Pathologic Predictive Factors for Late Recurrence of Hepatocellular Carcinoma in Chronic Liver Disease

Abstract

Background & Aims:

Late recurrence of hepatocellular carcinoma (HCC) is regarded as de novo HCC from chronic hepatitis. This study investigated clinicopathological and molecular factors to develop a nomogram for predicting late HCC recurrence (> 2 years after curative resection).

Methods:

The training and validation cohorts included HCC patients with a major etiology of hepatitis B who underwent curative resection. Clinicopathological features including lobular and porto-periportal inflammatory activity, fibrosis, and liver cell change were evaluated. Proteins encoded by genes related to late recurrence were identified using a reverse phase protein array of 95 non-tumorous liver tissues. Immunoexpression of phosphorylated signal transducer and activator of transcription 3 (pSTAT3), plasminogen activator inhibitor-1, phosphorylated extracellular signal-regulated kinase 1/2 (pERK1/2), and spleen tyrosine kinase (SYK) was measured.

Results:

Late recurrence occurred in 74/402 (18%) and 47/243 (19%) in the training and validation cohorts, respectively. Cirrhosis, moderate/severe lobular inflammatory activity, and expression of pSTAT3, pERK1/2, and SYK proteins correlated to the gene signature of hepatocyte injury and regeneration were independently associated with late recurrence, with odds ratios (95% confidence intervals) of 2.0 (1.2–3.3), 21.1 (4.3–102.7), and 6.0 (2.1–17.7), respectively (p < 0.05 for all). A nomogram based on these variables (histological parameters and immunohistochemical marker combinations) showed high reliability in both the training and validation cohorts (Harrell’s C index: 0.701 and 0.716; 95% confidence intervals: 0.64–0.76 and 0.64–0.79, respectively).

Conclusions:

The combination of pSTAT3, pERK1/2, and SYK immunoexpression with high lobular inflammatory activity and cirrhosis (fibrosis) predicts late HCC recurrence.

Lay Summary

• Immunoexpression of pSTAT3, pERK1/2, and SYK, which are correlated to the gene signature of hepatocyte injury and regeneration, along with moderate or severe lobular inflammatory activity and cirrhosis (fibrosis) in non-tumorous liver tissue predict late HCC recurrence (> 2 years after curative resection).

• The nomogram developed using these parameters (histological parameters and combination of immunohistomarkers) can be useful for predicting the risk of late HCC recurrence and establishing individualized treatments and follow-up plans after curative resection of HCC.

1. Introduction

Hepatocellular carcinoma (HCC) generally develops through a multistep process in a background of chronic liver disease^1,2. The cumulative 5-year recurrence rate of HCC after curative resection is as high as 79% and is caused by two types of recurrence³. Early HCC recurrence (within 2 years of curative resection) is considered to be intrahepatic metastasis and has been reported to be associated with tumor size, microvascular invasion, and satellite nodules of the primary HCC^4,5. By contrast, late HCC recurrence (> 2 years after curative resection) is considered to represent metachronous multicentric occurrence related to chronic hepatitis of the background liver as a field effect³.

The 5-year cumulative incidence of late HCC recurrence is 25–35%³, which is higher than the 5-year cumulative HCC occurrence rate of 5–30% in chronic hepatitis patients^3,6. Moreover, curative resection increases survival and, thus, the risk of late HCC recurrence also increases owing to prolonged survival. Therefore, predicting late HCC recurrence is helpful for individualized treatment in high-risk patients.

Several studies have shown that the risk of developing HCC increases in chronic hepatitis patients with or without a history of HCC^7–9. Histopathologic features of advanced fibrosis stage and high-grade inflammatory activity in the background liver tissue have been reported as risk factors for late HCC recurrence^3,4,9–11. Liver cell change (LCC) is also considered as a risk factor for developing HCC in cirrhosis patients^8,12. Notably, LCC, especially small LCC, shares genetic alterations with HCC (e.g., telomere shortening, cell cycle check point inactivation, chromosomal instability, and DNA damage in chronic liver disease) and thus might be a very early precursor of HCC^13,14. In addition, the clinicopathological features including total tumor number, serum alpha-fetoprotein (AFP) levels, hepatitis B virus surface antigen (HBsAg) level, and hepatitis B virus deoxyribonucleic acid (HBV-DNA) level were reported to be related to late HCC recurrence^3,4,15–17.

The risk of multiple occurrence or late recurrence of HCC has been predicted based on the expression profiles of specific genes in non-tumor liver tissues after surgical resection^18,19. The molecular signatures of hepatic injury and regeneration (HIR) including 233 genes in background non-tumorous liver tissue were recently reported to be closely associated with late HCC recurrence and poor prognosis²⁰. However, they are too complex to apply in daily pathologic practice.

Thus, this study aimed to investigate the histopathological parameters and expression of immunohistochemical (IHC) markers in the background liver that are associated with late HCC recurrence after curative resection and to develop and validate a model for predicting late HCC recurrence. To our knowledge, this is the first nomogram based on histopathological parameters and IHC protein markers for predicting late HCC recurrence.

2. Material and methods

2. 1. Case selection and clinical information

The training cohort included consecutive patients who underwent curative resection for HCC at Severance Hospital between March 2006 and February 2011. The validation cohort comprised consecutive patients who underwent curative resection for HCC at Seoul National University Bundang Hospital between January 2003 and February 2013 (Figure 1). Clinical data regarding age, sex, etiology of chronic liver disease, serum AFP level, presence of HBsAg, HBsAg level, HBV-DNA level, tumor number, and time to HCC recurrence were retrospectively collected from electronic medical records. Early HCC recurrence and late HCC recurrences were defined as recurrence within or after 2 years, respectively, after curative resection for HCC¹⁰.

Figure 1.

Study design. Non-tumorous liver tissue samples (n=402) of patients with hepatocellular carcinoma (HCC) in the training cohort were evaluated and used to generate a predictive model that was tested using data from an independent validation cohort involving 243 HCC patients.

This study was approved by the Institutional Review Boards of Severance Hospital, Yonsei University College of Medicine and Seoul National University Bundang Hospital. The need for patient consent was waived due to the retrospective nature of the study.

All patients were regularly followed up after curative resection according to the 2014 and 2018 Korean Liver Cancer Association practice guidelines for the management of HCC^21,22. Patients visited the hospital about 4–6 weeks after surgery for assessment of postoperative complications. Thereafter, they were required to attend regular follow-up visits every 3–6 months during the first 2 years to assess for tumor recurrence using serum AFP level and computed tomographic (CT) scan. Patients with no evidence of HCC recurrence within 2 years after resection were then followed up every 6 months. Those presenting with features of HCC recurrence, such as an elevated serum AFP level or the appearance of suspicious nodules on CT, underwent further evaluations with magnetic resonance imaging and/or positron emission tomography-CT to detect recurrent lesions.

2.2. Evaluation of histopathological parameters

For histological evaluation, representative sections of the non-tumorous liver parenchyma stained with hematoxylin and eosin and Masson’s trichrome were examined. Lobular and porto-periportal inflammatory activity and fibrosis stage in chronic hepatitis were assessed (Supplementary Table 1 and Supplementary Figure 1A–C). Stage 4 fibrosis (cirrhosis) was further divided into three subcategories according to septal thickness: 4A, 4B, and 4C. Small and large LCC (SLCC and LLCC, respectively) were evaluated according to previously established criteria^23,24 (Supplementary Figure 1D). The proportion of LCCs in the representative section was estimated; a positive result was defined as the presence of at least 5% of each type of LCC.

2.3. Evaluation of immunohistochemical markers

2.3.1. Proteomic data from reverse phase protein array (RPPA) experiments

Protein expression profiling was performed with an RPPA platform using archived non-tumorous fresh liver tissue samples of 95 HCC patients who underwent resection as previously described^25,26. Briefly, protein was extracted from the tissue using the RPPA lysis buffer composed of 1% Triton X-100, 50 nmol/L HEPES (pH 7.4), 150 nmol/L NaCl, 1.5 nmol/L MgCl₂, 1 mmol/L EGTA, 100 nmol/L NaF, 10 nmol/L Na pyrophosphate, 10% glycerol, 1 nmol/L phenylmethylsulfonyl fluoride, 1 nmol/L Na₃VO₄, and 10 mg/mL aprotinin. Lysis buffer was used to lyse the frozen tumors by homogenization using a Precellys® homogenizer (Bertin Instruments, Montigny-le-Bretonneux, France). The protein concentration in the lysate was adjusted to 1 μg/μL, as determined with the bicinchoninic acid assay, and the samples were boiled in 1% sodium dodecyl sulfate. Serial 5-fold dilutions were prepared with lysis buffer. An Aushon Biosystems 2470 arrayer (Burlington, MA, USA) was used to print samples onto nitrocellulose-coated slides (Grace Bio-Labs, Bend, OR, USA). These were then probed with 172 validated primary antibodies followed by corresponding secondary antibodies (goat anti-rabbit or -mouse or rabbit anti-goat IgG). Signals were visualized with a diaminobenzidene colorimetric reaction using a DakoCytomation catalyzed signal amplification system (Dako, Glostrup, Denmark). Slides were scanned with CanoScan 9000F (Canon, Tokyo, Japan). Spot intensity was analyzed and quantified using Arraypro (http://www.mediacy.com/index.aspx?page=ArrayPro) (Level 1 data). SuperCurveGUI software²⁷ (available at http://bioinformatics.mdanderson.org/Software/supercurve/) was used to estimate the half-maximal effective concentrations of proteins in each dilution series (in log2 scale). The RPPA data were processed and normalized as previously described^{25,26,28–30}. Final antibody selection was dictated by the availability of high-quality antibodies that consistently passed a strict validation process³¹. These antibodies were assessed for specificity, quantifiability, and sensitivity (dynamic range) using protein extracts from cultured cells or tumor tissue. The HIR signature was determined based on the mRNA expression data of the same tissue samples, as previously reported²⁰.

2.3.2. Immunohistochemical analyses

Representative paraffin-embedded sections of the non-tumor liver parenchyma were used for tissue microarray construction and IHC analyses. Microarray construction and IHC staining were performed as previously described^32,33. Details of the antibodies are summarized in Supplementary Table 2. In all markers, positive expression was defined as staining of ≥ 1% of the hepatocytes.

2.4. Development and validation of predictive model for late HCC recurrence

2.4.1. Selection of significant risk factors

Data were analyzed using the t-test and Mann–Whitney U-test for continuous variables and the χ² test or Fisher’s exact test for categorical variables. Univariate Cox proportional hazards regression analyses were performed to identify significant risk factors among the histologic parameters and IHC markers. To confirm the significance of differences in late HCC recurrence-free survival between the IHC marker combinations (all negative or positive expression of one, two, or three markers), survival curves were generated using the Kaplan–Meier method, and late HCC recurrence-free survival was compared using the log-rank test.

2.4.2. Construction of a predictive model and visualization using a nomogram

The following three potential predictive models were considered: Model 1 used only histological parameters (lobular inflammatory activity and cirrhosis) (base model); Model 2 used histological parameters + individual weighted IHC markers; and Model 3 used histological parameters + IHC marker combinations. Final risk factors were selected using a stepwise variable selection method, and multiple Cox proportional hazards regression analyses were performed for each of the three models.

For constructing a predictive model and visualizing it with a nomogram, the regression coefficient of each variable was calculated, and a score was assigned for each point based on the weighted (relative) importance of the individual risk factors. Individual probabilities for 3- and 5-year recurrence-free survival of HCC patients were calculated.

2.4.3. Discrimination and calibration

The predictive accuracy of the final model was assessed by discrimination evaluated by calculating the concordance index (Harrell’s C index)³⁴ and measured as the time-dependent incremental area under the receiver operating characteristic curve Heagerty’s integrated AUC (iAUC)³⁵ for the three models. Differences and 95% confidence intervals (CIs) between the outcome and model were calculated using a bootstrapping method. To identify improvements in the predictive capability by adding IHC markers in the base model (histological parameters only), the net reclassification improvement (NRI) and the integrated discrimination improvement (IDI)^36,37 at 3 and 5 years were computed. Calibration draw to the agreement between observed outcomes and predictive probabilities, and a bootstrap method was used to reduce bias.

2.4.4. Statistical analyses

Two-sided p-values of < 0.05 (excluding 0) in the 95% CI were considered to indicate statistical significance. All statistical analyses were performed using SAS v.9.2 (SAS Institute, Cary, NC, USA), and R v. 3.2.5 (survival, rms, and risksetROC packages) (R Foundation for Statistical Computing, Vienna, Austria).

3. Results

3.1. Histological features of the background non-neoplastic liver

The training and validation cohorts involved 402 and 243 HCC patients, respectively. The major etiology of hepatitis was hepatitis B virus infection (335/402, 83.3% in the training cohort, and 188/243, 77.4% in the validation cohort), followed by hepatitis C virus infection and alcohol use, in both cohorts. There was no significant difference in baseline clinicopathological characteristics between the two cohorts. There were 132 cases of early recurrence (33%) and 74 cases of late recurrence (18%) during a median follow-up period of 82 months (range, 1–135 months), in the training cohort. There were 75 cases of early recurrence (31%) and 47 cases of late recurrence (19%) during a median follow-up period of 56 months (range, 1–155 months) in the validation cohort (Supplementary Table 3). In both training and validation cohorts, background livers with late HCC recurrence showed a higher frequency of cirrhosis (fibrosis stage 4) and more extensive LLCC and SLCC compared to those with no recurrence (p < 0.05 for all) (Table 1 and 2). Moderate to severe lobular inflammatory activity was more frequent in those with late HCC recurrence than in those with no recurrence in the validation cohort (p < 0.001), but not in the training cohorts. In the training cohort, the HBsAg level was lower and the number of tumors was higher in late HCC recurrence cases compared to those with no recurrence (p < 0.05 for both), but such differences were not evident in the validation cohort.

Table 1.

Clinicopathological characteristics of the patients with no and late recurrence in the training cohort

	No recurrence (n = 196)	Late recurrence (after 2 years) (n = 74)	p-value
Clinical characteristics
Age (years, mean ± SD)	55.28 ± 10.63	54.78 ± 9.73	0.729
Sex (male/female)	156 (80%)/40 (20%)	62 (84%)/12 (16%)	0.436
Etiology (HBV/HCV/Alcohol/Unknown)	163 (83%)/5 (3%)/7 (4%)/21 (10%)	63 (85%)/6 (8%)/1 (2%)/4 (5%)	0.085
AFP level (ng/mL, mean ± SD)	1820.29 ± 8764.71	918.29 ± 4479.85	0.399
HBsAg† (positive)	142 (76%)	58 (83%)	0.261
HBsAg level† (IU/mL, mean ± SD)	548.59 ± 923.55	261.36 ± 478.02	0.004*
HBV DNA level† (IU/mL, mean ± SD)	1110671.19 ± 7884762.85	707453.65 ± 2004088.26	0.716
Tumor number (1/2/3/4)	183 (93%)/12 (6%)/1 (1%)/0 (0%)	62 (84%)/12 (16%)/0 (0%)/0 (0%)	0.031*
Histological features
Grade (inflammation)
Lobular activity(no/minimal/mild/moderate/severe)	12 (6%)/42 (21%)/142 (72%)/0 (0%)/0 (0%)	4 (5%)/10 (14%)/58 (78%)/2 (3%)/0 (0%)	0.062
Porto-periportal activity (no/minimal/mild/moderate/severe)	12 (6%)/50 (26%)/118 (60%)/16 (8%)/0 (0%)	4 (5%)/13 (18%)/45 (61%)/12 (16%)/0 (0%)	0.182
Stage (fibrosis)
Cirrhosis (stage 4)	81 (39%)	48 (65%)	0.001*
Stage (0/1/2/3/4A/4B/4C)	5 (3%)/8 (4%)/36 (18%)/66 (34%)/22 (11%)/50 (26%)/9 (5%)	0 (0%)/1 (1%)/10 (14%)/15 (20%)/11 (15%)/32 (43%)/5 (7%)	0.033*
Liver cell change
Liver cell change (No/LLCC/SLCC/LLCC and SLCC)	63 (32%)/84 (43%)/1 (1%)/48 (25%)	17 (23%)/26 (35%)/1 (1%)/30 (41%)	0.055
LLCC proportion(%, mean ± SD)	8.41 ± 10.05	13.39 ± 12.43	0.003*
SLCC proportion(%, mean ± SD)	2.44 ± 5.39	4.39 ± 7.14	0.035*
Immunohistochemical marker expression
pSTAT3 (positive/negative)	21 (11%)/175 (89%)	18 (24%)/56 (76%)	0.005*
pERK1/2 (positive/negative)	49 (25%)/147 (75%)	35 (47%)/39 (52%)	<0.001*
SYK (positive/negative)	50 (26%)/146 (74%)	25 (34%)/49 (66%)	0.176
PAI-1 (positive/negative)	114 (58%)/82 (42%)	53 (72%)/21 (28%)	0.042*

^†Some patients had unavailable data. Only patients with complete data were included in the analysis.

SD, standard deviation; HBV, hepatitis B virus; HCV, hepatitis C virus; AFP, Alpha-fetoprotein; HBsAg, Hepatitis B virus surface antigen; HBV DNA, Hepatitis B virus deoxyribonucleic acid; LLCC, large liver cell change; SLCC, small liver cell change; pSTAT3, phosphorylated signal transducer and activator of transcription 3; pERK1/2, phosphorylated extracellular signal-regulated kinase 1/2; SYK, spleen tyrosine kinase; PAI-1, plasminogen activator inhibitor-1

Table 2.

Clinicopathological characteristics of the patients with no and late recurrence in the validation cohort

	No recurrence (n = 121)	Late recurrence (after 2 years) (n = 47)	p-value
Clinical characteristics
Age (years, mean ± SD)	57.46 ± 10.02	58.60 ± 10.77	0.542
Sex (male/female)	83 (69%)/38 (31%)	39 (83%)/8 (17%)	0.061
Etiology (HBV/HCV/Alcohol/Unknown)	90 (74%)/7 (6%)/8 (7%)/16 (13%)	35 (75%)/1 (2%)/4 (8%)/7 (15%)	0.755
AFP level (ng/mL, mean ± SD)	1817.64 ± 6236.37	787.04 ± 2066.27	0.060
HBsAg† (positive)	50 (41.3%)	26 (55.3%)	0.102
Tumor number (1/2/3/≥4)	98 (81%)/15 (12%)/6 (5%)/2 (2%)	42 (90%)/3 (6%)/1 (2%)/1 (2%)	0.418
Histological features
Grade (inflammation)
Lobular activity (no/minimal/mild/moderate/severe)	5 (4%)/18 (15%)/98 (81%)/0 (0%)/0 (0%)	1 (2%)/21 (45%)/23 (49%)/2 (4%)/0 (0%)	<0.001*
Porto-periportal activity (no/minimal/mild/moderate/severe)	7 (6%)/38 (31%)/71 (59%)/5 (4%)/0 (0%)	3 (6%)/11 (24%)/26 (55%)/7 (15%)/0 (0%)	0.097
Stage (fibrosis)
Cirrhosis (stage 4)	56 (46%)	32 (68%)	0.011*
Stage (0/1/2/3/4A/4B/4C)	3 (2%)/8 (7%)/24 (20%)/30 (25%)/15 (12%)/35 (29%)/6 (5%)	0 (0%)/1 (2%)/2 (4%)/12 (26%)/13 (28%)/18 (38%)/1 (2%)	0.029*
Liver cell change
Liver cell change (No/LLCC/SLCC/LLCC and SLCC)	39 (32%)/51 (42%)/2 (2%)/29 (24%)	12 (25%)/13 (28%)/1 (2%)/21 (45%)	0.016*
LLCC proportion (%, mean ± SD)	9.67 ± 9.28	14.68 ± 12.57	0.049*
SLCC proportion (%, mean ± SD)	3.14 ± 6.28	5.72 ± 7.94	0.041*
Immunohistochemical marker expression
pSTAT3 (positive/negative)	31 (26%)/90 (74%)	20 (43%)/27 (57%)	0.032*
pERK1/2 (positive/negative)	33 (27%)/88 (73%)	21 (45%)/26 (55%)	0.030*
SYK (positive/negative)	24 (20%)/97 (80%)	11 (23%)/36 (77%)	0.609

^†Some patients had unavailable data. Only patients with complete data were included in the analysis.

SD, standard deviation; HBV, hepatitis B virus; HCV, hepatitis C virus; AFP, Alpha-fetoprotein; HBsAg, Hepatitis B virus surface antigen; LLCC, large liver cell change; SLCC, small liver cell change; pSTAT3, phosphorylated signal transducer and activator of transcription 3; pERK1/2, phosphorylated extracellular signal-regulated kinase 1/2; SYK, spleen tyrosine kinase

2. Protein markers associated with late HCC recurrence

We generated proteomic data from 95 non-tumor liver tissue samples using RPPA, and 7 proteins were significantly correlated with HIR signature probability (p < 0.05): tyrosine kinase (SYK), neurofibromin 2 (NF2), phosphorylated signal transducer and activator of transcription 3 (pSTAT3) (Y705), plasminogen activator inhibitor (PAI)-1, RAB25, phosphorylated estrogen receptor α (pERα) (S118), and 14-3-3ε. Among these, the levels of SYK, NF2, pSTAT3 (Y705), PAI-1, and RAB25 were positively correlated whereas those of pERα (S118) and 14-3-3ε were negatively correlated with HIR signature (Figure 2A, Supplementary Table 4).

Figure 2.

(A) Seven proteins significantly correlated with hepatic injury and regeneration (HIR) signature using reverse phase protein array. (B) Expression of immunohistochemical markers. Hepatocytes in the background non-tumorous liver expressing pSTAT3, pERK1/2, SYK, and PAI-1 (original magnification, ×400). (C) Kaplan–Meier curve of late HCC recurrence-free survival according to the combinations of immunohistochemical protein markers (pSTAT3, pEKR1/2, and SYK) in the training cohort.

pSTAT3, SYK, and PAI-1 showed high AUC values for late HCC recurrence (> 0.7) and were thus selected for IHC analysis. We also examined phosphorylated extracellular signal-regulated kinase 1/2 (pERK1/2), as the MAPK/ERK pathway is a well-known marker of stress response and proliferative activity, and pERK1/2 is known to activate pSTAT3^38,39. The nuclear immunoexpression of pSTAT3 and pERK1/2 and the cytoplasmic expression of SYK and PAI-1 were evaluated (Figure 2B).

In the training cohort, the expressions of pSTAT3, pERK1/2, and PAI-1 in the background liver were higher in those with late HCC recurrence than in those with no recurrence (p < 0.05 for all) (Table 1). In the validation cohort, the expressions of pSTAT3 and pERK1/2 in the background liver were significantly higher in those with late HCC recurrence than in those with no recurrence (p = 0.032 and 0.030, respectively) (Table 2).

3. Development and validation of a predictive model for late HCC recurrence

3.1. Selection of significant risk factors

Univariate Cox regression analysis showed that tumor number; lobular inflammatory activity; porto-periportal inflammatory activity; cirrhosis; LCC; and IHC expression of pSTAT3, pERK1/2, and SYK significantly increased the risk of late HCC recurrence (p < 0.05 for all) (Table 3). Of these, cirrhosis and lobular activity were retained as histologic parameters and pERK1/2 and SYK were statistically significant as IHC markers via multiple Cox regression analysis (p < 0.05 for both) (Table 3). The hazard ratios (95% CI) of pSTAT3, pERK1/2, and SYK by multivariate analyses were 1.6 (0.9–2.8), 1.8 (1.1–3.0), and 1.7 (1.0–2.7), respectively (Table 3), showing similar levels of increasing risk for late HCC recurrence. The various combinations of pSTAT3, pERK1/2, and SYK showed differences in predictive capability for late HCC recurrence on log-rank test (all-negative vs. one-positive, p = 0.005; all-negative vs. two-positive, p < 0.001; all-negative vs. three-positive, p < 0.001; one-positive vs. two-positive, p = 0.405; one-positive vs. three-positive, p = 0.616; two-positive vs. three-positive, p = 0.034) (Figure 2C).

Table 3.

Univariate and multivariate Cox proportional hazards regression analyses for the predictors of late HCC recurrence in the training cohort (n = 402)

Parameters		Univariate analysis		Multivariate analysis
				[Model 1] Histological parameters only		[Model 2] Histological parameters + individual weighted IHC markers		[Model 3] Histological parameters + IHC marker combinations
		HR (95% CI)	p-value	HR (95% CI)	p-value	HR (95% CI)	p-value	HR (95% CI)	p-value
Clinical characteristics
Age (ref ≤ 60 years)	≥ 60	0.8 (0.5–1.3)	0.358
Sex (ref = male)	Female	0.7 (0.4–1.4)	0.375
Etiology (ref = HBV)	HCV	2.9 (1.2–6.7)	0.014*
	Alcohol	0.4 (0.1–3.2)	0.416
	Unknown	0.6 (0.2–1.7)	0.347
AFP level (ng/mL) (per 1000 unit)		0.97 (0.9–1.0)	0.253
HBsAg† (ref = negative)	Positive	1.2 (0.6–2.3)	0.551
HBsAg level† (IU/mL) (per 100 unit)		0.96 (0.9–1.0)	0.081
HBV DNA level† (IU/mL) (per 100000 unit)		0.99 (0.9–1.0)	0.671
Tumor number (per 1 unit)		1.93 (1.1–3.4)	0.023*	1.5 (0.9–2.8)	0.147	1.4 (0.8–2.6)	0.241	1.4 (0.8–2.6)	0.238
Histological features
Grade (inflammation)
Lobular activity (ref = no, minimal, mild)	Moderate, severe	31.9 (6.7–150.6)	<0.001*	22.4 (4.7–107.0)	<0.001*	19.4 (3.9–95.6)	<0.001*	21.1 (4.3–102.7)	<0.001*
Porto-periportal activity (ref = no, minimal, mild)	Moderate, severe	2.0 (1.04–3.6)	0.036*
Stage (fibrosis)
Cirrhosis (ref = fibrosis stage < 4)	≥ stage 4	2.3 (1.4–3.7)	<0.001*	2.2 (1.4–3.6)	0.001*	2.0 (1.2–3.3)	0.005*	2.0 (1.2–3.3)	<0.001*
Liver cell change
Liver cell change (ref = no)	LLCC only	1.03 (0.6–1.9)	0.914
	SLCC only	2.3 (0.3–17.5)	0.412
	LLCC and SLCC	1.9 (1.0–3.4)	0.038*
IHC marker expression
pSTAT3 (ref = negative)	Positive	2.3 (1.3–3.8)	0.003*			1.6 (0.9–2.8)	0.115
pERK1/2 (ref = negative)	Positive	2.3 (1.5–3.7)	<0.001*			1.8 (1.1–2.9)	0.016*
SYK (ref = negative)	Positive	1.8 (1.1–2.9)	0.022*			1.7 (1.1–2.8)	0.035*
PAI-1(ref = negative)	Positive	1.6 (0.9–2.6)	0.073
Combination of 3 IHC markers	One-positive	2.1 (1.2–3.7)	0.009*					2.0 (1.1–3.4)	0.018*
(pSTAT3 or pERK1/2 or SYK)	Two-positive	3.4 (1.9–6.3)	<0.001*					2.7 (1.4–5.1)	0.002*
(ref = all negative)	Three-positive	5.1 (1.8–14.8)	0.003*					6.0 (2.1–17.7)	0.001*

^†Some patients had unavailable data. Only patients with complete data were included in the analysis.

HCC, hepatocellular carcinoma; HR, Hazard ratio; CI, confidence intervals; HBV, hepatitis B virus; HCV, hepatitis C virus; AFP, Alpha-fetoprotein; HBsAg, Hepatitis B virus surface antigen; HBV DNA, Hepatitis B virus deoxyribonucleic acid; LLCC, large liver cell change; SLCC, small liver cell change; IHC, immunohistochemical; pSTAT3, phosphorylated signal transducer and activator of transcription 3; pERK1/2, phosphorylated extracellular signal-regulated kinase 1/2; SYK, spleen tyrosine kinase; PAI-1, plasminogen activator inhibitor-1.

3.2. Selection of the predictive model with the most discrimination and prediction improvement

Model 2 and Model 3 showed a higher predictive capability than Model 1 in the training cohort dataset (differences in Heagerty’s iAUC between Model 2 and Model 1 = 0.075 and between Model 3 and Model 1 = 0.074; p < 0.05 for both). Model 2 showed a slightly higher discriminatory capability than Model 3 (Harrell’s C index of Model 2 = 0.704, 95% CI = 0.645–0.763; Heagerty’s iAUC of Model 2 = 0.679, 95% CI = 0.620–0.738; Harrell’s C index of Model 3 = 0.701, 95% CI = 0.642–0.760; Heagerty’s iAUC of Model 3 = 0.678, 95% CI = 0.619–0.737). However, the differences were not significant (differences in Harrell’s C index between Model 2 and Model 3 = −0.003 [CI = −0.023–0.017], p = 0.764; differences in Heagerty’s iAUC between Model 2 and Model 3 = −0.001 [CI = −0.026–0.024], p = 0.939) (Figure 3A and Supplementary Table 5).

Figure 3.

Discrimination of the nomogram with the comparison of time-dependent areas under the receiver operating characteristic curve (iAUC). (A) Discrimination of the nomogram in the training cohort. Model 2 and Model 3 showing higher predictive value than Model 1. (B) Similar predictive values in the validation cohort.

Adding individual weighted IHC markers (Model 2) or IHC marker combinations (Model 3) to the base model (Model 1) yielded significant positive NRI values for 3- and 5-year late HCC recurrence. Model 3 showed a higher improvement in predictive capability (NRI of Model 2 = 0.280 and 0.207, NRI of Model 3 = 0.401 and 0.250, p < 0.05 for all). IDI exhibited significant positive values in Models 2 and 3 compared with that in Model 1 for 5-year late HCC recurrence (0.045 and 0.048, respectively; p < 0.05 for both) (Supplementary Table 6).

3.3. Visualization using a nomogram and calibration

Model 3 (histological parameters and IHC marker combinations) was selected as the final predictive model, and the risk factors were proportionally assigned points on a scale of 0–100 in the nomogram: 23 points for cirrhosis; 100 points for moderate or severe lobular inflammatory activity; and 23, 34, and 58 points for the number of expressed IHC markers (Figure 4 and Supplementary Table 7). The equation S(t, X) = [S₀(t)]^exp(LP) was generated based on the results of the multiple Cox proportional hazards regression analysis of Model 3. The letters in the equation represent the following: S = score; t = time (3- or 5-year); X = three variables associated with late recurrence (cirrhosis, lobular activity, and IHC marker combinations); S₀(t) = constant value for 3- or 5-year prediction [S₀ (3) = 0.9395265; S₀ (5) = 0.8375704]; LP, linear predictor = $\sum _{\mathrm{i}=1}^{\mathrm{p}}{\mathrm{\beta }}_{\mathrm{i}}\times ({\mathrm{x}}_{\mathrm{i}}-\stackrel{-}{\mathrm{x}})$; LP = 0.7054 × (a − 0.4777778) + 3.0500 × (b − 0.007407407) + 0.6889 × (c1 − 0.3444444) + 1.0328 × (c2 − 0.1555556) + 1.7651 × (c3 − 0.02592593); a = cirrhosis (0, no cirrhosis; 1, cirrhosis); b = lobular activity (0, no or minimal or mild; 1, moderate or severe); c1 = one positive marker (0, all-negative; 1, one positive); c2 = two positive markers (0, all-negative or one positive; 1, two positive); and c3 = three positive markers (0, all-negative or two positive; 1, three positive).

Figure 4.

Nomogram of the predictive model (Model 3) for the 3- and 5-year risk of late HCC recurrence.

The nomogram comprised seven rows. The first row involved point assignment. The following three rows were assigned to cirrhosis, lobular inflammatory activity, and IHC marker combinations. The total number of points (row 5) was the sum of all the assigned points for three variables. The predictive probability for 3- and 5-year late HCC recurrence (rows 6 and 7) was obtained by drawing a vertically matched point from the total point (Figure 4).

The calibration plot of the model represented a parallel smooth line close to the reference (diagonal) line in the training cohort, suggesting high agreement between the predicted and observed probabilities of 3- and 5-year HCC recurrence (Figure 5A).

Figure 5.

Calibration plots of the predictive model (Model 3) for the predictive capability of the nomogram. (A) The calibration plot in the training cohort showed that it reliably predicted 3- and 5-year recurrence of HCC, as shown by its parallel position close to the diagonal (ideal) line. (B) A similar parallel line close to the reference line is observed for the validation cohort (gray, ideal line; black, observed line).

3.4. Validation

Model 3 showed the best predictive capability for the validation cohort (Harrell’s C index = 0.716, 95% CI = 0.643–0.789; Heagerty’s iAUC = 0.704, 95% CI = 0.630–0.778). (Figure 3B and Supplementary Table 5). The IDI of Model 3 for the 5-year late HCC recurrence had a significant positive value (0.060, p = 0.010) (Supplementary Table 6). The calibration plot represented good agreement with a parallel line close to the reference line of 3- and 5- year probability in the validation cohort (Figure 5B).

3.5. Predictive capability of the model for early recurrence

To investigate whether the models could predict early HCC recurrence, we assessed the discrimination and calibration plot for early HCC recurrence in the validation cohort (Supplementary Table 8). For Model 3, the Harrell’s C index was 0.567 (95% CI = 0.516–0.618) and Heagerty’s integrated AUC was 0.566 (95% CI = 0.519–0.613), showing lower predictive capability for early HCC recurrence than that for late HCC recurrence (Supplementary Figure 2A and Supplementary Table 8). The calibration plot was not close or parallel to the diagonal line (Supplementary Figure 2B) and did not fit well with the predictive model.

Discussion

We constructed a predictive model for the 3- and 5-year probability of late HCC recurrence based on histologic features and IHC markers of the background non-neoplastic liver. Moderate or severe lobular inflammatory activity and cirrhosis (stage 4) were independent histologic parameters for late HCC recurrence in this study. Background liver inflammation and fibrosis have been reported to be risk factors of late HCC recurrence^3,4,10. Interestingly, lobular inflammatory activity, rather than porto-periportal inflammatory activity, was a risk factor for late HCC recurrence in this study, indicating that persistent lobular necroinflammation destroys and further accelerates the regeneration and accumulation of genetic alterations in hepatocytes, promoting hepatocarcinogenesis¹.

A previous analysis of HIR signatures identified 233 differentially expressed genes in hepatic injury and regeneration of human liver tissue to be a strong genomic predictor for late HCC recurrence²⁰. In this study, seven proteins [SYK, NF2, pSTAT3 (Y705), PAI-1, RAB25, pERα (S118), and 14-3-3ε] were found to be significantly correlated with the HIR signature, using RPPA. Among them SYK, pSTAT3 (Y705), and PAI-1 had high AUC values for late HCC recurrence and were selected for IHC evaluation. pERK1/2 was added to further evaluate pSTAT3 activity. STAT3 has been identified as the central gene within the HIR signatures through gene network analysis, and its expression is increased in HCC patients with late recurrence²⁰. STAT3, usually activated with phosphorylation at tyrosine 705 (Y705) by Janus family tyrosine kinase (JAK), upregulates cytokine expression such as that of interleukin-6 by acting as a transcription factor and promotes carcinogenesis^40,41. pERK1/2 can phosphorylate STAT3 at serine 727 (S727) because it has serine/threonine kinase activity and might interact with the JAK/STAT pathway in this way. The role of pSTAT3(S727) is controversial^39,42, but it might play an important role in cellular growth control and even oncogenesis in bladder cancer and melanoma^43,44 and in earlier stages of hepatocarcinogenesis⁴⁵. The stress-activated MAPK cascade also plays an important role in hepatocarcinogenesis from chronic hepatitis⁴⁶. SYK modulates inflammation and fibrosis during cross-talk between stellate cells and hepatocytes by increasing the level of cytokines^47,48. In this study, IHC evaluation revealed that pSTAT3, pERK1/2, and SYK significantly increased the risk of late HCC recurrence.

We initially built three models for predicting late HCC recurrence, and we assumed that the predictive capability could be increased when IHC markers were added (Model 2 and Model 3) compared to that of the baseline model with histological features only (Model 1). This was based on a higher Harrell’s C index and Heagerty’s iAUC and on positive values of NRI and IDI. Model 2 (histological parameters + individual weighted IHC markers) and Model 3 (histological parameters + IHC marker combinations) showed similar discriminative capabilities. We performed additional analyses of NRI and IDI for identifying the improvement in predictive performance in each model. Model 3 showed higher improvement in predictive capability compared with Model 2 based on the NRI. Moreover, the Cox regression hazard ratio of each IHC marker was similar. Further, compared with individual weighted IHC markers, a combination of IHC markers might yield a higher predictive accuracy for outcomes because a combination of IHC markers compensates for the association among different carcinogenic pathways by avoiding multicollinearity among markers^49,50. Accordingly, Model 3 was selected, and a nomogram was developed for visual explanation of the prediction of late HCC recurrence. Calibration plots for 3- and 5-year late HCC recurrence showed optimal agreement between the predictions and actual observations. Thus, we confirmed that the nomogram based on Model 3 had good performance with accurate and reliable predictive capability.

In the validation cohort, Model 3 had high predictive capability, and the calibration plot showed good agreement. The Harrell’s C index and Heagerty’s iAUC were higher in the validation cohort than those in the training cohort were. In general, the C index or iAUC of the validation cohort is lower than that of the training cohort because the model is fit with the training cohort. However, a higher C index and iAUC have been observed in the validation cohort in rare cases where variables would be more fit in the validation cohort. A relatively shorter follow-up period could have resulted in the higher predictive capability of the model in the validation cohort. Therefore, further evaluation is needed using a larger validation cohort and longer follow-up period to obtain better results.

Model 3 revealed a lower predictive capability for early recurrence than for late recurrence in the validation cohort; the calibration plot was not close or parallel to the diagonal line and did not fit well with the model. This result was expected, as early HCC recurrence is associated with the moleculo-pathological factors of tumors, including tumor size, microvascular invasion, and satellite nodules. In contrast, our model was developed using moleculo-pathological factors of non-tumorous liver tissue.

To our best knowledge, this is the first predictive model using histological features and IHC markers for late HCC recurrence. We could not consider all histological parameters and used only several parameters to avoid multi-collinearity. Thus, more predictive markers and models should be developed.

In conclusion, the combinations of the IHC markers pSTAT3, pERK1/2, and SYK in addition to histological features of high lobular inflammatory activity and cirrhosis can predict the risk of late HCC recurrence. The model developed based on these parameters showed good predictive capability for late HCC recurrence. The established nomogram can be useful for predicting the risk of late HCC recurrence and establishing an individualized treatment and follow-up plan after curative resection of HCC.

Abbreviations

HCC

Hepatocellular carcinoma

pSTAT3

Phosphorylated signal transducer and activator of transcription 3

pERK1/2

Phosphorylated extracellular signal-regulated kinase 1/2

SYK

Spleen tyrosine kinase

LCC

Liver cell change

HIR

Hepatic injury and regeneration

IHC

Immunohistochemical

SLCC

Small liver cell change

LLCC

Large liver cell change

AFP

Alpha-fetoprotein

HBsAg

Hepatitis B virus surface antigen

HBV DNA

Hepatitis B virus deoxyribonucleic acid

RPPA

Reverse phase protein array

AUC

Area under the receiver operating characteristic curve

iAUC

integrated area under the receiver operating characteristic curve

CI

Confidence intervals

NRI

Net reclassification improvement

IDI

Integrated discrimination improvement

NF2

Neurofibromin 2

PAI-1

Plasminogen activator inhibitor-1

pERα

Phosphorylated estrogen receptor α