3.1. CTCs enumeration by CellSearch
CTCs were isolated and enumerated with the CellSearch® system in 20 mCRPC patients in three different time points: before chemotherapy (Visit 1, V1), after one cycle of docetaxel therapy (Visit 2, V2) and at radiological progression determined by CT-Scan (Visit 3, V3). At V1, 1 sample (#UM164) was not evaluated due to insufficient amount of blood. Among the 19 patients analysed at V1, 13 patients (68.5%) had ≥ 5 CTCs per 7.5 mL blood and 6 patients (31.5%) had 1-4 CTCs. After 1 cycle of docetaxel (V2), we observed a decrease in the CTCs counts in 84.2 % of the patients (p=0.002). Thus, 6 patients (30%) had ≥ 5 CTCs and 3 patients (15%) had 1-4 CTCs, while in 11 patients (55%) no CTCs were identified. At clinical progression (V3), among the 13 patients analysed, we observed an increase in the CTCs counts when compared with V2 (p=0.006): 7 patients (54%) showed an increase in the CTCs numbers and had ≥ 5 CTCs; and 3 out of 7 switched from good prognostic to the bad prognostic group; 3 patients (23%) had 1-4 CTCs and no CTCs were detected in 3 patients (23%) (Figure 1).
We identified significant differences between samples with < 5 and ≥ 5 CTCs/7.5 mL of blood being high CTCs account data associated with high PSA levels and receiving ≤ 2 previous hormone treatments. There is an imbalance in the number of cases previously treated with ≥ 2 hormone-deprivation therapy that may be biasing the result. Other clinical data as age, Gleason score, the type of castration, LDH or ALK-P levels did not show a statistical difference between having < 5 CTCs or not (Table 1).
Table 1. Associations between CTCs number and the clinical-pathologic characteristics of the cohort of patients at the diagnostic of metastasis. P-values were calculated using the Fisher exact test. For LDH, ALK-P and PSA levels, the median value of the analysed cohort was considered as a threshold for low or high category (see Table S1).
Next, we analysed if CTCs count was able to predict patient´s outcome as previously described in CRPC. To address this aim, a survival analysis was performed to evaluate OS and PFS in relation to CellSearch® data, considering the previously defined good prognosis (≥ 5 CTCs) or bad prognosis (< 5 CTCs) cut off [22]. Patients with ≥ 5 CTCs tended to a shorter OS and PFS both at V1 (PFS, p=0.16; OS, p= 0.6, log-rank test), V2 (PFS, p=0.2; OS, p= 0.05, log-rank test) and V3 (OS, p= 0.09, log-rank test) compared with patients with < 5 CTCs, however statistical significance was not reached (Figure S1).
Taken together, these analyses indicate that the number of EpCAM+ CTCs is reduced after chemotherapy and it increases again at clinical progression but it did not predict outcome in this patient cohort.
3.2. Gene expression profile in the CTCs enriched fraction
In parallel to the CellSearch® enumeration, we performed a gene expression analysis on CTCs isolated with a negative enrichment approach in 20 patients (n, V1=20, V2=20, V3=13). We analysed the expression of a gene panel in the CTCs enriched fraction in relation to the expression found in the paired PBMCs’ fraction.
CTC-positive samples were defined as those with at least one epithelial (CDH1, EpCAM or KRT19), one mesenchymal (SNAI1, VIM or ZEB1) or one stem (ALDH1A1 or PROM1) cell marker with higher expression regarding to the paired PBMCs expression (ΔΔct ≥ 1.5). Only one patient (#UM30) was considered CTC-negative both at V1 and V2 following this criterion which agrees with CellSearch® data (V1=3 CTCs; V2=0 CTCs). CTC-positive samples (n=51) exhibited different frequency of expression for the epithelial (CDH1: 88.2%, EpCAM: 45.1%, KTR19: 45.1%), mesenchymal (SNAI1: 58.8%, VIM: 45.1% or ZEB1: 41.1%) and stem cell markers (ALDH1A1: 41.1%, PROM1: 45.1%). Interestingly, the CTC-positive samples showed a very heterogeneous pattern of expression, displaying a hybrid epithelial-mesenchymal phenotype in most of the analysed samples (70.6%) while 21.6% were exclusively epithelial and 7.8% expressed only mesenchymal markers.
Regarding the other analysed genes, we observed variable expression among the samples (in percentages of CTCs positive samples): for the cell cycle-associated genes, we observed an increased relative expression of CCND1 (56.8%), CDK4 (50.9%), E2F4 (78.4%) and RB1 (68.6%). The PC related gene KLK3 was expressed in more than half the samples (54.9%). BCL2 and the proto-oncogene MYC showed lower relative expression in 90.2% and 92.1% respectively, although MYCL had higher relative expression in 64.7% of them. In addition, CTNNB1 and GDF15 showed also higher expression in 62.7% and 78.4% of the samples, respectively.
Next, we studied the association between gene expression and the number of CTCs (CellSearch® data) for each sample. For that, we consider a high expression as ≥ 1.5 fold change. We identified differential gene expression profiles depending on whether the samples had ≥ 5 or < 5 CTCs. Patients with ≥ 5 CTCs had a higher relative expression of the epithelial marker KRT19 or the prostate-specific marker KLK3 (p<0.0001) while EpCAM higher expression was near significance (p=0.057) (Figure 2) and KLK3, KRT19 and EpCAM relative expression associated also among them (p<0.01). Interestingly, MYCL relative expression was associated with the good prognosis group (< 5 CTCs determined by CellSearch®) (p=0.004) and also with high relative expression of stemness markers (ALDH1A1 and PROM1) and CTNNB1, as well as mesenchymal markers as SNAI1, VIM but inversely to ZEB1. All the other markers were expressed independently of the CTCs counts and, remarkably, CDH1 and GDF15 are widely expressed in a high percentage of the samples (Figure 2).
CDH1, GDF15, RB1, E2F4, MYCL and CTNNB1 genes showed increased relative gene expression in CTCs in most of the patients during the three visits, while in visit 3 the majority of the patients showed EpCAM, CCND1 or CDK4.
We next analysed if the absolute relative expression was different between patients having < 5 CTCs or those having ≥ 5 CTCs. We found that E2F4, KLK3 and KRT19 were statically different between both groups, showing higher values in the poor prognosis one.
Lastly, we studied the relationship between the clinic-pathological characteristics of the patients’ and their gene expression data. We found that CTC-positive samples by gene expression analysis from patients with low Gleason score showed a higher relative expression of CCND1 (p=0.019) and RB1 (p=0.004) at V1. No other association was found with clinical data.
Hence, these results suggest that CTCs in circulation display a hybrid epithelial-mesenchymal phenotype, showing also high expression of cell cycle regulation and PC-related genes. Although there is an association of some of the analysed genes with CTCs numbers, the expression profile of the CTCs did not group the samples by visits or patients except for sporadic cases.
3.3. CTCs gene expression and its prognostic value
In order to determine if the gene expression of CTCs of the analysed markers had prognostic value, we applied the Cox regression for each of the visits independently, considering only CTC-positive samples (by our gene expression approach criterion). We defined 2 groups depending on CTCs expression value relative to the PBMCs as high or low. Overall, the mean follow-up period was 467 days, 19 out of 20 patients eventually developed progression and 15 out of 20 died during the follow-up.
Of note, at V1, those patients with CTCs displaying epithelial characteristics had shorter PFS. Thus, a high expression of KRT19 (p=0.03, log-rank test, 139 vs 180 days) (Figure 3A) predicted a worse prognosis. Furthermore, EpCAM showed the same trend (Table S3). In contrast, patients whose CTCs showed a higher relative expression of stem markers, CTNNB1, or those CTCs with a hybrid phenotype (that includes epithelial, mesenchymal and stem cell characteristics) showed a later disease progression (Figure 3B-D and Table S3). In addition, high MYCL also tended to predict better outcome (PFS, p=0.07; OS, p=0.06, log-rank test) (Table S3).
Table S3. Prognostic value of CTCs gene expression levels in mCRPC patients of the indicated markers. P-values were calculated using the log-rank test.
To evaluate early response to therapy, CTCs were analysed after 1 cycle of docetaxel. Thus, patients with CTCs with high expression of ZEB1 in V2 had both shorter PFS (p=0.03, 119 vs 190 days, log-rank test) and OS (p=0.04, 260 vs 426 days, log-rank test) (Figure 4A-B). In addition, CTCs with high expression of CDK4 were also linked to poorer outcome (PFS, p=0.03, log-rank test)(Table S3).
Lastly, we studied variations in CTCs gene expression at the clinical progression to identify if their molecular profile could predict the patients’ outcome in terms of OS. Indeed, at V3, high relative expression of either KRT19 or KLK3 in the CTCs enriched fraction was able to predict shorter OS (p=0.008, 174 vs 391 days, log-rank test and p=0.02, 183 vs 720 days, log-rank test, respectively)(Figure 5A-B).
Since high MYCL and high SNAI1 expression was associated with low expression of KRT19 (p=0.02, Fisher´s Exact test) at V3, we next explored whether the high expression of both MYCL and SNAI1 was able to predict clinical outcome. We found that patients with high expression of these genes had a mean increase of 219 days in OS (p=0.02, log-rank test) (Figure 5C) and that MYCLhighSNAI1highKRT19low gene signature predicts OS with a superior statistical power (p=0.008, log-rank test).