Abstract
Small cell lung cancer (SCLC) is a pulmonary neuroendocrine cancer with very poor prognosis and limited effective therapeutic options. Most patients are diagnosed at advanced stages, and the exact reason for the aggressive and metastatic phenotype of SCLC is completely unknown. Despite a high tumor mutational burden, responses to immune checkpoint blockade are minimal in patients with SCLC. This may reflect defects in immune surveillance. Here we illustrate that evading natural killer (NK) surveillance contributes to SCLC aggressiveness and metastasis, primarily through loss of NK-cell recognition of these tumors by reduction of NK-activating ligands (NKG2DL). SCLC primary tumors expressed very low level of NKG2DL mRNA and SCLC lines express little to no surface NKG2DL at the protein level. Chromatin immunoprecipitation sequencing showed NKG2DL loci in SCLC are inaccessible compared with NSCLC, with few H3K27Ac signals. Restoring NKG2DL in preclinical models suppressed tumor growth and metastasis in an NK cell–dependent manner. Likewise, histone deacetylase inhibitor treatment induced NKG2DL expression and led to tumor suppression by inducing infiltration and activation of NK and T cells. Among all the common tumor types, SCLC and neuroblastoma were the lowest NKG2DL-expressing tumors, highlighting a lineage dependency of this phenotype. In conclusion, these data show that epigenetic silencing of NKG2DL results in a lack of stimulatory signals to engage and activate NK cells, highlighting the underlying immune avoidance of SCLC and neuroblastoma.
This study discovers in SCLC and neuroblastoma impairment of an inherent mechanism of recognition of tumor cells by innate immunity and proposes that this mechanism can be reactivated to promote immune surveillance.
Introduction
Small cell lung cancer (SCLC) has a very poor prognosis, with little change in treatment over decades causing the NCI and U.S. Congress to designate it as “recalcitrant cancer” (1). Treatment for SCLC for the last 30 years has involved use of cisplatin and etoposide with most tumors relapsing within 1 year (2). Recently nivolumab (anti-PD1 antibody) was approved for third-line treatment and combination of atezolizumab (anti-PDL1 antibody) with carboplatin and etoposide was approved for frontline treatment for metastatic SCLC; however, the survival gains were measured in only a few months and it is not yet clear such immune checkpoint blockade impacts long-term survival of patients with SCLC (3).
The majority of the SCLCs (over 90%) have inactivating mutations in RB1 and TP53 but do not have targetable driver mutations. A smaller subset (about 8%) has inactivation of RBL2 (P130; ref. 4). SCLC has been classified into subtypes based on expression of four transcription factors: ASCL1, NEUROD1, POU2F3, and YAP1. ASCL1high and NEUROD1high groups make up of 80% of all SCLC and, with expression of a panel of neuroendocrine genes, are referred to as “neuroendocrine” SCLCs. In contrast, the other two transcription factor subtypes do not express this neuroendocrine gene panel and are referred to as “non-neuroendocrine” SCLCs (5). Some of the low neuroendocrine SCLCs have a different morphology referred to as “variant.” Despite these biological differences among the SCLC subtypes, currently there are no treatments tailored for each of these subtypes and the immune landscape of these subtypes has not been fully explored.
Other neuroendocrine tumors such as neuroblastomas share similar features with SCLC such as expression of neural transcription factor ASCL1. Neuroblastoma has two subtypes according to superenhancer-associated differentiation states, adrenergic (ARDN) and mesenchymal (MES). The two states can exist in one tumor and interconvert to each other (6). A recent study showed that ASCL1 is higher in ARDN neuroblastoma compared with MES neuroblastoma (7). Besides these primary neuroendocrine tumors, neuroendocrine transformation has been observed after acquisition of resistance to targeted therapies such as with EGFR inhibitors (8) or checkpoint blockade (9) in lung adenocarcinomas, and resistance to castration therapy in prostate cancer (10).
Responses to checkpoint blockade were shown to positively correlate with the total number of somatic mutations or potential neoantigens in several cancers (11). Despite the high mutational burden in SCLC (12), for reasons we do not fully understand, SCLCs respond poorly to immunotherapies. PD-1 immune checkpoint blockade has been approved as first line in combination with cisplatin and etoposide in SCLC even though it is only effective in a small subset (∼10%) patients (13). There is little known for SCLC about tumor immune interactions and innate immunity. Both in experimental models and patient tumors, SCLC tumors exhibit fewer total immune cells within the tumor microenvironment, compared with NSCLCs potentially accounting for poor responses to immune checkpoint blockade (14).
Antitumor activity of the immune system largely depends on cytotoxic cells: T and natural killer (NK) cells. While T cells are an important component of adaptive immunity and depend on specific antigens, NK cells are part of innate immunity and recognize tumors by germline-encoded patterns (15). NK cells are critical in preventing lung tumor growth, as depletion of NK cells was shown to facilitate lung cancer initiation in experimental models (16). NK cells are widely circulating lymphocytes, specialized in eliminating virus-infected cells and malignant cells. They attack cancer cells by secreting cytotoxic proteins, such as perforin and granzymes, and exosome membrane-bound death ligands. Antitumor activity of NK cells is stimulated by NKG2D ligands (NKG2DL) present on the surface of cancer cells (17, 18). In humans, NKG2DLs can be classified into two subsets, MICA/B and RAET1 (also known as ULBP). While mice do not have genes corresponding to human MICA/B, mice do have orthologs of human RAET1 genes including Rae1α/β/γ/δ/ϵ, MULT1, and H60a/b/c, which were shown to encode mouse NKG2DLs that also bind to NKG2D (19). Cytotoxic T cells also express NKG2D receptor, and NKG2D ligands can stimulate cytotoxicity of T cells (20).
Previous reports suggested lower expression of MHC-1 expression in SCLC (21). While low MHC expression would make SCLC resistant to adaptive immunity, this state should make SCLCs susceptible to NK killing (22) yet these tumors grow aggressively in mice and patients. Here we studied SCLC NK- and T-cell interactions and report lack of surface ligands for NK-activating receptor NKG2D in both preclinical and clinical samples of neuroendocrine SCLC and neuroblastoma. In our preclinical models, restoration of NKG2D ligand expression restored immune responses and clearance of primary SCLC tumors and metastasis highlighting this critical immune evasion mechanism by SCLCs as well as a roadmap to new immune treatment strategies.
Materials and Methods
Computational analysis of data
Analysis of cell line expression data
RNA samples were submitted to paired-end RNA sequencing (RNA-seq). Reads were aligned to the human reference genome GRCh38 using STAR-2.7 (https://github.com/alexdobin/STAR) and fragments per kilobase of transcript per million reads (FPKM) values were generated with cufflinks 2.2.1 (http://cole-trapnell-lab.github.io/cufflinks/). These were then normalized (top quartile normalization) and log transformed. Data were deposited to dbGaP under accession no. phs001823.v1.p1 (23).
Analysis of patient tumor data
Cancer Cell Line Encyclopedia
Log2 mRNA values were obtained from the Cancer Cell Line Encyclopedia (CCLE) database (25) and were graphed.
Chromatin immunoprecipitation sequencing data
Superenhancers were determined using published H3K27Ac chromatin immunoprecipitation sequencing (ChIP-seq) data (H69 data are from GSE62412; and A549, 3122, and PC9 data are from GSE89128; and H2087 data are from GSE72956) and unpublished data (for H510 and H2107) calculated by HOMER using bound regions that were significant at Poisson P value threshold of 1 × 10−9 for the superenhancer prediction.
RNA-seq
RNA was extracted using Qiagen RNeasy Mini Kit (catalog no. 74106). RNA was submitted to University of Texas Southwestern Medical Center (UTSW, Dallas, TX) Mcdermott Sequencing Core for library preparation and sequencing. Samples were tested for integrity and concentration before library preparation. A total of 1 μg of DNAse treated total RNA was prepared with the TruSeq Stranded mRNA Library Prep Kit from Illumina. Poly-A RNA was purified and fragmented before strand-specific cDNA synthesis. cDNA were then A-tailed and indexed adapters were ligated. After adapter ligation, samples were PCR amplified and purified with beads. Samples were quantified then run on the Illumina NextSeq 500 using V2.5 reagents. Raw FASTQ files were analyzed using FastQC v0.11.2 and FastQ Screen v0.4.4, and reads were quality trimmed using fastq-mcf (ea-utils/1.1.2-806). The trimmed reads were mapped to the hg19 assembly of the human genome (the University of California, Santa Cruz, CA, version from igenomes) using STAR v2.5.3a. Duplicated reads were marked using Picard tools (v1.127; https://broadinstitute.github.io/picard/). Differential expression analysis was performed using edgeR with statistical cutoffs of FDR ≤ 0.05 and log2CPM ≥ 0. Gene ontology analysis was done by PantherGO classification system. Data were deposited to Gene Express Omnibus under accession number GSE161313.
ELISA for MICA
All patient samples were obtained from subjects providing written informed consent for blood in accordance with the Declaration of Helsinki and studies were approved by an Institutional review board at Osaka University (Suita, Japan; no.11122, 16450). Sampling was performed during routine clinical procedures before the initiation of treatment. A total of 69 serum samples from 25 patients with SCLC and 44 patients with NSCLC, who initiated the treatment in Osaka University Hospital (Osaka, Japan) between October 2012 and March 2017, were investigated. Patients with SCLC are consisted of 12 extensive disease-SCLC and 13 limited disease-SCLC. Patients with NSCLC are consisted of 15 squamous cell cancer and 29 adenocarcinoma (Ad) with respective clinical stage: stage IIA (n = 2), IIB (n = 2), IIIA (n = 4), IIIB (n = 4), and IV (n = 32). A total of 11 Ad cases had EGFR mutations and two Ad cases had ALK translocation. Detailed information related to patient samples are in Supplementary Table S1.
MICA levels in the blood from patients with lung cancer, healthy people, or cell lines were quantified using the Human MICA DuoSet ELISA Kit (R&D systems). For cell lines, 1 × 106 cells were plated and media was collected for ELISA 24 hours after plating.
Cell culture
All human lung cell lines were obtained from Hamon Center for Therapeutic Oncology Research lines (UTSW). Cell lines were DNA fingerprinted using PowerPlex 1.2 Kit (Promega) and confirmed to be Mycoplasma free using e-Myco kit (Boca Scientific). Cells were maintained in RPMI1640 (Life Technologies) with 5% FBS at 37°C in a humidified atmosphere containing 5% CO2. Outside of the Hamon Center the cells were maintained in RPMI supplemented 10% FBS and with 1% penicillin-streptomycin (10,000 U/mL). Human neuroblastoma cell lines SK-N-BE (CRL-2271) and SK-N-SH (HTB-11) were obtained from ATCC and maintained in DMEM (Thermo Fisher Scientific, catalog no. 10569010) supplemented with 10% FCS and 1% penicillin-streptomycin. Human neuroblastoma cell lines IMR32 (CCL-127) was also obtained from ATCC and maintained in Eagle Minimum Essential Medium supplemented with 10% FBS and 1% penicillin-streptomycin. Mouse SCLC cell line RP984 was maintained in DMEM supplemented with 10% FBS and 1% penicillin-streptomycin. HEK293FT was maintained in DMEM supplemented with 10% FBS and 1% penicillin-streptomycin. Rb1/p53/p130 (RPP) cell lines were maintained in DMEM supplemented with 10% FBS and 1% penicillin-streptomycin. Human NK92 cell line (CRL-2407) was purchased from ATCC. NK92 was maintained in α-MEM (Gibco, catalog no. 12571063) without ribonucleosides and deoxyribonucleosides but with 2 mmol/L l-glutamine (Gibco, catalog no. 25030081) and 1.5 g/L sodium bicarbonate (Gibco, catalog no. 25080094). To make the complete growth medium, following components were added to the base medium: 0.2 mmol/L inositol (Sigma, catalog no. I7508); 0.1 mmol/L 2-mercaptoethanol(Gibco, catalog no. 21985023); 0.02 mmol/L folic acid (Sigma, catalog no. F8758); 10 U/mL recombinant IL2 (PerpoTech, catalog no. GMP200-02); adjust to a final concentration of 12.5% horse serum (Gibco, catalog no. 16050122) and 12.5% FBS (Gibco, catalog no. 16000044).
Stable expression cell lines
MICA and Rae1 delta cDNA were purchased from Origene (MICA: catalog no. SC1200303, Rae1 delta: catalog no. MR221464). MICA and Rae1 delta fragments were cloned from plasmids by Phusion High-Fidelity DNA Polymerase (Thermo Fisher Scientific, catalog no. F530). The fragments and backbone (pLvx-EF1a-IRES-puro, Addgene, catalog no. 85132) were double digested with restriction enzyme and then ligated by T4 DNA ligase (Thermo Fisher Scientific, catalog no. EL0012). Ligation products were transformed to NEB Stable Competent E. coli (catalog no. C3040) and then spread on selection Lysogeny broth (LB) agar plate. After overnight culture at 37°C, clones were chosen for culture in LB. Plasmids were extracted from the cells and validated by Sanger sequencing. Validated plasmids (pLvx-EF1a-IRES-MICA-puro/pLvx-EF1a-IRES-Rae1d-puro), pMD2.G (Addgene, catalog no. 12259) and psPAX (Addgene, catalog no. 12260) or empty vectors were transfected to HEK293FT cells by lipofectamine 2000 transfection reagent (Thermo Fisher Scientific, catalog no. 11668030). After transfection, HEK293FT was maintained in opti-MEM (Thermo Fisher Scientific, catalog no. 31985088) overnight. Then change the medium to virus production medium (DMEM, 10% FBS, Glutamax, 1% penicillin-streptomycin, and 1% BSA). Media were collected at 24 and 48 hours after production. Then media were filtered through 0.45 μmol/L polyvinylidene difluoride filter. Both empty vector control or MICA/Rae1d overexpressing virus medium supplemented with polybrene (Sigma, catalog no. TR-1003-G, 2 μg/mL for human SCLC and 8 μg/mL for mouse SCLC) was added to cells. Cells were cultured with virus for 48 hours and then selected with complete culture medium containing 1 μg/mL puromycin for 72 hours. Selected cells were stained with phycoerythrin-conjugated flow antibodies and FACS sorted.
NK92/Tumor cell coculture
NK92 cells were activated with 1,000 U/mL of recombinant human IL2 (PeproTech, catalog no. 50813175) for 48 hours. SHP77 cells were seeded into 96-well plate at the density of 5,000 cells per well. H69 cells were treated with vehicle (DMSO; Sigma, catalog no. D2650) or 1 μmol/L trichostatin (TSA; MedChemExpress, catalog no. HY-15144) for 24 hours and then were dissociated with Accumax (Innovative Cell Technologies, catalog no. AM105-500) and then seeded into 96-well plate at the density of 5,000 cells per well. NK92 and tumor cells were cocultured at the ratio of NK92/tumor cells = 10:1 for 4 hours in the presence of TSA. After 4 hours coculturing, 50 μL of medium was harvested for cytotoxicity assay using CyQUANT LDH Cytotoxicity Assay Kit (Thermo Fisher Scientific, catalog no. 20300) following manufacturer's instructions. Final percentage of lysis was calculated by the formula % lysis = 100*(Release − Tumor spontaneous release − PBMC spontaneous release)/(Tumor maximum release − Tumor spontaneous release). Cells were later harvested for flow cytometry analysis as described. NK92 cells were gated as CD45+ population.
Mouse xenografts and allografts
All mice were housed in a barrier facility and maintained on standard chow. Human cell lines were implanted into athymic nude mice (The Jackson Laboratory, catalog no. 002019) of 4–8 weeks and mouse cell lines RP984 were implanted into hybrid mice of C57B6/J and 129S1 [C57B6/J (The Jackson Laboratory, catalog no. 000644) and 129S1 (The Jackson Laboratory, catalog no. 002448)] of 6–8 weeks. Both male and female mice were included in all studies. A total of 5 million human SCLC cells or 1 million mouse SCLC cells were injected subcutaneously with Matrigel (Corning, catalog no. 354230) and injected into both sides of mouse flanks. Width (shorter dimension) and length (longer dimension) of tumor were measured by digital caliper and volume was calculated using following formula. Volume (mm3) = width (mm) × width (mm) × length (mm)/2. Mice were randomized for treatments. In wild-type mice, NK cells were depleted by intraperitoneally injecting 200 μg isotype control or mouse anti-NK1.1 antibody every 3 days. CD4+ T cells were depleted intraperitoneally injecting 200 μg isotype control or mouse anti-CD4+ antibody every 3 days. CD8+ T cells were depleted intraperitoneally injecting 200 μg isotype control or mouse anti-CD8+ antibody every 3 days. In nude mice, NK cells were depleted by intraperitoneally injecting 200 μg isotype control antibody or mouse anti-sialo GM1 antibody every 3 days. For metastasis studies, 5 × 104 mouse cells were injected by intravenous route. All animal work described in this article has been approved by the University of Texas Southwestern (Dallas, TX) Institutional Animal Care and Use Committee.
Drug treatment
For in vitro assays, all drugs were dissolved in 100% DMSO. For TSA, human SCLCs were treated with 0, 1, and 5 μmol/L for 48 hours and mouse SCLC was treated with 0, 0.1, 0.2, and 0.5 μmol/L for 48 hours. For vorinostat, human SCLCs human SCLCs were treated with 0, 0.5, 1, and 5 μmol/L for 48 hours and mouse SCLC was treated with 0, 0.5, and 1 mol/L for 48 hours. For pracinostat, human SCLCs were treated with 0, 1, and 5 μmol/L for 48 hours and mouse SCLC was treated with 0, 0.25, 0.5, and 1 μmol/L for 48 hours.
For in vivo treatment, a stock solution of TSA was prepared in DMSO (Sigma, catalog no. D2650) at 5 mg/mL concentration. A total of 10% TSA stock solution in DMSO, 5% Tween-80 (MedChemExpress, catalog no. HY-Y1891), 40% PEG300 (MedChemExpress, catalog no. HY-Y0873), and 45% PBS were mixed for final administration. The vehicle control and drug was administrated by oral gavage at 5 mg/kg five times a week.
RNA extraction and qRT-PCR
RNA was extracted by Zymo Direct-zol RNA Miniprep (catalog no. R2051). A total of 1 μg of total RNA was reverse transcripted to cDNA by Applied Biosystems TaqMan High-Capacity RNA-to-cDNA Kit (Thermo Fisher Scientific, catalog no. 43-874-06) in 20 μL system. A total of 1 μL product was used for qPCR in 10 μL system by Taqman mastermix (Thermo Fisher Scientific, catalog no. 4369016). Amplification was assessed by QuantStudio 3 Real-Time PCR System.
Flow cytometry analysis
Cell lines were first stained with fixative live/dead cell stain (Thermo Fisher Scientific, catalog no. 50-112-1528) room temperature for 8 minutes to stain the dead cells. They were then stained with fluorophore-conjugated antibodies (in FACS buffer) for 20 minutes on ice. Samples were washed with FACS buffer (2% FBS in PBS). Stained samples were analyzed using BDFACS Canto. Data were analyzed using FlowJo.
Mouse tissues were minced and digested at 37°C for 1 hour (100 units/mL collagenase, Thermo Fisher Scientific, catalog no. 17104019, 10 μg/mL DNase I Sigma, catalog no. DN25, 10% heat-inactivated FBS in RPMI) to dissociate cells. Red blood cells were lysed in the dissociated tissue with ACK lysis buffer (Thermo Fisher Scientific, catalog no. A1049201). Then tissues were passed through 70 μm cell strainer to generate single-cell suspension. Cells were stained with fixative live/dead cell stain at room temperature for 8 minutes. Then they were incubated with CD16/32 antibody (BioLegend, catalog no. 101320) for 20 minutes on ice. Next, they were incubated with fluorophore-conjugated antibodies diluted in FACS buffer for 20 minutes on ice to stain for surface markers. Samples were then washed with FACS buffer after every incubation with fixative dye or antibody. For the intracellular markers, intracellular staining protocol was followed after surface staining.
Intracellular staining was performed by eBioscience Foxp3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific, catalog no. 00-5523-00). Samples were permeabilized with fixation/permeabilization buffer at 4°C overnight. Then they were incubated with fluorophore-conjugated antibodies (IFNγ, Ki67, granzyme B) diluted in FACS buffer for 20 minutes on ice. Samples were washed with permeabilization buffer after every incubation with antibody. All antibodies were diluted in permeabilization buffer.
To determine T-cell activation, lymphocytes were enriched using Ficoll-Paque (GE Healthcare) following the protocol. Enriched samples were incubated with PMA/ionomycin/Golgi plug for ex vivo stimulation for 6 hours. Cells were then first stained with surface markers (for lineage markers) then intracellular markers (for IFNγ, Ki67, and granzyme B) as detailed above. Samples were run on BDFACS Canto and flow data were analyzed using FlowJo. Source and catalog numbers of all antibodies used for flow are listed in Supplementary Table S2.
Results
SCLCs have reduced MICA/B RNA and lack MICA/B on cell surface
To begin to understand the SCLC tumor and innate immune cell interactions, we analyzed antigen presentation by MHC-1 expression in NSCLC versus SCLC cell lines at the protein level. As reported by several studies, neuroendocrine-SCLC lines [SCLC-A (ASCL1 high) or N (NEUROD1)] tested showed reduced MHC-1 expression compared with NSCLC lines as determined by flow cytometry (Supplementary Fig. S1). Lack of MHC expression is one of the reasons for reduced immunogenicity despite the high tumor mutational burden. Next, we explored SCLC-innate immune cell interactions. We analyzed expression of ligands for the most potent NK cell–activating receptor NKG2D (MICA, MICB ULBP1, 2, and 3) in clinical and preclinical samples. We analyzed NKG2DL expression from RNA-seq data of human lung adenocarcinomas from TCGA (24) and SCLC tumors from George and colleagues (4), MICA/B and ULBP1,2,3 expression was significantly decreased in primary human SCLCs compared with NSCLC (Fig. 1A; Supplementary Fig. S2). Shed MICA has also been reported to be found in the circulation of patients with cancer with a potentially distinct role than surface MICA (26). We found that while late-stage patients with NSCLCs show significantly elevated MICA protein in blood, patients with SCLC with late stage or extensive disease have similar levels of MICA in blood as healthy controls (Fig 1B; n = 44 patients with NSCLC, 25 patients with SCLC, and 16 healthy controls). SCLC lines showed a similar pattern of reduced NKG2DLs (MICA, MICB, ULBP2, and 3; all SCLC subtypes included; Fig. 1C; Supplementary Fig. S2). We also determined protein expression of NKG2DLs and found human SCLC-A lines lacked detectable surface MICA/B expression (Fig. 1D) and soluble MICA/B (Fig. 1E) while NSCLCs expressed these proteins. Overall, patient-derived SCLC compared with NSCLC lines showed a significantly lower level of total NKG2DLs (Supplementary Fig. S3). To determine whether the observations in patient samples and patient-derived materials are applicable to mouse models, we examined NKG2DL expression in genetically engineered mouse models (GEMM), the NSCLC KrasG12D lung adenocarcinoma model, and the SCLC RPP triple knockout model. Epcam positive tumor cells isolated by flow cytometry from SCLC GEMMS showed very little or no surface NKG2DL proteins as compared to the NSCLC GEMMs (Fig. 1F and G). These findings argue that SCLCs, particularly SCLC-A's, have a reduced “visibility” to adaptive and innate immunity. There was also a reduction in overall immune cell infiltration (Fig. 1H), and NK-cell infiltration in the SCLC tumors as compared with NSCLCs (Fig. 1I) in the syngeneic tumors of KrasG12D p53−/− NSCLC or Rb1/p53 SCLC GEMMs implanted into wild-type animals. NK-cell activation was also reduced in SCLCs as compared with NSCLC tumors as indicated by CD107 staining (Fig. 1J).
Restoration of MICA in SCLC caused NK-cell activation and promoted tumor regression in an NK cell–dependent manner
To determine the functional role of NKG2DL expression in SCLC tumor growth, shaping the tumor immune environment, and preventing metastasis, we expressed NKG2DL in SCLC models (Figs. 2 and 3). To determine the changes in direct killing of tumor cells we first cocultured MICA (human NKG2DL)-expressing stable human SCLC line (SHP77; Fig. 2A) and the corresponding vector control with human NK cell line NK92. MICA expression caused increased cytotoxicity and activation as indicated by IFNγ+ staining (Fig. 2B).
To confirm our observations in patient-derived models, we utilized human SCLC lines representing the SCLC-A transcription factor subtype (SHP77, NCI-H510, and NCI-H69). Expression of MICA suppressed the growth of subcutaneous xenografts in athymic nude mice with functional NK cells (Fig. 2C; Supplementary Fig. S4A and S4B) owing to the fact that human MICA interacts with mouse NKG2D ligands (27). MICA expression does not affect the growth rate of these cells in vitro (Supplementary Fig. S4C). Inhibition of SCLC tumor growth in nude mice was dependent on NK cells as, the tumor rejection phenotype was partially reversed when NK cells were depleted (Fig. 2D). To determine the immune-mediated mechanism of tumor clearance after restoration of MICA expression, we generated single-cell suspensions of control and MICA-expressing tumors and analyzed by flow cytometry (Supplementary Fig. S5A). In xenografts formed by SCLC SHP77, we saw an increase in infiltrating total immune cells (CD45+) and NK cells (Fig. 2E). In addition, NK cells in MICA-expressing tumors were more activated as determined by CD107a and IFNγ production (Fig. 2E). Similar results were seen in xenografts formed by SCLC models NCI-H510 and NCI-H69 (Supplementary Fig. S5B and S5C). These results demonstrate that restoring MICA expression on SCLCs can boost the antitumor effect of NK cells.
To confirm our observations in a fully immunocompetent setting, we used a syngeneic line RP984, that was derived from SCLC tumors from a Rb1/p53 double mutant mouse (28). This model when implanted into strain-matched wild-type animals allows us to study the tumor immune microenvironment in a fully immunocompetent context. Expression of a mouse NKG2DL-Rae1d in RP984 syngeneic SCLC model caused significant reduction in the growth of primary tumors in mouse flanks (Fig. 3A–C) while it did not change their in vitro growth rate (Supplementary Fig. S4C). One of the common clinical features of SCLCs is the ability to metastasize. When RP984 cells were implanted intravenously into mice, tumors metastasized to livers and lungs with prominent macroscopic tumors apparent in livers (Fig. 3D and E). Expression of Rae1d significantly reduced the metastatic ability of mouse SCLCs to livers (Fig. 3E). Inhibition of tumor growth in this setting was dependent on NK cells as tumors in the Rae1-expressing group grew significantly larger in NK cell–depleted mice while non-Rae1d-expressing SCLC tumors did not show a significant difference in growth in either NK-depleted versus nondepleted tumors (Fig. 3F). Depletion of CD4 T cells did not influence the growth of Rae-expressing tumors while depletion of CD8T cells partially reduced tumor rejection (Fig. 3G).
In the immunocompetent syngeneic model using the RP984 SCLC model, we observed a significant increase in T-cell infiltration with Rae1d expression (Fig. 3H). Proliferation of both cytotoxic T cell (CD3+CD8+) and NK cells were increased in the Rae1d-expressing group, which was defined by Ki67+ staining (Fig. 3H). In addition, IFNγ production of NK cells was increased (Fig. 3H). Because the NKG2D receptor is also expressed on T cells and T cells can be activated by NKG2D–NKG2DL interaction, both T cell and NK cells were activated in an immunocompetent host. These observations demonstrate that restoring NKG2D ligands in SCLC preclinical models with initial low levels of NKG2DL expression promotes antitumor immunity.
SCLCs downregulate MICA/B epigenetically
NKG2DLs are regulated at epigenetic, transcriptional, posttranscriptional, and posttranslational levels. Chemotherapy is a known transcriptional inducer of NKG2DL through activation of the cGAS/STING pathway (29, 30). We treated SCLCs with the standard-of-care chemotherapy agents, cisplatin, etoposide or a PARP inhibitor (olaparib) that was previously shown to induce the STING pathway in SCLC. However, these treatments did not induce NKG2DL expression in human SCLC lines (Supplementary. Fig. S6A and S6B). We thus speculated that these genes may be epigenetically silenced. We analyzed H3K27Ac ChIP-seq data from multiple NSCLC and SCLC cell lines and found that MICA/B enhancers are hypoacetylated in ASCL1 high SCLC lines as compared with NSCLC lines (Fig. 4A and B). Because H3K27ac usually marks active enhancers and promoters, it appeared that MICA/B was downregulated epigenetically in SCLCs. To see whether we could reverse this epigenetic inhibition, we utilized commonly used pan-histone deacetylase (HDAC) inhibitors—TSA and vorinostat (SAHA). Treatment with these HDAC inhibitors resulted in an increase in MICA/B mRNA (by qRT-PCR) in human SCLC lines H510 and H69 (Fig. 4C). Because HDAC inhibitors affect gene expression globally, we performed RNA-seq in H69 cells treated with TSA to determine changes across the transcriptome and also validated findings with qRT-PCR for MICA/B, ASCL1, and INSM1. The most significantly affected pathways included genes involved in neural transcription (ASCL1, INSM1) and cytotoxic T-cell regulation (MICA, MICB; Fig. 4D and E). Looking more broadly at the significantly changed genes, interestingly, we observed loss of expression of neuroendocrine genes ASCL1, INSM1, CHGB, BEX1, and DDC (Fig. 4F; Supplementary Table S3). After TSA treatment, there was an increase in expression of genes involved in recognition of cancer cells by T cells: all NKG2DLs, structural subunit of MHC-1 (B2M), and MHC-1 genes were significantly elevated. Prosurvival genes (BCL2, MCL1) were decreased and proapoptotic genes (BAK, BAX) were significantly increased in the treated cells (Fig. 4F). E2F family of transcription factors are upregulated in SCLCs due to inactivation of RB and promotes tumor proliferation and the expression of these E2F transcription factors was downregulated by TSA treatment (31). Finally, Notch genes were increased (Fig. 4F) in TSA-treated cells, potentially contributing to shutdown of the expression of a neuroendocrine program. Treatment of human neuroendocrine SCLC lines (H69, H510, H82, and H209) with TSA increased MICA/B protein expression (Fig. 4G and H). Additional HDAC inhibitors SAHA and pracinostat also induced MICA in H69 and H510 at the protein level (Fig. 4H). All these experiments suggest that epigenetic regulation contributes to neuroendocrine differentiation, deregulated proliferation, and reduced immune visibility in SCLCs and that epigenetic targeting agents such as HDAC inhibitors can reverse this.
To determine the functional role of NKG2DL restoration by HDAC inhibitors, we first performed a coculture assay with NK cell line NK92 and human SCLC line H69 in presence of TSA. Cancer cells were pretreated and NK cells were added in the culture with TSA present. TSA treatment caused increased cytotoxicity and increased activation of NK92 cells as compared with vehicle-treated control (Fig. 5A). For the mouse SCLC model RP984, we performed flow cytometry for all NKG2DLs and found that HDAC inhibitors increased NKG2DL expression (Fig. 5B and C). In vivo, we tested the HDAC inhibitor TSA in immunocompetent mice with RP984 allografts. TSA treatment (5 mg/kg) caused significant decrease in tumor growth (Fig. 5D), associated with increased NKG2DL expression on the surface of tumor cells (Fig. 5E). TSA treatment also increased NK-cell recruitment into the tumor and activation of cytotoxic T cells (IFNγ+ Granzyme B+; Fig. 5E). Depletion of NK cells significantly reduced the therapeutic effect of TSA in vivo (Fig. 5F), indicating that this particular HDAC inhibitor functions through activating NK cells given in this dose and schedule to this model. Our results indicated that NKG2DL expression can be induced by HDAC inhibitors in SCLC models and this can suppress SCLC growth in vivo and modulate tumor microenvironment to trigger the antitumor immunity by NK cells and T cells.
Low neuroendocrine SCLC subtypes express MICA, while neuroblastomas also exhibit reduced or no MICA expression
We wished to further understand whether these observations in neuroendocrine SCLCs are common to all SCLC subtypes and additionally whether these observations can be applied to other neuroendocrine tumors. MICA/B expression is significantly negatively associated with ASCL1 in all cancer lines and specifically SCLC lines (Supplementary Fig. S7A). The majority of SCLC lines express high ASCL1 (70%). While SCLCs in general have lower levels of MICA/B compared with NSCLC, there is heterogeneity between the SCLCs. Neuroendocrine SCLCs expressed significantly lower levels of MICA mRNA as compared with non-neuroendocrine SCLCs (Fig. 6A). When classified into different subtypes driven by transcription factors (32), YAP1 high SCLCs have significantly higher levels of MICA as compared with ASCL1 high, NEUROD1 high, or POU2F3 lines (Fig. 6B). Expression level of MICA/B in additional cell lines (NEUROD1high line H82 and YAP1high H841) were validated at the protein level (Supplementary Fig. S7B).
To determine whether there are other tumors that lack NKG2DL expression, we analyzed the CCLE dataset (25). SCLCs were second lowest NKG2DL-expressing cells while neuroblastoma had the lowest among all cancer lines used in this study (Fig. 6B). Interestingly, these tumor types were the top two tumor types expressing ASCL1, indicating a potential lineage dependency of this NK-cell avoidance phenotype. We examined ASCL1 and MICA/B expression in three neuroblastoma lines. CCLE data showed that neuroblastoma line SK-N-SH had low ASCL1 and high MICA/B while the two other ASCL1high neuroblastoma lines (IMR32, SK-N-BE) expressed low levels MICA/B, which was validated by flow cytometry (Fig. 6C). When treated with HDAC inhibitors, these two ASCL1high neuroblastoma lines, which had no detectable surface MICA/B, showed increased MICA mRNA, and protein levels (Fig. 6D and E). This suggest that an increase in innate immune visibility induced by HDAC inhibitors can be applicable to both SCLC and neuroblastoma.
Finally, we isolated high and low neuroendocrine phenotype exhibiting isogenic cell pairs using the GEMM RPP model. It was previously shown that in mixed suspension cultures of SCLC lines adherent population represents the variant phenotype (33). This led us to separate suspension (S) and adherent (A) tumor cell populations, from the mouse RPP SCLC line (derived from Rb1/p53/p130 triple knockout SCLC GEMM). RPP-A expressed less ASCL1 mRNA compared with RPP-S, and higher levels of NKG2DLs at the protein level (Supplementary Fig. S8A and S8B). Demonstrating the plasticity of SCLC, we found the nonattached RPP-S SCLC subtype always gave rise to an adherent population (Supplementary Fig. S8C), explaining the residual NKG2DL expression in the RPP-S cells. Together, these observations support the hypothesis that low NKG2DL expression is associated with high ASCL1 expression and neuroendocrine phenotype, while high NKG2DL expression is associated with the low ASCL1, low neuroendocrine expression phenotype.
Discussion
We show here that human and mouse SCLCs expressing ASCL1, exhibiting neuroendocrine phenotype, lack expression of surface NKG2DL providing a mechanism for their escape from NK immune surveillance. Another neuroendocrine cancer type neuroblastoma showed similar lack of NKG2DL expression in ASCL1 high cell lines. Restoring NKG2DL expression in SCLC tumor cells that lack NKG2DL was able to suppress tumor growth. We found that epigenetic changes through histone deacetylation led to lack of NKG2DL expression and thus appears to be a mechanism of deregulated innate recognition in SCLCs. Therapeutically, pan-HDAC inhibitors induced NKG2DL expression in SCLC in vitro and in vivo, and SCLC tumors were sensitive to HDAC inhibition in vivo, triggered antitumor immunity by recruiting and activating NK cells and T cells in the preclinical models. Interestingly, HDAC inhibitor also reduced expression of lineage oncogene ASCL1 and several other neuroendocrine genes expression in SCLCs further highlighting a role of epigenetic regulation of neuroendocrine gene program.
Paucity of total immune infiltrates in SCLC tumor microenvironment (14) and as we show here specifically paucity of NK cells combined with reduced NK- and T-cell visibility of the tumor cells may contribute to immune resistance of SCLCs. Increasing NK-cell recognition of tumors by pharmacologic means such as epigenetic regulators, may sensitize them to standard treatments and T-cell killing. Our studies argue for determination of MICA expression as a prognostic factor in SCLC patient tumors and a subset of SCLCs may be considered for HDAC inhibitor treatments. HDACs are potential therapeutic targets in both NSCLC and SCLC and have effects beyond modulating NKGD2DLs. An HDAC inhibitor suppressed cell proliferation, induced cell-cycle arrest, and activated Notch signaling in SCLCs in vitro (34). Inhibitor of HDAC6, a cytoplasmic HDAC was shown in a prior study to cause NK cell–dependent cytotoxicity in a preclinical SCLC model though the phenotype was not related to NKG2DLs (35).
Previous trials have shown that NKG2DL stimulating drugs improved responses in patients receiving NK-cell infusion (36). Thus, an HDAC inhibitor could be combined with adoptive NK-cell transfer for patient treatment. Even though we observed increased NK-cell cytotoxicity with HDAC inhibitors in our in vitro and in vivo models, others have shown that HDACs can negatively impact NK-cell viability (37). HDACs also can increase (38) or decrease (39) NKG2D receptor expression on NK cells and modulate NK-cell cytotoxicity or other immune cell functions independent of direct effect on tumor cells.
NKG2DLs are not only found on the surface of tumor cells but also in the circulation in patients with cancer. We observed similarly low levels of circulating NKG2DLs in patients with SCLC while patients with NSCLC showed increased NKG2DL in blood. The role shed NKG2DL plays in mediating NK responses was shown to be highly variable. Some reports indicated that presence of soluble NKG2DLs were shown to cause internalization of NKG2D receptor and suppression of NK cells (40–42). There are also reports about lack of NK-cell suppression with NKG2DLs (43) or activation of NK cells with soluble NKG2DLs (44). SCLCs show reduced levels of soluble NKG2DLs. While in mouse SCLCs, this did not result in increased NK activity possibly due to lack of surface NKG2DL and fewer NK cells in the tumor. The lack of shed NKG2DLs can be exploited in future studies as part of NK-cell activation therapies in SCLCs.
Neuroendocrine tumors are also detected after transformation of other tumor types such as lung adenocarcinoma and prostate cancer. About 5% of the EGFR mutant, tyrosine kinase inhibitor resistant, lung adenocarcinomas acquire targeted therapy resistance by transforming to SCLC (45). There are also several reported cases on NSCLC to SCLC transformation after PD1/PD-L1 blockade (9). Selective pressure imposed by treatments may drive the transformation or allow the rare subpopulations of neuroendocrine clones to expand under such treatment pressure. SCLC has an immune cold tumor microenvironment with few infiltrated cytotoxic lymphocytes compared with NSCLC (14). Thus, transformation to SCLC with reduced immune visibility may protect the tumor cells from immune cytotoxic T cells and NK cells contributing to treatment resistance. Our findings indicate the need to explore pharmacologic epigenetic regulation of SCLC with monitoring for reexpression of NK ligands such as NKG2DL and potentially other less characterized NK cell–activating ligands (46) and determining their effect on cellular and innate immune responses in clinical samples for this hard to treat lung cancer.
Authors' Disclosures
J.D. Minna reports personal fees from NIH and University of Texas Southwestern Medical Center during the conduct of the study. No disclosures were reported by the other authors.
Authors' Contributions
M. Zhu: Conceptualization, data curation, formal analysis, investigation, writing–original draft, writing–review and editing. Y. Huang: Formal analysis, investigation. M.E. Bender: Investigation. L. Girard: Data curation, formal analysis. R. Kollipara: Formal analysis. B. Eglenen-Polat: Investigation. Y. Naito: Formal analysis, investigation. T.K. Savage: Investigation. K.E. Huffman: Formal analysis. S. Koyama: Formal analysis, investigation. A. Kumanogoh: Resources, funding acquisition. J.D. Minna: Resources, supervision, funding acquisition, writing–review and editing. J.E. Johnson: Supervision, funding acquisition, writing–review and editing. E.A. Akbay: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, writing–original draft, writing–review and editing.
Acknowledgments
The authors thank Lauri Knox for administrative help; David McFadden for sharing RP984 cells; Victor Stastny for SCLC cell line work; McDermott sequencing Core for RNA and Chip-seq analysis. E. Akbay is a Cancer Prevention and Research Institute of Texas (CPRIT) Scholar in Cancer Research. This work was supported by CPRIT Scholar Award RR160080, NIH CA070907, and Welch Foundation grant (1975-20190330) to E.A. Akbay. J.D. Minna is supported by CA070907, CA213338, and CA213274, and J.E. Johnson is supported by CA213338.
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.