Address reprint requests to Han Chu Lee, M.D., Department of Gastroenterology (Asan Liver Center), Asan Medical Center, University of Ulsan College of Medicine, 388-1 Poongnap-2dong, Songpa-gu, Seoul, Korea 138-736. Telephone: +82-2-3010-3190; FAX: +82-2-485-5782; E-mail: firstname.lastname@example.org
In patients with hepatocellular carcinoma (HCC), liver transplantation (LT) remains the only therapeutic option that offers an opportunity to cure both the tumor and any underlying cirrhosis.[1, 2] The increasing number of HCC candidates for LT has stimulated discussion about living donor LT, particularly in Asian countries, and has justified the expansion of the tumor burden limit beyond the established criteria (ie, the Milan criteria).[1, 3, 4] However, the increased risk of HCC recurrence after transplantation is a major obstacle to the use of expanded criteria, despite several reports of comparable outcomes.[2, 5] Even though the recurrence of HCC is the main cause of patient death after transplantation, there are no widely accepted parameters predictive of the biological behavior of tumors other than the tumor size and number.
Several studies have suggested that serum alpha-fetoprotein (AFP) is a significant prognostic factor for the risk of posttransplant recurrence.[6-8] In the process of hepatic development and carcinogenesis, the activity of the AFP promoter is tightly regulated by members of the hepatocyte nuclear factor 1 (HNF1) family (more so by HNF1β than HNF1α), and these factors also regulate hepatic gene transcription.[9, 10] In vitro experiments have shown that the morphological dedifferentiation of HCC cells leading to a malignant phenotype is accompanied by reduced levels of HNF1α and increased levels of HNF1β.[10-12] More recently, experiments with an HCC xenograft model found that HNF1α gene therapy led to functional and phenotypic suppression of tumorogenicity.[10, 11, 13]
The consideration of all these findings led us to speculate that the expression of members of the HNF1 family as activators of AFP production might be related to outcomes following LT in patients with HCC. To test this hypothesis, we investigated the relationship between AFP and HNF1 expression in explanted HCC specimens at the protein level, and by means of immunohistochemical analysis using tissue microarrays, we inquired whether the expression of these molecules influenced posttransplant outcomes.
PATIENTS AND METHODS
Patients and Samples
The study sample consisted of 159 patients with HCC who underwent deceased or living donor LT at the Asan Medical Center between January 1999 and December 2004. Archival formalin-fixed and paraffin-embedded HCC specimens from all the patients were obtained from our pathology tissue bank, and information on pathology before and after transplantation with complete survival data was collected from the database of the Asan Liver Center. For all the patients, HCC was pathologically confirmed in the explanted livers. Cases of perioperative mortality, which was defined as patient death within 3 months of LT, were excluded from the study: none of these patients died because of cancer-related causes. None had extrahepatic spread or gross vascular invasion of the tumor. Also, before LT, none of the patients had other primary malignancies or uncontrolled or significant cerebrocardiovascular disease that could have led to death for reasons unrelated to HCC or liver disorders. Eighty-seven patients (54.7%) underwent LT for initial tumor control without any pretransplant treatment for HCC. The remaining 72 patients (45.3%) received transarterial chemoembolization (n = 70) and/or radiofrequency ablation (n = 7) as bridging therapy before LT (Table 1).
Table 1. Baseline Clinicopathological Profiles of the 159 Transplant Patients
The data are presented as the median and interquartile range.
The pathological tumor stage was based on the pathological tumor-node-metastasis classification proposed by the International Union Against Cancer and the American Joint Committee on Cancer.
The use of human tissue specimens and the associated data for this research were approved by the institutional review boards of our hospital.
Pretransplant Workup and Posttransplant Follow-Up
Specific approaches to assessment before and after LT by our LT team have been described in detail previously. In addition to measurements of hepatitis serology and serum AFP with an immunoradiometric assay (RIA-gnost AFP, Cis-Bio International, Schering, Switzerland), the pretransplant evaluation of HCC routinely included multidetector dynamic liver computed tomography (CT) with 3-dimensional reconstruction, positron emission tomography, bone scintigraphy, and chest CT to ensure a patient's eligibility for LT with respect to the extent of the tumor and hepatic vascularization. All LT recipients were followed up periodically by means of serum AFP assays, chest X-rays, and dynamic liver CT scans until the documentation of HCC recurrence. Different follow-up protocols for HCC surveillance were established and depended on the level of risk of posttransplant HCC recurrence; hence, CT examinations were performed more frequently in patients outside the Milan criteria versus patients within the Milan criteria. Positron emission tomography and/or chest CT scans were performed immediately in patients suspected of HCC relapse on the basis of routine radiological or serological analysis.
Immunosuppressive Treatment Protocol
The peritransplant primary immunosuppression protocols used for adult LT recipients at our institution consisted of an interleukin-2 receptor inhibitor (basiliximab) on days 0 and 4, an intraoperative steroid bolus (5-10 mg/kg), and a calcineurin inhibitor (CNI). The corticosteroid was gradually tapered and usually stopped within 3 months of transplantation. Mycophenolate mofetil was given to patients who showed CNI-associated side effects such as renal dysfunction, neurological complications, and impaired glucose tolerance, and it was usually combined with a CNI reduction. The CNI (tacrolimus or cyclosporine) was started in an intravenous form, and this was switched to an oral form after the restoration of gastrointestinal motility.
Tissue Microarray Technique
Tissue microarrays were constructed as described previously. Representative areas of each HCC tissue sample paired with corresponding adjacent noncancerous tissue were identified on the relevant hematoxylin and eosin–stained slides. Tissue cylinders with a diameter of 1 mm were punched from each donor tissue block and entered into a recipient paraffin block with a tissue microarrayer (MTAII, Beecher Instruments, Silver Spring, MD). The recipient paraffin block was subsequently cut, and slices were transferred with adhesive tape to coated slides. Next, the slides were dipped in paraffin to prevent oxidation. Each sample was arrayed in duplicate to minimize tissue loss and to overcome the effect of tumor heterogeneity.
Immunohistochemistry and Evaluation
The tissue microarray sections were immunohistochemically stained for AFP, HNF1α, and HNF1β. Briefly, the microarray slides generated from paraffin-embedded tissue blocks were deparaffinized and rehydrated for 5 minutes. After the standard cell conditioner 1 (CC1) protocol (Ventana Medical System, Tucson, AZ) was applied for antigen retrieval, the slides were immersed in 0.3% (vol/vol) hydrogen peroxide for 20 minutes to block endogenous peroxidase activity, and then they were washed and incubated at 37°C for 32 minutes with primary antibodies against AFP (1:200 dilution; Neomarkers, Fremont, CA), HNF1α (1:50 dilution; Epitomics, Burlingame, CA), and HNF1β (1:50 dilution; ProteinTech, Chicago, IL). After incubation with a biotinylated anti-goat antibody, the slides were incubated with peroxidase-labeled streptavidin, and the reaction products were visualized through the immersion of the slides in diaminobenzidine tetrachloride and counterstaining with Harris hematoxylin. All immunostaining was performed with a BenchMark XT automated immunostaining device (Ventana Medical System, Tucson, AZ).
All immunostained samples were scored independently in a blind manner by 2 pathologists. Immunostaining for AFP was interpreted as positive when at least 10% of the tumor cells showed strong cytoplasmic staining (Fig. 1). Cases were considered positive for HNF1α or HNF1β when strong nuclear staining was observed in at least 10% of the examined tumor cells. In all other cases, the immunostaining results were considered negative. The duplicate tumor samples showed a good level of agreement with respect to the pattern of immunostaining (>98% for all markers). Because in a few cases adjacent tissue was positively stained (1.9% for AFP, 18.9% for HNF1α, and 5% for HNF1β), we analyzed only intratumoral expression of the target proteins.
Relationships between biomarker expression and other clinicopathological parameters were compared with the χ2 test or Fisher's exact probability test as appropriate. The Kaplan-Meier method was used to estimate HCC recurrence, and differences were analyzed with the log-rank test. We used a competing-risk analysis to assess more accurately whether biomarker expression predicted the risk of cancer-specific death; thus, we estimated and then compared cumulative incidence rates calculated with cause-specific hazards under the assumption that the competing events (ie, HCC-specific death and competing risks) were independent of each other. Univariate and multivariate Cox regression models for the hazards of recurrence and cause-specific mortality were used to evaluate the effect of biomarker immunoreactivity, with adjustments made for covariates. In addition, we used measures of the Harrell C statistic and area under the receiver operator curves (AUROCs) to quantitatively assess the improvement in prediction accuracy achieved by the addition of the expression of a new biomarker to the existing Milan criteria.
The median age of the 159 patients was 52 years (range = 18–64 years), and all but 29 patients were male. More than 90% of the patients underwent LT with grafts from living related donors. One hundred fifty patients (94.3%) had liver disease accompanied by hepatitis B virus (HBV), and 154 of the patients (96.9%) had a cirrhotic liver according to histology. Patients were classified as Child-Pugh class A (n = 16), B (n = 61), or C (n = 82). For 60.4% of the patients (n = 96), the Model for End-Stage Liver Disease (MELD) score was >18, which is the cutoff for high-risk patients according to the MELD system. The median serum AFP level was 39.2 ng/mL (interquartile range = 10-168.5 ng/mL). More than half of the patients had tumor diameters less than 3 cm and solitary tumors in the explant livers. Pathology examinations of the explants revealed microscopic vessel invasion in only 13 patients (8.2%), but the majority of the patients (88.1%) had poorly differentiated tumors corresponding to Edmonson grade III or IV. Thirty-six patients (22.6%) had HCCs exceeding the Milan criteria, and the remaining 123 patients (77.4%) met the Milan criteria according to explant pathology reports.
The immunohistochemical analysis revealed the expression of AFP, HNF1α, and HNF1β in 22.6% (n = 36), 46.5% (n = 74), and 61.0% (n = 97) of the tumors, respectively. The proportions of positive cells per tissue microarray core were as follows: less than one-third in 63.9%, one-third to two-thirds in 13.9%, and more than two-thirds in 22.2% of the AFP-positive tumors; less than one-third in 79.5% and one-third to two-thirds in 20.5% of the HNF1α-positive tumors; and less than half in 92.8% and half or more in 7.2% of the HNF1β-positive tumors.
Time to Recurrence According to AFP and HNF1 Expression
Within the median follow-up period of 99.9 months (interquartile range = 61.4–120.4 months) for living patients, tumors recurred in 34 patients (21.4%), and death related to HCC recurrence occurred in 29 (18.2%). The cumulative risk of recurrence for the overall population was 18.2% at 3 years, 21.2% at 5 years, and 27.8% at 10 years. The prognostic effects of clinicopathological and immunohistochemical variables on the times to recurrence and HCC-specific death were evaluated with univariate and multivariate Cox regression analyses (Table 2). In the univariate analysis, the time to recurrence was inversely correlated with Child-Pugh class C, a serum AFP level ≥ 40, 200, or 400 ng/mL, microvascular invasion, tumors outside the Milan criteria, a history of pre-LT locoregional therapy, and tumor expression of AFP and HNF1β (P < 0.05 in all cases). However, HNF1α expression in the tumors did not influence the time to recurrence (P = 0.95). After adjustments for other significant covariates (including serum AFP at any cutoff level), only HCC lesions outside the Milan criteria [hazard ratio (HR) = 7.48, 95% confidence interval (CI) = 3.65–15.32, P < 0.001], the expression of AFP ((HR = 2.42, 95% CI = 1.21–4.85, P = 0.01), and the expression of HNF1β (HR = 2.69, 95% CI = 1.16–6.24, P = 0.02) remained significant. Patients outside the Milan criteria had higher rates of 2- and 5-year recurrence (48.2% and 56.1%) than patients within the Milan criteria (5.2% and 11.9%, P < 0.001; Fig. 2A). A significant difference in recurrence rates was also observed with HNF1β immunostaining (18.8% and 27.6% for HNF1β positivity versus 7.1% and 10.9% for HNF1β negativity, P = 0.02; Fig. 2B).
Table 2. Cox Regression Models for Identifying Risk Factors Related to the Times to Recurrence and HCC-Specific Death
Time to Recurrence
Time to Cancer Death
HR (95% CI)
HR (95% CI)
Multivariate analyses adjusted for serum AFP with cutoff values of 40, 200, and 400 ng/mL gave the same results for the times to recurrence and cancer-specific death.
Time to HCC-Specific Death According to AFP and HNF1 Expression
During the follow-up period, 23 of the 52 nonsurvivors (44.2%) died without recurrence, and this constitutes a competing risk for HCC-specific survival: 29 (55.8%) died as a result of HCC. The 3-, 5-, and 10-year cumulative risks of HCC-specific death for the overall population were 12.0%, 15.5%, and 22.8%, respectively. Just as for the time to recurrence, tumors exceeding the Milan criteria (HR = 3.99, 95% CI = 1.88–8.47, P < 0.001) and positive HNF1β expression (HR = 3.22, 95% CI = 1.22–8.53, P = 0.02) were independently associated with the time to HCC-specific death along with previous locoregional treatment (HR = 2.96, 95% CI = 1.24–7.05, P = 0.01), whereas immunopositivity for HNF1α and even AFP were not associated with the time to HCC-specific death. The results were the same when the serum AFP cutoff value included as a covariate in the multivariate analysis was defined as 40, 200, or 400 ng/mL (Table 2). As shown in Fig. 3A, the 5- and 10-year cancer death rates for the patients who exceeded the Milan criteria were both 45.3%; this was much higher than the rates of 7.0% and 14.8%, respectively, for the patients within the Milan criteria (P < 0.001; Fig. 3A). The 5- and 10-year cumulative incidence estimates for HCC-specific mortality were 20.2% and 30.3%, respectively, in the HNF1β-positive group and 7.7% and 8.4%, respectively, in the HNF1β-negative group (P = 0.007; Fig. 3B).
Association of HNF1β Expression With Clinicopathological and Immunohistochemical Parameters
The associations between HNF1β expression and various patient and tumor factors are presented in Table 3. The positive expression of HNF1β was correlated with higher serum AFP levels at cutoffs of 40, 200, and 400 ng/mL (P = 0.001, P = 0.04, and P = 0.04, respectively) along with microvascular tumor invasion (according to pathology) and a history of pre-LT local therapy (P = 0.01 and P = 0.047, respectively). This correlation was substantiated when serum AFP was expressed on a logarithmic scale (1.95 ± 0.99 and 1.41 ± 0.79 for patients positive and negative for HNF1β expression, respectively, P < 0.001; Fig. 4). The expression of HNF1β was also correlated with the expression of AFP (P = 0.04), but it was not correlated with the expression of HNF1α (P = 0.11). No other parameters, including the tumor burden and the histological differentiation grade, differed between the HNF1β-postive and HNF1β-negative groups (P > 0.05 for all).
Table 3. Clinicopathological Parameters of HCC According to HNF1β Expression
HNF1β Expression [n (%)]
Positive (n = 97)
Negative (n = 62)
A or B
Prior locoregional therapy
Microscopic vascular invasion
I or II
III or IV
Effect of HNF1β Expression on Post-LT Outcomes for Tumors Within and Outside the Milan Criteria
To test the robustness of the effect of HNF1β immunopositivity on the times to HCC recurrence and death after transplantation in homogeneous tumor groups, we divided the patients into those with tumors within the Milan criteria and those with tumors outside the Milan criteria. For the patients meeting the Milan criteria without histological expression of HNF1β, the 10-year rates were only 2.4% for recurrence and 2.1% for HCC-specific death, which were much lower than the rates of 23.3% and 23.1%, respectively, for the patients meeting the Milan criteria who were positive for HNF1β (P = 0.007 for both; Fig. 5A,B). However, there was no significant difference in the time to recurrence or HCC-specific death according to HNF1β expression in the group with HCC exceeding the Milan criteria (P = 0.27 and P = 0.38, respectively), although the cumulative incidence function curves were separated (Fig. 5C,D).
Additional Value of HNF1β Expression in Risk Prediction
In the receiver operating characteristic analysis of the time to recurrence, the C statistic for the model using only the Milan criteria was 0.705 (95% CI = 0.632–0.791), and this increased when HNF1β expression was added [0.749 (95% CI = 0.686–0.825), P for improvement = 0.03]. An effect of a similar magnitude was observed for the time to cancer death (the C statistic increased from 0.674 to 0.734 when HNF1β was added to the Milan criteria; P = 0.04). The AUROCs for the times to recurrence and cancer death also improved after the inclusion of HNF1β (P < 0.05 for all; Table 4). When the histological expression of AFP was added to the Milan criteria, a significant increase was observed in the AUROCs for the time to recurrence at 2 and 5 years (P = 0.045 and P = 0.04, respectively) but not in the C statistics (P = 0.09). There was no significant change in C statistics or AUROCs with respect to cancer death (P > 0.05 for all; Table 4).
Table 4. Accuracy of Risk Prediction Using the Milan Criteria With and Without the Histological Expression of HNF1β or AFP
Milan Criteria Plus HNF1β
Milan Criteria Plus AFP
NOTE: Ninety-five percent CIs are shown in parentheses.
Time to recurrence
0.749 (0.686–0.825), P = 0.03
0.746 (0.686–0.830), P = 0.09
AUROC at 2 years
0.755 (0.684–0.828), P = 0.009
0.751 (0.682–0.830), P = 0.045
AUROC at 5 years
0.746 (0.672–0.811), P = 0.01
0.745 (0.678–0.816), P = 0.04
Time to cancer death
0.734 (0.660–0.816), P = 0.04
0.706 (0.627–0.787), P = 0.26
AUROC at 5 years
0.724 (0.653–0.800), P = 0.03
0.697 (0.621–0.774), P = 0.19
AUROC at 7 years
0.724 (0.654–0.799), P = 0.03
0.693 (0.617–0.770), P = 0.20
We have shown that after LT in patients with HCC, the expression of HNF1β in tumor tissue is strongly associated with the time-dependent risks of recurrence and HCC-specific death, and this is independent of the Milan criteria. Patients with AFP-positive tumors were also at increased risk for tumor recurrence but not for cancer death. The expression of HNF1β in explants had a more decisive effect on the clinical endpoints in the group meeting the Milan criteria versus the group exceeding the Milan criteria (although the difference may have been due to the small number of cases beyond the Milan criteria); only 1 of 50 patients who were negative for HNF1β experienced recurrence of HCC and died during follow-up. Importantly, the use of HNF1β measurements in combination with the conventional Milan criteria had added benefits in terms of the accuracy of predicting post-LT times to recurrence and cancer death, with significant increases in both C statistics and AUROCs.
LT is one of the best hopes for achieving a cure in select cases of HCC and cirrhosis. It is currently standard practice to select transplantable patients according to the tumor burden, mainly with the Milan criteria, although the details of the selection process differ between countries and even between centers.[1, 4] Although patients within the Milan criteria achieve more than 70% survival 5 years after LT, tumors recur in approximately 20% of patients during the same period.[1, 21] On the other hand, the tumor-based criteria exclude from LT some patients who could potentially benefit from it. It is most important that candidates who are likely to benefit from LT should be accurately selected because of the limited availability of donor livers and the possibility of living donor morbidity and even death in addition to transplant-related complications. Therefore, there is an urgent need for additional predictors of post-LT tumor biology as well as the identification of potential therapeutic targets to improve post-LT outcomes.
Microscopic vascular invasion has emerged as a putative predictor of HCC recurrence in posttransplant patients without gross vascular and distant tumor involvement. Because an increased tumor size and an increased number of tumors have been found to be important predictors of occult vascular invasion, candidacy for LT has usually been limited to patients with early-stage HCC, who are less likely to have microvascular invasion. Moreover, not all patients with microvascular invasion experience postresection tumor recurrence and poor survival, as we also observed in the present study. Accordingly, microvascular invasion may not always be a powerful indicator of a poor prognosis after LT for HCC. Interestingly, the expression of HNF1β was closely associated with microvascular invasion in our transplant cohort. The invasive nature of HNF1β-positive HCC may be responsible for the unfavorable post-LT prognosis.
According to a recent systematic review, patients with high levels of serum AFP before LT have poor outcomes as well as enhanced malignant behavior. In addition, a study by Fujiki et al. found a significant correlation between pre-LT AFP values and the risk of vascular invasion. In agreement with this result, the serological and immunohistological AFP status was also significantly correlated with the microscopic vessel invasion of tumors in our series (data not shown). A previous immunostaining study found that HCC patients whose tissues were positively stained for AFP had a lower rate of recurrence-free survival and displayed a trend toward a lower rate of overall survival after surgical resection. Similarly, our immunoprofiles of the liver explants revealed that AFP-positive tissue was clearly related to the time to recurrence, but the statistical significance of this relationship disappeared after multivariate correction in terms of the time to cancer death, and this corresponded to the pattern of AUROC results. We investigated the status of HNF1β, a critical modulator of AFP promoter activity in the hepatocarcinogenetic process, because this might explain the correlation between the expression of HNF1β in tumor tissue and AFP in the sera and tumor tissues of our cohort of LT recipients. Interestingly, we found that HNF1β played a pivotal role in predicting tumor recurrence and even subsequent death independently of the tissue and serum AFP status. This may indicate that upstream regulation of AFP expression by HNF1β at the transcriptional level operates specifically during the course of HCC progression following recurrence.
In the present study, HNF1α expression was not closely connected to post-LT outcomes. It has been suggested that inactivation of HNF1α may contribute to hepatic carcinogenicity.[13, 25] In contrast, a previous examination by northern blot analysis revealed that HNF1α was up-regulated in HCC samples versus nontumorous liver tissues. These differences in the expression profiling of HNF1α, which may result from multiple regulatory pathways being involved in the development of HCC, may account for its poor predictive power in clinical settings.
The main limitation of this study was that HNF1β expression was assessed only at the tissue level. This biomarker seemed to be useful for selecting candidates for LT per se and also for pre-LT treatment when it was checked before LT. Currently, the utility of pre-LT biopsy for an accurate diagnosis and for further characterization of the tumor is a matter of debate because the risk of seeding and disseminating malignant cells may outweigh the contribution to selecting patients suitable for LT. An examination of HNF1β expression at the plasma level is needed to improve its potential as a useful prognosticator of LT outcomes. Another consideration is that the majority of patients included in this study had HBV-related liver disease. Further studies are needed to assess an extension of the clinical use of HNF1β expression.
In conclusion, we have shown that the hepatic expression of HNF1β is a useful aid for predicting the recurrence of HCC after LT; it evidently influences survival after transplantation and can supplement the standard Milan criteria for the pre-LT selection and post-LT risk stratification of patients. HNF1β protein merits further preclinical study with respect to its role in hepatocarcinogenesis and as a potential novel therapeutic target in HCC. A prospective validation study, including serological measurements of HNF1β, is being planned by our research team.