Survival outcomes from different clinicopathologic classifications.
The clinicopathologic classifications of the 79 HCC cases were summarized in Table 1. Survival outcomes from these cases were compared based on their classifications of BCLC stages, recurrence status, and E-S grades. The Kaplan-Meier analysis showed that cases in BCLC stage A had longer survive than cases in BCLC stage B and C (P<0.001) (Figure 1A). Cases without recurrence had better prognosis than cases with recurrence (P < 0.001) (Figure 1B). Cases in E-S grade III seemed to have longer survival than cases in grade II (P=0.022) (Figure 1C). When we took the survival analysis on different combinations of stages and recurrence, cases without recurrence in stage A had the best prognosis, while cases with recurrence in stage C had the worst prognosis (Figure 1D). Considering the interactions between the three clinical classifications, we further stratified all cases into different BCLC stages A and C and evaluated the influence on survival by recurrence status and E-S grades. Cases without recurrence had better prognosis than those with recurrence in both stages A and C (P<0.001) (Supplemental Figure 1A/B). Cases in E-S grade III had better prognosis than grade II only in stage A (P=0.018) but not in stage C (Supplemental Figure 1C/D). It was also noted that cases in E-S grade III was significantly associated with non-recurrence in BCLC stage A (P<0.001) but had no significant association between recurrence and non-recurrence in BCLC stage C (P=1.00) (Table 1). These results indicated that the BCLC stages and recurrence status could be an independent predictor for prognosis and E-S grades might be a modifying predictor affected by stage and recurrence. However, the overall size sample size from this case series was limited and thus this preliminary observation would need further validation from a large case series.
Genomic profile of CNAs and correlation of CNAs with clinicopathologic classifications
Of the 79 cases subjected for aCGH analysis, 95% (75/79) were successful to have results from both tumor and adjacent nontumor DNA pairs. All DNAs from adjacent nontumor tissues showed normal results and CNAs detected in tumor DNAs were summarized in Supplemental Table 1. The genomic profile of CNAs in all 75 cases was shown in Figure 2A. Of the CNAs from these 75 cases, more than 50% cases had gains of 1q and 8q and a loss of 16q, more than 40% cases had losses of 4q and 17p, and a gain of 5p, and more than 30% cases had losses of 8p and 13q.
The genomic profiles of CNAs for cases in E-S grades II and III were shown in Figure 2B/C, respectively. There were significantly more copy number losses of 1p36.31p22.1 (chr1:6,334,098-93,461,630, 87.1 Mb) (P=0.004), 4q13.2q35.2 (chr4:68,288,998-190,706,331, 122.4 Mb) (P=0.001) and 10q22.3q26.13 (chr10:78,314,145-124,106,670, 45.8 Mb) (P=0.010), and more copy number gain of 20p13p11.1 (chr20:1,211,324-25,680,554, 24.5 Mb) (P=0.001) for cases in grade III than in grade II. The genomic profiles of CNAs for cases in different BCLC stages and recurrence status were shown in Supplemental Figures 2 and 3, respectively. There was no significant difference of CNAs between cases in different BCLC stages A and C and with or without recurrence.
We further used ROC curve to check the association of the percentage of genomic CNAs of each case with clinicopathologic classifications. We found that the area under the curve (AUC) were 0.68 (95% CI = [0.56, 0.81]), 0.57 (95% CI = [0.42, 0.72]), 0.53 (95% CI = [0.39, 0.66]), and 0.61 (95% CI= [0.48, 0.74]) for E-S grades, recurrence status, BCLC stage A versus stage B+C, and stage A+B versus stage C, respectively (Supplemental Figure 4). These results indicated that there were more CNAs in E-S grade III than grade II (P=0.003, one tail Wilcoxon rank test) but the percentage of genomic CNAs had no significant association with E-S grades, BCLC stages and recurrence.
During the analysis of survival outcome and clinicopathologic classifications, we found the impact of E-S grades and recurrence on BCLC stages (Supplemental Figure 1). We further stratified cases into BLCL stage A and C to evaluate any difference in CNAs. The significant difference between E-S grades II and III in BCLC stage A was more losses of 4q13.2q35.2 (P=0.003 in BCLC A; P=0.80 in BLCL C) and 10q22.3q26.13 (P=0.033 in BCLC A; P=0.55 in BCLC C) in grade III; the significant difference between E-S grades II and III in BCLC stage C was more gains of 2q11.2q21.2 (chr2:98,228,328-134,727,485, 36.5 Mb) (P=1 in BCLC A, P=0.008 in BCLC C) and 20p13p11.1 (P=0.59 in BCLC A; P=0.005 in BCLC C) in grade III (Supplemental Figure 5).
Cluster analysis of CNAs for clinical classifications
We used k-means method to cluster patterns of CNAs into three clusters for all cases and cases in BCLC stages A and C. The CNAs from all 75 cases were divided into three clusters as shown in Figure 3. The 28 cases in cluster 1 had more CNA losses, the 39 cases in cluster 2 had less CNAs, and the eight cases in cluster 3 had more CNA gains. There were more CNAs in clusters 1 and 3 than cluster 2. The E-S grades, BCLC stages and recurrence status were mixed in clusters 1 and 2, while the eight cases in cluster 3 were all E-S grade III and further divided into BCLC stage A without recurrence and BCLC stage C with recurrence. Further clustering analysis was performed on cases in BCLC stage A and C. For the 32 cases in BCLC stage A, 14 cases were in cluster 1 with more CNA losses and 12 of them were classified as E-S grade III and nonrecurrence, 15 cases were in cluster 2 with less CNAs and mixed E-S grades and recurrence statues, and three cases in cluster 3 with more CNA gains were all in E-S grade III and nonrecurrence (Supplemental Figure 6). For the 31 cases in BCLC stage C, 12 cases were in cluster 1 with more CNA losses, 14 cases were in cluster 2 with less CNAs, and five cases were in cluster 3 with more CNA gains. It was noted that cases in clusters 1 and 2 were mixed in E-S grades and recurrence status, while cases in cluster 3 was associated with E-S grade III and recurrence (Supplemental Figure 7).
Survival differences between the three clusters in all three data sets were shown in Supplemental Figure 8. For all 75 cases, cases in cluster 3 had worse survival than cases in clusters 1 and 2 (P=0.021, Supplemental Figure 8A). For cases in BCLC stage A, cases in cluster 1 showed better survival than cases in clusters 2 and 3 (P=0.047); for cases in BCLC stage C, cases in clusters 3 showed the worst prognosis than cases in clusters 1 and 2 (Supplemental Figure 8B/C). Further analysis focused on a comparison between cases with the best prognosis by BCLC stage A and nonrecurrence and cases with the worse prognosis by BCLC stage C and recurrence (Figure 1D). Of the 21 cases classified as BCLC stage A, E-S grade III and nonrecurrence in Table 1, there were 12, six and three cases in clusters 1, 2 and 3, respectively. The losses of 4q13.2q35.2 and 10q22.3q26.13 associated with E-S grade III were seen in cases of cluster 1, which suggested these CNAs might play a protective or modify role for survive outcome. Other gains of chromosomes 3, 6, 11p, 12q and 20 in cluster 3 probably related to poor prognosis (Supplemental Figure 6). Of the seven cases classified as BCLC stage C, E-S grade III and recurrence in Table 1, there were one case each in clusters 1 and 2 and five cases in cluster 3. The loss of 1p36.31p22.1 and gains of 2q11.2q21.2 and 20p13p11.1 associated with E-S grade III and additional gains of chromosomes 6, 7 and 20q were seen in cluster 3 and thus likely correlated with worst survival (Supplemental Figure 7). Combing clustering and survival results noted that specific CNAs could be indicators for clinicopathologic classifications and survival prediction.