HBs_371-379 and HBs_370-379 epitope sequences were retrieved from 5855 unique full-length HBV surface antigen (HBs) polyproteins (389, 399 and 400AA HBs polyprotein sequences) collected from UniProt around May 2019. First, there were 134 unique nonameric or 190 unique decameric peptides extracted respectively from HBs_371-379 and HBs_370-379 epitope sequences with Strawberry Perl 5.26.1.1 (http://strawberryperl.com) after removing duplication. Second, the most common 20 peptides of HBs_371-379 and HBs_370-379 were statistically analyzed from unique 5657 (97% of 5855) and 5528 (94% of 5855) full-length HBs polyprotein sequences separately. Finally, the most common 20 peptides that already categorized the HBV genotypes in UniProt were statistically analyzed with HBV genotypes. The corresponding HBV genotypes of the most common eight peptides of HBs_371-379 (vrt1 to vrt8) and HBs_370-379 (vrt0, vrt9 to vrt16) were analyzed with Package ‘pheatmap’

(https://cran.r-project.org/web/packages/pheatmap/index.html) based on statistical software R V.3.6.1 (https://www.r-project.org/).

Potential cross-reactive human peptides were identified by Expitope webserver 2.0 (http://webclu.bio.wzw.tum.de/expitope2/) designed to predict the similar peptides and its potential presentation by the HLA allele of interest.36 37

For identifying vrt1 and vrt0 related human cross-reactive peptides, we combined the results of the alanine scanning to determine the critical TCR binding position of vrt1 and vrt0. The maximum allowed number of mismatches were set as three for vrt1 and four for vrt0 for the primary screening. The chosen allele was HLA-A*02:01. The default parameters were used for the other options. We also carried out a second screening without fixing the critical positions, and the maximum allowed number of mismatches were set two for vrt1 and three for vrt0. Those potential peptides derived from proteins with expression data were proceeded the third step of T cell activation assays.

The 16 most frequent HBs_371-379 (HBs371) and HBs_370-379 (HBs370) antigenic peptides were retrieved from 5855 full-length HBsAg polyproteins. There were eight HBs371 nonameric peptides, that is, vrt1-ILSPFLPLL, vrt2-ILSPFIPLL, vrt3-ILSPFMPLL, vrt4-ILNPFLPLL, vrt5-IVSPFIPLL, vrt6-TLSPFLPLL, vrt7-ILNPFIPLL and vrt8-IVRPFIPLL, and eight HBs370 decameric peptides, that is, vrt9-NILSPFLPLL, vrt10-NILSPFIPLL, vrt11-SILSPFLPLL, vrt12-NILSPFMPLL, vrt13-NILNPFLPLL, vrt14-SILSPFIPLL, vrt0-SIVSPFIPLL and vrt16-NILNPFIPLL, that were analyzed (online supplemental figure S1, tables S1.1 and S1.2).

The irrelevant peptide-HLA complexes (pHLAs) (online supplemental table S2) were used to verify the specificity of the TCR mutants. Alanine-scanning peptides and potentially cross-reactive human peptides related to vrt1 and vrt0 (online supplemental tables S3.1 and S3.2) were used to test the binding site and TCR specificity, respectively. All peptides were purchased from Genscript and validated with a mass spectrometer, and the purities were confirmed to be greater than 95%.

T2, J.RT3-T3.5 cell, 293T and HepG2 were purchased from ATCC. HCC cell lines, PLC/PRF/5 were provided by Dr Xiaoping Chen from Guangzhou Institutes of Biomedicine and Health, Chinese Academy of Sciences. PBMC, provided by informed consent healthy donors, are listed in online supplemental table S4 and 12 sets of human primary cells from ScienCell are listed in online supplemental table S5, including human bronchial smooth muscle cells, human aortic smooth muscle cells, human meningeal cells, human renal mesangial cells, human gastric smooth muscle cells, and human renal epithelial cells. Cell culture dish was treated with Poly-L-Lysine before culturing. All cell lines were authenticated using short tandem repeat profiling and regularly tested negative for mycoplasma contamination throughout the whole duration of this study.

HLA-A2 negative HCC cell line PLC/PRF/5 has multiple, random integrations of HBV DNA and can express HBsAg. PLC/PRF/5 cells were transduced with HLA-A*02:01 subtype with pLenti-CMV-puro lentivirus vectors to constructed HLA-A*02:01⁺ cell lines. The transduced cells were maintained with the selection of 2 µg/mL puromycin (BD Biosciences). Cells were stained with APC-conjugated anti-human-HLA-A2 monoclonal antibody (BioLegend) and positive cells were isolated by sorting on a FACSAria III flow cytometer (BD Biosciences).

We constructed Vα and Vβ chain single-chain TCR variable domain (scTv) evolution libraries using TCR0, which was isolated previously with vrt0-SIVSPFIPLL. The nonapeptide variant vrt1-ILSPFLPLL HBs371-HLA-A*02:01 complex was used as the antigen for bio-panning to isolate affinity-improved TCR variants from the libraries using methods described previously.22 38 Briefly, the scTv libraries were displayed on the surface of the filamentous bacteriophage M13, and affinity-improved scTv variants were selected with biotinylated vrt1-pHLA captured on streptavidin-coated Dynabeads (Life Technologies). Binding improved scTvs were analyzed in soluble heterodimeric TCR molecules that were purified after refolding Escherichia coli-produced inclusion bodies. The TCR affinities were measured by surface plasma resonance with ProteOn and Biacore.

TCR α/β chain transgene lentiviral vectors contained a picornavirus 2A sequence. In addition, the mouse TCR constant domains replaced the human constant domains to reduce mispairing.

PBMCs, provided by healthy human volunteer donors under Institutional Review Board approved informed consent, were prepared with human Lymphoprep reagent (Axis-Shield). Peripheral blood lymphocytes (PBLs) prepared with the PBMCs were activated with Human T-Activator CD3/CD28 Dynabeads (Life Technologies) at a bead-to-cell ratio of 1:1 in RPMI 1640 medium supplemented with 10% fetal bovine serum and 100 IU recombinant human IL-2. Lentiviral vectors for WT HBsAg-TCR, affinity-improved HBsAg-TCRs and control-TCR-A6 (a TCR specific for the HTLV-1 Tax_11-19 (LLFGYPVYV)-HLA-A*02:01 complex) were transduced into PBLs, which were supplemented with protamine (10 µg/µl). After proliferation, T cells were stained with FITC-conjugated or APC-conjugated antibodies (BioLegend) for human CD3, CD8, TCR Vβ13.1 or mouse TCR β-chain, and a PE-conjugated HTLV-1 Tax_11-19-pHLA or vrt1-pHLA tetramers to verify TCR expression.

Interferon gamma (IFN-γ) secretion was measured with ELISPOT kits (BD Biosciences). Briefly, 2×10³ positive TCR-T cells were incubated with T2 cells (HLA-A*02:01⁺) pulsed with vrt0-vrt16 or irrelevant peptides from human antigens (ranging from 10⁻⁶ to 10⁻¹³ M), HCC cells (HLA-A*02:01⁻ PLC/PRF/5 or HLA-*02:01⁺ PLC/PRF/5 cells), PBMCs or human normal primary cell lines at an effector-to-target (E:T) ratio of 1:10. To compare specificity between WT and affinity-improved TCRs, TCR-T cells were incubated with T2 cells pulsed with the irrelevant antigen SLLMWITQC-HLA-A*02:01 in a range of 10⁻⁶ to 10⁻¹³ M. The other two negative controls were unpulsed T2 cells and HLA-A*02:01⁺ HBsAg⁻ HepG2 cells.

Cell lysis was determined by releasing lactate dehydrogenase (LDH) with a CytoTox 96 Non-Radioactive Cytotoxicity Assay kit (Promega). T2 cells (5×10⁴) pulsed with vrt0-vrt16 (ranging from 10⁻⁶ to 10⁻¹³ M) were incubated with an equal amount of TCR-T cells overnight. Unpulsed T2 and 10⁻⁶ M irrelevant SLLMWITQC-peptide-pulsed T2 cells were set as negative controls. Cell-line assays were carried out with the HCC cell line PLC/PRF/5 transduced with HLA-A*02:01 at an E:T ratio of 1:1 and 3:1, and the controls were HLA-A*02:01⁻ PLC/PRF/5 cells and HBsAg⁻ HLA-A*02:01⁺ HepG2 cells. Experiments were performed in triplicate.

Immunodeficient male NOD/SCID IL2rg^−/− (NSI) mice aged 6 to 9 weeks were provided by Dr Peng Li at the Guangzhou Institutes of Biomedicine and Health, CAS.39 Mice were maintained with daily monitoring under controlled sterile environment conditions on a 12 hours light/dark cycle, and received a rodent diet sterilized with 60Co-γ-ray irradiation and sterilized water. Experimental protocols were approved by the Institutional Animal Care and Use Committee (IACUC). Each mouse was injected subcutaneously with 3.0×10⁶ HLA-A*02:01⁺ PLC/PRF/5 HCC cells on day 0. After the tumors reached nearly 60 mm³ (day 13), the mice were randomly assigned to groups and treated with phosphate-buffered solution (PBS), 3.0×10⁷ TCR-A6-T cells, 3.0×10⁷ TCR0-T cells or 3.3×10⁶, 1.0×10⁷ or 3.0×10⁷ TCR15-T cells via tail vein injection. Tumor volumes were measured every 3 days. From the criteria of IACUC, for subcutaneous tumors the maximum allowable size is 20 mm in diameter for a mouse. So, on day 34, the tumor-bearing mice were euthanized by cervical dislocation, and tumor tissues were resected and weighed.

Histological HCC tumor sections were stained with H&E or antibodies for IHC. Primary anti-human CD3 (Dako) and anti-human CD8 (ZSGB-BIO) monoclonal antibodies and Dako REAL EnVision Detection System were used for IHC, and DAPI (blue) stained nuclei. The sections were imaged with a LAIKA2500, and cell numbers were calculated with Image-Pro Plus 6.0.

Commonly, phage display bio-panning for affinity improvement applies the antigens used for the isolation of WT molecules to maintain specificity. However, TCR binding shows a certain degree of promiscuity. Therefore, the nonameric antigen vrt1-ILSPFLPLL pHLA, the variant of decameric antigen vrt0-SIVSPFIPLL used to isolate the WT TCR, was employed for affinity evolution, and we named this kind of selection flexi-panning. Nineteen TCR variants with CDR3α mutations were identified and showed affinity-improved binding to the vrt1-pHLA (table 1).

Compared with TCR0-T cells (WT), TCR6-T, TCR11-T, TCR14-T, TCR15-T, TCR17-T and TCR19-T cells were Ai-TCR-T cells that could be specifically activated by T2 cells pulsed with relatively low-level antigens (figure 1A,B). Four TCR variants demonstrated stronger capabilities than WT TCR0 for redirecting T cells, with TCR14-T, TCR15-T, TCR17-T and TCR19-T cells showing significantly better half maximal effective concentration (EC₅₀) values. TCR14-T cells and TCR15-T cells showed the best activation with the HLA-A*02:01⁺ PLC/PRF/5 cell line (figure 1C) despite showing relatively little improvement in the EC₅₀ by peptide titration (online supplemental table S6).

TCR14-T and TCR15-T cells exhibited strong cytotoxicity, as measured by assessing the lysis of vrt1-pulsed T2 cells (figure 2A,B). However, only TCR15-T cells showed specific cytotoxicity with minimal background cytotoxicity. The EC₅₀ of vrt1-pulsed T2 cell lysis by TCR15-T cells was nearly 38-fold better than that by TCR0-T cells (online supplemental figure S2). When HBsAg-specific cytotoxicity was investigated by targeting HLA-A*02:01⁺ PLC/PRF/5 HCC cell lines, both WT-TCR-T cells and Ai-TCR-T cells exhibited dose-dependent target cell lysis, and TCR15-T cells caused significantly enhanced LDH release compared with TCR0-T cells (figure 2C).

These data demonstrated TCR15 transduced T cells with acceptable activation, killing potency and specificity. Peptide vrt0-isolated TCR0 was compared with vrt1-pHLA-isolated TCR15 for binding to the two distinct antigens by Biacore. The mutant TCR15 had over sevenfold and fivefold higher binding affinities for the variant vrt1-pHLA and the original vrt0-pHLA antigens, respectively (table 2). TCR15 also demonstrated modest affinity enhancement for binding to vrt0-pHLA, despite vrt1-pHLA being used as the ligand for molecular evolution. This result revealed that flexi-panning could maintain the original affinity pattern of the TCR.

The specificity of TCR0 and TCR15 was verified, and these TCRs showed no binding with a panel of irrelevant pHLAs (online supplemental table S2). A three-step approach for an in-depth study was carried out for potential cross reactivity on relevant pHLAs; the first step involved alanine scanning of vrt0 and vrt1, the second step involved searching for cross-reactive antigens in the available human proteome expression data (online supplemental tables S3.1 and S3.2), and the third step involved verifying these antigens with pulsed T2 cells. Safety was measured with a panel of normal cells. Based on the cell activation showed in the alanine scanning, the residue positions F5 and L8 of vrt1 and F6 and L9 of vrt0 are crucial for maintaining the interactions between the TCR and HBsAg peptide-pulsed HLA-A*02:01⁺ T2 cells (online supplemental figure S3A,B). We identified a total of 33 vrt1-related potentially cross-reactive human peptides, 1 of which contained 2 mismatch residues and 32 of which contained 3 mismatches plus invariable F5 and L8 residues. For vrt0, we identified 2 sequences containing 2 or 3 mismatches and 10 sequences containing 4 mismatches with fixed F6 and L9 residues. All 45 potentially cross-reactive candidates were verified for agonist capabilities on peptide-pulsed T2 cells over a concentration range from 10⁻⁶ to 10⁻⁸ M. We found that most of these peptides, except high concentration (10⁻⁶ and 10⁻⁷ M) vrt1-L1, vrt1-L5, vrt1-L12, and vrt1-L19, which mainly derived from poorly expressed proteins, showed no agonist capability (figure 3A,B, online supplemental figure S4). TCR0-T cells but not TCR15-T cells appeared to be activated by T2 cells pulsed with vrt1-L13 derived from a poorly expressed protein.

To verify TCR-T cell specificity with primary cells, TCR15-T and TCR0-T cells were tested and showed no activation by any PBMCs from 30 different healthy donors (figure 3C, online supplemental table S4). Further verification of the TCR15-T cell interactions with antigens presented by normal cells was carried out with 12 sets of human primary cells (online supplemental table S5). These normal cells could not activate TCR0-T or TCR15-T cells (figure 3D).

In summary, these results demonstrated that TCR15-T cells showed HBsAg specificity similar to that of WT-TCR-T cells and at least an equivalent safety profile under a range of measurements.

The specificity of flexi-panning-generated TCRs was measured with HBs371 and HBs370 variants (online supplemental figure S1, tables S1.1 and S1.2). Approximately 5855 unique full-length HBsAg protein sequences were collected from UniProt, and 16 of the most frequent HBs371- and HBs370-variant peptides represented approximately 90% of the sequences, that is, vrt1 to vrt8 were nonamer HBs371 variants, and vrt0, vrt9 to vrt16 were decamer HBs370 variants. Results for IFN-γ (figure 4A,B) and LDH release (figure 4C,D) showed that both WT and affinity-improved TCRs were able to endow T cells with the ability to recognize 11 of the 16 variants, excluding vrt4, vrt7, vrt8, vrt13 and vrt16. Epitope sequence analysis of these five non-stimulating peptides indicated that the Ser at position three in the nonapeptide and position four in the decapeptide would be the key in the epitope. Ser is allowed at this position, while Asn or Arg will abolish the binding to the TCR. Additionally, comparing EC₅₀ values made it very clear that the variant-mediated activation and cytotoxicity of TCR15-T cells were better than those of TCR0-T cells. This was consistent with the TCR affinity improvement (online supplemental table S7). Especially for vrt6, TCR15-T cells showed over 1000-fold enhancement than TCR0-T cells, which indicated no control of infection with HBV containing vrt6 variant. In conclusion, TCR15-T cells exhibited enhanced killing of T2 cells presenting the 11 most frequent variant peptides.

To verify whether improving TCR affinity affects in vivo tumor suppression, TCR0-T and TCR15-T cells were evaluated in xenograft mouse models established with HLA-A*02:01⁺ PLC/PRF/5 HCC cells, which expressed the HBs371 nonapeptide antigen vrt2 and HBs370 decapeptide antigen vrt10 (confirmed with reverse transcription-PCR, data not show) as the agonist antigens for both TCRs (figure 4). The mice bearing tumors were injected with TCR-T cells on day 13 (figure 5A). All tumors volume enlarged on the third day after the injection; however, on the ninth day, they showed different changes. The full-dose of TCR0-T cells (3.0×10⁷) produced only slight tumor-suppressive effects that were similar to those found in the mice receiving the low-dose of TCR15-T cells (3.3×10⁶). However, the medium-dose of TCR15-T cells (1.0×10⁷) produced significant tumor suppression. The mice in this group had only residual tumors detected on day 25, despite the tumors increased to approximately 107 mm³ on day 34. On the other hand, more than 50% (7/14) and 85% (12/14) of the mice that received the full-dose of TCR15-T cells (3.0×10⁷) showed no detectable tumors on day 23 and day 25, respectively. All tumors were eliminated by day 28 in the full-dose group, and no relapse occurred during an additional 17 days of monitoring. Thus, TCR15-T cells demonstrated superior efficacy and dose-dependent tumor suppression (figure 5B–E). These results confirmed that Ai-TCR-T cells (TCR15-T cells) had a significant advantage over WT-TCR-T cells (TCR0-T cells) in the suppression and elimination of HCC in vivo.

Human CD3⁺ and CD8⁺ T cell infiltration was detected in tumor sites by IHC. Examining H&E-stained (figure 6A) and IHC (figure 6B,C) samples revealed that significant CD3⁺ and CD8⁺ T cell staining was observed in the tumor sections from mice treated with 3.3×10⁶ (CD3⁺:~25.9% and CD8⁺:~16.8%) or 1.0×10⁷ (CD3⁺:~57.1% and CD8⁺:~38.6%) TCR15-T cells. In contrast, the positive rates for CD3⁺ and CD8⁺ T cell populations detected in tumors were only ~12.1% and~4.1%, respectively, for mice receiving 3-fold or ~10-fold more TCR0-T cells (figure 6D). It was obvious that elevated TCR15-T cell accumulation occurred in a dose-dependent manner. Furthermore, we found it intriguing that the percent of CD8⁺ cells in the CD3⁺ cells population was significantly different between the groups. Only 35% of the human T cells detected in the tumor sections of the mice infused with 3.0×10⁷ TCR0-T cells were CD8⁺ cells. On the other hand, 66% or 72% of the human T cells detected in the tumors of the mice infused with 3.3×10⁶ or 1.0×10⁷ TCR15-T cells, respectively, were CD8⁺ cells (figure 6E). However, the initial CD8⁺ percentages of all tested TCR-T cells, including TCR-A6-T cells, were very similar, with an approximate average of 65.4% (online supplemental figure S5). The increased CD8⁺ populations were consistent with the tumor reduction data for the models. We concluded that a large quantity of CD8⁺ TCR15-T cells accumulated in tumor sites.