Characteristics of HCC patients

The baseline characteristics of 48 patients with HCC who underwent TACE are listed in Table 1. The average age was 64.2 ± 7.6 years and 85.4% (n = 41) were men. The main etiology was alcohol-related liver disease (75.0%, n = 36), followed by chronic hepatitis C (39.6%, n = 19), chronic hepatitis B (12.5%, n = 6), and non-alcoholic fatty liver disease (10.4%, n = 5). Most of patients had Child–Pugh class A cirrhosis (83.3%, n = 40). The median model for end-stage liver disease score was 9 (interquartile range (IQR) 7–11). The median serum concentration of alpha-fetoprotein was 10.7 ng/mL (IQR 3.6–40.0). A single HCC nodule was observed in 52.1% (n = 25) of patients. Patients with multinodular disease had two nodules (29.2%, n = 14), three nodules (10.4%, n = 5), or four nodules (8.3%, n = 4). The total tumor diameter (i.e., the sum of the diameter of all nodules) was 3.6 cm (IQR 3.2–5.2), being less than 5 cm in 66.7% (n = 32) of patients. The median diameter of the largest tumor nodule was 3 cm (IQR 2.1–3.8), being less than 3 cm in 41.7% (n = 20) of patients. Tumor hypervascularity was observed in 64.6% of patients (n = 31). Therapies received thereafter were as follows: liver transplantation (n = 13, 27.1%), liver resection (n = 2, 4.2%), locoregional therapy (n = 15, 31.3%), or systemic therapy with sorafenib/lenvatinib (n = 9, 18.8%). Five patients (10.4%) were referred directly to palliative care after tumor progression without receiving other treatments. The remaining patients (n = 4, 8.3%) did not require additional therapies after the first TACE.

Table 1 Clinicopathological characteristics of patients. Data from 48 prospectively enrolled patients with hepatocellular carcinoma (HCC) undergoing transarterial chemoembolization (TACE) as a first-line therapy are includedMolecular characterization of CTCs

Before starting to evaluate the role of CTCs in the response of HCC patients to TACE treatment, we analyzed the genomic profile of 8 samples of EpCAM+ CTCs in order to confirm the presence of tumor cells in the population of isolated cells. We identified a total of 59 genomic variants in 27 genes. Most of variants (45.8%, n = 27) were found only once in any one patient and affected genes AKT1 (variant with rsID number rs146875699), APC (rs35031194, rs78919815, and rs2149778544), APC/ENSG00000258864 (rs143927847), ATM (rs3218674 and rs747750958), DERL3/SMARCB1 (rs367593276), ERBB2 (rs779727777), FGFR2 (chr10:123310878_9 insC), FIP1L1/PDGFRA (rs139913632), GNAS (rs1004902), HNF1A (chr12:121,437,046 C⇒T), KIT (chr4:55,592,213 delA), NF1 (rs1490145865), NOTCH1 (rs2133333012 and rs1377515876), PTCH1 (rs149398794 and rs528001004), PTCH1/ENSG00000271155 (rs1805153), PTCH1/LOC100507346 (rs758487789 and rs202007968), PTEN (rs755953294), RB1 (rs774525913), SMAD4/ENSG00000267699 (rs377767334) and TERT (rs35033501 and rs762113246). The variants that were represented in ≥ 50% of total patient samples are shown in the Supplementary Table 1. Among the recurrent mutations, 17 hotspots were identified in all 8 samples and located in 9 different genes, including ATM, BRCA1, CSF1R, MSH6 and SMAD4, and the most represented genes BRCA2, NF1, PTEN and RB1, with 3 positions each one. We also show those variants affecting exclusively ≥ 50% patients from a single response group. Genomic variants rs1135402852 and rs757338143 in the NF1 gene were identified in 3/5 CTC samples from the non-response group and none from the response group. In contrast, the variants rs751132262 and rs1838016977 affecting to the RET gene were exclusively identified in 2/3 and 3/3 patients from the response group, respectively.

CTC count and clinicopathological features of the patients

All patients included in the study had detectable CTCs at baseline, at post-TACE day 1 (D1), and at post-TACE day 30 (D30). The median CTC count at baseline, at D1, and at D30 after TACE were 44.0 (IQR 22.0–105.0), 44.0 (IQR 26.3–112.8) and 28.5 (IQR 11.5–47.8), respectively. When we analyzed the CTC count according to clinical parameters (Supplementary Table 2), we found that hypervascular tumors showed higher CTC counts at baseline (hypervascular: 53.0 [IQR 28.0–119.0] vs. medium/hypo-vascular: 22.0 [IQR 9.5–54.5], p = 0.009) and at D1 after TACE (46.0 [IQR 36.0–143.0] vs. 27.0 [IQR 13.5–91.0], p = 0.026), but not at D30 (27.0 [IQR 16.0–47.3] vs. 30.0 [IQR 10.8–49.5], p = 1.000). Younger patients had increased CTC counts at D1 (< 65 years: 56.0 [IQR 36.5–377.5] vs. ≥ 65 years: 40.0 [IQR 22.0–64.0], p = 0.046), but not at baseline and at D30. CTC counts did not correlate with any other clinical characteristic including tumor burden and liver function. Regarding clusters of CTCs, these were detected more frequently in patients with a single nodule (68.4%) compared to patients with multinodular disease (23.5%) (p = 0.007) at D30 (Supplementary Table 3).

Impact of TACE on CTC count

When we expressed the CTC count as a percentage with respect to the basal situation, we found an increase of 14% at D1 (114.0% [IQR 76.5%−178.0%], p = 0.019; n = 48) that decreased to baseline levels at D30 after TACE (76.5% [IQR 41.3%−131.8%], p = 0.263; n = 36). Compared to D1, the drop in CTC count at D30 was of 43% (57.0% [IQR 28.0%−117.3%], p = 0.033; n = 36) (Fig. 1). We did not observe differences in the count of CTC clusters at any study time (data not shown).

Fig. 1

Differences in circulating tumor cell (CTC) count. CTC kinetics is expressed as a percentage referred to baseline CTC count (left panel) or to day 1 post-TACE (D1) CTC count (right panel). Data represented as the median and 95% confidence interval. Statistically significant differences with *p ≤ 0.050

Patient characteristics and response to TACE

Among the 48 patients receiving TACE, 43.7% of patients (n = 21) obtained complete radiological response at D30 (referred as responders henceforth). The remaining 27 patients obtained either partial response (n = 14, 29.2%), stable disease (n = 4, 8.3%), or had disease progression (n = 9, 18.8%), and they are referred as non-responders. When comparing the patient characteristics between responders and non-responders, we found significant differences in the baseline diameter of the largest HCC nodule (2.5 cm [IQR 2.0–3.2] vs. 3.4 cm [IQR 2.2–4.3], p = 0.044) and the total tumor diameter (3.5 ± 1.0 cm vs. 4.9 ± 2.2 cm, p = 0.005).

TACE responders had a non-significant trend towards a progressive clearance of CTCs whereas non-responders showed a pronounced and early release of CTCs 24 h after the procedure (Fig. 2). Indeed, non-responders to TACE showed an increase in the percentage of CTCs by 38% after the procedure (138.0% [IQR 100.0%−196.0%], p = 0.007; n = 27). After 30 days, the CTC count in the non-response group fell to baseline levels (74.0% [44.0%−119.0%], p = 0.199, n = 19), and to 49.0% (IQR 28.0%−103.0%) compared to D1 (p = 0.009; n = 19). No differences were observed in the CTC cluster distribution at any study time (data not shown).

Fig. 2

Differences in CTC count between the groups of responders and non-responders to TACE. CTC kinetics is expressed as a percentage referred to baseline CTC count (left panel) or to D1 CTC count (right panel). Data represented as the median and 95% confidence interval. Statistically significant differences are marked with **p ≤ 0.010

Based on the CTC kinetics for the entire cohort at D1 after TACE (Fig. 1), we set 15% as the threshold for a clinically meaningful CTC increase. CTC count increase > 15% at D1 was strongly associated with the absence of complete radiological response one month after TACE (p = 0.009): 62.5% of patients (n = 15) with stable CTC count at D1 had complete radiological response to TACE while 75.0% of patients (n = 18) showing an early increase of CTCs after TACE were non-responders. CTC increase by 15% or more at D1 predicted the absence of radiological response at day 30 in 18 out of 27 patients [OR: 5.0 (95% CI: 1.4–17.3), p = 0.011] (Table 2). The decline of CTC count at D30 was not associated with the likelihood of complete radiological response within the first month after TACE.

Table 2 Univariate and multivariate logistic regression comparing patients with and without complete radiological response to TACE. Significance levels, odds ratios (OR), and 95% Confidence Intervals (CI) are shown for predictors of non-response

The absence of CTC clusters at baseline (p = 0.027) or at D30 (p = 0.047) was more frequent among non-responders to TACE. Specifically, the absence of CTC clusters at baseline predicted the non-response to TACE in 15 out of 27 patients [OR: 4.0 (95% CI: 1.1–14.1), p = 0.031)]. In the multivariate analysis, a larger tumor diameter [OR: 1.9 (95% CI: 1.1–3.3), p = 0.020] and CTC increase at D1 by ≥ 15% [OR: 5.3 (95% CI: 1.3–21.0), p = 0.017] were independent predictors of poor response to TACE (Table 2). The multivariate model composed by total tumor diameter and early release of CTCs had an area under receiver operating characteristic curve (AUROC) of 0.796 (95% CI: 0.7–0.9, p < 0.001), with sensitivity of 70.0% (95% CI: 47.4–92.3), specificity of 75.0% (95% CI: 57.2–92.8), positive predictive value of 66.7% (95% CI: 44.1–89.2), and negative predictive value of 77.8 (95% CI: 60.2–95.3) to predict suboptimal response to TACE (Fig. 3).

Fig. 3

ROC curves showing the best diagnostic ability to predict patient response to TACE. A Individual ROC curves for the predictors total tumor diameter (blue), absence of CTC clusters at baseline (red), and CTC increase ≥ 15% at D1 (green). B ROC curve for the combination of total tumor diameter and CTC increase ≥ 15% at D1 by logistic regression. Data for AUROC (95% confidence interval) are shown

When only patients with hypervascular tumors were considered in the analysis (n = 31), larger tumor diameter [OR: 2.5 (95% CI: 1.1–5.6), p = 0.022] and CTC increase at D1 by ≥ 15% [OR: 7.6 (95% CI: 1.1–54.4), p = 0.044] remained independent predictors of poor response to TACE. In patients with non-hypervascular lesions these variables had no predictive capacity for TACE response.

CTC dynamics and patient survival after TACE

The median follow-up after TACE was 22 months (95% CI: 11.7–32.3 months) during which 9 patients (23.1%) underwent liver transplantation and 17 patients (43.6%) died, 15 of them (88.2%) due to tumor progression. We excluded 9 patients from the OS analysis after TACE because they underwent liver resection or liver transplantation within the first 6 months after the procedure. The transplant-free survival rate was not associated with the radiological response to TACE (p = 0.451), nor was the CTC increase ≥ 15% at D1 (p = 0.150) or the presence of CTC clusters at baseline (p = 0.084).