Background
Hepatocellular carcinoma (HCC) is one of the leading causes of cancer mortality worldwide [
1]. The majority of HCC occur in patients with underlying liver disease, such as hepatitis B virus (HBV) infection and cirrhosis [
2]. Over half of patients with HCC are diagnosed at advanced stages, preventing the possibility of curative therapies. Alpha-fetoprotein (AFP) is a widely used, yet imperfect, biomarker for HCC early diagnosis. It has been reported that AFP (at a threshold level of 20 ng/mL) showed low sensitivity of 40–60% with specificity of 80–90% [
3]. Low sensitivity, false negativity (e.g., a small HCC with normal AFP level), and false positivity (e.g., liver function damage and certain gastrointestinal tumors) of AFP could lead to decreased chance of early diagnosis and thus poor clinical outcomes, highlighting the requirement for more effective approaches for HCC detection.
Cancer-associated autoantibodies (AAbs) may develop early during carcinogenesis when cancer-associated antigens appear in premalignant or malignant lesions. The immune system can effectively amplify and memorize immune responses to those antigens, thereby making AAbs as appealing cancer biomarkers. For example, DHCR24 AAb was identified as a novel biomarker for disease progression of hepatitis C [
4]. Likewise, it has been reported that AAbs against HCC1, CDKN2A, p53, CIP2A, and survivin could indicate the presence of HCC prior to clinical diagnosis [
5]. In another study, AAbs against NPM-1, 14-3-3 zeta, and MDM2 were suggested to have diagnostic value for AFP-negative HCC patients (AFP < 20 ng/mL; AFP
− HCC) [
6]. Serum AAbs against EIF3A [
7] and SF3B1 [
8] were also reported as potential diagnostic biomarkers for HCC. However, the sensitivity and specificity of those selected AAbs remain limited, and further high-throughput unbiased screening with a large cohort and independent validation are still required. In addition, the heterogeneity of human biology in cancer suggest that combined use of the cancer biomarkers in parallel or in tandem in algorithms such as artificial neural network (ANN) are necessary [
9,
10].
Protein microarrays are capable of presenting thousands of tumor-associated antigens to rapidly and globally identify AAb responses in serum (seromics) [
11,
12]. Known and predicted tumor antigens have been employed in a comprehensive protein array to profile cancer immune response, such as p53 [
13], GPR78 [
14], HER2 [
15], and HSP60 [
16]. In this regard, global AAb screening has identified high-performance AAb panels for early diagnosis of lung cancer [
13] and Behcet disease [
17]. Herein, the HuProt arrays, comprising of 21,154 unique full-length proteins, were first employed to survey serum AAbs using HCC samples. Subsequently, HCC-focused arrays were fabricated with the candidate proteins identified in the HuProt arrays. A large cohort of 1253 serum samples, including HCC patients, liver cirrhosis (Cirrhotic) patients, and healthy controls (Healthy), were screened to develop a diagnostic model. A novel panel of 7 proteins including CIAPIN1, EGFR, MAS1, SLC44A3, ASAH1, UBL7, and ZNF428 were discovered and evaluated for the early detection of HCC.
Methods
Human serum sample
The cohort was comprised of 1253 serum samples from 611 HCC patients, 249 cirrhotic patients, and 393 healthy controls. Between January 2019 and August 2019, these samples were collected at Zhongshan Hospital of Fudan University, Eastern Hepatobiliary Surgery Hospital, and Cancer Hospital of Guangxi Medical University. All blood samples were processed identically to obtain serum. Briefly, 5 mL venous blood was drawn from each individual (before any treatments and surgery), placed in room temperature (RT) for 1 h until coagulated. Serum was recovered by centrifugation at 3000 rpm for 10 min and stored in aliquots at − 80 °C until used. The informed consent and agreement of all samples used in this study have been obtained. The ethical regulations have been approved from each hospital.
Inclusion criteria for HCC patients in this study were (1) pathological diagnosis of HCC (
n = 446); or (2) diagnosis of HCC by enhanced computed tomography, enhanced magnetic resonance imaging, or contrast-enhanced ultrasonography in combination with AFP or des-gamma carboxyprothrombin for patients without pathological diagnosis (
n = 165); (3) without autoimmune diseases. Patients were all free of hepatic encephalopathy and ECOG/WHO/Zubrod performance status scored as 0~1. Child-Pugh score, BCLC staging [
18], TNM staging, and Chinese Liver Cancer staging [
19] were individually estimated; (4) patients with other cancerous history were excluded from our study.
Diagnosis of liver cirrhosis was confirmed by enhanced magnetic resonance imaging or pathology. Healthy controls had normal liver biochemistry and were in the absence of liver diseases and alcohol abuse.
Serum AAb profiling on HuProt arrays
HuProt™ Human Proteome Microarray v3.0 was provided by CDI Laboratories, Inc (Mayaguez, PR). Each HuProt array is comprised of 21,154 unique proteins. A total of 100 serum samples from discovery phase (I) was applied to HuProt arrays, including 50 HCC and 50 healthy controls. The microarray was taken out from − 80 °C and then incubated in blocking buffer (3% BSA in PBS) at RT for 3 h. Then a serum sample diluted at 1:200 in binding buffer (1% BSA in PBST) was added to the microarray and incubated at 4 °C overnight. After washing with PBST, the microarray was incubated with 1:1000 diluted Fluor conjugated goat anti-human IgG (532 nm) and donkey anti-human IgM (635 nm) (Jackson ImmunoResearch, West Grove, PA) at RT for 1 h in the dark. After washing with PBST, the microarray was rinsed with ddH
2O and dried. The microarray was scanned with the LuxscanTM 10 K-A (CapitalBio Corporation, Beijing, China). The GenePix Pro 6.0 (Axon Instruments, Foster City, CA) was used for foreground and background intensity extraction for each spot. The signal for each spot (SNR) was defined as the ratio of the foreground to the background median intensity as previously described [
20].
HCC-focused arrays
After serum incubation on the HuProt arrays, autoantibody signals were detected, normalized [
21], and quantified. For selection of candidate proteins, three criteria should be satisfied after comparing HCC vs
. Healthy: (1)
p values obtained from the
t test ≤ 0.05; (2) fold change (FC) ≥ 1.2; (3) the positive ratio ≥ 10% (The HCC positive reactivity was defined as greater than the mean plus 2 × SD of the healthy controls. The positive ratio was calculated as the number of HCC positive reactivity to its sum [
22]). According to the criteria above, 81 proteins were identified. The extra 19 AAbs including CTRL, DCAF4L2, BIRC5, CCNB1IP1, GPR78, HM13, HSPA2, IMP3, KDM1A, MAPK1, RALA, RPLP0, SARNP, SF3A3, TSPAN13, TUBB6, XRCC5, CENPF, and CDKN2A were selected based on cancer literature in general. We aimed to fabricate the HCC-focused arrays using more candidate proteins from our own experiment and the literature. Thus, a total of 100 proteins were picked to fabricate the HCC-focused arrays, which contained 14 identical subarrays on each slide (BC-BIO, Foshan, China). The subsequent assay process was similar to that described for HuProt array, with an exception that the dilution of serum samples was 1:100 per subarray.
Model development for HCC detection
For ANN model, we determined the number of hidden neurons based on previous literature [
23]. Using the model
Nh = (4n
2 + 3)/(n
2 − 8) [
Nh, the number of hidden neurons;
n, the number of input neurons],
Nh was set at 5 in our study. Thus, fully connected feedforward neural-networks including 7 input nodes, 5 neurons in the hidden layer, and 2 output nodes were chosen. Back propagation of error algorithm was used as the learning rule, and the average committee vote was used to classify samples [
24‐
26]. For the test phase (II), 576 samples were randomly split into 10 equally sized groups. One ANN model was constructed using 90% of cases as training set and the remaining 10% as verification set. This procedure was repeated 10 times to obtain 10 ANN models. After repeating 50 times, 500 ANN models were developed. Each ANN model provided the outputs 0 for control or 1 for HCC. The committee vote was performed by averaging all outputs and then to classify the samples. The samples in the validation phase (III) used 500 ANN models for the blind test. Both ANN models and AFP were tested using receiver operating characteristic (ROC) curve analysis.
Discussion
Although pathological and radiological examination remains the “gold standard” for clinical diagnosis of cancers, liquid biopsy has shown appealing potential for early detection of HCC [
29]. In this regard, tremendous efforts have been made on the early diagnostic potential of circulating micro-RNA signature [
30], cell-free DNA [
31], metabolites [
32], glycans [
33], and DNA methylation pattern [
34]. However, AFP is still the only widely used clinical protein biomarker for HCC diagnosis, although approximately 40% of HCC cases harbored a normal AFP level. Due to the nature of stability and easy detection, efforts have also been made to evaluate novel protein biomarkers for HCC detection, such as Dickkopf-1 [
35] and Aldo-keto reductase family 1 member B10 [
36].
Based on three steps for biomarker classifier development [
37], we focused on CIAPIN1, EGFR, MAS1, SLC44A3, ASAH1, UBL7, and ZNF428, which are mainly involved in activation of signaling cascades and apoptotic/metabolic processes. The molecular function of UBL7 is polyubiquitin modification-dependent protein binding, and loss of ubiquitin-proteasome players were suggested to lead to protein expression alteration and hepatocarcinogenesis [
27]. CIAPIN1 was reported to play an important role in HCC proliferation through regulating the expression of cell cycle-related proteins [
38]. EGFR is a transmembrane receptor tyrosine kinase and plays a key role in HCC development and progression [
39]. The biological functions of MAS1, SLC44A3, ASAH1, and ZNF428 in HCC were rarely reported. Here, we provided autoantibody clues for further exploring their biological significance in HCC.
It has been reported that neural network analysis was potentially more useful than traditional statistical techniques when the relationship among variables was complex and non-linear [
10]. The performance of ANN-based 7-AAb model could be further improved due to continuous learning of neural networks in future clinical application. However, there are several limitations in the present study. First, AAbs were reported to appear in multiple cancer types due to immune surveillance. Alternatively, it may indicate the potential of AAbs for monitoring various cancer types, similar to the pan-cancer diagnostic value of cfDNA alterations [
40]. Based on previous literature, AAb against ASAH1 could be applied to monitor the progression of melanoma [
41]. However, there were no significant differences in AAb against EGFR between patients with breast cancer and controls [
42]. Thus, further analyses are required to evaluate the diagnostic value of our 7-AAb panel in diverse cancers. Second, this study was conducted using most of patients with HBV-related HCC from China and HCC patients with high ANN value featured HBsAg positivity (
p value < 0.001, Pearson’s chi-squared test). A prospective multi-nation validation is necessary for further application. Third, the panel contained 7 biomarkers for ANN-model and it is more complex than single marker detection in clinic. Albeit its complexity, ANN could perform better when subclasses are separated by a non-linear boundary.
Conclusions
In summary, a comprehensive seromic survey was performed for discovering and validating serum diagnostic biomarkers in HCC. Based on ANN-model, we identified a 7-AAb panel that was generally superior to AFP for HCC detection, and performed well for AFP-negative HCC and HCC at early stage. The 7-AAb panel provides potentially clinical value for non-invasive early detection of HCC, and brings new clues on understanding the immune response against hepatocarcinogenesis.
Acknowledgements
We thank Mr. Sixian Yang and Mr. Yang Li (BC-BIO, Foshan, China) for the assistance of array analysis.
Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.