Introduction

Circulating tumor cells (CTCs) have been isolated from metastatic colorectal cancer (mCRC) patients,1, 2 and CTC count has been associated with survival.3, 4, 5 However, circulating cancer cells are rare within the bloodstream and exhibit a phenotypic diversity, which may impede their reliable detection. Using the CellSearch assay, CTCs are detected in only 30%–50% of mCRC patients.1, 2, 3, 4, 5 CTCs can undergo the epithelial–mesenchymal transition (EMT)6, 7, 8, 9 and assume a stem cell-like6 phenotype. Owing to such heterogeneity, current CTC isolation methods that emphasize epithelial traits may not capture their plasticity through activation of EMT and stem cell pathways.10 Refining the molecular characterization of CTCs may improve the clinical utility of CTC detection and provide insight into mechanisms directing the metastatic process as well as therapeutic resistance. To this end, several CTC markers3, 4, 5, 11, 12, 13, 14, 15, 16, 17 have been explored in an attempt to improve diagnostic yield and predict outcomes, but the optimal CTC gene expression signature remains to be defined.

We recently validated a technique combining immunomagnetic enrichment with quantitative real-time PCR (qRT-PCR) to isolate EpCAM+/CD45− CTCs, which express CK20 and/or survivin mRNA in mCRC patients receiving standard and experimental chemotherapy regimens.18 However, within the cohort of patients with CK20 and/or survivin-expressing CTCs, there was still considerable variability with regards to survival. Furthermore, we hypothesized that a proportion of epithelial–mesenchymal transitioned CTCs may not express cytokeratin or survivin and, therefore, evade detection. The PI3Kα/Akt-2/mTOR pathway helps regulate the EMT transition by repressing cell adhesion proteins, therefore facilitating cancer cell migration as well as survival and apoptosis.19 In preclinical studies, Akt-2 has been shown to be critical to the development of colorectal metastasis.20 Clinically, CTC expression of PI3Kα, Akt-2, Twist1 has been significantly associated with outcomes in cancer patients.8 Similarly, circulating cancer cells may acquire a stem cell phenotype,19 marked by aldehyde dehydrogenase 1 (ALDH1) upregulation, and CTC ALDH1 expression has demonstrated prognostic value in patients undergoing chemotherapy.8, 21

Here, we aim to refine the utility of our approach by measuring the mRNA expression of EMT (PI3Kα, Akt-2, Twist1) and stem cell (ALDH1) markers in circulating cancer cells. We found a proportion of CK20− and survivin− cells express PI3Kα, Akt-2 or ALDH1, and that this gene expression panel may prognostically stratify subgroups of mCRC patients receiving chemotherapy.

Materials and methods

Patient population and study design

Patients consented for peripheral blood collection and received standard FDA-approved therapies (including fluoropyrimidines, oxaliplatin, irinotecan, bevacizumab, cetuximab, panitumumab, regorafenib), or received experimental agents being tested in three phase I or II clinical trials examining 5-FU plus brivanib (NCT01046864), PRI-724 (NCT01302405) and celecoxib plus EPO906 (NCT00159484). Patients were enrolled at the Norris Comprehensive Cancer Center-University of Southern California or the Los Angeles County-University of Southern California Medical Center, between June 2009 and April 2014.18 The Institutional Review Board at the University of Southern California approved the study. Twenty healthy blood donors (age ⩾18 years), who had no known medical illness or history of malignant disease, served as control subjects. All the CTC studies were performed without knowledge of patients’ clinical status.

Sample collection and tumor cell enrichment

The sample collection and tumor cell enrichment were performed as previously described.18 An 8 ml blood sample was drawn from each patient into one Vacutainer CPT Tube with Sodium Citrate (BD, Franklin Lakes, NJ, USA). Negative immunomagnetic selection using anti-CD45 specific antibodies (Dynabeads M-450 CD45 pan Leukocyte, Invitrogen, Waltham, MA, USA) was performed to enrich for tumor cells following the manufacturer’s instructions. The CD45-negative (CD45−) supernatant was transferred to 15 ml tubes for immune separation using Dynabeads (Dynabeads Epithelial Enrich, #161.02, Invitrogen). Using Dynabeads coated with a monoclonal antibody towards human EpCAM, tumor cell selection was performed following the manufacturer’s instructions.

qRT-PCR and multiplex-PCR

CK20 and survivin mRNA expression levels were analyzed by the iTaq Fast SYBR Green Supermix (Bio-Rad #172-5101; Bio-Rad, Hercules, CA, USA) and an Applied Biosystems 7500 PCR Detection System (Applied Biosystems, Foster City, CA, USA). ALDH1, Akt-2 and PI3Kα mRNA expression levels were analyzed using the HotStarTaq Master Mix (Qiagen, Qiagen GmbH, Germany, #203443) and a thermocycler, and determined by using the Agilent 2100 Bioanalyzer on a DNA LabChip (Agilent Technologies, Santa Clara, CA, USA).

Analysis of mRNA expression

Primers for EMT markers and stem cell markers were provided by AdnaGen (AdnaGen, Langenhagen, Germany). The analysis of tumor cell-derived mRNA was performed by qRT-PCR for the following transcripts: PI3Kα, Akt-2, Twist1 and ALDH1. Thermal profiles were used as per the recommendations of the supplier (AdnaGen). The AdnaTest EMT-2/StemCell Detect Kit (AdnaGen), containing oligo(dT)25-coated beads, was used to isolate mRNA from tumor cells. The PrimerMix StemCell was used to amplify ALDH1, and the PrimerMix EMT-2 was used to amplify three EMT-related genes (Akt-2, PI3Kα, Twist1) and one control gene (Actin). Visualization of PCR fragments was carried out with the Agilent 2100 Bioanalyzer, with a limit of detection of ⩾0.15 ng μl−1.

Statistical analysis

The mRNA levels of CTC ALDH1, PI3K and Akt-2 were categorized into low and high values at the cutoff values provided by AdnaGen. The associations between the three markers and progression-free survival (PFS) and overall survival (OS) were analyzed by Kaplan–Meier curves and log-rank test in the univariable analysis and Cox regression model in the multivariable model, adjusting for baseline characteristics and treatment administered after CTC collection that were associated with PFS or OS at a significance level of 0.1.

Results

Patient and tumor characteristics

Patient and tumor characteristics are described in Table 1. Among 78 mCRC patients, the median duration of follow-up was 23.5 months (range: 2.0, 38.6 months), median PFS was 3.1 months (95% confidence interval: 2.1, 5.0 months) and median OS was 10.4 months (95% confidence interval: 8.9, 17.2 months). Patients received a median of three prior lines of therapy (range 0–5). Most patients had received fluoropyrimidines (97.4%), oxaliplatin (89.7%), irinotecan (75.6%) and bevacizumab (88.5%) before CTC collection. After CTC collection, 62.8% of patients received experimental therapies on clinical trials. There was an even distribution of primary tumor site across all the patients, and most did not have liver-limited metastases.

Table 1 Metastatic colorectal cancer patient baseline characteristics (n=78)
Full size table

CTC ALDH1, PI3Kα and Akt-2 expression in mCRC patients and healthy donors

We first compared biomarker expression in 78 mCRC patients and 20 healthy donors. As shown in Figure 1a, the expression of ALDH1 (69 versus 0%), PI3Kα (4 versus 0%) and Akt-2 (49 versus 0%) was significantly higher in the mCRC patients compared with the healthy donors (P<0.001 for all comparisons). Cutoff values for detection were provided by AdnaGen: 0.25 ng μl−1 for PI3Kα, Akt-2, Twist1 and 0.15 ng μl−1 for ALDH1. On the basis of these cutoff points, the sensitivity and specificity for detecting CTCs by the presence of either ALDH1, PI3Kα or Akt-2 expression was 83% and 100%, respectively. There was no significant Twist1 gene expression detected in mCRC patients or in healthy donors (data not shown).

Figure 1
figure 1

(a) CTC ALDH1, PI3kα, Akt-2 gene expression levels in 78 mCRC patients. (b) Comparison of EMT markers, ALDH1 in mCRC patients. CTC, circulating tumor cell; EMT, epithelial–mesenchymal transition; mCRC, metastatic colorectal cancer.

PowerPoint slide

Full size image

CTC ALDH1, PI3Kα and Akt-2 expression in mCRC patients

Next, we compared the expression of ALDH1, PI3Kα and Akt-2 in patients with and without CK20/survivin-expressing CTCs using the AdnaGen cutoff values. Seventy-eight percent of patients (n=61) had CK20 and/or survivin-expressing CTCs. Among patients without CK20 or survivin-expressing CTCs (n=17), 29%, 23% and 29% had expression of ALDH1, PI3Kα and/or Akt-2 markers, respectively; 55% had expression of any of three markers. Among patients without CK20-expressing CTCs (n=21), 14, 33, 29 and 29% had expression of survivin, ALDH1, PI3Kα and Akt-2, respectively; 62% had expression of any of four markers. The expression of ALDH1 (80 versus 29%), PI3Kα (50 versus 23%) and Akt-2 (54 versus 29%) was significantly higher in patients with CK20 and/or survivin-expressing CTCs relative to those with CTCs negative for both CK20 and survivin expression (P<0.05 for all comparisons; Figure 1b).

Prognostic utility of CTC PI3Kα, Akt-2 and ALDH1 expression

We examined the prognostic relevance of CTC ALDH1, PI3Kα and Akt-2 expression in the mCRC patients with and without CK20/survivin-expressing CTCs.

In univariate analysis, patients with positive CTC Akt-2 expression had a significantly shorter PFS (3.0 versus 4.0 months, hazard ratio (HR)=1.63, log-rank P=0.034) compared with those without CTC Akt-2 expression (Table 2, Figure 2a). This association maintained significance in a multivariate model accounting for baseline performance status and therapy received (standard versus experimental) after CTC measurement (HR=1.70, log-rank P=0.041). There was no significant association between CTC Akt-2 expression and OS (Table 3). Patients with positive CTC ALDH1 expression had a significantly shorter OS (10.0 versus 38.6 months, HR=2.04, log-rank P=0.021; Figure 2b and Table 3) and trend towards inferior PFS (3.0 versus 4.5 months, HR=1.50, log-rank P=0.079; Table 2) compared with CTC ALDH1-negative patients in univariate analysis. However, CTC ALDH1 expression was not significantly associated with survival in multivariate analysis. Patients with positive CTC PI3Kα expression trended towards a shorter PFS, but this difference was not statistically significant (2.9 versus 3.3 months, HR=1.46, log-rank P=0.10; Table 2) and did not withstand correction in multivariable analysis. There was no significant association between CTC PI3Kα expression and OS (Table 3).

Table 2 CTC marker expression and progression-free survival (PFS) in mCRC patients
Full size table
Figure 2
figure 2

Kaplan–Meier cumulative probability of survival based on (a) CTC Akt-2 and (b) CTC ALDH1 gene expression. CTC, circulating tumor cell; OS, overall survival; PFS, progression-free survival.

PowerPoint slide

Full size image
Table 3 CTC marker expression and overall survival (OS) in mCRC patients
Full size table

In univariate analysis, patients with any positive CTC marker (ALDH1, PI3Kα or Akt-2) had a significantly inferior median PFS (3.0 versus 7.7 months, HR=1.88, log-rank P=0.015; Table 2) and OS (10.0 versus 26.8+ months, HR=2.25, log-rank P=0.050; Table 3) compared with those with no elevated CTC markers (Figure 3). These associations, however, were not significant in multivariate analysis.

Figure 3
figure 3

Kaplan–Meier cumulative probability of (a) progression-free survival (PFS) and (b) overall survival (OS) based on circulating tumor cell (CTC) ALDH1, PI3kα, Akt-2 gene expression. EMT, epithelial–mesenchymal transition.

PowerPoint slide

Full size image

Discussion

We evaluated the clinical relevance of PI3Kα, Akt-2 and ALDH1 gene expression in the colorectal circulating cancer cells. To our knowledge, we are the first to report a significant association between CTC Akt-2 expression and PFS in mCRC patients receiving different standard and experimental therapeutics. Moreover, we found that patients whose CTCs do not express CK20 or survivin may express PI3Kα, Akt-2 and/or ALDH1 and that this gene expression signature may serve as a clinically useful prognostic marker.

Although the EMT program is thought to be integral to the ability of CTCs to form metastases, CTCs may also revert back to an epithelial state through the mesenchymal to epithelial transition.22 The position along which a CTC resides within the epithelial–mesenchymal axis is malleable, influenced by disease progression and cancer-directed therapy.10 Furthermore, it has been postulated that a subpopulation of CTCs exists in a semi-mesenchymal state, exhibiting both epithelial and mesenchymal traits.23

On the basis of these observations, we hypothesized that incorporating both epithelial and mesenchymal genes into a multimarker model would optimize the prognostic value of CTC detection. In our cohort, 62% of patients without CK20-expressing CTCs were found to have expression of survivin, PI3Kα, Akt-2 or ALDH1. Over half of patients whose CTCs did not express CK20 or survivin expressed PI3Kα, Akt-2 or ALDH1.

PI3Kα and Akt-2 expression have been implicated in EMT and stem cell renewal24 and have been reliably detected in CTCs.8 In cancer cell lines, PI3Kα-Akt-2 signaling has been shown to promote cell motility, invasiveness and metastatic potential,25 partly through its interaction with Twist1 and downstream effectors. In a landmark study by Ryachahou et al.,20 the Akt-2 isoform, in particular, was found to have sustained overexpression across all stages of CRC. In Akt-2 knockout CRC mouse models, the formation of liver tumors was greatly repressed, further supporting an essential role for Akt-2, in conjunction with PI3K, in the development and growth of CRC metastases.20

Clinically, evidence from breast cancer patients supports the use of PI3Kα, Akt-2 and Twist1 as sensitive markers for CTC identification,8 as well as predictors of response to endocrine and cytotoxic chemotherapy.8 In our study, 47%–51% of mCRC (n=78) and 52%–57% patients (n=70) with CK20/survivin-expressing CTCs also overexpressed PI3Kα and/or Akt-2, respectively, which is comparable to findings in other metastatic cancer patients.8 CTCs also frequently overexpress stem cell markers24 and acquire many of their stem cell properties during the EMT, allowing them to re-seed distant organs. In CRC patients, the stem cell marker, ALDH1, has been associated with EMT signaling.26 In our study, 70% of mCRC patients (n=78) and 80% patients (n=70) with CK20/survivin-expressing CTCs had ALDH1 overexpression.

The evidence for a prognostic effect of CTC EMT-related and stem-like markers has been inconsistent, with some8, 27, 28 but not all studies29, 30 demonstrating a significant relationship with disease recurrence or survival. The prognostic utility of adding stem cell (for example, CD133) to epithelial markers in advanced CRC patients was demonstrated by Linuma et al.,17 though this study did not use a negative enrichment step, which may have limited assay sensitivity. In our study, patients whose CTCs expressed at least one elevated marker (ALDH1, PI3Kα, Akt-2) had inferior survival, suggesting that an increasingly mesenchymal or stem-like CTC phenotype may serve as a surrogate for tumor invasiveness and refractory disease to affect survival.

Our study has several limitations, the first of which is the small sample size of our retrospective cohort. Furthermore, there was heterogeneity with respect to the type of treatment received and the time point at which blood was drawn from patients, which may have limited the power of our analysis. Notably, we did not detect Twist1 expression in any of our patients, and this may be due to Twist1 downregulation in CD45-depleted cells. However, the expression of PI3Kα, Akt-2, or ALDH1 was able to distinguish cancer patients from healthy controls and patients with or without CK20/survivin-expressing CTCs with high sensitivity and specificity. Importantly, a proportion of CTCs may lose their EpCAM expression during the EMT30, 31 and may, therefore, not be detected by our method.

In summary, we examined CTC ALDH1, PI3Kα and Akt-2 mRNA expression on CTC and found CTC Akt-2 expression to predict PFS in mCRC patients receiving different standard and experimental regimens. Larger prospective studies are needed to validate this gene expression model and assess its utility to predict disease progression and response to cytotoxic and targeted therapeutics.