kaplan–Meier survival analyses of cohorts of patients with breast cancer.

To generate KM plots for Singapore TTSH discovery cohort tumor samples, a cutoff of a copy-number ratio of 1.5 for the S100A8 gene was used to create two groups. To generate KM plots for Denmark OUH early-stage and Singapore NUHS neoadjuvant cohort blood samples, both subjects positive and negative for the 1q21.3 amplification were analyzed for progression-free and overall survival. KM plots were generated for the respective groups using GraphPad Prism version 6.0.

Genomic DNA extraction and real-time PCR analyses.

Fresh frozen tumor samples were homogenized with Qiagen TissueLyzer II, and genomic DNA was extracted with a QIAamp DNA Mini kit (Qiagen, Hilden, Germany), as described by the manufacturer. DNA was isolated from formalin-fixed, paraffin-embedded (FFPE) samples after extraction of 5-μm-thick paraffin sections in xylene and by using the QIAamp DNA FFPE Tissue kit (Qiagen, Hilden, Germany) DNA-extraction protocol as described by the manufacturer. FFPE slides were stained with H&E to evaluate tumor cell content. qPCR assays were performed using the KAPA SYBR FAST qPCR kit (KAPA Biosystems, Wilmington, MA). 10 ng of genomic DNA was used for each reaction. In determining the relative fold change in genomic DNA, the RPP30 level was used as an internal control for normalization. All reactions were analyzed on an Applied Biosystems PRISM 7500 Fast Real-Time PCR system in 96-well plate format.

ALDEFLUOR assay and fluorescence-activated cell sorting.

The ALDEFLUOR assay was performed using the manufacturer's recommended protocol (ALDEFLUOR kit, STEMCELL Technologies, cat. no. 01700). Gating was set up based on the manufacturer's recommended protocol. In brief, 1 million cells in single-cell suspension were centrifuged at 200g and resuspended in the ALDEFLUOR assay buffer supplied in the kit. The cells for each sample were incubated with or without an ALDH-specific inhibitor, diethylaminobenzaldehyde (DEAB; 15 mM), in the presence of 0.15 mM ALDH substrate. For fluorescence-activated cell sorting (FACS) analysis only, ALDEFLUOR staining was detected using the fluorescein isothiocyanate (FITC) channel of a FACSCalibur Flow Cytometry System (BD Biosciences) after 30 min of incubation at 37 °C. A control sample with DEAB was used as a sorting gate to reflect background fluorescence levels for each cell line. BD CellQuest was used for analysis with the FACSCalibur system. For cell sorting, ALDH-positive and ALDH-negative cell populations were sorted after ALDEFLUOR staining using a BD FACSAria II, BD FACSAria Fusion, or Beckman MoFlo. Sorted cells were pelleted and genomic DNA was extracted using the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany).

DNA fluorescence in situ hybridization.

In brief, cells were trypsinized and harvested for DNA FISH. The cell pellet was washed once with PBS and treated with 75 mM KCl for 15 min before fixing with modified Carnoy's fixative. After fixation of the cells, a drop of the fixed cell suspension was placed onto a glass microscope slide and air-dried. The slide was dehydrated through an ethanol series (70%, 85%, and 100%) for about 2 min for each step at room temperature and allowed to dry. FISH assays were carried out using a DNA probe mixture consisting of a 1q21.3 probe labeled in Texas Red and a 1p32.3 probe labeled in FITC (Cytocell Aquarius, Cambridge, UK, catalog no. LPH 039-A). The DNA probe mixture was applied to the target area and co-denatured, followed by overnight hybridization at 37 °C. Washes were performed, and the slide was counterstained with DAPI antifade solution (Vectashield, Vector Laboratories) and analyzed under an epifluorescence microscope. Signals from 100 non-overlapping nuclei were quantified for copy-number changes, and the normal signal pattern was defined as two copies.

Animal work.

Surgical procedures and experiments were conducted in compliance with animal protocols approved by the Agency for Science, Technology, and Research (A*STAR) Biopolis Institutional Animal Care and Use Committee of Singapore (IACUC). Female NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG) and NOD/MrkBomTac-Prkdcscid (NOD-SCID) mice aged 6–8 weeks were purchased from InVivos (Singapore). For the limiting-dilution assay, trypsinized patient-derived tumorspheres were serially diluted from 10,000 to 100 cells before resuspension in reduced Matrigel (BD Bioscience, cat. no. 356230) and injected with 10 μl of resuspended cells into the mammary fat pad of the NSG mice. For PDX orthotopic xenotransplantation, tissue suspensions were first prepared from whole-tumor explants from the mammary fat pads of immunodeficient NOD-SCID mice. Whole tumors were washed in PBS, minced with a sterile scalpel, and triturated until cells could pass through a needle bore. The PDX cell suspension was resuspended in reduced Matrigel (BD Bioscience, cat. no. 356230), and 10 μl was then injected into the mammary fat pad. Autoclips was used for primary wound closure. Mice were observed post-procedure for 1–2 h, and their body weights and wound healing were monitored weekly. Tumors were measured by Vernier caliper twice weekly, and tumor volume was calculated with the following formula: V = W × W × L/2. Randomization was performed by equally dividing tumor-bearing mice of similar tumor burden into control and experimental groups for drug treatment. No statistical method was used to predetermine sample size. No experimental samples were excluded in this study, with the exception of animals that died from surgery or unexpected illness. For the in vivo pacritinib sensitivity test, 1 million MDA-MB-231 or HCC70 cells mixed 1:1 in PBS:Matrigel were injected into the mammary fat pad of mice using standard procedures. After the tumors reached 100 mm3, the mice were randomized into four groups: treatment with vehicle (n = 10), 50 mg per kg body weight pacritinib (n = 10), 100 mg per kg body weight pacritinib (n = 10), and 150 mg per kg body weight pacritinib (n = 10). Mice were then dosed daily with treatment by oral gavage for 21 d. Tumor growth was monitored for 1 month. A similar procedure was performed for MDA-MB-468, MCF-7, and BT-474 xenografts, for which mice were randomized into two groups: treatment with vehicle and 150 mg per kg body weight pacritinib.

For the in vivo PDX model, each mouse bore one PDX tumor on one side of the mammary fat pad. After the tumors reached 100 mm3, the mice were randomized into two groups: vehicle-treated control tumor (n = 4) and paclitaxel-treated tumor (n = 27). Treated mice were given 20 mg per kg body weight paclitaxel (P-9600, LC Lab) every other day for two consecutive weeks via tail vein injection to induce tumor regression. Upon completion of paclitaxel treatment, the tumors were surgically removed. The mice were given 1 week to recover before randomization into four groups: vehicle alone (n = 4), paclitaxel (20 mg per kg body weight) treatment alone (n = 4), pacritinib (150 mg per kg body weight) treatment alone (n = 4), and paclitaxel + pacritinib combination treatment (n = 4). Paclitaxel treatment was given by tail vein injection every other day for two consecutive weeks while pacritinib was administered daily via oral gavage for 2 weeks. Following drug treatments, the tumors were monitored for approximately 1 month. Tumors were excised upon reaching 1,000 mm3 or at the end of the study. Mice were euthanized according to ethical guidelines, and the dissected tumors were analyzed for change in phosphorylated IRAK1 by western blot and gene expression changes by real-time PCR.

Serum and plasma collection.

All the serum and blood samples used in this study were archived, and frozen samples were stored at −80 °C. For serum preparation, blood was allowed to clot by leaving undisturbed at room temperature for 30 min, followed by centrifugation at 2,000g at 4 °C for 10 min within 1 h of blood collection to remove the clot. The resulting supernatant portion of blood serum was further aliquotted and stored at −80 °C. Plasma isolation was performed as previously described46. In brief, whole blood was collected in tubes containing sodium citrate as an anticoagulant. Then, the blood was centrifuged at 1,300g at 4 °C for 10 min within 1 h of blood collection and immediately processed through a 13-mm filter to remove potential cell contamination and blood-related debris. The upper layer of clear plasma was further aliquotted and stored at −80 °C.

Isolation and quantification of circulating cfDNA.

cfDNA was extracted from archived serum or plasma aliquots stored at −80 °C. On the day of cfDNA isolation, approximately 500 μl of frozen plasma or serum was quickly thawed in a warm water bath immediately followed by a further centrifugation at 20,000g for 10 min at 4 °C. cfDNA was isolated using the QIAamp circulating nucleic acid kit (Qiagen, Hilden, Germany, cat. no. 55114) from the cleared supernatant according to the manufacturer's instructions. Purified cfDNA was quantified using the Quant-iT PicoGreen dsDNA Assay Kit (Thermo Fisher Scientific, Carlsbad, CA, cat. no. P7589), and the DNA fragment-size distribution was further visualized using a High Sensitivity DNA Analysis Chip run on a 2100 Bioanalyzer instrument (Agilent Technologies, Waldbronn, Germany, cat. no. 5067-4626) to determine the genomic DNA background.

Droplet digital PCR conditions and optimization.

Digital PCR was performed on a QX200 ddPCR system (Bio-Rad) using EvaGreen chemistry with primers at a final concentration of 100 nM. For optimization of primer sets, we performed a gradient PCR to determine the optimal annealing temperature for all primer pairs in a single-plate reaction. PCR reactions were prepared with ddPCR Supermix for EvaGreen (Bio-Rad) and partitioned into a median of 20,000 droplets per sample in a QX-200 droplet generator following the manufacturer's instructions. All genomic DNA samples were treated with HindIII restriction enzyme (cat. no. R0104L, New England BioLabs) in reaction following the manufacturer's instructions. At least two negative-control wells with no DNA template were included in every batch. Emulsified PCR reactions were run on a 96-well plate thermal cycler (C1000 Touch, Bio-Rad), and plates were incubated at 95 °C for 5 min followed by 40 cycles of 95 °C for 30 s and extension at 60 °C for 90 s, with incubation for 5 min at 4 °C followed by 5 min at 98 °C. The temperature ramp increment was 2 °C per second for all steps. Plates were read on a Bio-Rad QX200 droplet reader using QuantaSoft version 1.4.0.99 software from Bio-Rad to assess the number of droplets positive for DNA. Analysis of the ddPCR data was also performed with QuantaSoft software (Bio-Rad).

To assess the reproducibility and detection performance of our assay with low DNA input amounts, we performed serial dilution of known positive-control (1q21.3-amplified cell line: HCC70) and negative-control (1q21.3-non-amplified cell line: fibroblasts) genomic DNA and evaluated the lowest input at which ddPCR could still accurately call a result (Supplementary Fig. 3a). The range of reproducibility for a small amount of DNA was determined using serial dilutions starting from 10 ng of the positive and negative controls. Amplification could be correctly detected using as little as 0.156 ng (~32–40 copies per reaction) of DNA input (Supplementary Fig. 3a).

To assess the sensitivity of our assay for detection of amplification in heterogeneous tumor samples, we performed a spike-in experiment and measured the averaged copy-number ratio of the three genes (TUFT1, S100A7, and S100A8) in a 1q21.3-amplified cell line (HCC70) serially diluted into normal diploid fibroblast DNA. As little as 5% amplified genomic DNA could be detected using 2 ng of total genomic DNA as input (Supplementary Fig. 3b).

To examine 1q21.3 amplification in tumor tissue, 5 ng of genomic DNA from tumor tissue was used for each reaction. To set a cutoff for definition of positive 1q21.3 amplification in genomic samples, an averaged copy-number ratio of the three genes (TUFT1, S100A8, and S100A7) relative to the reference gene, RPP30, from 30 normal breast tissues adjacent to breast cancer tumors was determined. A cutoff for a positive sample was set at 3 s.d. above the mean (cutoff threshold, 1.141) (Supplementary Fig. 5a).

Owing to the limited availability of the archived blood samples (300–500 μl), we first determined the ability of our assay to detect 1q21.3 amplification using a low amount of cfDNA. We spiked 1q21.3-amplified genomic DNA (HCC70) into normal control cfDNA through serial dilutions. At 2 ng of total DNA input per ddPCR reaction, we were able to detect 5% amplified tumor DNA in cfDNA (Supplementary Fig. 6a).

To examine 1q21.3 amplification using cfDNA, 2 ng of cfDNA from plasma or serum was used for ddPCR analysis. To set a cutoff for definition of positive 1q21.3 amplification in cfDNA, the averaged copy number ratio of the three genes (TUFT1, S100A8, and S100A7) relative to the reference gene, RPP30, from blood for 30 normal, healthy females was determined. The cutoff for a positive sample was set at 3 s.d. above the mean (cutoff threshold, 1.148) (Supplementary Fig. 6b). The specific primers used for ddPCR and qPCR of genomic DNA and cfDNA are summarized in Supplementary Table 15.

Immunoblotting.

All immunoblots shown in this paper represent at least two independent experiments. In brief, cells were washed with PBS and lysed in RIPA buffer supplemented with protease and phosphatase inhibitors and further sonicated using an XL2000 Microson Ultrasonice Processor (Misonix). Equal amounts of protein extract were separated on SDS–polyacrylamide gels and transferred to polyvinyl difluoride (PVDF) membranes. Membranes were further blocked with 5% milk or BSA and then probed with the following antibodies: anti-IRAK1 was purchased from Santa Cruz Biotechnology (Santa Cruz, CA; cat. no. sc-7883), anti-phospho-IRAK1 (Thr209) was purchased from Assay Biotech (Sunnyvale, CA; cat. no. A1074), anti-JAK2 (Danvers, MA; cat. no. 3230) and anti-phospho-JAK2 (Tyr221) (Danvers, MA; cat. no. 3774) were purchased from Cell Signaling Technology, and anti-actin was purchased from Sigma-Aldrich (cat. no. A5441). For protein extraction of xenograft tumors, snap-frozen samples were resuspended in RIPA buffer supplemented with protease and phosphatase inhibitors and homogenized with Qiagen TissueLyzer II according to the manufacturer's instructions, followed by immunoblotting as described above.

Quantitative RT–PCR analyses.

Total RNA was isolated using TRIzol reagent (Life Technologies, Carlsbad, CA; cat. no. 15596026) and purified with Direct-zol RNA MiniPrep (Zymo Research, Irvine, CA; cat. no. R2050). RT–PCR and qPCR assays were performed using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Carlsbad, CA; cat. no. 4368813) and the KAPA SYBR Fast qPCR kit (KAPA Biosystems, Wilmington, MA; KK4601). For quantification of mRNA levels, GAPDH level was used as an internal control. All reactions were analyzed on an Applied Biosystems PRISM 7500 Fast Real-Time PCR system in 96-well plate format. For RNA extraction of xenograft tumors, snap-frozen samples were resuspended in TRIzol reagent and homogenized with Qiagen TissueLyzer II as described by the manufacturer, followed by purification with Direct-zol RNA MiniPrep. The specific primers used for RT–PCR are summarized in Supplementary Table 15.

IHC staining for clinical samples.

Paraffin-embedded sections of primary and recurrent tumors were obtained from Tan Tock Seng Hospital, Singapore. Staining and image analyses of clinical samples were performed by the Histopathology Department of the Institute of Molecular and Cell Biology, A*STAR, Singapore. Briefly, paraffin-embedded tissue sections were deparaffinized and rehydrated, and antigens were retrieved using proteinase K solution; sections were then incubated in 3% H2O2 at room temperature, to block endogenous peroxidase. Slides were incubated in phosphorylated IRAK1 (Ser376) antibody from Genetex (cat. no. GTX60149, 1:500 dilution) or S100A8 (also known as MRP8) antibody from Abcam (cat. no. ab92331, 1:500 dilution) overnight, followed by incubation for 30 min with anti-mouse labeled polymer (cat. no. K4000, Dako). Specificity of the immunostaining was determined by the inclusion of isotype-specific IgG as a negative control. The detection system was DAB+ substrate–chromogen solution (Dako). The sections were counterstained with hematoxylin. Slides were scanned at 20× magnification using a Leica SCN400 slide scanner (Leica Microsystems, Germany). Images were exported to the Slidepath Digital Image Hub (Leica Microsystems) for viewing. The total cellular H-score was then further normalized and expressed as a Z-score after conversion with the following formula:Z = (total cellular H-score of each tumor – mean H-score)/s.d. of all tumors. The scoring process was carried out in a blinded fashion. Scanning and image analysis were performed by the Advanced Molecular Pathology Laboratory, Institute of Molecular and Cell Biology (IMCB), Singapore.

Statistical analyses.

All statistical analyses were performed with GraphPad Prism version 6.0 or IBM SPSS Statistics version 20. The Kruskal–Wallis test was used to test the correlation of CNV with mRNA gene expression in the TCGA data set, and the Mann–Whitney test was used to test the difference in Time to Death of subjects with amplified and non-amplified 1q21 and 8q. The McNemar test was used to test the change in the number of subjects with 1q21.3 amplification in matching primary tumor and recurrent tumor samples using GraphPad Prism. The chi-squared independence test used to find the relationship of CNV in primary tumor and relapse samples was done with SPSS. Correlation analyses were done using Spearman's correlation coefficients in GraphPad Prism. ROC curves were generated and AUC values were calculated using MedCalc statistical software. All P values were two-sided, and significance was accepted at P < 0.05 unless otherwise stated. In all box plots, median value (center line), minimum, 25th percentile, 75th percentile, and maximum value lines (box lines) are presented. For Kaplan–Meier survival analysis of the patient cohort, ddPCR analysis was carried out in a blinded fashion without knowledge of patients' clinical outcomes. For animal experiments, no blinding was done for treatment and analyses.

Study approval.

Human tissue and blood samples were provided by TTSH and National University Cancer Institute, Singapore, the National Cancer Center of Singapore, the John Wayne Cancer Institute (USA), and Odense University Hospital (Denmark). Studies with these samples were approved by the corresponding institutional or regional review board of each institution. Informed written consent was obtained from each individual who agreed to provide tissue and/or blood samples for research purposes. This work was performed in concordance with Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK) guidelines47.

Data availability.

The RNA-seq data for the primary breast cancer tumors and tumorspheres in Figure 1a were deposited into the GEO database with accession code GSE84054. Uncropped western blots, along with molecular weight standards, are included in the Supplementary Data. A Life Sciences Reporting Summary is available.

URLs.

Kaplan–Meier plotter, http://www.kmplot.com/.

Additional information

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Accession codes

Primary accessions

Gene Expression Omnibus

References

  1. Al-Hajj, M., Wicha, M.S., Benito-Hernandez, A., Morrison, S.J. & Clarke, M.F. Prospective identification of tumorigenic breast cancer cells. Proc. Natl. Acad. Sci. USA 100, 39833988 (2003).
  2. Nguyen, L.V., Vanner, R., Dirks, P. & Eaves, C.J. Cancer stem cells: an evolving concept. Nat. Rev. Cancer 12, 133143 (2012).
  3. Mitra, A., Mishra, L. & Li, S. EMT, CTCs and CSCs in tumor relapse and drug-resistance. Oncotarget 6, 1069710711 (2015).
  4. Ginestier, C. et al. ALDH1 is a marker of normal and malignant human mammary stem cells and a predictor of poor clinical outcome. Cell Stem Cell 1, 555567 (2007).
  5. Marcato, P. et al. Aldehyde dehydrogenase 1A3 influences breast cancer progression via differential retinoic acid signaling. Mol. Oncol. 9, 1731 (2015).
  6. Marcato, P. et al. Aldehyde dehydrogenase activity of breast cancer stem cells is primarily due to isoform ALDH1A3 and its expression is predictive of metastasis. Stem Cells 29, 3245 (2011).
  7. Cerami, E. et al. The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data. Cancer Discov. 2, 401404 (2012).
  8. Ciriello, G. et al. Comprehensive molecular portraits of invasive lobular breast cancer. Cell 163, 506519 (2015).
  9. Gao, J. et al. Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal. Sci. Signal. 6, pl1 (2013).
  10. Gyoőrffy, B., Surowiak, P., Budczies, J. & Lánczky, A. Online survival analysis software to assess the prognostic value of biomarkers using transcriptomic data in non-small-cell lung cancer. PLoS One 8, e82241 (2013).
  11. Pereira, B. et al. The somatic mutation profiles of 2,433 breast cancers refines their genomic and transcriptomic landscapes. Nat. Commun. 7, 11479 (2016).
  12. Curtis, C. et al. The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature. 486, 346352 (2012).
  13. Carreira, S. et al. Tumor clone dynamics in lethal prostate cancer. Sci. Transl. Med. 6, 254ra125 (2014).
  14. Gevensleben, H. et al. Noninvasive detection of HER2 amplification with plasma DNA digital PCR. Clin. Cancer Res. 19, 32763284 (2013).
  15. Miotke, L., Lau, B.T., Rumma, R.T. & Ji, H.P. High sensitivity detection and quantitation of DNA copy number and single nucleotide variants with single color droplet digital PCR. Anal. Chem. 86, 26182624 (2014).
  16. Pearson, A. et al. High-level clonal FGFR amplification and response to FGFR inhibition in a translational clinical trial. Cancer Discov. 6, 838851 (2016).
  17. Romanel, A. et al. Plasma AR and abiraterone-resistant prostate cancer. Sci. Transl. Med. 7, 312re10 (2015).
  18. Shoda, K. et al. Monitoring the HER2 copy number status in circulating tumor DNA by droplet digital PCR in patients with gastric cancer. Gastric Cancer 20, 126135 (2016).
  19. Alix-Panabières, C. & Pantel, K. Clinical applications of circulating tumor cells and circulating tumor DNA as liquid biopsy. Cancer Discov. 6, 479491 (2016).
  20. Chan, D. et al. Phase II study of gemcitabine and carboplatin in metastatic breast cancers with prior exposure to anthracyclines and taxanes. Invest. New Drugs 28, 859865 (2010).
  21. Wang, X. et al. S100A14, a mediator of epithelial–mesenchymal transition, regulates proliferation, migration and invasion of human cervical cancer cells. Am. J. Cancer Res. 5, 14841495 (2015).
  22. Tanaka, M. et al. Co-expression of S100A14 and S100A16 correlates with a poor prognosis in human breast cancer and promotes cancer cell invasion. BMC Cancer 15, 53 (2015).
  23. Bresnick, A.R., Weber, D.J. & Zimmer, D.B. S100 proteins in cancer. Nat. Rev. Cancer 15, 96109 (2015).
  24. Zhang, Q. et al. Downregulation of 425G>A variant of calcium-binding protein S100A14 associated with poor differentiation and prognosis in gastric cancer. J. Cancer Res. Clin. Oncol. 141, 691703 (2015).
  25. Cho, H. et al. The role of S100A14 in epithelial ovarian tumors. Oncotarget 5, 34823496 (2014).
  26. Wee, Z.N. et al. IRAK1 is a therapeutic target that drives breast cancer metastasis and resistance to paclitaxel. Nat. Commun. 6, 8746 (2015).
  27. Beauverd, Y., McLornan, D.P. & Harrison, C.N. Pacritinib: a new agent for the management of myelofibrosis?. Expert Opin. Pharmacother. 16, 23812390 (2015).
  28. Singer, J.W. et al. Comprehensive kinase profile of pacritinib, a nonmyelosuppressive Janus kinase 2 inhibitor. J. Exp. Pharmacol. 16, 1119 (2016).
  29. Phase 1/2 study of pacritinib, a next generation JAK2/FLT3 inhibitor, in myelofibrosis or other myeloid malignancies. Hematol. Oncol. 9, 137 (2016).
  30. Crowley, E., Di Nicolantonio, F., Loupakis, F. & Bardelli, A. Liquid biopsy: monitoring cancer-genetics in the blood. Nat. Rev. Clin. Oncol. 10, 472484 (2013).
  31. Siravegna, G. & Bardelli, A. Genotyping cell-free tumor DNA in the blood to detect residual disease and drug resistance. Genome Biol. 15, 449 (2014).
  32. Dawson, S.J. et al. Analysis of circulating tumor DNA to monitor metastatic breast cancer. N. Engl. J. Med. 368, 11991209 (2013).
  33. Thress, K.S. et al. Acquired EGFR C797S mutation mediates resistance to AZD9291 in non–small cell lung cancer harboring EGFR T790M. Nat. Med. 21, 560562 (2015).
  34. Siravegna, G. et al. Clonal evolution and resistance to EGFR blockade in the blood of colorectal cancer patients. Nat. Med. 21, 795801 (2015).
  35. Garcia-Murillas, I. et al. Mutation tracking in circulating tumor DNA predicts relapse in early breast cancer. Sci. Transl. Med. 7, 302ra133 (2015).
  36. Cancer Genome Atlas Network. Comprehensive molecular portraits of human breast tumours. Nature 490, 6170 (2012).
  37. Marusyk, A., Almendro, V. & Polyak, K. Intra-tumour heterogeneity: a looking glass for cancer? Nat. Rev. Cancer 12, 323334 (2012).
  38. Yates, L.R. et al. Subclonal diversification of primary breast cancer revealed by multiregion sequencing. Nat. Med. 21, 751759 (2015).
  39. Forshew, T. et al. Noninvasive identification and monitoring of cancer mutations by targeted deep sequencing of plasma DNA. Sci. Transl. Med. 4, 136ra68 (2012).
  40. Murtaza, M. et al. Non-invasive analysis of acquired resistance to cancer therapy by sequencing of plasma DNA. Nature 497, 108112 (2013).
  41. Beaver, J.A. et al. Detection of cancer DNA in plasma of patients with early-stage breast cancer. Clin. Cancer Res. 20, 26432650 (2014).
  42. Duffy, M.J., Shering, S., Sherry, F., McDermott, E. & O'Higgins, N. CA 15-3: a prognostic marker in breast cancer. Int. J. Biol. Markers 15, 330333 (2000).
  43. Harris, L. et al. American Society of Clinical Oncology 2007 update of recommendations for the use of tumor markers in breast cancer. J. Clin. Oncol. 25, 52875312 (2007).
  44. Sturgeon, C. Practice guidelines for tumor marker use in the clinic. Clin. Chem. 48, 11511159 (2002).
  45. Mermel, C.H. et al. GISTIC2.0 facilitates sensitive and confident localization of the targets of focal somatic copy-number alteration in human cancers. Genome Biol. 12, R41 (2011).
  46. Ono, S. et al. A direct plasma assay of circulating microRNA-210 of hypoxia can identify early systemic metastasis recurrence in melanoma patients. Oncotarget 6, 70537064 (2015).
  47. McShane, L.M. et al. Reporting recommendations for tumor marker prognostic studies (REMARK). J. Natl. Cancer Inst. 97, 11801184 (2005).

Download references

Acknowledgments

This work was supported by the Agency for Science, Technology and Research of Singapore (A*STAR), the Margie Petersen Breast Cancer Program, and the Association of Breast Cancer (D.S.B.H.), Singapore Ministry of Health's National Medical Research Council (NMRC) Clinician Scientist Individual Research Grants (NMRC/CIRG/1464/2016 and NMRC/CG/017/2013 to E.Y.T.; NMRC/CSA/015/2009 and NMRC/CSA-SI/0004/2015 to S.C.L.), and the Danish Cancer Society (R99-A6362 and R146-A9164 to H.J.D.). This work was also supported by an A*STAR Graduate Academy (A*GA) SINGA (Singapore International Graduate Award) scholarship to G.O. We thank the Histopathology Department from the Institute of Molecular and Cell Biology, A*STAR, for their service in IHC staining and analysis.

Author information

  1. These authors contributed equally to this work.

    • Jian Yuan Goh,
    • Min Feng &
    • Wenyu Wang

Affiliations

  1. Genome Institute of Singapore, Agency for Science, Technology and Research (A*STAR), Biopolis, Singapore.

    • Jian Yuan Goh,
    • Min Feng,
    • Wenyu Wang,
    • Gokce Oguz,
    • Siti Maryam J M Yatim,
    • Puay Leng Lee,
    • Yi Bao,
    • Wai Leong Tam,
    • Suman Sarma,
    • Alexander Lezhava &
    • Qiang Yu
  2. Department of Physiology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore.

    • Gokce Oguz &
    • Qiang Yu
  3. Cytogenetics Laboratory, Department of Pathology, Singapore General Hospital, Singapore.

    • Tse Hui Lim &
    • Alvin S T Lim
  4. Cancer Research Institute and School of Pharmacy, Jinan University, Guangzhou, China.

    • Panpan Wang &
    • Qiang Yu
  5. Cancer Science Institute of Singapore, National University of Singapore, Singapore.

    • Wai Leong Tam &
    • Soo Chin Lee
  6. Department of Oncology, Odense University Hospital, Odense, Denmark.

    • Annette R Kodahl &
    • Henrik J Ditzel
  7. Department of Cancer and Inflammation Research, Institute of Molecular Medicine, University of Southern Denmark, Odense, Denmark.

    • Maria B Lyng &
    • Henrik J Ditzel
  8. Department of Translational Molecular Medicine, John Wayne Cancer Institute, Santa Monica, California, USA.

    • Selena Y Lin &
    • Dave S B Hoon
  9. Division of Medical Oncology, National Cancer Centre Singapore, Singapore.

    • Yoon Sim Yap
  10. Department of Haematology–Oncology, National University Cancer Institute, National University Health System, Singapore.

    • Soo Chin Lee
  11. Department of General Surgery, Tan Tock Seng Hospital, Singapore.

    • Ern Yu Tan
  12. Institute of Molecular and Cellular Biology, A*STAR, Biopolis, Singapore.

    • Ern Yu Tan
  13. Cancer and Stem Cell Biology, Duke-NUS Medical School, Singapore.

    • Qiang Yu

Contributions

Q.Y. supervised the project and contributed to the design and interpretation of all experiments. J.Y.G. contributed to the design, conduct, and interpretation of all experiments. M.F. performed tumor sample preparation, genomic DNA and cfDNA extraction, RNA-seq, gene knockdown, and tumorsphere assays. Y.B. performed overexpression and tumorsphere assays. W.W. and P.L.L. performed western blot analyses. G.O. performed bioinformatics and statistical analyses. W.W., M.F., and S.M.J.M.Y. performed in vivo experiments. T.H.L. and A.S.T.L. performed DNA FISH analysis. P.W., A.R.K., M.B.L., and S.Y.L. contributed to collection of patient blood samples and clinical information. A.L. and S.S. contributed digital PCR analyses. W.L.T., Y.S.Y., D.S.B.H., H.J.D., S.C.L., and E.Y.T. provided crucial reagents and patient samples and contributed to clinical data analyses and interpretation. J.Y.G. and Q.Y. wrote the manuscript with input from all co-authors.

Competing financial interests

The authors declare no competing financial interests.

Corresponding authors

Correspondence to:

Author details

Supplementary information

PDF files

  1. Supplementary Text and Figures (1,634 KB)

    Supplementary Figures 1–11 and Supplementary Tables 1–15.

  2. Life Sciences Reporting Summary (133 KB)
  3. Supplementary Data (772 KB)

    Uncropped Immunoblots.

Additional data