Engineering In Vitro Organ‐Structured Tumor Model for Evaluating Neoantigen‐Specific T Cell Responses in Hepatocellular Carcinoma

Neoantigens derived from somatic mutations in cancer cells can induce antigen‐specific T‐cell immune response for cancer immunotherapy. However, the 3D models for assessing neoepitope immunogenicity and efficacy of anti‐tumor T‐cell immune response to neoantigens are less than perfect. Here, a 3D tumor model based on recellularized liver matrix is leveraged with HepG2 cells to investigate T cell cytotoxic reactivity toward hepatocellular carcinoma (HCC) neoantigens. The whole exome sequencing (WES) data of 364 HCC patients in The Cancer Genome Atlas database are collected and 25 highly potential immunogenic neoantigens to human leukocyte antigen (HLA)‐A*02:01 molecule in silico are predicted. Six of the HCC neoantigen candidates are functionally validated with high immunogenicity by measuring cellular interferon‐γ secretion and cytotoxicity during neoantigen‐specific T‐cell immune responses in vitro. Then, the minigene of six functionally identified neoantigen peptides is constructed and the minigene‐modified GFP‐HepG2 cells are generated. Neoantigen‐specific immune response is observed with highly secreted Granzyme B, IFN‐γ, and PD‐1 when targeting the minigene‐modified GFP‐HepG2 cells in the 3D RLM HCC tumor model. Overall, the 3D RLM tumor model provides a novel strategy for preclinical assessment of the efficacy of neoantigen‐specific T cell immune response, which helps develop personalized cancer vaccines and immunotherapy treatments for HCC patients.


Introduction
Hepatocellular carcinoma (HCC), as the predominant form of liver cancer, is one of the most common malignant neoplasms worldwide and represents the third leading cause of cancerrelated death. [1][2][3][4] Despite the substantial progress in drug development for liver cancer, the overall survival (OS) benefits remain minimal for patients with advanced-stage HCC. [1] Recently, cancer immunotherapies have revolutionized the management of advanced-stage HCC with a benchmark median overall survival, [5] which sparks a new line of research to apply immune oncology for cancer treatment.
However, the development of appropriate in vitro models to assess the potential efficacy of immunotherapies has lagged far behind. [6] Currently, in vitro models applied in the evaluation of immunotherapies of solid tumors mainly include human tumorderived two-dimensional (2D) cell lines, [7] patient-derived organotypic www.advancedsciencenews.com www.advmatinterfaces.de cultures, [8] and three-dimensional (3D) tumor spheroids/organoids. [9] Due to the inability to replicate the 3D tumor microenvironment (TME) of the 2D monolayer culture models, the 3D tumor models have been established to incorporate more advanced tumor-immune cell interactions but generally apply simplified structures that are restricted by apoptotic effects due to the lack of perfusable vascular networks. [10,11] Therefore, there is an urgency in developing 3D preclinical models with appropriate tumor-mimicking structures to enable the investigation of basic immune responses such as immune infiltration. Here, we proposed to engineer a 3D in vitro tumor model using an organ-structured decellularized extracellular matrix (dECM) featuring unique perfusable vasculatures for evaluating the immunotherapy of HCC treatment. Previous studies showed that decellularized liver matrix (DLM) could be recellularized with HepG2 cells to derive a recellularized liver matrix (RLM) platform, providing a physiomimetic in vitro model for pharmaceutical tests. [12][13][14][15] We hypothesized that the patency of abundant vasculatures in the RLM model would enable the incorporation of endovascular perfusion of tumorinfiltrating lymphocytes to replicate the dynamic migration and infiltration processes of cytotoxic T cells, the crucial step to induce cytotoxicity in solid tumors.
Cytotoxic T lymphocytes (CTLs)-based immunotherapy is a type of adoptive cellular therapy (ACT) that uses CTLs derived from cancer patients with advantageous high specificity, safety, and identification of individuals. [16,17] Data from several clinical trials have proven that activated CTLs can effectively alleviate multiple cancers such as melanoma, nasopharyngeal carcinoma, renal cell carcinoma, and breast cancer. [17][18][19] As for the CTL immunotherapy in the treatment of HCC, tumor-associated antigens (TAAs, expressed low levels in normal cells, but overexpressed in the cancer cells) could be identified through the serological analysis of recombinant cDNA libraries (SEREX) in HCC. [20,21] The HCC-associated antigen HCA587 317-325 was identified as one of the specific human leukocyte antigen (HLA)-A*02:01-restricted CTL epitopes. [22] CTLs ex vivo activated with TAAs (TP53, hTERT, and Survivin)-pulsed DCs were used to evaluate the clinical response of immunotherapy in HCC patients, and DC-CTLs treatment may provide an effective therapy on HCC. However, the CTL-based immunotherapy studies on HCC produce relatively poor outcomes, and the curative effect is not ideal due to immune tolerance and autoimmunity. [23] Therefore, screening more effective neoantigens is critical for developing specific CTL treatments for HCC.
In this study, we generated a 3D preclinical RLM tumor model with HepG2 to validate the neoantigen-specific T cell immune response for HCC immunotherapy. We predicted the HLA-A*02:01 restricted neoantigen candidates from whole genome sequencing (WES) data of 364 HCC patients from the Cancer Genome Atlas (TCGA) database. The neoantigen candidates with high binding affinities were used to stimulate antigen-specific T cells in vitro, and six immunogenic neoantigens were functionally identified for constructing the minigene-modified GFP-HepG2 cells.
To develop a 3D RLM HCC tumor model, we recellularized the DLM by perfusing with minigene-modified GFP-HepG2 cells. Then, we successfully validated the CTLs immune response towards six HCC neoantigens in the 3D RLM HCC tumor model. This study provides a novel strategy with RLM scaffolds for preclinical validation of neoantigen-based CTL immunotherapy and can advance the therapeutic HCC vaccines with higher efficiency in clinical applications.

In Silico Prediction of HCC Neoantigen Candidates in TCGA Database
To predict the potential HCC neoantigens, we first selected the whole exome sequencing (WES) datasets of 364 HCC patients in the TCGA database ( Figure 1A). The single-nucleotide variants (SNVs) detected in at least two samples were retained for mutation annotation. Subsequently, the population of 9-to 11-mer peptides with high-frequency SNVs was obtained and checked in the protein database (https://www.uniprot.org/). After filtering out the epitopes that are in the protein database, we obtained the tumor mutation derived-peptides for predicting the binding affinity of antigens restricted to HLA-A*02:01, which is an important HLA restriction element for peptide presentation to induce an immune response in disease and cancer. [24] The high binding affinity of predicted peptides to HLA-A*02:01 with IC50<50 nm were selected as the HCC neoantigen candidates. The predicted HCC neoantigens were used for assessment of immunogenicity by using antigen peptide-specific functional CTLs stimulation in vitro.
The functionally identified HCC neoantigens were then used for constructing the minigene to generate the minigene-modified HepG2 cells ( Figure 1A). Subsequently, a 3D RLM tumor model was established with the minigene-modified HepG2 cells to validate the HCC neoantigen-specific immune response. Particularly, a DLM was first obtained from fresh rat liver following a modified decellularization protocol reported previously, [12][13][14][15] which provided a unique organ-structured template with wellpreserved liver-specific bile duct (BD) and vasculature systems as well as the extravascular extracellular matrix (ECM). The DLM was then recellularized with the minigene-modified HepG2 cells through the BD due to the comparatively porous wall of the BD allowing the infiltration of the cancer cells into the ECM. The   (Table S1, Supporting Information). Subsequently, we created the 9-mer peptides containing SNVs that were used for predicting HCC neoantigen candidates. To enhance the prediction accuracy of HCC neoantigens, the artificial neural network-based method (NetMHC-4.0, [25] NetMHCpan-3.0, [26] NetMHCpan-4.0 [27] ), Position Specific Scoring Matrix (PSSM)-based method [28] (PSSMHCpan-1.0), PickPocket-1.0, SMM-1.0, and our pipeline [29] were performed as the benchmark to predict the antigen binding affinity restricted to HLA-A*02:01 molecules. The results showed that our pipeline displayed the highest sensitivity (53.1%) and accuracy (ACC) (70.7%) for HCC neoantigen candidate prediction compared to other methods. However, the SMM-1.0 method showed the highest specificity (98.8%) and false omission rate (FOR) (51.8%) compared to other methods. NetMHCpan-4.0 method demonstrated the highest positive predictive value (PPV) (97.7%) and area under the curve (AUC) (89.2%) of the predicted neoantigen candidates compared to other methods ( Figure 1B; Table S2, Supporting Information). The parameters for selection of HCC neoantigen require: 1) The neoantigens must be predicted with at least two different methods by IC50 < 500 nm, and the lower IC50 from two methods were used as the final binding affinity for neoantigen candidates; 2) The IC50 for neoantigen candidates must be lower than wild-type peptides. Finally, the top six neoantigen candidates were selected with high antigen binding affinity (IC50 < 50 nm) restricted to HLA*A02:01 for the first batch ( Table 1).
To increase the number of predicted neoantigen candidates and cover more HCC patients, we optimized the neoantigen prediction pipeline. The top 20 most high-frequency missense SNVs detected in at least two HCC patients were selected, which covers a 5.5% accumulating mutation ratio in the population of HCC patients. Then, the 9-11 mer peptides containing SNV mutation sites were created as candidate neoepitopes (Table S3, Supporting Information). In addition to the methods (NetMHC-4.0, NetMHCpan-3.0, NetMHCpan-4.0, PSSMHCpan-1.0, PickPocket-1.0, SMM, and our pipeline) used for the prediction of neoantigen candidates, we also used the antigen presentation software the Epitope Presentation Integrated Prediction (EPIP) to predict the efficiency of antigen presenting on the cell surface. [30] The parameters of selected neoantigen candidates in this batch require: 1) The neoantigen candidates must be predicted by at least three methods with IC50 <500 nm, and the lowest IC50 score among these three methods were used as the final binding affinity parameter for HCC neoantigen candidates; 2) The scores of the efficiency antigen presentation by EPIP is EPIP score ≥ 0.5; 3) The HCC neoantigen candidates with high binding affinity (IC50 < 50 nm) restricted to HLA*A02:01 were selected, and ordered by the gradually reduced antigen binding affinities; and 4) The ELscore of neoantigen candidates must be <2. In summary, the final parameters for the second batch of HCC neoantigen candidates with high binding affinity restricted to HLA*A02:01 is IC50 <50 nm, EPIP score ≥0.5, and ELscore <2 ( Table 2). Together, a total of 25 HCC neoantigen candidates were selected for further in vitro validation of neoantigen immunogenicity.
We checked the bioavailability of the associated wild-type peptide sequences from two common mass spectrometry (MS)based databases, i.e., HLA Ligand Atlas (https://hla-ligand-atlas. org/welcome) and Immune Epitope Database (https://www. iedb.org/), to evaluate the probability of the peptide presentation of the predicted 25 neoantigens (Table S4, Supporting Information). We found one corresponding wild-type peptide (CTNNB1) in the HLA Ligand Atlas, and six wild-type peptides (CTNNB1, TP53, GMPS, SLC22A18, TRPC4AP, and GAA) in the Immune Epitope Database. We further performed Google searches and found that 15 related wild-type peptide sequences (SI, AP1G1, UGT2B10, TPST1, ROBO2, TP53, GMPS, OR2G2 (2-8), TRPC4AP, SUCNR1, GPR137B, SI (2-16), PIK3CA, ZD-HHC22, and TGIF2LX) have been reported in the literature. As expected, none of the peptide sequences of the predicted neoantigens could be detected in the reported wild-type peptides, suggesting the specificity of the neoantigens with tumor somatic mutations in patients.

Identification of Immunogenicity of HCC Neoantigen Candidates with In Vitro CTLs Stimulation
To validate the immunogenicity of predicted HCC neoantigens, we applied the approach of in-vitro stimulation of antigenspecific CTLs by DCs loaded antigen peptides. [5,31] Figure S1, Supporting Information). Second, CD8positive T cells and CD14-positive monocytes were isolated from PBMCs for neoantigen peptide presentation on mature DCs and T cell stimulation. After 2 weeks of stimulation, we validated the antigen-specific T cell immune response by functionally cytokine secretion. The ability of interferon-(IFN-) secretion was detected in the stimulated CTLs when targeting human T2 lymphoblastoid cells (T2 cells) loaded with neoantigen peptides. Using an enzyme-linked immunospot (ELISpot) assay, we observed that the numbers of the secreted IFN-spots were significantly more when the stimulated CTLs responding to T2 cells loaded with six individually predicted neoantigens (peptide SI, AP1G1, UGT2B10, TPST1, OR2G2, and TGIF2LX) compared to DMSO (Figure 2A,B). Three previously validated peptides, cytomegalovirus (peptide CMV), [32] virus-associated antigen (peptide HCV: NS3), [33] and HCC-associated antigen (peptide VEGFR2-190), [34] were recruited as positive control peptides. We also detected the immunogenicity of other candidate neoantigen peptides, and most of them did not show significant differences compared to the control group ( Figure S2, Supporting Information). The LDH assay is used to detect the cytotoxicity of the stimulated CTLs when targeting neoantigen peptides. The results showed that the stimulated CTLs demonstrated significantly increasing cytotoxicity when targeting T2 cells loaded with neoantigens than DMSO. In detail, the cytotoxicity of stimulated CTLs reactivity to targeting neoantigens was at ≈92.9% for peptide SI, ≈60.5% for peptide AP1G1, ≈62.7% for pep-tide UGT2B10, ≈72.5% for peptide TPST1, ≈71.3% for peptide OR2G2, and ≈75.9% for peptide TGIF2LX. And the cytotoxicity of matured CTLs with HCC-associated antigen was at ≈42.8% for peptide VEGFR2-190 ( Figure 2C). Collectively, these results showed that the CTLs stimulated with immunogenic HCC neoantigens and HCC-associated antigens could secrete functional IFN-cytokines and have antigen-specific cytotoxicity.
To validate the binding affinity of the six functional neoantigens, the neoantigen-specific MHC class I tetramers were used to detect the positive frequencies of antigen-specific CD8 + T cells in CTLs. We synthesized tetramers with the QuickSwitch™ Quant HLA-A*02:01 Custom Tetramer-PE kit through switching the existing peptide, and detected the mean intensity of FITC in the existing peptide by flow cytometry (see Experimental Section). The switching efficiency of these neoantigen peptides was all over 90% ( Figure S3, Supporting Information). Then, we applied these synthetic tetramers to detect the proportions of neoantigenspecific CTLs. The results showed that the positive frequencies of antigen-specific CD8 + T cells were 0.5% for peptide SI, 3.2% for peptide AP1G1, 0.8% for peptide UGT2B10, 1.8% for peptide TPST1, 0.3% for peptide OR2G2, 0.8% for peptide TGIF2LX, 3.2% for peptide 25, and 0.5% for peptide VEGFR2-190, suggesting that all six functional HCC neoantigens exhibited the high binding affinity to HLA-A*02:01 ( Figure 2D).
We further validated the predicted affinity of the top six neoantigen candidates in an HLA-peptide binding assay, all of which showed a high binding affinity (mean fluorescence intensity (MFI) >150%, Figure S4, Supporting Information). In addition, we analyzed the gene expression level of 22 genes that carry the 25 SNVs derived from 364 HCC cases on the TCGA database, showing 68% of the genes with a considerable tumor abundance (transcripts per million, TPM >1; Figure S5, Supporting Information). For the top six neoantigen candidates, half of the genes (i.e., AP1G1, TPST1, and UGT2B10) are highly expressed (TPM >10), while the rest three genes (i.e., OR2G2, SI, TGIF2LX) exhibit a relatively low expression (TPM <1), suggesting that the tumor abundance can be an important restrictive factor associated with the peptide immunogenicity.

Generation of DLM and RLM
To better investigate the CTLs' response to neoantigens presented on cancer cells in a 3D tumor model, we established in vitro DLM and RLM with HepG2 cells. Decellularized wholeliver scaffolds were successfully fabricated using a combination of physical destruction and chemical detergents. Fresh livers ( Figure 4A,i) were frozen at −80°C and thawed at room temperature, and then were continuously perfused through the portal vein using Triton X-100/SDS. After a decellularization process, the native rat liver became translucent with remaining ECM components and easily visualized a well-preserved vascular network with PV for blue-ink and HV for red-ink ( Figure 4C,i). Subsequently, we introduced roughly 50 million HepG2 cancer cells mainly via BD infusion for recirculation within 30 min. After the seeding process, the recellularized liver was placed in the incubator for 2 h static culture-making cells to attach to the DLM structures, and then it was transferred into a sterile perfusion bioreactor for in vitro culture for 48 h. We found that the decellularization almost completely removed the DNA content from the fresh liver, while the recellularization efficiently repopulated about half of the cellular components back into the liver matrix ( Figure S6, Supporting Information). In addition, the immunofluorescent images confirm that the extracellular matrix compositions, including collagen I, collagen IV, fibronectin, and laminin, were well retained during the decellularization process ( Figure S7, Supporting Information). The H&E staining images and TUNEL images of native liver ( Figure 4A,ii,iii) compared to recellularized liver scaffolds ( Figure 4C,ii,iii) revealed an approximate cell distribution and activity, and the preservation of large vessels and patency and the existence of the vast majority of reseeded cells within RLM. After the HepG2 cells successfully attach with ECM, the matured CTLs were injected via PV with in vitro culture for 24 h.
Albumin, as a representative liver-specific marker, was evaluated to assess metabolic activity and synthetic functions of reseeding HepG2 cells using immunofluorescence. The extensive albumin production in fresh liver and the 3D recellularized scaffolds suggested the ability of HepG2 cells to maintain their hepatic function at a high level for 87.2% ± 3.7% and 84.6% ± 1.2% ( Figure 4B(i),D(i),E). Proliferation was evaluated by immunofluorescence staining for Ki67, and a large number of positive cells were detected in RLM for 94.8% ± 2.5%, while in the native liver was only 4.2% ± 1.0% ( Figure 4B(ii),D(ii),F). In addition, betacatenin ( -cantenin) as a cell-cell interaction indicator showed high expression at the boundary of cells, the native liver, and the RLM kept the function of hepatic cells at a close high level for 77.2% ± 3.5% and 74.5% ± 4.3% ( Figure 4B(iii),D(iii),G). Alphafeto protein ( -FP) as a valuable serum marker of hepatocarcinoma was detected with remarkable differences in the normal liver and RLM for respectively 16.7% ± 4.0% and 87.6% ± 2.6% ( Figure 4B(iv),D(iv),H). Therefore, the recellularized HepG2 cells kept high metabolic activities in the 3D RLM model, which suggested the suitability of mimicking tumor microenvironment and the HepG2 cell-recellularized liver model potentially could be used in immunotherapy.
Due to the maintenance of native vasculature systems, the HepG2 cells embedded in the DLM could well maintain the cell viability (Figure 5A,B). The 3D model remains the essential structural component and maintains the multiple-set hepatic vasculature, [35] which enables this model to be the most attractive approach for bio-mimicking the microenvironment and the intricate dynamics of a native liver tumor. [36] Thus, a large number of internal cells died during the continuous enlargement of the wildly used 3D spheroids ( Figure 5C), while RLM with adequate nutrition and oxygen supply maintained a significantly high cell survival rate ( Figure 5D). Moreover, this RLM holds the potential for the cytotoxic validation of CTLs in some way analogous to in vivo model. [37,38] Thus, RLM scaffolds provide an ideal strategy for mechanism-based therapy development, assessing antitumor toxicity of a variety of immune factors and cells.

Immunotherapy in the 3D RLM Liver Tumor Scaffolds
The 3D tumor model of DLM scaffolds with perfusion of HepG2 cells in the DLM scaffolds closely resembles the in vivo condition of the liver. We fabricated two biomimetic 3D micro-physiological liver cancer models by reseeding wild type or the mini-gene modified GFP-HepG2 cells for capturing in vivo pathological features and dissecting the potential immunogenetic mechanisms to drive CTLs killing progression and immunotherapy evaluation ( Figure 6A). As the H&E staining of neoantigenstable recellularized liver shown ( Figure 6D), CTLs (yellow arrow) were more evenly distributed in the refused minigene modified   GFP-HepG2 cell mass. The TUNEL staining of the corresponding region confirmed the activation of CTLs for 76.5% ± 5.2% and the inactivation of stable HepG2 cells for 23.5% ± 5.2% ( Figure 6E,H,I). In contrast, in the H&E staining image of wild-type RLM ( Figure 6B), CTLs (yellow arrow) were partial to concentrated beside the reseeded wild HepG2 cell mass. Furthermore, as the TUNEL staining of the corresponding region showed ( Figure 6C,H,I), the HepG2 cell-mass growth in normal status was measured in sharp contrast to the stable HepG2 cell group, with the activation of CTLs for 16.5% ± 3.4% and the inactivation of stable HepG2 cells for 83.5% ± 3.4%. Granzyme B, as a neutral serine protease in specific CTLs, localized in the cytotoxic granules of T cells and natural killer (NK) cells mediates commitment to T cell apoptosis and suppression of proliferation. [39] In addition, IFN-is a highly effective antiviral bioactive substance produced by mitogen-stimulated T lymphocytes, also a kind of lymphokine with extensive immunomodulatory effects. [40] PD-1 is an important immunosuppressive molecule that regulated the immune system and promoted self-tolerance by inhibiting the inflammatory activity of T cells. [41] To investigate the neoantigenspecific CTLs immune response targeting mini-gene modified HepG2 cells, we detected the distributions of the infiltrated T cells, PD-1 protein expression, Granzyme B, and IFN-secretion in RLM liver scaffolds seeded with two different HepG2 cells. The results showed that the extensive expressions of Granzyme B, CD3, IFN-, PD-1 were detected in the RLM liver scaffolds seeded with stable HepG2 cells, suggesting the presence of an immune killing response in stable HepG2 cells seeded RLM scaffolds ( Figure 6F) compared to wild type HepG2 cells seeded RLM scaffolds ( Figure 6G). Moreover, the cell fluorescence percentage of Granzyme B and IFN-in the immunofluorescence image of stable HepG2-seeded RLM scaffolds was significantly higher than that in the wild-type HepG2-seeded RLM scaffolds ( Figure 6J). Collectively, these results showed that the 3D RLM liver cancer model seeded with stable HepG2 overexpression with neoantigen minigenes provides a novel preclinical research model for neoantigen-specific T cells immunotherapy for HCC.

Discussion
Based on the advancement of biotechnological tools for cancer patients, biological therapy can counteract side effects in systemic cancer treatment compared with chemotherapy. [19] There are two main limitations in the development of antigen-specific cytotoxic T lymphocytes (CTL) therapy for HCC. One limitation is the lack of valuable therapeutic targets for HCC. The antigens with high binding affinities present effective and wide therapeutic potentials for cancers by effectively promoting T cells to recognize cancer cells, thereby avoiding immune tolerance and enhancing immune response. [18,42] Another limitation involves difficulties in screening highly specific and effective antigenic peptides as targets for HCC treatments. Traditional antigens that are selected from tumor-associated antigens (TAAs) can lead to prominent side effects, such as immune tolerance and low immune responses. [43] Therefore, screening and identifying TAAs can overcome the disadvantages of traditional antigens. In recent years, cancer immunotherapy has gained acceptance among medical practitioners. These immune therapies including immune-checkpoint antibodies, vaccines and adoptive cell immunotherapy (ACT) have been adopted into the medical treatment routines. [2,19,44,45] Immune checkpoints with high expression levels in tumor cells enable tumors to escape from the immune surveillance. Therefore, immune checkpoint antibodies or inhibitors can reactivate T cells to eliminate tumor cells, which have been applied in HCC immunotherapy. [17,[46][47][48] Conventionally, the 2D coculture of suspension T cells with the adherent HepG2 cells makes it challenging to use imaging tools to characterize and confirm the immune responses between the two cell types. [49] In addition, the previously reported 3D cell models for immunotherapy are mainly restricted by apoptotic effects due to the lack of perfusable vascular networks, hardly providing a tumor-mimicking environment for immune infiltration evaluation. [50,51] Therefore, the immune infiltration of cytotoxic T lymphocytes (CTLs) into solid tumors, which plays a vital role in the immunotherapy efficacy, can only be detected in in vivo assay, but not in vitro assay. [52][53][54] Here we for the first time solved the challenge by developing a 3D pre-clinical cancer cell model incorporating with complex perfusable vasculatures that allow the cytotoxic T cells perfuse through the vasculatures, extravasate from the vessels into tumor interstitium, and finally infiltrate inside the tumor tissues to elicit immune responses. Specifically, using the 3D RLM model, we not only confirmed the CTL cytokine secretion and cytotoxicity to the predicted neoantigens as observed in 2D cultures, but also directly detected the infiltration and destruction processes of the CTLs that have only been reported in the animal models. [55] In particular, we found that the CTLs were highly active and exhibited profound tumor infiltration capability in the minigene-modified HepG2 tumor model, but remained in the peripheral regions in response to the control tumor model not expressing the cognate antigens. Therefore, the 3D model provides a unique organ-structured platform that enables the CTLs to infiltrate the tumor, recognize the neoantigens, and kill the cancer cells, replicating the key steps for the CTLs to execute their functions in vivo. Clearly, the 3D model developed in this work successfully builds the missing link between conventional in vitro 2D cultures and in vivo animal models.
We understand that some limitations still exist in this study. First, the recellularized liver matrix-based 3D tumor model, which consists of HepG2 cells and cytotoxic T cells, cannot fully replicate the complexity and heterogeneity of the tumor microenvironment found in patient tumors. The vasculature systems in the organ-structured tumor model remain intact but probably become more permeable after the removal of endothelial cells upon the decellularization process, which may allow the cytotoxic T cells to extravasate easier into the tumor tissue. Yet this would not affect the tumor infiltration of the cytotoxic T cells. Due to the flexibility of the organ-structured tumor model, further endeavors can extend to recellularizing endothelium in the vasculatures to offer a better bio-mimicking scaffold for immunological evaluation. Moreover, the stromal environment of HCC, mainly comprising the cellular components of hepatic stellate cells, macrophages, and endothelial cells, as well as stroma-related extracellular matrix, plays a key role in providing a chronic inflammation environment that causes liver fibrosis, the major risk factor associated with up to 90% of HCC patients. [56] The tumor-stroma interaction in HCC is critical but remains poorly understood. [57] Currently, there is still a lack of suitable in vitro models for evaluating the interplay between the stromal environment and the tumor cells. [11] The 3D organstructured model developed in our study can potentially provide an advantageous in vitro platform to recapitulate the tumor microenvironment by reseeding a variety of cells into the decellularized matrix to achieve a better understanding of the underlying interactions and mechanisms for various immune therapeutics.
Second, the 3D tumor model lacks a complete representation of the immune system. While the model allows for the interaction between tumor cells and immune cells, it does not fully capture the systemic immune response and the complex interplay of different immune cell populations. The absence of other immune organs and systemic immune modulation factors may limit the comprehensive evaluation of neoantigen-specific immune responses. Further development may be achieved by incorporating microphysiological systems through organs-on-chips into the recellularized 3D tumor model for a meticulous reconstruction of the in vivo environment. [58] Third, the generation of the recellularized 3D tumor model requires liver organs from animals, which may restrict the widespread application and scalability due to ethical considerations. Moreover, the organ resources feature inherent variability, making it more challenging to establish standardized protocols as compared to conventional in vitro models. Therefore, further investigation may be needed to test the consistency and reproducibility of the organ-derived tumor model for the evaluation of different immune therapeutics.

Conclusion
In summary, our study could provide a novel biotechnological strategy to predict and evaluate neoantigens in HCC patients or other cancer patients in 3D in vitro model. Antigenic immunogenicity and cytotoxicity in CTLs could be first identified with ELISpot assay and LDH assay in vitro. The percentage of antigenspecific CD8 + T lymphocytes could be detected with peptide-MHC class I tetramer staining and flow cytometry. In addition, the stable-neoantigens cells proved the enhanced antigenic immunogenicity and cytotoxicity in CTLs with the tandem antigenic genes of peptide SI, AP1G1, UGT2B10, TPST1, OR2G2, and TGIF2LX. Furthermore, RLM was used to evaluate the immunotherapy killing efficiency which was mutually verified for in vivo immune treatment, which to some extent predicts the results of animal experiments. The RLM remains blood supply vessels (IVC, inferior vena cava), recellularized vessels (BD, bile duct), and CTLs injected vessels (PV, portal vein). A perfusable device with large vascular-like channels and capillary structures was fabricated to facilitate 3D tissue regeneration, which maintained proper tissue functions and substance exchange ability. [15,59] Finally, these results revealed that the RLM-neoantigens model had the potential in vaccine applications with high efficiency and specificity on personalized immunotherapy for cancer patients in the future.

Experimental Section
Ethics: The whole animal experimental procedures were approved by the Institutional Animal Care and Use Committee at the Southern University of Science and Technology (SUSTC-JY2018034) and completed in a sterile environment. Human PBMCs were approved by the Ethics Com-mittee of Shanghai Zha Xin Hospital of Integrated Traditional Chinese and Western Medicine (202 007). All patients provided written informed consent.
Mutation Selection and Epitope Prediction: Screening of high-frequency mutation sites in liver cancer were based on the TCGA database. The top 10 high-frequency mutation sites were selected. Cell somatic mutation data of 364 liver cancer whole exome sequencing results were downloaded from the TCGA database (https://portal.gdc.cancer.gov/projects/TCGA-LIHC) (version 20170929). The top 10 high-frequency mutation peptides with the highest coverage ratio and HLA typing A0201 were selected (Table  S1, Supporting Information), and the top six high-frequency mutation peptides with the strongest affinity were obtained after peptide-MHC affinity prediction ( Table 1).

Development of a Peptide Epitope Prediction Program and Prediction of Tumor-Specific Peptides:
Based on the 10 high-frequency mutation peptides selected, a script was used to generate 9-mer peptide sequences corresponding to the peptide sites. To identify potential candidates for developing a new vaccine against cancer, PSSMHCpan, NetMHC-4.0, NetMHCpan-3.0, PickPocket, and SMM were included in the process to predict new antigens in the TCGA liver cancer samples. PSSMHCpan, NetMHC-4.0, NetMHCpan-3.0, PickPocket, and SMM were used to predict new antigens. Finally, candidate new antigens were selected that met the following criteria: 1) at least 2 software predictions are binding complexes (IC50 <500 nm), and the minimum value of IC50 values of the software that predicts binding complexes is used as the final prediction result; 2) the IC50 value of the new antigen derived from the given SNV must be smaller than the IC50 value of its corresponding wild-type peptide.
A program was developed for peptide epitope prediction, which utilizes bioinformatics analysis methods to predict the tumor-specific antigen peptides corresponding to the mutation sites. The whole exome sequencing results of 364 liver cancer cell mutation data were downloaded from the TCGA database (https://portal.gdc.cancer.gov/projects/TCGA-LIHC). Missense SNVs loci were screened that were at least replicated in two patients and obtained the top 20 most frequent mutation sites in the liver cancer patient population with HLA-A0201 typing as an example (Table S3, Supporting Information). Based on the variation information, the amino acid sequence was truncated into short peptides with a length of 9-11 bp and containing the variant amino acids. Then, short peptides containing the mutated amino acids with a length of 9-11 bp were generated using the script. To identify potential candidate markers for developing anti-tumor vaccines and optimize prediction accuracy, NetMHC-4.0, NetMHCpan-3.0, NetMHCpan-4.0, PSSMHCpan-1.0, PickPocket-1.0, and SMM were incorporated into the workflow for predicting antigen affinity. Also, the cell surface antigen delivery prediction software EPIP was used to predict antigen delivery ability, and detect new antigens in TCGA liver cancer samples. Finally, the following criteria for candidate new antigens were selected: 1) at least 3 software predicted results as complexes (IC50 <500 nm), and the minimum IC50 value of the software predicting results as complexes was taken as the final prediction result; 2) EPIP score > = 0.5; 3) ELscore < 2. Based on the IC50 score, IC50 <500 nm indicates affinity, IC50 <50 nm indicates high affinity, and peptides with IC50 <500 nm were selected. The lower the IC50 value, the higher the affinity. Based on the delivery ability score, a higher score indicates a stronger delivery ability.
Using Software to Predict the Tumor-Specific Peptides and the Affinity and Delivery Ability of the MHC Molecules: A new antigen prediction was performed on the 9-11 bp short peptides for the HLA typing HLA-A*02:01, and based on the results of delivery ability and affinity prediction, the top 20 highly frequent mutated peptides with the strongest antigen delivery ability and affinity were selected (IC50 <500 nm and EPIC score ≥ 0.5 and ELscore < 2) ( Table 2).
Cells: HLA-A*02:01-restricted human purified peripheral blood mononuclear cells (PBMCs) were provided by Milestone Biotechnology company (MT-BIO, Shanghai, China), 50 million cells/a cryopreserved vial. PBMCs were thawed in a water bath at 37°C, diluted with 20 mL RPMI-1640 medium (Gibco, NY, USA), and collected with a 10 min centrifugation at 300 × g. Subsequently, PBMCs were cultured in RPMI-1640 medium supplemented with 10% Fetal Bovine Serum (FBS, Gibco, NY, USA) for 6-12 h. The human T2 cell line was provided by BGI. T2 cells www.advancedsciencenews.com www.advmatinterfaces.de with expression of HLA-A*02:01 molecules can be loaded with exogenous peptides and act as target cells for immunogenicity assay and cytotoxicity assay. T2 cells were cultured in Iscove's Modified Dubecco's Medium (IMDM, Gibco, NY, USA) supplemented with 10% FBS in an incubator containing 5% CO 2 at 37°C.
Reagents: Neoantigen peptides, virus-associated peptides, tumorassociated peptides, and CMV peptide (sequence: NLVPMVATV, positive control) were synthesized by Beyotime Biotechnology (Beyotime Biotec, Shanghai, China) according to commercially manufacturer's procedures. The synthetic peptides of predicted neoantigens are listed in Table 1 and  Table 2, and the information of virus-associated peptides and tumorassociated peptides are listed in Tables S1 and S2 (Supporting Information). The purity of the peptides was over 95%. Peptides were dissolved in dimethyl sulfoxide (DMSO, Shanghai, China) at the concentration of 5 mg mL −1 , and stored at −20°C. CD8 + T Cells Selection Using Magnetic Beads: After 6-12 h incubation, PBMCs were washed with 20 mL MACS buffer. Cells were collected with a 10 min centrifugation at 300 × g and counted with a cell counter (Bio-Rad, CA, USA). Then, 80 μL MACS buffer and 20 μL CD8 microbeads were added per 10 7 cells, and incubated at 4°C for 15 min with shaking once per 5 min. After incubation, cells were washed with 20 mL MACS buffer and collected with a 10 min centrifugation at 300 × g. Then, CD8 + T cells were sorted by the magnetic frame and MS column and cryopreserved until use. CD8 negative cells were collected with a 10 min centrifugation at 300 × g, and counted with a cell counter. Then, 80 μL MACS buffer and 20 μL CD14 microbeads were added per 10 7 cells, and incubated at 4°C for 15 min with shaking once per 5 min. Cells were washed with 20 mL MACS buffer, and collected with a 10 min centrifugation at 300 × g. Then, CD14 + cells were sorted by the magnetic frame and MS column, and cultured in Cellgenix DC (CG-DC) medium supplemented with 5% FBS, 10 ng mL −1 human recombinant IL-4, and 80 ng mL −1 human recombinant GM-CSF in an incubator containing 5% CO 2 at 37°C for 48 h.
Culture of CD14 and CD8 + Cells: After 48 h, the half culture medium was changed by CG-DC medium supplemented with 5% FBS, 20 ng mL −1 IL-4, and 160 ng mL −1 GM-CSF for another 48 h. Then, all culture medium was changed, and cells were stimulated by 10 ng mL −1 IL-4, 10 ng mL −1 LPS, 10 ng mL −1 IFN-, and 80 ng mL −1 GM-CSF in CG-DC medium supplemented with 5% FBS at 37°C for 24 h. Peptides were added with 10 μg mL −1 to each well individually in the incubator with 5% CO 2 at 37°C for 16 h. Then, CD8 + T cells were thawed, and cultured in CG-DC medium supplemented with 5% FBS, 10 ng mL −1 IL-7 in an incubator with 5% CO 2 at 37°C.
Co-Culture of CD8 + Cells with Matured CD14 + Cells: The co-culture system of CD8 + T cells with matured CD14 cells was developed at a ratio of 4:1, and cultured in CG-DC medium supplemented with 5% FBS, and 30 ng mL −1 IL-21 in an incubator with 5% CO 2 at 37°C for 72 h. Half culture medium was changed by CG-DC medium supplemented with 5% FBS, 10 ng mL −1 IL-2, IL-7, and IL-15 for 48 h. Then, all culture medium was changed by CG-DC medium supplemented with 5% FBS, 10 ng mL −1 IL-2, IL-7, and IL-15 for another 48 h. The cell passages were performed according to the cell density.
IFN-ELISpot Assay: The IFN-ELISpot assay kit (Mabtech, Nacka, Sweden) is a classical method of immune enzyme technology with high sensitivity that can realize the relatively quantitative detection of cytokines secreted by effector cells in a 96-well plate pre-coated with anti-human primary antibody (IFN-). The detailed procedures were introduced as following. When CD8 + T cells (effector cells) were already activated, T2 cells were collected and loaded with neoantigen peptides, virus-associated peptides, tumor-associated peptides, and CMV peptide individually, and incubated at 37°C for 4 h. The concentration of neoantigen peptides used to stimulate the antigen-specific immune response in each individual peptide and peptide mixture groups was 10 μg mL −1 . The ELISpot plate was washed by PBS with 150 μL per well for five times, 5 min each time, blocked by 100 μL IMDM medium plus 10% FBS per well, and incubated at 3°C for 30 min. Then, 1 × 10 4 T2 cells and 2.5 × 10 4 CD8 + T cells were resuspended in CG-DC medium supplemented with 2% FBS with 100 μL per well, respectively. The negative control group (loaded with DMSO) was also added into the plate, and incubated at 37°C for another 16-20 h. All cells were removed from the plate, and the plate was washed with 150 μL PBS per well five times, 5 min each time. The secondary antibody (7-b6-1-ALP, Biotinated Alkaline Phosphatase) was added into each well at a ratio of 1:200 with 100 μL PBS plus 0.5% FBS per well, and incubated at 37°C for 2 h. The plate was washed with 150 μL PBS per well five times, 5 min each time. Subsequently, 100 μL substrate (NBT/BCIP, Nitro-Blue-Tetrazolium/5-Bromo-4-Chloro-3-Indolyl Phosphate) was added with per well, followed by reaction in the dark for 2-8 min. Lastly, the plate was rinsed quickly with water and stored in a drying box. Finally, the AID ELISpot Reader (iSpot, AID, Germany) was used to count the number of purple spots in each well.
Peptide-MHC Class I Tetramer Detection of Antigen-Specific CD8 + T Cells: The peptide-MHC class I tetramer staining with flow cytometry is a powerful and standard diagnostic technology to detect antigen-specific CD8 + T lymphocytes in cancer and autoimmune diseases. [60,61] The QuickSwitch Quant HLA-A*02:01 Custom Tetramer-PE kit was purchased from MBL Biotechnology Company (Nagoya, Japan). 2 mm of neoantigen peptides, virus-associated antigen peptides or tumor-associated antigen peptides were dissolved with DMSO, which can exchange the existing peptide on the control tetramer-PE. The exchange efficiency (higher than 75% was better) was detected with the FACSCanto Analyzer. Moreover, the QuickSwitch calculator was shown on the website (http://www.mblintl. com/quickswitch-peptide-exchange-calculator/). Then, 1 × 10 6 stimulated CD8 + T cells were stained with tetramer-PE (1:100 in PBS) and the CD8-FITC antibody (HIT8 , 1:1000 in PBS) at 4°C for 30 min in the dark. Cells without staining were used as control. Finally, CD8 + T cells were washed with PBS plus 0.5% FBS and detected with the FACSCanto Analyzer. FlowJo software V10 was used for data analysis.
LDH Cytotoxicity Assay: The LDH assay kit (Promega, WI, USA) was used to detect the cytotoxicity of effector cells (CTLs) stimulated by antigenic polypeptides, as described in previous studies. [62,63] CD8 + T cells were stimulated by antigens presented by DCs, while T2 cells were used as target cells, which were collected and incubated with antigen peptides at 37°C for 4 h. Then, 1 × 10 4 T2 cells and 5 × 10 4 CTLs were co-cultured with a ratio of 1:5 at 37°C for another 4 h. When antigen-specific CTLs reactivated to T2 target cells, the permeability of the cell membrane increased, and LDH was released into the supernatant. The released LDH change was presented as formazan compound with a high absorption at a wavelength of 490 nm. The microplate reader (Synergy H1, BioTek, USA) was used for OD value measurement, which represents efficiency of cytotoxic activity of CTLs to target cells. The percentage of cytotoxicity in CD8 + T lymphocytes was determined using the following formula: cytotoxicity (%) = [(effector/target LDH release − effector spontaneous LDH release − target release)/(maximum LDH release − target release)] × 100%.
Peptide Binding Assay: Peptide binding to HLA-A*02:01 was assessed using a FACS-based MHC-I stabilization assay. Incubation of TAP-deficient cell line T2 with peptides stabilizes surface MHC-I expression, and an increase in surface MHC class I levels can be measured by flow cytometry. In this study, peptide binding to HLA-A*02:01 could be evaluated by an HLA-A*02:01-specific antibody staining for an increased HLA-A expression on T2 cells. Briefly, T2 cells were incubated in a 24-well plate at 1 × 10 6 cells in a 1 mL IMDM medium with 100 μg mL −1 peptide per well for 18 h at 37°C, and the same volume of DMSO for the control group. Cells were then washed and surface levels of HLA-A*02:01 were assessed by staining with HLA*A02:01 specific antibody (BB7.2-PE, 343306, 1:1000, Biolegend) at 4°C for 30 min in the dark. Cells were then washed twice and suspended in 200 μL PBS with 0.5% FBS for flow cytometry on BD FACS Aria II. The CMV peptide (NLVPMVATV), which binds with high affinity to HLA-A*02:01, was used as a positive peptide. The peptide binding affinity was evaluated by fluorescence index (FI) values, and calculated based on the relative mean fluorescence intensity (MFI) using the following equation.
Preparation of 3D HepG2 Spheroids: Eighty-one 400 μm diameter rubber micro-molds were purchased from Micro Tissues Inc. (Providence, RI, USA). A 2% (g/mL) solution of agarose powder (Sigma) mixed with DMEM was autoclaved for sterilization. Five hundred microliters of the liquid solution was pipetted into each mold and allowed to solidify. The micro-well plates were removed and equilibrated in six-well culture Petri dishes (Corning, NY, USA) with 2 mL DMEM containing 1% P/S for 24 h. The medium was then removed and 200 μL of a 5 × 10 5 cells mL −1 cell suspension was carefully seeded into each well. Two milliliters of DMEM, 10% FBS, and 1% P/S were added to the culture Petri dishes and incubated for 4 days at 37°C and 5% CO 2 . The spheroids formed were harvested for further experiments or transferred to new non-adhesive plastic Petri dishes for imaging. For the soft agar assay, the collected spheroids were pretreated for 20 min with 0.25% EDTA trypsin to prepare a cell suspension, and 1 × 10 4 cells were seeded into a medium containing 0.6% agarose. The culture dishes were incubated at 37°C and 5% CO 2 and colonies were recorded after 7 days of growth. The culture medium was changed every 2 days and contained DMEM, 10% FBS, and 1% P/S. Donor Liver Decellularization: Male Sprague-Dawley rats (weight>500 g) were provided for liver harvesting after being euthanized by carbon dioxide asphyxiation, and underwent laparotomy. A three-dimension of liver scaffold was prepared with vessels involving the portal vein (PV), hepatic vein (HV), and BD, which were cannulated with flat pinhead needles. The fresh livers were stored in autoclaved Petri dishes sealed with parafilm, and frozen at −20°C until decellularization. The liver was decellularized with the next three steps. First, when the liver was returned to room temperature, the distilled water was used to perfuse for 30 min at a flow rate of 8 mL min −1 . Second, 4% Triton X-100/0.02% EGTA aqueous solution was used to perfuse for 1 h at 10 mL min −1 . Finally, 1% sodium dodecyl sulfate (SDS) solution was used to perfuse for 3 h at 5 mL min −1 until the liver becomes transparent. All reagents were autoclaved and perfused through PV.
HepG2 Cells Recellularization: DLM was rinsed with PBS for 2 h at 5 mL min −1 to remove residual fixation fluid before the recellularization process. Then, the DLM was transferred into a 60 mm Petri dish and flushed with 15 mL of complete DMEM medium (containing 10% FBS and 100 IU penicillin/streptomycin) via the PV and stored in an incubator at 37°C for 30 min. Before liver recellularization, the HV and PV of DLM were ligated during the perfusion. HepG2 cells were re-suspended in an injection syringe with a total of 80 million cells (≈10 mL cell suspension) via BD manually within 30 min. The injection process was repeated two to three times. After perfusion, the liver scaffold was transferred into an incubator for 2 h. After culturing, the needles on inferior vena cava (IVC) and PV were removed. The harvest of recellularized liver was kept in a bioreactor connected to a peristaltic pump (Longer Pump) at 37°C and lasted for 48 h. The co-culture T cells were proceeding perfused through PV with 5 million cells, and then incubated in the bioreactor for 24 h with DMEM solution cycle. The samples were then fixed in PFA for 24 h before H&E staining and immunostaining, which was used to assess the toxic effects of cells to T cells in RLM.
Statistical Analysis: Results were shown as mean ± standard deviation (SD) of three independent experiments, unless otherwise specified. A pvalue < 0.05 was considered as statistical significance.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.

Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.