Targeting Triple-Negative Breast Cancer with Combination Therapy of EGFR CAR T Cells and CDK7 Inhibition

Triple-negative breast cancer (TNBC) comprises 10% to 20% of all breast cancers and is associated with an aggressive phenotype and a high incidence of recurrence (1). To date, there are no approved targeted therapies for TNBC. Proof-of-principle studies have indicated the potential benefits of immunotherapy (2). Antibodies targeting programmed cell death protein 1 (PD-1) and programmed death-ligand 1 (PD-L1) have been approved for clinical use (3). Adoptive cell therapy, particularly chimeric antigen receptor (CAR)–modified T-cell therapy, has gained much attention in the past decade (4). Previous studies have demonstrated that EGFR is a potential therapeutic target for TNBC (5). We and others have reported that third-generation EGFR CAR T cells exhibit potent and specific cytotoxicity against TNBC (6, 7).

Despite the unprecedented, durable response rates observed, the majority of patients do not benefit from immunotherapies (primary resistance), and some responders relapse after treatment (acquired resistance; ref. 8). The immune system can paradoxically constrain or promote tumor development and progression. This process is referred to as cancer immunoediting and, in its most complex form, proceeds through three phases: elimination, equilibrium, and escape (9). Both intrinsic and extrinsic factors in tumor cells weigh into this balance (8). Therefore, curative immunotherapy must not only break immunotolerance and generate responses to tumor antigens, but also circumvent the evolving barrage of acquired escape mechanisms (10).

The commonly cited impediments to effective CAR T-cell therapy include the loss or modulation of the target antigen, lack of CAR T-cell persistence, cytokine-release toxicity, and product manufacturing failures (11). However, tumor-associated impediments that lead to resistance to CAR T-cell therapy remain elusive. Effective and prolonged immunotherapy requires synergy to eliminate impediments from both intrinsic and extrinsic factors of tumor cells (12). Emerging evidence indicates that epigenetic modulation can robustly sensitize patients to immunotherapy. Epigenetic modifiers, such as inhibitors of histone deacetylase, DNA methyltransferase, lysine-specific histone demethylase 1, enhancer of zeste homolog 2, and bromodomain and extraterminal (BET) proteins, have been shown to regulate the presentation and generation of neoantigens and immune checkpoints, secretion of cytokines, and activation of immune cells (13, 14). Enhancers, particularly superenhancers, which participate in regulating the expression of key genes in tumor cells, are particularly sensitive to treatment and show great promise for therapeutic intervention. Enhancer-associated transcription depends on a plethora of transcription factors and cofactors (15). Accumulating evidence suggests that BET inhibition (BETi) can suppress PD-L1 transcription in cancer and immune cells to boost anticancer immune responses (16, 17). However, the function of other proteins associated with enhancers in modulating immune responses remains to be characterized.

In this study, we report that TNBC tumors in one third of mice acquired resistance to third-generation EGFR CAR T-cell treatment. Transcriptomic analysis of EGFR CAR T-cell–resistant tumors revealed the activation of a large set of genes associated with immune suppression, which were presumably activated by IFNγ released by CAR T cells. These CAR T-cell–induced immunosuppressive genes were associated with CAR T-cell–activated enhancers and were especially sensitive to THZ1, a CDK7 inhibitor. Accordingly, combination therapy with THZ1 and EGFR CAR T cells suppressed immune resistance, tumor growth, and metastasis of TNBC in mice.

HEK293T and EMT6 cells were purchased from the American Type Culture Collection in 2018, and MDA-MB-231 and MDA-MB-468 cells were purchased from the Cell Bank of the Chinese Academy of Sciences in 2018. HEK293T and MDA-MB-231 cells were maintained in DMEM high glucose (Biological Industries) supplemented with 10% heat-inactivated FBS (Gibco). MDA-MB-468 cells were maintained in RPMI 1640 (Biological Industries) medium supplemented with 10% heat-inactivated FBS, and EMT6 cells were maintained in Waymouth's Medium (Gibco) supplemented with 15% heat-inactivated FBS. All cells were maintained in a humidified incubator with 5% CO₂ at 37°C. Cell lines were not authenticated since purchase and were cultured for fewer than 10 passages. Cell lines were routinely tested for Mycoplasma using a Mycoplasma contamination detection kit (rep-pt1, InvivoGen).

Full-length human PD-L1 (NM_014143.4), HVEM (NM_001297605.2), and IDO1 (NM_002164.6) cDNAs were first cloned into the lentiviral vector pCDH-CMV-MCS-EF1-Puro (VT1480, Youbio). HEK293T cells were then transfected with the plasmid mixture of the lentiviral and packaging vectors, psPAX2 (12260, Addgene) and pMD2.G (12259, Addgene), using polyethylenimine (Mw 40,000, 24765-2, Polysciences) following the manufacturer's instructions. Lentivirus-containing supernatants were harvested 60 hours after transfection. MDA-MB-231 cells were then transduced with these lentiviruses, and positive clones were selected in a culture medium containing puromycin (1 μg/mL; ant-pr-5, InvivoGen) for 2 weeks and further confirmed by qRT-PCR, immunoblotting, or flow cytometry analysis.

STAT1- or CDK7-knockdown (KD) MDA-MB 231 cells were generated using the CRISPR/Cas9 system. Briefly, STAT1- or CDK7-targeting sgRNAs (sgSTAT1: 5′-TTCCCTATAGGATGTCTCAG-3′; sgCDK7: 5′-TTTACGGCGCCGGATGGCTC-3′) and a control sgRNA (sgCTL: 5′-GTAGCGAACGTGTCCGGCGT-3′) synthesized by Sangon were cloned into the lentiviral vector (lentiCRISPR v2 vector; 52961, Addgene). Lentiviruses were produced as described above. MDA-MB-231 cells were transduced with a lentiviral vector containing sgCTL, sgSTAT1, or sgCDK7 for 48 hours, followed by puromycin selection. Knockdown of STAT1 and CDK7 was verified by qRT-PCR and immunoblotting.

Firefly luciferase (XM_031473197.1) was cloned into the lentiviral vector pCDH-CMV-MCS-EF1-Puro. Lentiviruses were produced as described above. MDA-MB-231 cells stably expressing firefly luciferase (MDA-MB-231-fluc) were obtained by lentivirus transduction, followed by puromycin selection.

Generation of EGFR CAR T cells (human) was performed as previously described (6). Experiments were performed with the understanding and informed written consent from two T-cell donors, and the study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board and Ethics Committee of Xiamen University. The mouse EGFR–targeted CAR construct (EGFR mCAR) consisted of an anti-human EGFR single-chain variable fragment (scFv) derived from cetuximab, spacer, a transmembrane domain derived from murine CD28, and intracellular signaling domains derived from murine CD28, 4-1BB, and CD3ζ. Sequence alignment between human and mouse EGFR revealed 87% similarity for the scFv region. This EGFR mCAR sequence was cloned into the MSCV-IRES-GFP retroviral vector (20672, Addgene). HEK293T cells were transfected with a plasmid mixture of retroviral vectors and pCL-Eco plasmid (12371, Addgene) using polyethylenimine following the manufacturer's instructions for 60 hours before collecting supernatants, which were further concentrated using Amicon Ultra-15 Centrifugal Filters (100 kDa, UFC910096, Merck). Concentrated retroviruses were stored at −80°C.

Spleens of female BALB/c mice ages 6 to 12 weeks (Shanghai SLAC Laboratory Animal Center) were disrupted using a 70 μm strainer to prepare a single-cell suspension. Murine T cells were then isolated using the EasySep Mouse T Cell Isolation Kit (19851, STEMCELL Technologies) according to the manufacturer's instructions, yielding over 90% purity of T cells. For activation of T cells, isolated T cells suspended in X-VIVO medium (04-418Q, LONZA) were stimulated on plates coated with mouse CD3 (1 μg/mL, 100302, BioLegend) and mouse CD28 antibodies (2 μg/mL, 102102, BioLegend) for 24 hours. Activated murine T cells were transduced with concentrated retroviruses in 24-well plates at a multiplicity of infection of 50 for 1 hour (37°C, 2,000 × g). After transduction, mCAR T cells were expanded in X-VIVO medium supplemented with mouse IL2 (40 ng/mL, 51061-MNAE, Sino Biological) and IL7 (5 ng/mL, 50217-MNAE, Sino Biological) for 3 days before use.

A TNBC tumor samples were collected with the understanding and informed written consent from a patient with TNBC, and the study was conducted in accordance with the Declaration of Helsinki and approved by the Review Board and Ethics Committee of Shanghai Cancer Hospital. Fresh tumor samples were collected in DMEM high-glucose medium on ice, cut into pieces, and digested into single tumor cells with collagenase (300 U/mL) and hyaluronidase (100 U/mL; 07912, STEMCELL Technologies), which were then transplanted s.c. into female nude mice ages 6 to 12 weeks (Shanghai SLAC Laboratory Animal Center). Mice were sacrificed when the tumor size reached 1,500 mm³. Tumors were collected and retransplanted into nude mice for in vivo passage. After the establishment of patient-derived xenograft (PDX), tumors were cut into pieces, digested into single cells, and cryopreserved at −80°C in storage buffer (85% FBS and 15% DMSO).

For MDA-MB-231–derived xenograft experiments, MDA-MB-231 cells stably expressing firefly luciferase (MDA-MB-231-fluc) were suspended in 100 μL of PBS (5.0 × 10⁵ cells/mouse) and inoculated s.c. into female SCID mice ages 6 to 12 weeks (Shanghai SLAC Laboratory Animal Center). Tumors were allowed to grow for 5 days until the mean flux reached approximately 1 × 10⁸ p/s/cm²/sr before CAR T-cell treatment (day 0). For EGFR CAR T-cell treatment, mice were injected i.v. with control T (CTL T) or EGFR CAR T cells (5.0 × 10⁶ cells/injection) once every other day for 31 days (3 mice/group). For combination treatment with EGFR CAR T cells and THZ1 (CDK7 inhibitor; HY-80013, MedChemExpress), tumors were allowed to grow until the mean flux reached approximately 1 × 10⁸ p/s/cm²/sr (day 0). Mice were then randomly assigned to four groups (3 mice/group): CTL T cells (5.0 × 10⁶ cells/injection, i.v. injection), THZ1 (10 mg/kg, i.p. injection), EGFR CAR T cells (5.0 × 10⁶ cells/injection, i.v. injection), and THZ1 (10 mg/kg, i.p. injection) combined with EGFR CAR T cells (5.0 × 10⁶ cells/injection, i.v. injection). Mice were injected with CTL T cells or CAR T cells once every other day, followed by THZ1 treatment for 65 days and observed for 130 days. Mice were sacrificed before the tumor burden exceeded the limit (∼1,500 mm³). Tissue samples were resected, formalin-fixed/paraffin-embedded, and stored at −20°C before histopathologic analysis (see below).

For rechallenge experiments, SCID mice inoculated with MDA-MB-231-fluc cells (5.0 × 10⁵ cells/mouse) were first treated with THZ1 and EGFR CAR T cells as described above. Cured mice and age-matched healthy SCID mice were s.c. injected with MDA-MB-231-fluc cells (5.0 × 10⁶ cells/mouse, 5 mice/group) after 65 days of treatment. Mice were sacrificed before the tumor burden exceeded the limit (∼1,500 mm³).

If tumor volume started and continued to increase at a similar rate as the control group with CAR T-cell treatment, and exhibited no response to CAR T-cell treatment when reimplanted into SCID mice, the tumors were considered resistant. Meanwhile, tumors in mice continuously showing decreased volume when treated with CAR T cells were considered sensitive. To treat CAR T-cell–resistant tumors, fresh tumor tissues were isolated from CAR T-cell–resistant tumor-bearing mice after 31 days of treatment. After lysis of erythrocytes by Red Blood Cell Lysis Buffer (R1010, Solarbio), cells were washed with PBS and digested into single-cell suspensions with collagenase (300 U/mL) and hyaluronidase (100 U/mL; #07912, STEMCELL Technologies) at 37°C according to the manufacturer's instructions. Single tumor cells (2 × 10⁵) were mixed with Matrigel (354234, BD Biosciences) and transplanted s.c. into SCID mice ages 6 to 12 weeks. When tumors reached 100 to 200 mm³ in size (day 0), mice were randomly assigned to four groups (5 mice/group) and treated with CTL T or EGFR CAR T cells (5.0 × 10⁶ cells/dose, i.v. injection) in the presence or absence of THZ1 (10 mg/kg/dose, i.p. injection) as described above. Treatment lasted 14 days, and the observation continued until day 35. Peripheral blood was collected (retro-orbital) on days 3, 5, 9, 13, and 20 for assessment of copy number (see below). Serum samples were obtained at the end of the experiment for assessment via ELISA (see below). Tumor tissues were resected, formalin-fixed, paraffin-embedded, and stored at −20°C before histopathologic analysis.

For TNBC PDX experiments, fresh tumor tissues were isolated from stable TNBC PDX–bearing mice and digested into single tumor cells with collagenase (300 U/mL) and hyaluronidase (100 U/mL; #07912, STEMCELL Technologies). Single tumor cells (2 × 10⁵) were mixed with Matrigel (354234, BD Biosciences) and transplanted into SCID mice ages 6 to 12 weeks. When tumors reached 100 to 200 mm³ in size (day 0), mice were randomly assigned to four groups (5 mice/group) and treated with CTL T or EGFR CAR T cells (5.0 × 10⁶ cells/dose, i.v. injection) in the presence or absence of THZ1 (10 mg/kg/dose, i.p. injection) as described above. Treatment lasted 29 days, and the observation continued until day 60.

For EMT6-derived allograft experiments, EMT6 cells were suspended in 100 μL of PBS (1.0 × 10⁶ cells/mouse) and inoculated s.c. into female BALB/c mice ages 6 to 12 weeks (Shanghai SLAC Laboratory Animal Center). When tumors reached 100 to 200 mm³ in size (day 0), mice were randomly assigned to four groups (5 mice/group) and treated with control murine T (CTL T) or EGFR mCAR T cells (2.0 × 10⁶ cells/dose, i.v. injection) in the presence or absence of THZ1 (10 mg/kg/dose, i.p. injection) as described above. Treatment lasted 14 days, and the observation continued until day 26.

Tumor progression was monitored daily by bioluminescence using the Xenogen IVIS Lumina imaging system (Caliper Life Sciences), and tumor size was measured daily with a caliper. Each mouse was injected i.p. with beetle luciferin (1.5 mg; E1605, Promega) and then imaged 6 to 8 minutes later with an exposure time of 3 minutes. Luminescence images were analyzed using the Living Image software (Caliper Life Sciences). Tumor volume (V) was calculated using the formula: V = 1/2 (length × width²). Mice were weighed every 1 to 3 days after infusion. The survival of mice was monitored daily. All animal experiments were conducted in accordance with a protocol approved by the Animal Care and Use Committee of Xiamen University.

Cytotoxicity assays were performed using the xCELLigence Real-time Cell Analyzer (RTCA) System (ACEA Biosciences) as described previously (6). MDA-MB-231 or EMT6 cells were seeded and cultured for 24 hours. Control T cells or EGFR CAR T cells with or without THZ1 (250 nmol/L) were added to the RTCA unit at different ratios. Impedance signals were recorded for 72 hours at 5-minute intervals.

To assess the persistence of the CAR T cells in recipient mice, peripheral blood mononuclear cells (PBMC) of SCID mice were collected, and genomic DNA (gDNA) was extracted using a DNeasy Blood & Tissue Kit (69506, QIAGEN). gDNA (20 ng) was then subjected to qRT-PCR analysis using a primer pair targeting the junction of the CD137 domain and adjacent CD3ζ chain (forward primer: 5′-GAAGAAGGAGGATGTGAACT-3′; reverse primer: 5′-TCCTCTCTTCGTCCTAGATT-3′) on an AriaMx Real-Time PCR System Module (G8831A, Agilent Technologies). gDNA extracted from PBMCs in CTL T-cell–treated mice was used as control. Standard curves were prepared using serial dilutions of the CAR plasmid, starting at 10⁶ copies/μL.

Surface expression of EGFR on MDA-MB-231 cells and TNBC PDX was detected using the phycoerythrin (PE)-conjugated mouse anti-human EGFR antibody (555997, BD Biosciences). Murine T cells were isolated from the spleens of BALB/c mice as described above. Surface expression of the mCAR construct on murine T cells was detected using a Myc-Tag (9B11) mouse mAb (PE-conjugated; 3739, Cell Signaling Technology). CD16/CD32 mAb (93)/FcR blocker (14-0161-82, eBioscience) was used to block nonspecific staining. Zombie Aqua Fixable Viability Kit (423102, BioLegend) was added to exclude dead cells. After using forward scatter area/side scatter area (FSC-A/SSC-A) to separate cells from debris, forward scatter area/forward scatter height (FSC-A/FSC-H) were used to gate single cells from doublets. Zombie-negative cells were further gated as live cells. Fluorescence was assessed using an Attune NxT Flow Cytometer (Thermo Fisher Scientific), and the data were analyzed using FlowJo vX.0.7 (BD Biosciences).

Mouse serum samples (20 μL) were diluted and detected for the presence of IL6 and Indoleamine 2,3-dioxigenase 1 (IDO1) using the Human IL6 ELISA Kit (EH004-96, ExCell Bio) and Human Indoleamine 2,3-dioxygenase/IDO ELISA Kit (Colorimetric; NBP2-62765, Novus), respectively, following the manufacturer's instructions. The absorbance was assessed using an Enzyme-label analyzer (infinite F50, Tecan). The concentrations were evaluated using a standard curve by plotting the absorbance (y-axis) against the protein concentration (x-axis).

Tissue samples from mice were resected, formalin-fixed, and paraffin-embedded. Tissue sections were deparaffinized and rehydrated before staining. For IHC staining, EDTA antigen retrieval solution (MSV-0098, Maxim Biotechnologies) and UltraSensitiveTM SP (Mouse/Rabbit) IHC Kit (Kit-9730, Maxim Biotechnologies) were used according to the manufacturer's instructions. Primary antibodies against human Ki67 (Kit-0005, Maxim Biotechnologies), human EGFR (RMA-0689, Maxim Biotechnologies), human CD8 (ab17147, Abcam), human PD-L1 (ab213524, Abcam), human PD-L2 (ab187662, Abcam), human IL6 (ab239482, Abcam), and human IL8 (ab18672, Abcam) were incubated at 4°C overnight. Stained tissue sections were developed using the Diaminobenzidine (DAB) kit (DAB-1031, Maxim Biotechnologies) for 1 minute and counterstained with hematoxylin solution (HHS16, Sigma-Aldrich) for 10 minutes. For hematoxylin and eosin staining, sections were stained with hematoxylin solution (HHS16, Sigma-Aldrich) for 5 minutes and eosin Y solution (318906, Sigma-Aldrich) for 1 minute. For TdT-mediated dUTP nick end labeling (TUNEL) assays, sections were treated with proteinase K (20 μg/mL; 0706, AMRESCO) at 37°C for 20 minutes and then washed in 1× phosphate buffer. The Colorimetric TUNEL Apoptosis Assay Kit (C1098, Beyotime Institute of Biotechnology) was used to detect apoptotic cells according to the manufacturer's instructions. Histopathologic images were obtained by Olympus BX51 microscope and analyzed using the Olympus cellSens Standard software. The histopathologic images were further converted using ImageJ software to quantify positive expression. Liver and lung metastases were determined by Ki67- and EGFR-positive staining.

Control siRNA (siCTL, siN000000-1-1-1) and siRNAs targeting STAT1 (siSTAT1, 5′-CTGGATATATCAAGACTGA-3′) or CDK7 (siCDK7, 5′-CATACAAGGCTTATTCTTA-3′) were synthesized by RiboBio. siRNA transfections were performed using Lipofectamine 2000 Transfection Reagent (11668500, Invitrogen) according to the manufacturer's protocol. Specifically, MDA-MB-231 cells transfected with siCTL or siSTAT1 were incubated with or without EGFR CAR T cells at a ratio of 2:1 or IFNγ (10 ng/mL) for 48 hours; MDA-MB-231 or MDA-MB-468 cells transfected with siCTL or siCDK7 were incubated with CAR T cells at a ratio of 2:1 or IFNγ (10 ng/mL) for 48 hours. Total RNA was isolated using the Eastep Super Total RNA Extraction Kit (LS1040, Promega) following the manufacturer's protocol. Total RNA (1 μg) was subjected to first-strand cDNA synthesis using the GoScript Reverse Transcription System (A5001, Promega), followed by qPCR using an AriaMx Real-Time PCR machine (Agilent Technologies). Beta-actin was used as an internal control. Expression data presented were the normalized value to control samples after normalization to the expression of beta-actin. Primer sequence for each gene is displayed in Supplementary Table S1.

MDA-MB-231 cells were incubated with CTL T or EGFR CAR T cells at a ratio of 2:1 and incubated with or without THZ1 (250 nmol/L) for 48 hours. The specific ratio was used to avoid a large number of dead cells. Alternatively, MDA-MB-231 cells were incubated with or without IFNγ (10 ng/mL; 554616, BD Pharmingen) in the presence or absence of THZ1 (250 nmol/L) for 48 hours. T cells in suspension were removed, and the adherent tumor cells were collected. Dead tumor cells were also removed using the Dead Cell Removal Kit (Miltenyi Biotec) before RNA sequencing (RNA-seq) analysis. Tumor tissues from the EGFR CAR T-cell–resistant and –sensitive groups (n = 3) were also subjected to RNA-seq analysis (dissociated as indicated above to obtain single-cell suspensions). The Eastep Super Total RNA Extraction Kit (LS1040, Promega) was used for RNA isolation. DNase I (2 U/μL, 5 μL) was included in the column digestion to ensure RNA quality. RNA library preparation was performed using the NEBNext UltraTM Directional RNA Library Prep Kit for Illumina (E7420L). Paired-end sequencing was performed using an Illumina HiSeq 3000 system. Sequencing reads were aligned to the hg19 RefSeq database using Tophat (http://ccb.jhu.edu/software/tophat/index.shtml). Cuff-diff was used to quantify the expression of RefSeq-annotated genes with the option -M (reads aligned to repetitive regions were masked) and -u (multiple aligned reads were corrected using “rescue method”). Coding genes with fragments per kilobase per million mapped reads (FPKM) larger than 0.5 were included in our analysis. Up- or downregulated genes were determined by the fold change of the gene FPKM between groups. The FPKM of a gene was calculated as mapped reads on exons divided by exonic length and the total number of mapped reads. Box plots and heat maps were generated using R software, and statistical significance was determined using the Student t test.

For chromatin immunoprecipitation (ChIP) assays, MDA-MB-231 cells were incubated with CTL T cells or EGFR CAR T cells at a ratio of 4:1 and then incubated with or without THZ1 (250 nmol/L) for 24 hours. Alternatively, MDA-MB-231 cells were incubated with or without IFNγ (10 ng/mL) in the presence or absence of THZ1 (250 nmol/L) for 24 hours. T cells in suspension were removed, and the adherent MDA-MB-231 cells were washed twice with PBS and then fixed with 1% formaldehyde (689316, Sigma) for 10 minutes at 25°C. Fixation was stopped by adding glycine (1610718, Bio-Rad; 0.125 mol/L), and the cells were incubated for 5 minutes at room temperature, followed by washing with PBS twice. Cells were lysed in lysis buffer (1% SDS, 10 mmol/L EDTA, and 50 mmol/L Tris-HCl), and chromatin DNA was sheared to an average size of 300 to 500 bp by sonication using Picoruptor (Diagenode, P-141406). The resultant protein and DNA mixture (∼500 μg) was then diluted with dilution buffer (1% Triton X-100, 2 mmol/L EDTA, 150 mmol/L NaCl, and 20 mmol/L Tris-HCl) and subjected to immunoprecipitation with anti–Pol II (2 μg), anti–Pol II ser5pho (2 μg), and acetylated histone H3 lysine 27 (H3K27Ac, 1.5 μg) antibodies overnight at 4°C, followed by incubation with protein G magnetic beads (0.4 mg/mL, 1614023, Bio-Rad) for an additional 4 hours. After washing, the protein–DNA complex was reversed by heating at 65°C overnight. Immunoprecipitated DNA was purified using the QIAquick PCR Purification Kit (28104, QIAGEN) and subjected to high-throughput sequencing. The following antibodies were used: anti–Pol II (A300-653A, Bethyl Laboratories), anti–Pol II ser5pho (ab5131, Abcam), and anti-H3K27Ac (ab4729, Abcam).

Immunoblotting was performed as described previously (6). Specifically, cells listed below were subjected to immunoblotting: MDA-MB-231 cells incubated with CTL T or EGFR CAR T cells at a ratio of 2:1 were incubated with or without THZ1 (250 nmol/L, HY-80013, MedChemExpress), JQ1 (100 nmol/L, HY-13030, MedChemExpress), C646 (20 μmol/L, HY-13823, MedChemExpress), MG149 (200 μmol/L, HY-15887, MedChemExpress), KDM5-C70 (5 μmol/L, T15648, TOPSCIENCE), or iCDK9 (5 μmol/L, HY-126251, MedChemExpress) for 72 hours; MDA-MB-231 or MDA-MB-468 cells were incubated with CTL T or EGFR CAR T cells at a ratio of 2:1 and then incubated with or without THZ1 (250 nmol/L) for 72 hours; MDA-MB-231 or MDA-MB-468 cells were incubated with IFNγ (10 ng/mL) in the presence or absence of THZ1 (250 nmol/L) for 72 hours; MDA-MB-231 or MDA-MB-468 cells transfected with siCTL or siCDK7 were incubated with EGFR CAR T cells at a ratio of 2:1 or IFNγ (10 ng/mL) for 72 hours; MDA-MB-231 cells were transduced with lentivirus-containing control vector (PCDH-CTL 231), human PD-L1 (PCDH-PDL1 231), IDO1 (PCDH-IDO1 231), or HVEM (PCDH-HVEM 231) for 48 hours; MDA-MB-231 cells were transduced with lentivirus-containing sgCTL, sgSTAT1, or sgCDK7 for 48 hours. Cells mentioned above were lysed in lysis buffer (50 mmol/L Tris-HCl, pH 7.4, 150 mmol/L NaCl, 1 mmol/L EDTA, and 1% Triton X-100) containing protease inhibitor cocktail (P2714-1BTL, sigma) on ice for 30 minutes followed by centrifugation. The resultant supernatant (20 μg) was directly boiled in SDS sample buffer (1% SDS, 5% glycerol, 50 mmol/L dithiothreitol (DTT), 30 mmol/L Tris-HCl, pH 6.8, and 0.25% bromophenol blue) for 5 minutes, resolved by 10% SDS-PAGE gel in SDS running buffer (25 mmol/L Tris, 250 mmol/L glycine, and 0.1% SDS), and transferred to nitrocellulose membrane (Bio-Rad). The membrane was then blocked at room temperature in blocking buffer [5% skim milk in Tris-buffered Saline with Tween 20 (TBST) buffer containing 10 mmol/L Tris, pH 8.0, 150 mmol/L NaCl, and 0.05% Tween 20] for 1 hour, incubated with primary antibody diluted in blocking buffer at 1:1,000 at 4°C for overnight, and washed 5 times with TBST. Blots were then incubated with horseradish peroxidase (HRP)–conjugated secondary antibody diluted at 1:5,000 at room temperature for 1 hour. Membranes were then rinsed with TBST extensively followed by imaging by ECL detection reagents (K-12045-D50, Advansta). The following primary antibodies were used: anti–PD-L1 (ab213524, Abcam), anti-IDO1 (ab55305, Abcam), anti-HVEM (10138-1-AP, Proteintech), anti-STAT1 (9172S, Cell Signaling Technology), anti-CDK7 (A300-405A, Bethyl Laboratories), and anti-actin (8432, Cell Signaling Technology). Rabbit (111-035-003, Jackson Immunoresearch) and mouse HRP-conjugated secondary antibody (115-035-003, Jackson Immunoresearch) were used.

Comparison of two groups or data points was performed using two-tailed t test. Statistical analyses of tumor growth data between multiple groups were performed using one-way ANOVA with Holm–Šídák multiple comparisons test or two-way ANOVA with Dunnett multiple comparisons test as needed. Survival curves were constructed according to the Kaplan–Meier method and compared using the log-rank (Mantel–Cox) test. Statistical significance was set at P < 0.05. Statistical analysis was performed using GraphPad Prism8 software (GraphPad Software Inc.).

The authors confirm that the data supporting the findings of this study are available within the article and/or its Supplementary Materials or can be made available upon reasonable request. Sequencing data were deposited in the Gene Expression Omnibus database under accession GSE164902.

Our previous study demonstrated that third-generation EGFR CAR T cells are potent and specific in suppressing TNBC (6). To further understand the molecular pathways activated by EGFR CAR T cells, RNA-seq in MDA-MB-231 cells revealed that EGFR CAR T-cell treatment altered the expression of a large cohort of genes (Fig. 1A). The IFNγ response was the most enriched hallmark gene set and gene ontology term among genes upregulated by EGFR CAR T-cell treatment, which is consistent with the fact that EGFR CAR T cells secrete high levels of IFNγ when coincubated with TNBC cells (ref. 6; Fig. 1B; Supplementary Fig. S1A). IFNγ-mediated cellular response is well known to be involved in adaptive immune resistance (6). Immunosuppressive genes, such as CD274 (PD-L1), PDCD1LG2 (PD-L2), TNFSF14 (HVEM), IL6, CXCL8 (IL8), CSF1, CXCL2, IDO1, and IL1B, were significantly induced by EGFR CAR T cells (Fig. 1C, left two columns; refs. 18–24). In addition to the IFNγ/cytokine-mediated signaling pathway, response to endoplasmic reticulum stress, apoptotic signaling pathway, and positive regulation of programmed cell death, which are known to play important roles in antitumorigenesis, were also among the most enriched terms (Fig. 1B; Supplementary Fig. S1A; refs. 25, 26).

To confirm that the primary cellular response induced by EGFR CAR T cells was activated by IFNγ, RNA-seq was performed with IFNγ treatment. The effect of IFNγ on the whole transcriptome correlated with that of EGFR CAR T cells (Supplementary Fig. S1B). A total of 1,090 overlapping genes were found to be induced, which were 46.3% and 57.9% of EGFR CAR T-cell– and IFNγ-induced genes, respectively (Fig. 1D–F). The immunosuppressive genes listed above were among the genes commonly induced by EGFR CAR T cells and IFNγ (Fig. 1C). The induction of these genes was validated, which appeared to be dependent on STAT1, a critical regulator of the IFNγ signaling (Fig. 1G and H). Similar observations were made in MDA-MB-468 cells (Supplementary Fig. S1C and S1D). Taken together, our data suggest that the primary cellular response to EGFR CAR T cells in TNBC was the activation of IFNγ signaling, which induced expression of immunosuppressive genes.

The activation of immunosuppressive genes suggests that TNBC tumor cells in mice receiving EGFR CAR T-cell treatment might become resistant. Approximately 30 days after treatment, resistance was observed in one third of the mice. Tumors from EGFR CAR T-cell–sensitive and –resistant groups were subjected to RNA-seq analysis (Fig. 2A–D). Compared with sensitive tumors, 262 and 266 genes were up- and downregulated in CAR T-cell–resistant tumors, respectively (Fig. 2C). The IFNγ response was among the top most enriched pathways for upregulated genes, including the immunosuppressive genes (Fig. 2E). Induction of immunosuppressive gene expression was confirmed by qRT-PCR analysis (Fig. 2F).

The observation that immunosuppressive genes are activated in EGFR CAR T-cell–resistant TNBC tumors led to our hypothesis that blocking such activation might relieve immune resistance. To test this hypothesis, we first characterized the mechanism by which EGFR CAR T cells/IFNγ activated downstream target genes, particularly the immunosuppressive genes. Enhancers, including superenhancers, have been suggested to play a vital role in signaling-induced gene transcriptional activation (27). We hypothesized that EGFR CAR T cells/IFNγ might also activate such an enhancer program. To define the active enhancer landscape upon EGFR CAR T-cell treatment, we performed ChIP sequencing (ChIP-seq) for H3K27Ac, a histone marker decorating both active promoters and enhancers (Fig. 2G). The majority of EGFR CAR T-cell–induced genes were associated with H3K27Ac-occupied enhancers nearby in the presence of CAR T cells. The occupancy of H3K27Ac was significantly higher than that in the group incubated with CTL T, suggesting that they were responsive to EGFR CAR T-cell treatment and might be involved in the activation of the EGFR CAR T-cell–induced genes (Fig. 2H and I). EGFR CAR T-cell–induced active enhancers on representative genes are shown in Fig. 2J and K (Supplementary Fig. S2A). EGFR CAR T-cell–activated enhancers were similarly activated by IFNγ, with similar effects on gene activation (Fig. 2L–O; Supplementary Fig. S2B). Taken together, our data revealed that the EGFR CAR T-cell–activated gene cohort was associated with the activation of nearby enhancers.

We next sought to interfere with EGFR CAR T-cell–activated enhancers and enhancer-associated immunosuppressive genes, aiming to overcome the associated immune resistance. Epigenetic modifiers, such as CDK7, BRD4, P300, TIP60, MOF, KDM5, and CDK9, are involved in the activation of enhancers and enhancer-associated genes (28–33). The effects of THZ1 (CDK7 inhibitor), JQ1 (BRD4 inhibitor), C646 (P300 inhibitor), MG149 (TIP60 and MOF inhibitor), KDM5-C70 (KDM5 subfamily inhibitor), or iCDK9 (CDK9 inhibitor) on the CAR T-cell–induced expression of immunosuppressive genes were then tested (Fig. 3A). Among all the inhibitors tested, THZ1 displayed the most consistent and dramatic attenuation of EGFR CAR T-cell–induced gene expression (Fig. 3A). The effects of THZ1 on the protein expression of representative genes were also demonstrated (Fig. 3B).

To extend our observations, RNA-seq results revealed that the expression of 67% of EGFR CAR T-cell–induced genes was attenuated by THZ1 (Fig. 3C–E). The immunosuppressive genes were among the genes that were most affected (Fig. 3F). EGFR CAR T-cell–induced enhancer activation was similarly attenuated, as seen from the occupancy of H3K27Ac (Fig. 2H–K). Similarly, THZ1 treatment attenuated the expression of 47% IFNγ-induced genes, including the immunosuppressive genes (Fig. 3G–J), which was further confirmed by qRT-PCR and immunoblotting (Fig. 3K–N). IFNγ-induced enhancer activation was attenuated (Fig. 2L–O). The inhibitory effects of THZ1 were also demonstrated in another TNBC cell line, MDA-MB-468 (Supplementary Fig. S2C–S2F). The “Achilles cluster” of superenhancer-associated, TNBC-specific genes reported previously was similarly affected by THZ1, which was further validated (Supplementary Fig. S2G and S2H; ref. 29). The effect of THZ1 on EGFR expression was also examined; the expression of membrane EGFR was slightly reduced by THZ1 after incubation for 6 days but did not change further after treatment for a longer duration (Supplementary Fig. S2I). Taken together, THZ1, a CDK7 inhibitor, could suppress EGFR CAR T-cell–induced enhancer and enhancer-associated immunosuppressive gene expression.

We next tested whether CDK7, the major target of THZ1, is required for the expression of EGFR CAR T-cell–induced immunosuppressive genes. The requirement of CDK7 for the expression of the set of immunosuppressive genes was demonstrated by qRT-PCR and immunoblotting (Fig. 4A and B; Supplementary Fig. S3A). Similarly, CDK7 was required for the IFNγ-induced expression of representative immunosuppressive genes (Fig. 4C and D). The requirement of CDK7 was also observed in MDA-MB-468 cells (Supplementary Fig. S3B–S3E).

CDK7, a component of the general transcription factor TFIIH, phosphorylates the carboxyl-terminal domain of RNA Pol II at serine 5 (RNA Pol II Ser5Pho) to promote transcription (34). We then tested whether THZ1 inhibited CDK7-mediated Pol II phosphorylation to inhibit the expression of immunosuppressive genes. Upon EGFR CAR T-cell treatment, RNA Pol II occupancy simultaneously increased in both the promoter and gene body regions of the genes induced by EGFR CAR T cells, and this was suppressed by THZ1, suggesting that EGFR CAR T cells regulated the transcriptional initiation of these genes (Fig. 4E). Indeed, the traveling ratio or promoter–proximal pausing index, which is defined as the relative ratio of Pol II density in the promoter–proximal region and the gene body and used to measure transcriptional elongation, exhibited no significant changes upon EGFR CAR T-cell treatment, supporting that EGFR CAR T cell regulated the transcriptional initiation of these genes rather than elongation (Supplementary Fig. S3F). The expression of RNA Pol II Ser5pho was significantly induced by EGFR CAR T cells, further supporting the EGFR CAR T-cell regulation of transcription initiation (Fig. 4F). Consistent with its inhibitory effects on EGFR CAR T-cell–induced transcriptional activation, THZ1 attenuated EGFR CAR T-cell–induced Pol II and Pol II Ser5pho (Fig. 4E and F). Similarly, THZ1 attenuated EGFR CAR T-cell–induced RNA Pol II and Pol II Ser5pho occupancy on EGFR CAR T-cell–activated enhancers (Fig. 4G–J). The effects of THZ1 were shown on representative immunosuppressive genes (Fig. 4K and L; Supplementary Fig. S3G). Taken together, our data revealed that THZ1 inhibited CDK7-mediated RNA Pol II phosphorylation to suppress EGFR CAR T-cell–induced enhancer and cognate gene activation.

The ability of THZ1 to suppress EGFR CAR T-cell–induced immunosuppressive genes prompted us to examine whether cotreatment with THZ1 would improve the efficacy of EGFR CAR T cells in killing TNBC cells in vitro. Combined treatment with THZ1 and EGFR CAR T cells exhibited stronger cellular toxicity than THZ1 or EGFR CAR T cells alone (Fig. 5A). To confirm that THZ1 inhibition of immunosuppressive genes was linked to the improved efficacy of CAR T cells, overexpression of CD274 (PD-L1), TNFRSF14 (HVEM), and IDO1 exhibited no significant impact on THZ1-induced cellular toxicity, whereas it attenuated that of EGFR CAR T-cell treatment alone or cotreatment with EGFR CAR T cells and THZ1 (Fig. 5A–C). THZ1 cotreatment also attenuated CAR T-cell–induced expression of immunosuppressive genes, and no additional effects were observed when CDK7 or STAT1 was knocked down via CRISPR/Cas9 (Fig. 5D–H). Accordingly, THZ1 improved the efficacy of CAR T cells in killing MDA-MB-231 cells, but knockdown of CDK7 or STAT1 did not further enhance THZ1's effects (Fig. 5I). These data suggest that THZ1 modulation of immunosuppressive genes is, at least partially, responsible for the improved efficacy of THZ1. It should be noted that THZ1 itself exhibited minor, yet significant, cellular toxicity toward CAR T cells (Supplementary Fig. S4).

The improved efficacy of THZ1 prompted us to examine whether cotreatment with THZ1 alleviated EGFR CAR T-cell–associated immune resistance in mice. SCID mice were subcutaneously implanted with MDA-MB-231 cells, and tumor growth/metastasis was monitored via bioluminescence imaging and caliper-based sizing (Fig. 6A–D; Supplementary Fig. S5A–S5D). Some mice in the control group survived for 45 days (Fig. 6B and E). In the THZ1-treated group, two thirds of the mice exhibited resistance (i.e., the tumor started to grow at a similar rate as the control group) and did survive until the end of the observation, and the remaining one third were alive with tumors (Fig. 6B and E). In the EGFR CAR T-cell–treated group, one third of the mice exhibited resistance to CAR T-cell treatment at approximately 30 days, one third were alive with tumors at the end of the observation, and the remaining one third showed no sign of tumors after treatment, which was consistent with our previous report (ref. 6; Fig. 6B and E). However, mice treated with a combination of THZ1 and EGFR CAR T cells showed no signs of tumors as early as 30 days after treatment. Even after 2 months of treatment withdrawal, when the observation ended, mice showed no sign of tumor recurrence, indicating that combination treatment eradicated MDA-MB-231 cell–originated tumors in mice (Fig. 6B and E). We also rechallenged surviving mice with 10-fold more tumor cells, and prolonged survival was observed, indicating that these mice attained long-term CAR T-cell memory responses (Supplementary Fig. S5E).

The effects of EGFR CAR T cells and THZ1 on tumor growth and metastasis were also evaluated by IHC staining (Fig. 6F–H). Ki67 and EGFR positively stained sections were used to define tumor regions, which were also used to identify and quantify tumor metastases in the tissues (Fig. 6F and G). For EGFR CAR T-cell–treated or THZ1-treated group, tumors exhibiting resistance were selected for staining. Ki67 expression in the primary tumors in the EGFR CAR T-cell–treated or THZ1-treated group was similar to that in the control group (Fig. 6F). The propensity of TNBC for visceral metastasis is higher to the lung than to the lymph nodes, bone, or liver. Consistently, Ki67 staining revealed lung metastasis and fewer liver metastases in the control group, which were abolished in mice cotreated with EGFR CAR T cells and THZ1 (Fig. 6F). EGFR staining showed strong signals (more than 50% cells were EGFR-positive) in primary tumors and lungs but relatively weaker signals (less than 20% cells were EGFR-positive) in the liver of the control group, which was consistent with Ki67 staining (Fig. 6G). EGFR staining was abolished when sections were cotreated with EGFR CAR T cells and THZ1 (Fig. 6G). We also performed CD8 staining to track the infiltration of CAR T cells. As expected, no CD8 staining was observed in both primary tumors and lungs in the control or THZ1-treated group, whereas it was evident in the EGFR CAR T-cell–treated group (Fig. 6H). Combination treatment with EGFR CAR T cells and THZ1 led to the eradication of tumor and tumor metastasis in both lung and liver, and therefore no CD8 staining (i.e., no residual T cells) was observed, further strengthening the specificity of EGFR CAR T cells towards TNBC cells (Fig. 6H). No notable tissue damage was observed in the mice treated with both EGFR CAR T cells and THZ1 (Supplementary Fig. S5F and S5G). Strong staining for representative immunosuppressive proteins was observed in the primary tumor as well as in the lung and, to a lesser extent, in the liver of EGFR CAR T-cell–resistant tumors, which was significantly attenuated by cotreatment with THZ1 (Supplementary Fig. S5H–S5K). All IHC staining was quantified (Supplementary Fig. S5L–S5R).

The effect of the combination treatment was further tested in an EGFR CAR T-cell–resistant MDA-MB-231 cell–derived xenograft model (Fig. 7A). EGFR CAR T cells exhibited no significant effects on tumor growth after approximately 14 days of treatment, suggesting that the resistant model was successful. However, THZ1 cotreatment resensitized the resistant tumor cells to CAR T-cell treatment (Fig. 7B–D), which was associated with the infiltration of T cells, represented by the expression of CD3 and CD8 (Fig. 7E–G). We observed a rapid increase in CAR gene copies after treatment for 3 days (>10,000 copies/μg gDNA), which decreased sharply in the EGFR CAR T-cell–treated group but decreased gradually and remained at a high level in the EGFR CAR T cells and THZ1 cotreated group even after 20 days (>5,000 copies/μg gDNA; Fig. 7H). EGFR CAR T-cell treatment led to increased expression of immunosuppressive genes, which was abolished in the presence of THZ1 (Fig. 7I; Supplementary Fig. S6). Similar observations were made for IL6 and IDO1 in the serum (Fig. 7J).

We also tested the effect of EGFR CAR T cells and THZ1 cotreatment on TNBC in a PDX model (Supplementary Fig. S7A and S7B). Cotreatment with THZ1 and EGFR CAR T cells displayed the most dramatic effects in suppressing the growth of the PDX (Supplementary Fig. S7C–S7E). The induction of immunosuppressive genes by EGFR CAR T-cell treatment was significantly attenuated when cotreated with THZ1 (Supplementary Fig. S7F).

Finally, the effect of EGFR CAR T cells and THZ1 cotreatment on TNBC was examined using an immunocompetent model. We first constructed a retroviral vector encoding EGFR-targeted CAR designed for mice (EGFR mCAR; Supplementary Fig. S8A), which had high expression in HEK293T cells after transfection (92.2%; Supplementary Fig. S8B). Mouse T cells were infected with retroviruses encoding EGFR mCAR to generate EGFR mCAR T cells (Supplementary Fig. S8C). EGFR mCAR T cells killed murine TNBC EMT6 cells in a dose- and time-dependent manner (Supplementary Fig. S8D). THZ1 enhanced the efficacy of EGFR mCAR T cells in killing EMT6 cells in vitro (Supplementary Fig. S8E). The EGFR mCAR T-cell–induced expression of immunosuppressive genes was attenuated by THZ1 (Supplementary Fig. S8F). Similar to what we observed in xenografts from human TNBC cell lines or patients, EGFR mCAR T cells and THZ1 cotreatment exhibited the most significant effects on EMT6-derived allografts (Supplementary Fig. S8G–S8J). The expression of immunosuppressive genes was attenuated by THZ1 (Supplementary Fig. S8K).

To overcome EGFR CAR T-cell–associated immune resistance, we sought to determine the molecular mechanisms by performing transcriptomic analysis in TNBC tumor cells from EGFR CAR T-cell–resistant xenografts, which revealed that a large cohort of immunosuppressive genes was strongly induced. These genes were found to be associated with EGFR CAR T-cell–induced enhancers and were especially sensitive to THZ1, a CDK7 inhibitor. Combination therapy with THZ1 and EGFR CAR T cells suppressed immune resistance, tumor growth, and metastasis in TNBC xenograft and allograft models in mice (Supplementary Fig. S9).

Small-molecule inhibitors or mAbs targeting EGFR are being evaluated in clinical trials for TNBC (35). We have shown that EGFR CAR T cells exhibit potent and specific cytotoxicity against TNBC in cultured cells and mice (6). Like many other immunotherapies, tumors soon acquire resistance after treatment, which limits their application in patients with cancer. Multiple factors contribute to the acquired resistance, such as the emergence of tumor-mediated immunosuppressive mechanisms (12). Through transcriptomic analysis, we found that a large cohort of immunosuppressive molecules was induced after EGFR CAR T-cell treatment in TNBC cells, which were primarily regulated by IFNγ signaling (6). As we reported previously, the cytotoxicity of EGFR CAR T cells was also mediated through IFNγ signaling. IFNγ signals are early immune responses in T-cell activation and differentiation, which predict clinical responses to immune checkpoint blockade therapy (36–38). However, in certain circumstances, such as chronic exposure, IFNγ can induce tumor progression and/or drug resistance (39–41). Therefore, IFNγ serves as a double-edged sword: killer and protector (39).

The observation that a large cohort of immunosuppressive genes was induced after EGFR CAR T-cell treatment prompted us to find a way to systematically interfere with this cohort rather than focus on individual genes, aiming to reduce the associated immune resistance. EGFR CAR T-cell–induced immunosuppressive genes were found to be associated with enhancers. By testing several inhibitors targeting known functional players on active enhancers, we found that THZ1 displayed the best effect on attenuating the expression of EGFR CAR T-cell–induced immunosuppressive genes. Although the link between the set of interferon signaling genes and the improved efficacy of THZ1 remains unclear, our data suggest that the effects of THZ1 on the set of IFN signaling genes account for, at least partially, the improved efficacy of EGFR CAR T cells by THZ1. It has been shown that an “Achilles cluster” of superenhancer-associated, TNBC-specific genes is especially sensitive to CDK7 inhibition/THZ1 (29), which was also seen in our RNA-seq analysis. The inhibitory effects of THZ1 on these genes may also contribute to the improved efficacy of THZ1. Although we demonstrated that CDK7 is, at least partially, responsible for the improved efficacy of THZ1, it is worth noting that THZ1 is a covalent inhibitor that also targets CDK12/13 (28, 42). CDK7 inhibition using YKL-5-124 activates IFNγ and TNFα signaling in small-cell lung cancer (43). Unlike THZ1, cells incubated with YKL-5-124 display little change in RNA Pol II phosphorylation (44). Here, EGFR CAR T-cell–induced interferon signaling genes were subjected to the regulation of Pol II phosphorylation, and THZ1 inhibited CDK7-mediated Pol II phosphorylation to suppress these genes. The different working mechanisms might be, at least in part, why THZ1 and YKL-5-124 exhibit different effects on gene expression. The IFN signaling genes are regulated by cell-type–specific enhancers, which might also account for the observed differences.

In summary, we discovered that tumor cells acquire immune resistance to CAR T-cell treatment due to CAR T-cell–activated enhancer and -cognate immunosuppressive genes, and we identified an effective way to disrupt the transcription of these genes to reduce immune resistance. Unlike the usual “divide-and-conquer,” we took advantage of a “unite-and-conquer” strategy to suppress the expression of a cohort of genes by using inhibitors targeting a transcriptional modulator. Our strategy may apply to other cancer immunotherapies. This strategy may also be effective in treating other difficult-to-treat cancers with EGFR overexpression. Due to the critical role of STAT in the IFNγ signaling pathway, combination therapy with EGFR CAR T cells and inhibitors of the JAK/STAT signaling pathway might also be effective in overcoming the acquired immune resistance associated with EGFR CAR T cells. Combining inhibitors targeting some of the critical immunosuppressive genes induced by EGFR CAR T-cell treatment, such as PD-L1/2, with EGFR CAR T cells may also be promising. Future investigations will be required to determine whether these therapeutic strategies can be translated into clinical settings.

No disclosures were reported.

L. Xia: Data curation, formal analysis, writing–original draft, writing–review and editing. Z. Zheng: Validation, investigation, visualization. J.-y. Liu: Validation, investigation, visualization. Y.-j. Chen: Validation, investigation, visualization. J. Ding: Software, formal analysis. G.-s. Hu: Software. Y.-h. Hu: Validation. S. Liu: Resources. W.-x. Luo: Resources. N.-s. Xia: Resources. W. Liu: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing.

This work was supported by the Natural Science Foundation of Fujian Province of China (2020J05018) and Fundamental Research Funds for the Central Universities (20720200104) to L. Xia. This work was also supported by the Ministry of Science and Technology of China (2020YFA0112300 and 2020YFA0803600), National Natural Science Foundation of China (81761128015, 81861130370, 31871319, and 91953114), Natural Science Foundation of Fujian Province of China (2020J02004), Fujian Province Health Education Joint Research Project (WKJ2016-2-09), Xiamen Science and Technology Project (2017S0091), Xiamen Science and Technology major projects (3502Z20171001-20170302), and the Fundamental Research Funds for the Central University (20720200002 and 20720190145) to W. Liu. The authors feel thankful for the discussions with Dr. Chi-Meng Tzeng.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.