Genetic alterations are related to clinicopathological features and risk of recurrence/metastasis of hepatocellular carcinoma

Abstract

Lack of efficient biomarkers and clinical translation of molecular typing impedes the implementation of targeted therapy for hepatocellular carcinoma (HCC). High-throughput sequencing techniques represented by next-generation sequencing (NGS) are tools for detecting targetable genes. The objective of this study is to explore the genetic alterations associated with clinicopathological features and the risk of recurrence/metastasis in HCC. NGS analysis was conducted on formalin-fixed paraffin-embedded tissues from 164 resected liver samples obtained from Chinese patients. Morphologic subtypes were reviewed based on hematoxylin-eosin and immunohistochemistry staining, Correlation to the acquired molecular features were analyzed with clinicopathological information. We also retrieved follow-up information of the 123 transplanted cases from 2017 to 2019 to screen recurrence/metastasis-associated factors by univariate analysis. Generally, the most frequently mutated genes include TP53 and CTNNB1 which showed a trend of mutually exclusive mutation. Copy-number variant with the highest frequency was detected in TAF1 and CCND1 in 11q13.3 loci. Correlation analysis showed that various genetic alterations were associated with morphologic subtypes and other pathologic features. While gene signatures of proliferation/nonproliferation class were correlated with differentiation, satellite foci and other invasive morphological features. Macrotrabecular-massive subtype, TSC2 (tuberous sclerosis complex 2) mutation, Ki-67 expression, and other six factors were found to be associated with recurrence/metastasis after liver transplantation. Genetic alterations detected by NGS show correlation with not only pathological and clinical features, but also with recurrence/metastasis after liver transplantation. Further gene-level molecular typing will be practical for targeted therapy and individual recurrence risk assessment in HCC patients.

Introduction

Over the past 30 years, liver cancer has been causing a huge disease burden worldwide, with the incidence and the mortality rate ranking seventh and third among all cancer types in the year of 2018 (Sia et al., 2017, Orcutt and Anaya, 2018). Moreover, a global disease report shows that between 2007 and 2017, liver cancer rose from third to second place in terms of the absolute year of lives lost caused by cancer (GBD 2015 Mortality and Causes of Death Collaborators, 2016).

Hepatocellular carcinoma (HCC) constitutes over 90% of primary liver cancer, for which surgical resection remains the main treating strategy. In recent years, in order to prolong the overall survival (OS) of HCC patients, especially those in advanced stage when surgical resection is no longer feasible, targeted drugs like tyrosine kinase inhibitors, and immune checkpoint inhibitors like PD-1/PD-L1 antibodies have been studied and tested as potential treating options. However, the results of clinical trials based on single or combined targeted therapy are seldom satisfactory (Qiao et al., 2022), which is partially due to the lack of biomarkers for target population screening and treatment effect predicting. In order to better serve the need of precisive diagnosis and treatment, a lot of researches rely on sequencing methods to study the molecular changes of HCC from genomic to transcriptomic level. A consensus conclusion reached among the studies based on transcriptome is that HCC can be roughly dichotomized into proliferation and nonproliferation subtypes (Li et al., 2017, Calderaro et al., 2019), with molecular changes in cell proliferation-related pathways and WNT/β-catenin signaling pathway, respectively, corresponding to two types of cases with poor and good survival.

As next-generation sequencing (NGS) is gradually being utilized for tumor research and clinical diagnosis for its cost-effective, timesaving, and accurate qualities (Morash et al., 2018, Morishita et al., 2018, Dominguez and Wang, 2020, Ke et al., 2021), related studies have revealed the roles of important gene mutations, such as TERT (telomerase reverse transcriptase) promoter, RB1, FGF19, ARID1A, TSC1/2, RPS6KA3, and KEAP1 in the occurrence and recurrence of HCC (Cancer Genome Atlas Research Network, 2017, Calderaro et al., 2019, Rebouissou and Nault, 2020, Demory and Nault, 2020). These genetic alterations serve as potential markers for patient stratification and targeted drug selection, and are also summarized into 10 driver pathways in the newly updated 5th edition of WHO tumor classification. However, the way of interpreting these findings still remains a difficulty. Combining vital pathological features, such as morphological subtypes with molecular alterations to form a molecular-morphologic typing (Niu et al.. 2016, Calderaro et al., 2017) may be a solution to propel this translational process. An exemplary study by Calderaro et al. supplements morphological subtypes, differentiation, and results of immunohistochemistry staining into the G1–G6 molecular typing first proposed by Boyault et al. (2007) and Calderaro et al. (2019), which partially reveals the molecular mechanism lied behind morphologic features. Some of these molecular annotations for morphologic subtypes are also introduced in the new WHO tumor classification, especially those for the rare subtypes such as macrotrabecular-massive (MTM) subtype and cirrhotic subtype.

The clinicopathological relevance of HCC molecular alterations awaits more researches to verify and discover, and the practical application value of NGS in HCC molecular diagnosis also needs to be carefully evaluated. Therefore, we performed NGS on 164 transplanted and resected HCC samples and interpreted the pathological and clinical relevance of the detected molecular alterations; we also analyzed the prognostic information to screen the molecular factors of early recurrence/metastasis, thus expanding the pretransplantation evaluation criteria from a molecular perspective.

Methods

Case retrieval and sample preparation

This was a hospital-based retrospective study conducted in Shanghai, China, which included patients from a range of provinces across the country. The following criteria were used to determine which cases were included in the study: (1) cases with a pathological diagnosis of HCC; (2) cases who underwent liver transplantation or partial hepatectomy according to the Chinese Management Criteria of Hepatocellular Carcinoma; (3) cases who did not undergo radiotherapy or chemotherapy; (4) cases who were aware of the objectives of the study and provided informed consent; (5) cases who were capable of completing the necessary surveys and questionnaires; (6) cases who had available cancer tissue samples and clinicopathological data.

The following cases were excluded: (1) cases with HCC that had not been confirmed by histopathological examination; (2) cases that had received chemotherapy or radiotherapy prior to surgical treatment; (3) cases that had been rejected, discontinued, or lost information. Between January 2017 and December 2019, Zhongshan Hospital of Fudan University recruited a total of 164 patients with HCC that met the aforementioned case inclusion and exclusion criteria. Of these, 123 were liver transplant patients and 41 were partial hepatectomy patients.

The hematoxylin-eosin (HE) and immunohistochemistry sections as well as one or more blocks of formalin-fixed paraffin-embedded (FFPE) tissue with sufficient tumor tissue and representative morphologic features were retrieved. Clinical-pathological information as well as follow-up information were extracted from digital medical record system including: pathological information such as numbers and diameters of lesions according to macroscopic record, general clinical information such as gender, age, date of surgery, pre-operation serum alpha-fetoprotein (AFP) level and of hepatitis B virus HBV/HBV infection, and follow-up information such as date of surgery and recurrence/metastasis.

Pathological examination of HE and immunohistochemistry sections

Microscopy of H&E and immunohistochemistry sections was performed by two pathologists independently. Pathological characteristics of all HE sections were recorded, including morphologic subtypes of each H&E section (macrotrabecular subtype, microtrabecular subtype, pseudoglandular subtype, compact subtype, steatohepatitis subtype, clear cell subtype, as well as special subtypes such as cirrhotic subtype, fibrolamellar subtype and lymphoepithelioma-like subtype). The composition ratio (%) of each subtype in a single case was calculated as a weighted average of all the sections with lesions of the case. Morphological components with a ratio >30% were regarded as existential components, among which those with a ratio >50% were regarded as dominant components. Microvascular invasion (MVI) was evaluated as M0–M2. The presence of satellite foci was also recorded.

The results of immunohistochemistry staining were recorded as follows: for PD-1/PD-L1 (PD-L1 antibody’s clone-number: 28-8), the positive expression rate of both tumor cells and inflammatory cells were recorded; for Ki-67, only the positive expression rate of tumor cells was recorded. The rates were then converted to a 0–2 classification variable: for PD-1/PD-L1, a positive rate of 0% was recorded as 0, one between 1 and 5% was recorded as 1, and one >5% was recorded as 2; for Ki-67, a positive rate ≤10% was recorded as 0, one between 11 and 49% was recorded as 1, and one >50% was recorded as 2. The stainings of AFP, CK7, CK19, Arg-1, GPC3, GS, Hepa, and Hsp70 were recorded as negative (0) or positive (1) using a binary classification, with a tumor cell positive staining rate <5% recorded as negative and one ≥5% as positive.

Next-generation sequencing and processing of detected alterations

NGS was performed to the FFPE tissue of 164 cases using a panel of 508 genes, and each targeted gene region was detected with a sequencing coverage >99.8% and a depth >568×. Single number variation (SNV), insertion or deletion (INDEL), and copy-number variant (CNV) of tumor tissue DNA were determined. In order to observe and analyze the HCC driver molecular alterations, we summarized 10 major HCC-related pathways (Fig. 1) according to the molecular typing established by Zucman-Rossi et al., the 5th edition of WHO tumor classification, and conclusions of other previous studies (Morishita et al., 2018), and counted the alterations of genes in each pathway; mutations of genes in six DNA damage repair (DDR)-related pathways were included for further correlation analysis; given that molecular mechanisms of HCC have not been fully studied, 19 gene mutations with a mutation rate >5% which lacked annotations from previous researches or public datasets were included; 39 genes of which CNV were detected were also included.

Fig. 1.

Genes with SNV or INDEL frequency greater than 5% in HCC. HCC, hepatocellular carcinoma; INDEL, insertion or deletion; SNV, single number variation.

Statistical analysis

Correlation analysis of genetic alterations and clinical-pathological data was performed using R (version 3.6.0; R Foundation for Statistical Computing, Vienna, Austria, https://www.R-project.org/): Wilcoxon rank sum test was used for ordered categorical variables; χ² test, adjusted χ² test, and Fisher’s exact test were used for disordered categorical variables. For follow-up information, univariate and multivariate Cox proportional hazard ratio regression models were applied sequentially. All hypothesis tests were two-sided, and P value <0.05 was the criterion for statistical difference.

Results

General features of clinical and pathological variables

Among the 164 cases included in the study, 123 (75.0%) cases underwent liver transplantation and 41 (25.0%) cases underwent partial liver resection. The cases show a man-dominant (89.0%) and hepatitis virus HBV infection-dominant (96.9%) trend, which is consistent with the epidemiology of HCC in China. Ages range between 30 and 81 years old with a median age of 54 years old. Preoperative AFP serum level did not exceed 20 ng/ml in almost half of the cases, while one third of cases reached a level over 200 ng/ml.

The numbers of lesions are mostly between 1 and 3 (148/160, 92.5%) according to the macroscopic record, among which single-nodule cases constitute almost half (75/160, 46.9%); there are 11 cases (6.8%) with more than 100 nodules which are considered as diffuse subtype. Edmondson–Steiner grades of all cases are II or III, while no cases are I and IV; and there are more grade III cases than grade II (57.7% vs 42.3%). MVI was detected in 64.6% of the cases, of which over half of the cases were classified as M2. Satellite foci was found in 16.9% of the cases. As for morphologic subtypes, macrotrabecular subtype, microtrabecular subtype, pseudoglandular subtype, compact subtype, steatohepatitis subtype and clear cell subtype were detected (case number sorted from large to small) with no rare subtypes detected; microtrabecular subtype is the most common morphologic subtype (57/116, 46.0%); the number of macrotrabecular subtype-dominant cases ranks second (37/116, 29.8%), which comply the criteria of the newly defined MTM subtype; microtrabecular subtype and MTM subtype are distributed in most cases (75.8%), and the two show a mutually exclusive trend (P < 0.01); cases dominant in the other four morphologic subtypes constitute less than 20% of total cases.

Immunohistochemistry staining of PD-1/PD-L1 in most cases were negative in tumor cells (87.3% and 93.7%, respectively), and less than 5% of cases had a positive rate over 5%. The majority of cases showed a positive rate of inflammatory cells between 1 and 5%; in addition, correlation between PD-L1 expression of tumor cells and inflammatory cells (P = 0.01), and between the expression of PD-1 and PD-L1 of inflammatory cells (P < 0.01) were statistically significant. Ki-67 positive rate of nearly half of the cases falls between 11 and 49%, with those ≤10% and ≥50% accounting for 21.0% and 24.7%, respectively; 76.0, 70.9, and 89.4% of the cases showed a negative AFP, CK7, and CK19 staining (Supplementary Table 1, Supplemental digital content 1, http://links.lww.com/EJCP/A523).

Genetic alterations detected by next-generation sequencing

Among these genes, there were 28 genes with SNV or INDEL frequency greater than 5%. Top 10 genes with the highest mutant frequency in ascending order were TP53, KMT2D, CTNNB1, KMT2C, ARID1A, AXIN1, OTOP1, ABCA13, LRP1B, and JAK1; top 10 genes with the highest number of mutant cases in ascending order were TP53, CTNNB1, KMT2D, ARID1A, KMT2C, AXIN1, OTOP1, ABCA13, JAK1, and LRP1B.

We specifically focused on 10 HCC-related pathways, 6 DDR pathways, and 19 other single-gene mutations with mutation frequency greater than 5% (see Table 1 for mutations in all pathways and genes). Although at least one gene mutation was detected by our panel in all cases, the mutations we focus on were not detected in 10 cases (6.1%).

Table 1.

Clinical and pathological features of HCC patients

Variables	Case number	Value	Frequency	Percentage (%)
Clinical variables
Sex	164	Male	146	89
		Female	18	11
Age	164	30–40 years	16	9.8
		40–60 years	106	64.6
		>60 years	42	25.6
Types of surgery	164	Transplantation	123	75
		Resection	41	25
AFP (pre-operation)	156a	≤20 ng/ml	77	49.4
		>20 ng/ml ≤200 ng/ml	27	17.3
		>200 ng/ml	52	33.3
Etiology	163b	HBV	158	96.9
		Alcohol	2	1.2
		HCV	3	1.8
Pathological variables
Morphology	124c	Macro	37	29.8
		Micro	46	37.1
		Pseudoglandular	5	4
		Compact	6	4.8
		Clear cell	7	5.6
		Steatohepatitis-like	3	2.4
Number of nodule(s)	160d	1	75	46.9
		2	30	18.8
		3	42	26.3
		>3	13	8.2
Satellite foci	160d	Present	27	16.9
		Absent	133	83.1
Edmondson–Steiner grade	129e	II	52	42.3
		III	77	57.7
MVI	127f	M0	45	35.4
		M1	34	26.8
		M2	48	37.8

AFP, alpha-fetoprotein; HBV, hepatitis B virus; HCV, hepatitis C virus; HCC, hepatocellular carcinoma; HE, hematoxylin-eosin; MVI, microvascular invasion.

^aEight cases without preoperative serum AFP test.

^bOne case excludes HBV/HCV infection and drinking history, clinically suspected as autoimmune hepatitis.

^cHE sections of 40 cases are unavailable.

^dNumbers of lesions and satellite foci of four cases are missing in the pathological system.

^eHE sections of 35 cases are unavailable, and differentiation is not recorded in the pathological system.

^fHE sections of 37 cases are unavailable, and MVI is not recorded in the pathological system.

A total of 144 cases (87.8%) had at least one mutation in 10 HCC-related pathways, and there were 54 (32.9%), 45 (27.4%), 31 (18.9%), 6 (3.7%), and 8 (4.9%) cases detected of mutations of one to five pathways respectively. Among the 10 pathways, cell cycle-related pathway was the most frequently mutated pathway (102/164, 62.2%). TP53 in this pathway is the most frequently mutated gene (79/164, 48.2%) among all detected genes, and 13.9% of these mutations are hotspot mutation related to aflatoxin B exposure (R249S). The mutation frequency of chromosome remodeling pathway ranks second (44/164, 26.8%), of which ARID1A is the most frequently mutated gene (18/164, 11.0%). WNT/β-catenin pathway mutated cases constitute 26.2% (43/164), and CTNNB1 ranks first in this single pathway and second among all detected genes with a mutation frequency of 15.9% (26/164). The top two mutated genes, TP53 and CTNNB1, bear a tendency of mutually exclusive mutation (P = 0.006). In addition, other genes of the 10 pathways whose mutation frequency reaches 10% include AXIN1 (17/164, 10.4%), MLL4 (22/164, 13.4%) and MLL3 (18/164, 11%). According to previous studies, mutations of Notch pathway, AKT/mTOR pathway, and copy number gain of CCND1/FGF19 belong to proliferation class signature, and mutations of WNT/β-catenin pathway (except for AXIN1 mutation) belong to nonproliferation class signature. Cases which can be classified as these two classes account for 33.5% (55/164) and 13.4% (22/164), respectively. Correlation tests between pathways show that cell cycle-related pathway mutation is statistically correlated with Notch and JAK/STAT pathway mutation (P = 0.03 and 0.05, respectively), while hepatocyte metabolism-related pathway mutation is statistically related to histone modification and JAK/STAT pathway (P = 0.01 and P < 0.01 respectively).

Mutations of 23 genes in six DNA damage response and repair gene pathways, namely checkpoint factors (CPF), fanconi anemia (FA), non-homologous end joining (NHEJ), homologous recombination repair (HRR) gene, base excision repair, and missmatch repair (MMR) gene, were detected in 2–6% of all cases. Five cases are mutated in two of the pathways and one case is mutated in three of them (CPF, MMR, and FA); no case is mutated in two or more genes in one single pathway. In these pathways, mutation frequency of a single gene does not exceed 3%, with the most frequently mutated ones being ATM and PARP1 (4/164, 2.4%); more than 50% of these genes (12/23) are detected of only one mutated case (Fig. 2).

Fig. 2.

Gene mutation frequencies in HCC. (a) Mutation frequencies of 10 HCC-related pathways and 6 DDR pathways. (b) Gene mutation frequencies in cell cycle-related pathway. (c) Gene mutation frequencies in WNT/β-catenin pathway. DDR, DNA damage repair; HCC, hepatocellular carcinoma.

Furthermore, we also focus on 19 genes that do not belong to HCC-related pathways or DDR pathways and have not been investigated by previous studies (Fig. 3). All of these genes bear a relatively high mutation frequency >5% (i.e. the number of mutated cases is >8).

Fig. 3.

Mutation frequencies of 19 uncategorized genes.

A total of 48 cases out of 164 cases are detected of CNV of at least one gene, and one of the cases bear CNV of 16 genes simultaneously. Among the genes detected of CNV, TAF1 is the most frequent (20/164, 12.2%), followed by CCND1 (12/164, 7.3%), while the other genes have a CNV frequency between 0.6 and 3.0%. The distribution of CNV on chromosomal arms is as follows: 8q has the largest number of genes involved (13 genes in total, with RSPO2 and ZFPM2 both located at 8q23.1; ATAD2 and NSMCE2 both located at 8q24.13), while Xq has the largest number of cases involved (21 cases in total, genes including TAF1, FLNA, GPC3, and ZIC34). A total of seven cases are detected with CNV at 11q13.3 loci (four genes including FGF3/4/19 and CCND1), and another five cases are detected with only CCND1 amplification (Table 2).

Table 2.

CNV frequency distribution on chromosomal arms, loci, and genes

Chromosome arm	Total case number	Locus	Gene	CNV	Frequency	Percentage
1q	3	1q21.2	MCL1	Gain	2	1.2%
		1q23.3	DDR2	Gain	1	0.6%
3p	1	3p22.1	CTNNB1	Gain	1	0.6%
5p	2	5p15.33	TERT	Gain	2	1.2%
6p	2	6p21.1	CCND3	Gain	2	1.2%
		6p21.1	VEGFA	Gain	1	0.6%
8q	7	8q11.23	PCMTD1	Gain	1	0.6%
		8q12.2	CHD7	Gain	1	0.6%
		8q13.2-q13.3	SULF1	Gain	2	1.2%
		8q21.13	TPD52	Gain	2	1.2%
		8q21.3	WWP1	Gain	2	1.2%
		8q22.3	FZD6	Gain	2	1.2%
		8q23.1	RSPO2	Gain	2	1.2%
		8q23.1	7FPM2	Gain	2	1.2%
		8q23.1-23.2	PKHD1L1	Gain	1	0.6%
		8q23.3	CSMD3	Gain	2	1.2%
		8q24.13	ATAD2	Gain	1	0.6%
		8q24.13	NSMCE2	Gain	1	0.6%
		8q24.21	MYC	Gain	4	2.4%
9p	2	9p21.3	CDKN2A	Loss	2	1.2%
		9p21.3	CDKN2B	Loss	1	0.6%
10p	1	10p15.2	KLF6	Gain	1	0.6%
11q	12	11q14.1	GAB2	Gain	1	0.6%
		11q13.3	CCND1	Gain	12	7.3%
		11q13.3	FGF19	Gain	7	4.3%
		11q13.3	FGF3	Gain	7	4.3%
		11q13.3	FGF4	Gain	7	4.3%
12q	2	12q14.1	CDK4	Gain	2	1.2%
13q	2	13q14.2	RB1	Loss	2	1.2%
14q	1	14q21.1	FOXA1	Gain	1	0.6%
20q	1	20q13.32	GNAS	Gain	1	0.6%
Xq	21	Xq28	FLNA	Loss	2	1.2%
		Xq13.1	TAF1	Loss	20	12.2%
		Xq26.2	GPC3	Loss	1	0.6%
		Xq26.3	ZIC3	Loss	1	0.6%
Xp	4	Xp21.1-21.2	DMD	Loss	1	0.6%
		Xp22.12	RPS6KA3	Loss	2	1.2%
		Xp22.2	GPR143	Loss	4	2.4%
		Xp22.31	VCX	Loss	4	2.4%

CNV, copy-number variant.

Correlation analysis of genetic alterations with clinical and pathological features

Univariate correlation analysis was performed on genetic alterations and clinicopathological features, and statistically significant results were summarized in Table 3. All the 10 HCC-related pathways have gene mutations related to clinicopathological features. Among them, cell cycle-related pathway has the largest number of genes with clinical and pathological correlations. As for the six DDR pathways, only HRR, FA, and NHEJ show significant correlations. Mutations of 9 uncategorized genes and CNV of 10 genes also show clinicopathological correlations.

Table 3.

Correlation between genetic alterations and morphologic subtypes

Pathway	Gene symbol	Macrotrabecular	MTM	Microtrabecular	Pseudoglandular	Compact	Clear cell	SH-like
HCC-related
Cell cycle	TP53	0.015	0.038	0.015
	CCNE1					0.010
	CDKN2A				0.006
	RB1	0.031
WNT	AXIN1	0.034	0.035	0.001			0.010
NOTCH	NOTCH1						0.002
	NOTCH2						0.001
Liver metabolism	CREBBP					0.035
	ARID2					0.010
Histone modification	KMT2C			0.026
	KMT2D					0.030
Oxidative stress	NFE2L2	0.042	0.030
	KEAP1						0.035
AKT/mTOR		0.031
Telomere	ATRX						0.010
	TERT			0.013
DDR
HRR	RAD51B				0.001
	BRCA2					0.003
FA	BRIP1
	XRCC5	0.042	0.030
Others
	ABCA13					0.012
	EYS	0.015	0.032	0.007
	CSMD3		0.031
	SDK2					0.002
	PCLO				0.043
	TAF1					0.002
CNV
11q13.3	CCND1	0.028	0.014
	FGF19/3/4		0.008

CNV, copy-number variant; DDR, DNA damage repair; FA, fanconi anemia; HCC, hepatocellular carcinoma; MTM, macrotrabecular-massive.

Correlation tests between genetic alterations and morphologic subtypes show that macrotrabecular subtype bears the largest number of corresponding genetic alterations. MTM subtype subdivided from the former corresponds to seven genetic alterations, sharing six common alterations with macrotrabecular subtype (including TP53, AXIN1, NFE2L2, XRCC5, EYS mutations and CCND1 amplification) except for CSMD3 mutation which is MTM-specific; SDK2 and RB1 mutations are related to macrotrabecular subtype, but not MTM subtype; amplification of 11q13.3 locus (including FGF3/4/19 and CCND1) is related to MTM subtype, while amplification of CCND1 alone is also related to MTM subtype as well as macrotrabecular subtype. Compact subtype corresponds to eight gene mutations (including CCNE1, CREBBP, ARID2, KMT2D, BRCA2, ABCA13, SDK2, and TAF1), among which SDK2 mutation is associated with both compact subtype and macrotrabecular subtype. Microtrabecular subtype is associated with TP53, AXIN1, EYS, KMT2C, and TERT mutations, among which TP53 and AXIN1 mutations show a negative correlation with microtrabecular subtype and a positive correlation with macrotrabecular subtype and MTM subtype, while EYS mutation shows a positive correlation with the former and a negative one with the latter. Clear cell subtype is associated with five genetic alterations, of which AXIN1 mutation is a shared one with macrotrabecular subtype and MTM subtype and the rest are specific (including KEAP1, ATRX, NOTCH1, and NOTCH2 mutations). The three gene mutations (including CDKN2A, RAD51B, and PCLO) related to pseudoglandular subtype are all specific. Steatohepatitis subtype is the only morphologic subtype which is not found to be associated with any specific genetic alteration.

Mutations of nine genes from four HCC-related pathways (cell cycle, hepatocyte metabolism, histone modification, and JAK/STAT pathways), two genes from one DDR pathway (NHEJ), and six uncategorized genes (LRP1B, EYS, SDK2, TAF1, PDE4DIP, and TCHHL1) as long as CNV of three genes (TAF1, GPR143, and VCX) are correlated with PD-1/PD-L1 expression. Among them, there are four genes (CREBBP, KMT2C, JAK2, and XRCC1) from HCC-related pathways and two uncategorized genes (PDE4DIP and TCHHL1) whose mutations are related to PD-1 expression of tumor cells, while mutations of one gene (ERCC3) of DDR pathway and two uncategorized genes (EYS and SDK2) are related to PD-1 expression of inflammatory cells. As for PD-L1 expression of tumor cells and inflammatory cells, mutations of CDKN2A and IRF2 in the cell cycle pathway, XRCC5 in the NHEJ pathway, and uncategorized PDE4DIP are related to the former, while mutations of RPS6KA3 in the cell cycle pathway, JAK1 in the JAK/STAT pathway, uncategorized LRP1B and TAF1, and CNV of TAF1, GPR143, and VCX are related to the latter. Among the 16 related genes, PDE4DIP mutation shows correlation with both PD-1 and PD-L1 expression in tumor cells, while the other mutations show a single correlation. In addition to PD-1/PD-L1, higher Ki-67 expression level corresponds to TP53 and PDE4DIP mutations while lower one corresponds to ZNF737 mutation (Table 4).

Table 4.

Correlation between genetic alterations and immunohistochemistry stainings

Pathway	Gene	PD-1 (tumor)	PD-1 (inflammatory cells)	PD-L1 (tumor)	PD-1 (inflammatory cells)	Ki-67
HCC-Related
Cell cycle	TP53					<0.001
	CDKN2A			0.034
	IRF2			0.004
	RPS6KA3				0.033
Liver metabolism	CREBBP	0.009
Chromosome remodeling	ARID1A
Histone modification	KMT2C	<0.001
JAK/STAT	JAK1				0.03
	JAK2	0.01
DDR
NHEJ	ERCC3		0.014
	XRCC5			<0.001
Others
	LRP1B				0.001
	EYS		0.045
	SDK2		0.021
	TAF1				0.032
	PDE4DIP	0.007		0.047		0.048
	ZNF737					0.036
	TCHHL1	0.037
CNV
	TAF1				0.046
	GPR143				0.006
	VCX				0.006

CNV, copy-number variant; DDR, DNA damage repair; HCC, hepatocellular carcinoma; NHEJ, non-homologous end joining.

TP53, RPS6KA3, and ABCA13 mutations are associated with higher Edmondson–Steiner grade, while AKT/mTOR pathway mutation and ATRX gene mutation are associated with MVI. MYC amplification is associated with higher preoperative serum AFP level. Furthermore, CNV of nine genes including TAF1 and CSMD3 are related to the newly defined macroscopic diffuse subtype (Supplementary Table 2, Supplemental digital content 2, http://links.lww.com/EJCP/A524).

Univariate/multivariate analysis of recurrence/metastasis after liver transplantation

The median follow-up duration of the transplanted cases is 423 days. There are a total of 15 recurrent cases and 18 metastatic cases, among which recurrence and metastasis sequentially happen in seven cases.

In the phase of univariate analysis, a total of nine statistically significant factors were screened (Table 5), including two morphologic variables, three genetic alterations, one pathologic feature, and three immunohistochemistry staining results. MTM subtype and microtrabecular subtype are correlated with a higher and a lower rate of recurrence/metastasis respectively. TERT and TSC2 gene mutations and AKT/mTOR pathway mutation (including TSC2) all show positive correlation with recurrence/metastasis. As for other pathological features, satellite foci is associated with higher rate of recurrence/metastasis. Higher Ki-67 expression level is correlated with higher rate of recurrence/metastasis, while higher CK7 or Hepa expression level are correlated with lower rate of recurrence/metastasis.

Table 5.

Results of univariate analysis of recurrence/metastasis in transplanted cases

	Recurrence/metastasis
Variate	Present	Absent	P value
AKT/mTOR pathway
Mutated	7	5	0.034
Unmutated	21	67
TSC1
Mutated	4	1	0.021
Unmutated	24	71
TERT
Mutated	3	0	0.020
Unmutated	25	72
MTM
Present	12	12	0.038
Absent	16	49
Microtrabecular
Dominant	12	36	0.019
Present	1	10
Absent	15	15
Satellite foci
Present	10	11	0.031
Absent	18	61
Ki-67
2	11	15	0.016
1	15	32
0	2	23
CK7
Positive	2	26	0.002
Negative	26	43
Hepa
Positive	14	50	0.020
Negative	12	13

MTM, macrotrabecular-massive.

In order to proceed with multivariate analysis using Cox’s proportional hazard ratio regression model, we performed a proportional hazard hypothesis test on the nine variables that were selected in univariate analysis. The test shows that six of the variables do not meet the proportional hazard hypothesis with only three variables remaining (including MTM subtype, CK7, and Hepa expression) for further analysis. Considering that most cases in our research cohort have a follow-up duration shorter than 2 years and the number of final events observed is also small, we decided to skip the phase of multivariate analysis and interpret the results of univariate analysis.

Discussion

Genetic alterations and clinicopathological correlations of pathways we emphasize, including HCC-related pathways, are consistent in part with the findings of previous researches. In our study, TP53 and CTNNB1 were identified as the top two genes with the highest mutation frequencies. Additionally, there was a notable tendency for mutually exclusive mutation between these two genes. The TP53 mutation is associated with a poorer degree of differentiation and a macrotrabecular morphological subtype. However, the mutation frequency of TP53 (48.2%) is significantly higher than that of CTNNB1 (15.9%) in our study. Furthermore, the CTNNB1 mutation does not demonstrate a correlation with better differentiation or prognosis, in contrast to the findings of most western cohort-based studies (Niu et al., 2016, Calderaro et al., 2019). The discrepancy in gene mutation frequency may be attributed to the different molecular mechanisms of HCC by different etiologies: western cohorts are dominated by alcohol or hepatitis C virus (HCV)-related HCC which have been found to be correlated with CTNNB1 mutation, while in our cohort, HBV-related HCC is the absolute dominance (96.9% of all cases, and 97.6% of transplanted cases). The impact of CTNNB1 mutations on pathological characteristics and prognosis is not statistically significant in our HBV-related HCC dominant cohort, but we may need to incorporate more alcohol and HCV-related cases into the study to further verify that.

The binary classification of proliferation/nonproliferation class is a representative and universal molecular typing: proliferation class is characterized by activation of various pathways including cell proliferation and survival-related pathways, AKT/mTOR pathway, Notch pathway, TGF-β pathway, and CCND1/FGF19 amplification, while nonproliferation class is mainly characterized by activation of the WNT/β-catenin pathway (Li et al., 2017, Calderaro et al., 2019, Wang et al., 2021). We defined the gene-level alterations of proliferation/nonproliferation class according to previous studies: for proliferation class, gene signature includes Notch pathway, AKT/mTOR pathway mutation, and CCND1/FGF19 amplification; for nonproliferation class, gene signature includes mutations of CTNNB1 and APC of WNT/β-catenin pathway. Although AXIN1 encodes proteins which serve as an inhibitory link of WNT/β-catenin pathway, a study has found evidence that loss-of-function mutation of AXIN1 does not induce HCC by activating classic WNT pathway, but by activating Notch pathway and YAP pathway (Abitbol et al., 2018), so AXIN1 mutation is classified as one of the genetic alterations of proliferation class. Under this classification, cases that have gene alterations of proliferation/nonproliferation class constitute about half of our cases, with cases of proliferation class accounting for 33.5% (55/164) and those of nonproliferation class accounting for 13.4% (22/164). Correlation analysis shows that proliferation class is associated with higher preoperative serum AFP level (≥400 ng/ml, P = 0.02), macrotrabecular subtype (P < 0.01), and MTM subtype (P = 0.02). Besides, cases of proliferation class tend to have more nodules (P = 0.08), satellite foci (P = 0.10), and are prone to be less differentiated (P = 0.14) and more likely to have recurrence/metastasis after transplantation (48.3% vs 20%, P = 0.15) than those of nonproliferation class, although the results are not statistically significant (Supplementary Table 3, Supplemental digital content 3, http://links.lww.com/EJCP/A525). Clinicopathological features of proliferation/nonproliferation class in our study are basically consistent with what was discovered by previous transcriptome-based studies. However, considering the diversity of mutation sites in a single gene, the functional impacts of all mutation sites on the encoded product (gain-of-function or loss-of-function) are hardly known, so the consistency between our gene-level classification signatures and transcriptome-based signatures needs further verification. Moreover, the molecular mechanism of proliferation/nonproliferation class has not been fully understood, which means there may be more pathways, genes, or immune response involved. Therefore, the proliferation/nonproliferation classification system in gene level needs to be further complemented by relevant studies.

Except for alterations mentioned above, our study also incorporates mutations of 19 genes whose impacts have not yet been defined on the development of HCC, and some of these gene mutations were found through statistical tests to be correlated with morphologic and other features. LRP1B (LDL Receptor Related Protein 1B) encodes protein that belongs to the low-density lipoprotein (LDL) receptor family. Studies have confirmed that deletion of LRP1B is related to poor prognosis of glioblastoma, and its mutation promotes growth and metastasis of colon cancer (Liu et al., 2021), therefore it is considered to be a tumor suppressor gene. There has not been any relevant study that depicts its role in HCC, except for one study that has found it to be one of the frequently integrated sites of hepatitis virus HBV (Xu et al., 2021), which may be related to HBV-related HCC. Although LRP1B mutation shows no correlation with invasive pathological features, it is found to be associated with higher expression of PD-L1 of inflammatory cells, which may indicate that LRP1B mutation is related to immune suppression status. The encoded product of TAF1 (TATA-Box Binding Protein Associated Factor 1) is a component of the basic transcription factor TFIID. Two related studies have shown that TAF1 mutation and methylation are related to progression of HCC (Cai et al., 2020). In our study, TAF1 mutation is associated with compact subtype and the expression of PD-L1 inflammatory cells; TAF1 copy number loss is found to be related to diffuse subtype and PD-L1 expression on inflammatory cells. Correlations of TAF1 genetic alterations with morphologic subtypes that have relatively poor prognosis and high PD-L1 expression level in immune cells indicate that TAF1 may function as a tumor suppressor and its mutation may influence immune microenvironment of HCC. PCLO (Piccolo Presynaptic Cytomatrix Protein) is mainly involved in neuronal synaptic activity. Its function in liver tissue has not been studied, but a research has found that mutation frequency of PCLO in well-differentiated HBV-related early stage HCC is higher than that of poor-differentiated ones (Gao et al., 2021); 6.7% of the cases in our HBV-related HCC dominated cohort are mutated in PCLO, and mutation of it is related to pseudoglandular morphology and lower histological grade, which corresponds to the conclusion of the study mentioned above.

Another gene that caught our attention is CSMD3 (CUB and Sushi Multiple Domains 3), for its mutation and CNV are both found to be related to specific morphologic subtypes which tend to have poorer prognosis than the other subtypes. So based on these pathological correlations, it seems reasonable to infer that genetic alterations of CSMD3 in HCC may indicate poorer prognosis. Although the role of genetic alterations of CSMD3 in tumorigenesis has not been clarified, a few studies have reported mutations or CNV of this gene in certain types of tumors: loss-of-function gene mutation of CSMD3 is reported in lung cancer, ovarian cancer, and dedifferentiated liposarcoma (Lu et al., 2021). CSMD3 is found to be one of the genes that show the most significant level of copy number loss in a research on colon cancer (Wolff et al., 2018); another animal model-based study of HBV-related HCC also points out that lowered expression of CSMD3 is related to the occurrence of HCC (Lai et al., 2016), thus speculating that CSMD3 may be a tumor suppressor gene. On the other hand, however, there are studies suggesting an opposite conclusion, for example, a study has found high level of CSMD3 amplification in phyllode tumor of the breast (Laé et al., 2016). What we found in our study is not sufficient enough to identify this gene as a tumor suppressor gene or oncogene of HCC, for that its mutation is associated with MTM subtype of poor prognosis, but its amplification is also associated with a morphologic subtype that has a higher recurrence rate than the others. So more research-based evidence is needed to determine the specific role of this gene. Other unreported molecular alterations that we have found of clinicopathological correlations include ABCA13 mutation related to compact subtype, EYS mutation related to MTM and macrotrabecular subtype, and SDK2 mutation related to compact and macrotrabecular subtype. In addition, CNV of GPR143, VCX, and FZD6 are related to PD-L1 expression and diffuse subtype. The functional impacts of these molecular alterations on HCC require more in vivo and in vitro experiments to study, and their feasibility as potential therapeutic targets need to be evaluated as well, because the most frequent mutations in HCC, such as TP53 and CTNNB1 mutations are not targetable.

The majority of previous studies have focused on the screening of prognostic markers of HCC based on transcriptome data. In our study, we managed to incorporate SNV, INDEL, and CNV obtained by NGS along with other pathological features into the prognosis of liver-transplanted cases. Univariate analysis screened nine relevant variables including three genetic alterations including TERT, TSC2, and AKT/mTOR pathway mutations. TERT promoter mutation has been confirmed to be an oncogenic hotspot mutation by various studies, and its mutation frequency (including promoter region and exon region) in HCC is between 10 and 20% in public database. In our study, the mutation frequency reaches only 5%, and all these cases are mutated in exons. Loss-of-function mutation in exons of TERT may lead to tumorigenesis by shortened telomere and chromosomal recombination (Falini et al., 2020) according to relevant studies, and the correlations of exon mutations with certain tumors have been revealed in several studies, for example, P308Hfs*43 is found to be associated with acute myeloid leukemia (Falini et al., 2020). We found that five transplanted cases with TERT exon mutation have higher frequency of recurrence/metastases than cases without TERT mutations (100.0% vs 25.8%, P = 0.02), and three of the mutated cases (60%) are MTM subtype, while none of the cases are microtrabecular subtype, which further suggests that exon mutation of TERT in HCC may be correlated with poor prognosis.

Mutations of AKT/mTOR pathway or TSC2 mutation alone are related to higher recurrence/metastasis rate. TSC2 encodes a protein that downregulates mTORC1, so targeted drugs that inhibit mTOR, such as everolimus, may be a treating option for HCC with mutations of AKT/mTOR pathway. Everolimus failed in improving the survival of HCC patients who progressed after sorafenib treatment or were intolerant of sorafenib in a phase III clinical trial as a targeted drug (Kang et al., 2021), but it can also be used as an immunosuppressant other than a targeted drug for tumors, playing a dual role in preventing immune rejection and tumor recurrence after liver transplantation. There are studies that have proved it to be an ideal treatment option for HCC patients who undergo liver transplantation: patients who received everolimus after liver transplantation have better 5-year survival than patients using other immunosuppressants (Sapisochin et al., 2022), and the combined application of everolimus and tacrolimus after transplantation can improve OS when compared with using the latter alone (Lee et al., 2021). The finding of our study that AKT/mTOR pathway mutation is correlated with higher risk of recurrence/metastasis after transplantation further indicates HCC patients with this mutation may benefit from mTOR inhibitors after liver transplantation.

In addition to the genetic alterations mentioned above, we also have found three pathological variables and three immunohistochemistry staining variables which are associated with the prognosis of transplanted cases. MTM subtype, satellite foci, and the expression of Ki-67 are factors that are related to higher recurrence/metastasis risk, while microtrabecular subtype, the expression of Hepa, and CK7 are related to lower risk. Even though multivariate analysis is not applicable, the factors can be a reference for screening candidates for liver transplantation. We only incorporated transplanted cases for prognostic analysis, eliminating the impacts from postoperative residues, intrahepatic metastasis, and viral infection, inflammation, and sclerosis of the peritumoral liver tissue. The pathological and molecular factors we screened can be complemented into the present criteria to predict the potential benefits from liver transplantation for each candidate, but further verification and complement will be needed for designing an NGS panel with fewer gene targets for resected liver in transplantation, so as to assess the risk of recurrence/metastasis and present targeted treating strategy after transplantation.

The principal limitations of this study are the relatively small number of cases included and the inevitable degree of selection bias.

Conclusion

In our study, genetic alterations which are detected using NGS correspond to various significant pathological and clinical features, which provide gene-level understanding of morphologic manifestation and display value of recurrence/metastasis risk assessment for transplanted cases. More practice will be needed for clinical application of NGS to better serve the goal of individual molecular-pathological diagnosis and prognosis for HCC patients.

Acknowledgements

This work was supported by the Immune and cellular property shifts in human liver disease (22JC1403002).

Conflicts of interest

There are no conflicts of interest.