HDAC Inhibitors Enhance T-Cell Chemokine Expression and Augment Response to PD-1 Immunotherapy in Lung Adenocarcinoma

This article is featured in Highlights of This Issue, p. 3985

Lung cancer is a leading cause of cancer-related death around the world, and the 5-year survival has remained unchanged for decades. Importantly, recent studies have demonstrated the considerable potential of immunotherapy in the treatment of lung cancer and other malignancies (1, 2). In particular, blockade of CTLA-4 and PD-1 checkpoint cell surface receptors on T cells is a promising approach (3, 4). CTLA-4 and PD-1 deliver inhibitory signals following binding to their ligands CD80/86 and PD-L1/2, respectively, and blocking binding of these ligands with antibodies augments antitumor T-cell responses (1, 2). PD-1 blockade is an especially promising approach (3, 4), yet response rates are relatively low at approximately 20% in lung cancer, indicating that combinatorial approaches are needed to enhance efficacy. Combinatory therapies currently being evaluated include blockade of multiple checkpoint receptors, as well as use of vaccines, radiation, and agonistic mAb (1, 5, 6). There is growing interest in efficacious combinations of small-molecule chemotherapeutics with immunotherapy to enhance response rates (6–8). Several traditional therapies are dependent upon immune activation, including induction of immunogenic cell death (e.g., by anthracyclines; ref. 9), an increase in granzyme B permeability of tumor cells (e.g., by taxol; ref. 10), and alterations in metabolite and amino acid levels within the tumor microenvironment (7, 11). In this study, we tested the hypothesis that strategies which increase expression of T-cell chemokines and T-cell infiltration to tumors will be especially efficacious in enhancing response to PD-1 blockade.

Previous studies have demonstrated that increased tumor expression of T-cell chemokines, such as CCL5 and CXCL10, is associated with a better response to immunotherapy (12). Furthermore, expression of T-cell chemokines is strongly and positively associated with increased T-cell infiltration and improved patient survival (13–15). Conversely, poor T-cell infiltration in tumors can be associated with resistance to immunotherapy. On the basis of these findings, we tested the hypothesis that oncology agents capable of enhancing expression of T-cell–attracting chemokines will augment the response to immunotherapy. To this end, we screened a library of FDA-approved oncology agents specifically for their ability to induce expression of T-cell chemokines. Interestingly, only a single class of agents, histone deacetylase (HDAC) inhibitors (HDACi), were found to robustly induce expression of multiple chemokines in tumor cells, tumor-infiltrating macrophages, and T cells. HDACi romidepsin treatment triggered a strong antitumor immune response in mice and enhanced PD-1 blockade immunotherapy in multiple tumor models. These findings indicate the significant potential benefit of pairing checkpoint blockade immunotherapy with HDACi for near-term clinical use. In addition, our findings indicate that therapeutic approaches which upregulate T-cell chemokine expression represent a novel strategy for augmenting the response to immunotherapy.

All mice were bred and housed in the animal facility at Moffitt Cancer Center under specific pathogen-free conditions. 129S4/SvJaeJ mice were obtained from Jackson Laboratories. All animal experiments were performed in accordance with the Institutional Animal Care and Use Committee. Mouse LKR-13 had KRAS G12D mutation (16) while 393P and 344SQ (kindly provided by Dr. J. Kurie, MD Anderson Cancer Center, Houston, TX) had KRAS G12D and TP53 mutations (R172H; ref. 17). For bioluminescence imaging, LKR cells were transduced with a replication-deficient lentivirus in which luciferase expression was driven by the thymidine kinase promoter. All lung cancer cell lines were maintained in RPMI1640 with 10% FBS. CD11b⁺ cells were isolated from LKR tumors using microbeads (Miltenyi Biotec) and cultured in DMEM media. Forty-eight hours later, nonadherent cells were removed and attached cells were stimulated with romidepsin. FDA-approved oncology agents were utilized from the Approved Oncology Drugs Set (97 agents) provided by the Developmental Therapeutics Program of NCI (Bethesda, MD). Anti-mCD4 (Clone GK1.5), mCD8 (Clone 2.43), mIFNγ (Clone R4-6A2), and mPD-1 (Clone RMP1-14) were purchased from Bio-X-Cell, along with matching isotype controls. Romidepsin and other HDACi were purchased from Selleckchem and used at the following concentrations: vortinostat 10 μmol/L, MS275 0.5 μmol/L, MGCD0103 1 μmol/L, and LBH-589 0.1 μmol/L. mIFNγ (25 ng/mL) was purchased from eBioscience and hIFNγ (100 ng/mL) from Pestka Biomedical. TAK-779 was purchased from Sigma and Collagenase D was from Roche.

FDA-approved oncology agents were utilized from the Approved Oncology Drugs Set (97 agents) from the Developmental Therapeutics Program of NCI. Beckman-Coulter Biomek FX^P robotic liquid handling system was used for drug treatment in 96-well plates following which cell viability was determined. Briefly, LKR cells were plated in 96-well plates at a concentration of 2,000 cells/100 μL of RPMI with 10% FBS. Twenty-four hours later, oncology agents were individually introduced in triplicate wells in DMSO to achieve final concentrations of 10 μmol/L, 1 μmol/L, 0.1 μmol/L, and 0.01 μmol/L. A viability assay was performed 48 hours after drug administration using Cell Counting Kit-8 (Dojindo Laboratories) as per the manufacturer's instructions. Absorbance for the standard curve was determined 24 hours after plating 2,000, 1,000, 500, 250, 125, and 62.5 LKR cells per 100 μL. Lethal dose 50 (LD₅₀) values for oncology agents were calculated using “Dose–response – Inhibition: log(inhibitor) versus normalized response” in GraphPad Prism 5 software. Agents which did not induce cell death were used at 10 μmol/L (see Supplementary Table S1). LD₅₀ of romidepsin for LKR at 48 hours was 23 nmol/L. However, the LD₅₀ range in different lung cancer lines used here was 23–35 nmol/L. To treat different cell lines with the same romidepsin concentration, we used it at 30 nmol/L.

RNA was isolated using a Qiagen RNeasy Kit, reverse transcribed, and subjected to quantitative PCR analysis, as described previously (13, 18). For tumor RNA analysis, tumors were frozen in liquid nitrogen and homogenized in a Bead Beater. Tumor RNA was extracted using Qiagen RNeasy Kit. Microarray analysis was performed using the Mouse Genome 430 2.0 Arrays as described previously (13).

Cells were harvested in logarithmic growth phase after being cultured for less than 2 weeks, washed once in injection medium (phenol-free DMEM supplemented with 2% FBS), and counted. A total of 1 × 10⁶ cells were injected subcutaneously and tumors were monitored for growth by measurements 2–3 times per week. The tumor volume was determined as length × length × width/2. Romidepsin was injected intraperitoneally (2 mg/kg) on days 14, 16, and 18 after tumor inoculation. Anti-PD-1 was injected intraperitoneally (300 μg/mouse per injection) on days 15, 17, and 19 after tumor inoculation. Isotype controls were injected in control mice. Anti-CD4 and anti-CD8 (300 μg/mouse) were injected intraperitoneally as indicated in Supplementary Fig. S3. TAK-779 (150 μg/mouse) or anti-IFNγ (200 μg/mouse) was injected on day 12 and then twice a week. Mice were sacrificed when subcutaneous tumors reached a diameter of 20 mm or when they showed signs of morbidity. Relative tumor size between treatment groups was analyzed using the t test with Welch correction. To determine presence of infiltrating T cells, tumors were chopped using forceps and scalpels, digested in the Collagenase D buffer with 2 mg/mL Collagenase D at 37°C for 45 to 75 minutes, passed through 70-μm strainer, and then subjected to FACS analysis as indicated in figures. In the orthotopic model, 50,000 tumor cells were injected percutaneously into the left lateral thorax in mice anesthetized with sodium pentobarbital (50 mg/kg body weight). For bioluminescence imaging (BLI) in the orthotopic model, the IVIS Imaging system was used as described previously (19). A conditional mutant KRAS^G12D autochthonous knock-in mouse model of lung cancer (20) was obtained from Jackson Laboratories. At approximately 6 weeks of age, KRAS^G12D mice were injected with 5 × 10⁶ PFU of adenovirus-expressing CRE (Ad-CRE) through the intratracheal (i.t.) route. Pretreatment tumor volume was determined by MRI to noninvasively monitor and assess tumor volumes longitudinally starting at 3 months after Ad-CRE injection. 2 rounds of romidepsin+anti-PD-1 treatment, as described in subcutaneous model, were given immediately and 2 weeks after determining pretreatment tumor volumes. Posttreatment MRI was performed 4 weeks after pretreatment MRI. These studies were performed in the Moffitt Small Animal Imaging Laboratory (SAIL) Core facility using a 7-T horizontal magnet (ASR 310, Agilent Technologies).

Mice were placed in an induction chamber and anesthetized with 2% isoflurane for transfer onto a mouse cradle and maintained under anesthesia. A respiration pad was used for monitoring respiration rate, where a steady state breathing rate of 50 to 60 breaths per minute was achieved. A fiber optic rectal thermometer was used to monitor the temperature, and a core body temperature of 37 ± 1°C was sustained. The MRI experiments were done on a 7 Tesla Agilent ASR 310 equipped with nested 205/120/HDS gradient insert in a bore size of 310 mm. Two RF coils, a 35-mm Litzcage coil (Doty Scientific, Inc.) and a 24-mm Litzcage coil (Doty Scientific, Inc), were used depending on the sizes and weights of the mice. Coronal multislice T₂-weighted fast spin echo images were acquired with TE = 30.05 ms, TR = 1497.52 ms, field of view of 90 × 40 mm², and, 20 slices of 1.2 mm thickness and a data matrix of 256 × 128 × 20. Rather than using respiratory gating, additional images were acquired to offset the effects of motion (32 averages). These images were acquired within 13 minutes. To quantify the tumor burden within the lungs, T₂-weighted images were used to manually draw regions of interest with the Image Processing Toolbox in MATLAB. These were done on a slice-by-slice basis, and care was taken to avoid the heart, blood vessels, and the mediastinum. Tumors were segmented using Otsu multi-thresholding, and total tumor volume was determined by summing those lesions for all slices within the total multislice volume of interest. This was an iterative process, and additional T₁-weighted images, and multislice maximum intensity projections were used to discriminate vessel from lesion.

ELISPOT was used to detect IFNγ produced by CD8 and CD4 T cells as described (21). Mouse tumor cells were processed as described previously (22) and subjected to the magnetic bead separation for isolating CD8 T and CD4 T cells according to manufacturer's recommendations (Miltenyi Biotec). Next, 4 × 10⁵/well CD8 or CD4 T cells and 1 × 10⁵/well of 5,000 rads irradiated tumor cells were plated in triplicate wells and incubated in 96-well plates at 37°C for 24 hours. LKR cells were stimulated with IFNγ to increase MHC expression. T cells were also cultured alone or with Concavalin A as negative and positive controls, respectively. Plates were washed 6 times with PBS + 0.05% Tween 20 and 100 μL/well of biotinylated anti-IFNγ detecting antibody (eBioscience, cat no. 13-7312-85) diluted to 1 μg/mL in PBS + 0.05% Tween 20 was added. Avidin-HRP (BD Biosciences, cat no. 554058) was used as detection reagent. Spot counting was done with an AID ELISPOT Reader System (Autoimmun Diagnostika GmbH).

Cells were incubated for 5 minutes at room temperature with Fc block and DAPI was added prior to analysis to assess viability. Flow cytometric analysis was performed on an LSR II cytometer (BD Biosciences). Data were acquired using CellQuest software (BD Biosciences) and analyzed using FlowJo software (Tree Star). For tumor-infiltrating T cells, cells were gated on strict forward and side scatter parameters to ensure single-cell analysis. In addition, the DAPI- population was used for viable cells. Finally, cells were assessed for CD3 and CD4 or CD8 expression. The following FACS antibodies were used: APC anti-mouse MHC Class I (H-2Kb; cat no. 17-5958-82) and PE-Cyanine 7 anti-Mouse CD45 (cat no. 25-0451-82) were purchased from eBioscience, PE CD274 (B7-H1, PD-L1) anti-mouse mAb (clone MIH5; cat no. A14764) was from Life Technologies, Biotin anti-mouse CD95 (Cat No. 554256), anti-mouse CD16/CD32 (Mouse BD Fc Block; cat no. 553142) PE anti-mouse CD3 (cat no. 553063), FITC anti-mouse CD4 (cat no. 553729), and PerCP-Cy5.5 anti-mouse CD8 (cat no. 551162), were from BD Biosciences. Cell culture supernatants were tested for presence of IFNγ using a CBA assay (BD Biosciences).

Statistical analysis was performed using two-tailed Student t test, Student t test with Welch correction, and Fisher exact test. GraphPad Prism 5 software (GraphPad Software Inc.) was used with significance determined at P < 0.05.

To identify agents capable of enhancing T-cell chemokine expression, we utilized the Approved Oncology Drugs Set consisting of 97 agents. In addition to conventional cytotoxic chemotherapy drugs, this set also includes epigenetic agents and targeted agents such as proteasome and kinase inhibitors. On the basis of the rationale that driver oncogenic mutations may also impact the response to treatment, we used the KRAS-mutant LKR-13 (LKR) lung cancer cell line (16) as a model system as KRAS is frequently mutated in lung adenocarcinoma. The approach to determine concentrations of different agents for equitable comparison was based on their LD₅₀ and is described in Supplementary Materials. Expression of T-cell–attracting chemokine genes (Ccl5, Cxcl9, and Cxcl10) is regulated by NF-κB and IFNγ-induced pathways (23, 24); consequently, agents which potentially module activity of these pathways may induce T-cell chemokine expression. Within the subset of agents that induced cytotoxicity (see Supplementary Table S1), several agents (vinrelobine, teniposide, bortezomib, vincristine, romidepsin, vinblastine, and docetaxel) induced Ccl5 mRNA (>10-fold) in the original screen (Fig. 1A; full list of agents is shown in Supplementary Fig. S1); however, only the HDAC inhibitor (HDACi) romidepsin additionally induced strong expression of Cxcl9 and Cxcl10 at LD₅₀ concentrations (30 nmol/L; Fig. 1B and C). For Cxcl10 in particular, this effect of romidepsin was evident across a range of different concentrations (5–30 nmol/L; Supplementary Fig. S2A). An additional HDACi in the Oncology Drugs Set, vorinostat, on the other hand increased Cxcl9 and Cxcl10 expression more strongly than Ccl5 expression (Supplementary Fig. S2B). Focusing on romidepsin, we further confirmed that romidepsin treatment also increased CCL5 and CXCL10 secretion in LKR cells by ELISA (Fig. 1D; CXCL9 was not tested). We further validated enhanced gene expression of Ccl5, Cxcl9, and Cxcl10 in the mouse lung cancer cell line 344SQ (17), bearing mutations in KRAS and TP53, and the human A549 lung cancer line (Fig. 1E and F). As Cxcl10 showed the highest fold induction (Fig. 1E and F), we evaluated changes in expression of this T-cell chemokine in further studies. Ability to induce CXCL10 gene expression was shared at LD₅₀ concentration dosing by several other clinically relevant HDACi MS-275, MGCD0103, LBH-589, and vorinostat in A549 cells (Fig. 1G), and with matching increases in CXCL10 protein levels demonstrated for both romidepsin and vorinostat (Fig. 1H). As myeloid cells are known to express Ccl5, Cxcl9, and Cxcl10, we also tested romidepsin ability to induce these chemokines in Raw 264.7 macrophages. Importantly, romidepsin potently induced gene expression of all 3 chemokines (Fig. 1I). Finally, romidepsin also triggered Cxcl9 and Cxcl10 expression in macrophages isolated from LKR tumors (Fig. 1J). These results indicate that among oncology agents in clinical use, HDACi are powerful activators of T-cell chemokine expression.

To determine the effects of romidepsin on tumor growth in vivo, we implanted LKR cells subcutaneously in syngeneic immunocompetent 129S4/SvJaeJ mice and allowed tumors to grow beyond 200 mm³ (day 12–14). Romidepsin given on days 14, 16, and 18 after tumor cell inoculation at 2 mg/kg, a dose well below the MTD of 5.6 mg/kg (25), induced a highly significant decrease in tumor growth in LKR (P = 0.0024; final tumor volume) (Fig. 2A). While tumor rejection was not observed with this treatment regimen, a prolonged period of stable disease was seen (Fig. 2A). Consistent with in vitro studies, we found that whole tumor gene expression of T-cell chemokines was significantly enhanced after romidepsin treatment (Fig. 2B). To determine whether T cells were necessary for the observed response to romidepsin, mice were pretreated with depleting mAbs against both CD4 and CD8 (Supplementary Figs. S3 and S4). Although T-cell depletion did not significantly impact the growth of LKR tumors alone, it completely reversed the antitumor effects of romidepsin (P < 0.0001; final tumor volume; Fig. 2C). Thus, a critical mechanism of action of this HDACi likely involves enhancement of antitumor T-cell responses. Furthermore, despite ability to induce cell death in vitro, these results indicate that romidepsin does not have a direct therapeutic effect on tumor cells.

Importantly, we observed increase in T-cell infiltration in pooled sets of day 17 LKR tumors after romidepsin treatment, for example, CD4: 0.6% to 1.4% and CD8: 0.9 to 2.7% in the experiment shown (Fig. 2D). However, combined results from three independent experiments showed that a significant increase in infiltration was evident in CD8 but not in CD4 T cells (Fig. 2E). Microarray studies showed that Ccl5 and Cxcl9/10 were the only T-cell chemokines induced after romidepsin treatment (Supplementary Table S2). To examine whether T-cell recruitment was necessary for the antitumor effect of romidepsin, we pretreated mice with TAK-779, an inhibitor of CCR5 and CXCR3 (26), the receptors of CCL5 and CXCL9/10/11, respectively. TAK-779 prevented the romidepsin-induced increase in T-cell density (CD4: 1.4% to 0.9% and CD8: 2.7% to 1.2%), but did not reduce steady-state T-cell numbers (CD4: 0.6% to 0.8% and CD8: 0.9 to 1.0%; Fig. 2D). As with a significant increase in CD8 T-cell infiltration after romidepsin treatment, a significant decrease in infiltration after TAK-779 was evident for CD8 but not CD4 T cells in combined results from three independent experiments (Fig. 2E). Although TAK-779 did slightly reduce LKR tumor growth when used alone, it also abrogated the effect of romidepsin treatment, indicating a crucial role for the CCR5 and/or CXCR3 receptors and their respective chemokines (Fig. 2F). Collectively, these findings suggest a mechanism whereby T-cell chemokine expression induced by romidepsin promotes T-cell recruitment, which in turn drives an antitumor response and promotes disease stabilization.

The response to PD-1 blockade has been associated with interrelated parameters of T-cell infiltration, an Th1/IFNγ signature, and tumor PD-L1 expression (4, 27–29), suggesting therapies that augment T-cell infiltration or effector function could synergize with checkpoint blockade. We first assessed PD-L1 levels in mouse LKR, 393P, and 344SQ lung cells (16, 17) in vitro with or without 24 hours of exposure to romidepsin or IFNγ. Importantly, romidepsin treatment increased PD-L1 levels although this did not compare in magnitude to the increase observed with IFNγ (Supplementary Fig. S5). We next determined whether LKR tumors were responsive to anti-PD-1 therapy by injecting anti-PD-1 on days 15, 17, and 19 after tumor inoculation. Tumor growth was significantly reduced by anti-PD-1 treatment, but complete rejection was seldom observed over the 30-day period (1/7 mice; Fig. 3A). While occasional tumor rejection was seen over approximately 30 days, longer term follow-up of 47 days in separate experiments showed that neither romidepsin nor anti-PD-1 caused tumor rejection beyond 30 days (Supplementary Fig. S6A). Instead, for both romidepsin and anti-PD-1 treatments, robust tumor growth resumed after a period of stasis (Supplementary Fig. S6A). To evaluate synergism between romidepsin and anti-PD-1 therapy, romidepsin was administered on days 14, 16, and 18, with anti-PD-1 administered on each subsequent day. Combined anti-PD-1 and romidepsin significantly reduced tumor growth compared with individual treatments with romidepsin (P = 0.0008) or anti-PD-1 (P = 0.0041; Fig. 3A) and rejection was noted in 9 of 11 tumors (Fisher exact test, P < 0.0001; Supplementary Fig. S6B). Furthermore, no regrowth of rejected tumors was observed in anti-PD-1 and romidepsin-treated mice (Supplementary Fig. S6C). We also evaluated combination therapy with romidepsin and anti-PD-1 in the 344SQ and 393P tumor models. Tumor rejection was not seen with 344SQ tumors, but combined anti-PD-1 and romidepsin treatment significantly delayed tumor growth compared with either individual treatment (Supplementary Fig. S7). A more pronounced response was observed with 393P tumors, as complete tumor rejections were observed with combination therapy in five of five animals, but in none of the tumor-bearing animals treated with individual agents (Fig. 3B). The striking difference in response in 344SQ and 393P tumor models is, however, not due to differences in oncogene/tumor suppressor mutations as they both harbor KRAS and TP53 mutations (17).

Luciferease-expressing LKR cells were also injected in the lung as an orthotopic tumor model. As with subcutaneous tumors, the greatest antiresponse was seen after combined anti-PD-1 and romidepsin treatment as determined by bioluminescence imaging (BLI; Fig. 3C and D). Finally, we evaluated the effect of combined romidepsin and anti-PD-1 treatment in a conditional mutant KRAS^G12D autochthonous knock-in mouse model of lung cancer (20). Tumor growth was initiated by intratracheal injection of replication-deficient adenovirus-expressing CRE recombinase, as described previously (20). Three months later, MRI was used to determine total baseline tumor volume, followed by two rounds of treatment with three doses each of romidepsin and anti-PD-1 treatment. Tumor volume evaluated 1 month after the first MRI increased more than twofold in untreated mice (Fig. 3E). In mice treated with romidepsin and anti-PD-1, tumor volume was significantly reduced by approximately 50% (Fig. 3E). As expected, H&E staining of untreated lung specimens showed significantly larger tumor burden than the lungs of treated mice (Fig. 3F). Together, these findings provide strong rationale for evaluating combined romidepsin and anti-PD-1 mAb therapy in the clinic.

Microarray studies showed that many of the romidepsin-induced genes were known targets of IFNγ (e.g., T-cell chemokines, MHC, and FAS; Supplementary Table S2). Interestingly, expression of the key IFNγ-induced transcription factor STAT1 was also increased by romidepsin (Supplementary Table S2), potentially linking effects of this HDACi to IFNγ signaling. Using additional microarray studies, we tested the possibility that romidepsin may impact IFNγ-induced gene expression. Interestingly, we found that expression of Ccl5 and Cxcl9/10 was synergistically increased after romidepsin and IFNγ treatment (Supplementary Fig. S8A). In addition, as with romidepsin alone treatment (Supplementary Table S2), we also found increase in expression of Stat1 (3 of 45 upregulated probe sets). Interestingly, MHC (H2k)and Fas gene expression was increased after romidepsin treatment (Supplementary Table S2) but which was not further enhanced by IFNγ. These results indicate that combined romidepsin and IFNγ treatment can potentiate expression of a subset of genes regulated by these agents, including T-cell chemokines and Stat1. We validated our microarray data demonstrating increased expression of Cxcl10 in LKR cells by real-time PCR (Supplementary Fig. S8B). Synergism between romidepsin and IFNγ was also pronounced in 393P cells (Supplementary Fig. S8C).

Previous studies have demonstrated that anti-PD-1 enhances T-cell chemokine expression and T-cell infiltration, and that this effect is mediated by IFNγ production (30). Based upon the ability of IFNγ to synergistically promote chemokine expression with romidepsin in vitro, we postulated that IFNγ was also a central regulator of the antitumor response seen with romidepsin and anti-PD-1 in vivo. Consistent with this interpretation, combination therapy in mice bearing LKR tumors increased tumor expression of Cxcl10 and Cxcl9 by 50-fold and 15-fold, respectively (Fig. 4A and B). This compared to only a 10-fold increase in tumor expression of Cxcl10 and a fivefold increase in Cxcl9 in the single-agent control groups (Fig. 4A and B). The chemokine expression pattern was closely matched by increased expression of Ifng within the tumors of mice treated with both therapeutic agents (Fig. 4C). Chemokine expression levels correlated with T-cell density, with the highest numbers of CD4 and CD8 T cells present within the tumors of animals treated with combination therapy (Fig. 4D and E).

We next determined whether IFNγ was functionally required to promote chemokine expression and mediate the antitumor effects of combination therapy. To this end, we used an anti-IFNγ antibody to neutralize endogenous IFNγ starting 2 days before initiation of romidepsin/anti-PD-1 treatment. IFNγ neutralization completely abrogated the induction of Cxcl9/Cxcl10 expression within tumors after romidepsin and anti-PD-1 treatment (Fig. 4F and G). Furthermore, neutralization of IFNγ also substantially reduced T-cell infiltration (Supplementary Fig. S9). IFNγ neutralization also blocked the induction of IFNγ gene expression (Fig. 4H), consistent with a feed-forward signaling pathway where production of IFNγ by T cells enhances tumor cell immunogenicity (31), in turn further enhancing T-cell recruitment and function. Finally, we found that IFNγ neutralization ablated the effects of combination therapy, with tumor growth comparable with untreated controls (Fig. 4I). As with reversion of romidepsin therapy alone (Fig. 2F), administration of the CCR5/CXCR3 inhibitor TAK-779 also prevented a response to combination therapy (Fig. 4I). Thus, romidepsin and anti-PD-1 treatments synergize by promoting a chemokine-IFNγ–positive feedback loop through tumor cell and T-cell intermediaries.

We next sought to determine potential direct effect of romidepsin on T cells in vitro and in vivo. First, in T cells stimulated with TCR/CD28 engagement, we did not see any significant impact of romidepsin on IFNγ secretion (Fig. 5A). Previous studies suggest that T cells can also express CCL5 and CXCL10 (32, 33). We therefore next tested the interesting possibility that romidepsin may also induce expression of these chemokines in T cells. Indeed, a substantial increase in expression of Ccl5 and Cxcl10 was seen after romidepsin treatment of both CD4 and CD8 T cells (Fig. 5B–D). Therefore, in addition to tumor cells and macrophages, romidepsin can also enhance expression of chemokines in T cells. Finally, we determined whether romidepsin also enhances T-cell functionality in vivo. Importantly, the frequency of tumor CD8 and CD4 cells that produce IFNγ ex vivo in response to LKR cells in an ELISPOT assay was significantly increased by combination therapy (Fig. 5E and F). This measure of T-cell functionality was not increased by either romidepsin or anti-PD-1 therapy alone, again highlighting the synergistic nature of the combination. Therefore, in addition to promoting T-cell infiltration, romidepsin can also enhance recruited T-cell function in manner that synergizes with PD-1 blockade.

Recent clinical studies have demonstrated the considerable potential of immunotherapy in cancer treatment. However, even with combinatorial approaches with anti-CTLA-4 and anti-PD-1, there remain a significant percentage of patients that do not derive a clinical benefit (34). We hypothesized that oncology agents that enhance T-cell recruitment would increase response to immunotherapy. We screened a library of FDA-approved oncology drugs to determine ability to induce expression of Ccl5, Cxcl9, and Cxcl10 and found that only a single-agent class, HDACi, was capable of inducing expression of these chemokines. Most importantly, the HDACi romidepsin induced a strong T-cell–dependent antitumor response and combinatory treatment with anti-PD-1–induced tumor regression or rejection across multiple lung tumor models. As multiple HDACi and PD-1 blockade agents (pembrolizumab and nivolumab) are now FDA-approved, combinatory therapy with these two agent classes may represent a promising approach for lung cancer treatment in the near term.

As suggested by recent clinical results (27), the response to anti-PD-1 is restricted by poor T-cell infiltration into tumors. We found that HDACi therapy alone increases tumor expression of Cxcl9 and Cxcl10, promoting T-cell recruitment and a period of tumor stasis (see Fig. 6). The increase in T-cell recruitment was typically more significant for CD8 T cells, the subset strongly implicated in benefit from PD-1 blockade therapy (27). Chemokine expression may depend on multiple cell types as our results indicate romidepsin ability to induce expression in tumor cells, macrophages, and T cells. However, HDACi treatment also upregulated PD-L1 expression, which may limit T -cell functionality (Fig. 6) (35). On the other hand, combination therapy unleashes an IFNγ-dependent response to promote maximal T-cell chemokine expression and T-cell recruitment without sensitization of T cells to the upregulation of PD-L1 on tumor cells (Fig. 6). Limiting CCR5/CXCR3-dependent recruitment of T cells or neutralizing IFNγ prevents response to combination therapy, implicating both steps as critical for therapeutic efficacy. We utilized multiple tumor models (LKR, 393P, and 344SQ) to show greater efficacy of combined treatment versus individual treatments. In the KRAS autochthonous model, however, only the effect of combined romidepsin and anti-PD-1 treatment was studied. Additional studies will therefore be required to determine whether combined treatment is also more efficacious in this model compared with single treatments. Interestingly, recent studies showed reduced immune surveillance in lung adenocarcinoma with mutations in the STK11/LKB1 tumor suppressor (36, 37). The impact of therapeutic regimens identified here on treatment response of KRAS/STK11 as well as the widely used KRAS/TP53 autochthonous model (38) will therefore also be important to determine.

Interestingly, we found that individual depletion of CD4 or CD8 T cells was sufficient to reverse the antitumor effect of romidepsin (Supplementary Fig. S10). In addition to increasing infiltration, an important finding was the increase in tumor-infiltrating CD4 and CD8 function after combination therapy (Fig. 5E and F). This increase may be a consequence of romidepsin-induced increase in tumor cell immunogenicity (e.g., through upregulation of MHC expression), coupled with direct augmentation of T-cell function by anti-PD-1 and potentially by romidepsin (Fig. 6). Interestingly, previous studies indicate that in addition to promoting T-cell recruitment, Cxcl10 also enhances effector T-cell function (33, 39). Therefore, another interesting possibility is that romidepsin-triggered Cxcl10 expression synergizes with anti-PD-1 to enhance T-cell function. Further studies will be required to determine specifically how HDACi ± anti-PD-1 modulate T-cell function. An important related issue is the effect of in vivo HDAC dose. While we used a dose below the MTD of romidepsin, it is possible that higher dosing may adversely impact function of T cells or other cell types crucial for the antitumor response and thereby reduce therapeutic efficacy. The HDACi dose capable of inducing beneficial epigenetic changes and gene expression but without cytotoxicity towards T cells will also be important to determine for clinical use of these agents when combined with immunotherapeutics. HDACi treatment has previously been reported to induce systemic myeloid-derived suppressor cells (MDSC) depletion leading enhancement of T-cell responses induced by checkpoint blockade (40). However, we found that romidepsin treatment of mice did not impact MDSC numbers (Supplementary Fig. S11), suggesting that effects of HDACi treatment observed here are not due to MDSC depletion.

Histone modification by histone acetyltransferases (HAT) is typically associated with increased gene expression while removal of acetyl groups by HDACs represses gene expression. Although overall NF-κB activity was crucial for romidepsin-induced T-cell chemokine expression, romidepsin did not enhance nuclear translocation of NF-κB (data not shown). In contrast, romidepsin increased Stat1 mRNA and further enhanced Stat1 protein levels after IFNγ treatment (Supplementary Fig. S12). A likely mechanism of synergy may therefore involve IFNγ activation of STAT1 coupled with romidepsin-induced histone modifications to maximize gene expression. It is also likely that genes uniquely regulated by romidepsin or IFNγ act in concert to promote antitumor responses. Interestingly, it was recently shown that romidepsin decreased levels of β-catenin by upregulating expression of its inhibitor SFRP1 (41). In addition, β-catenin signaling was recently shown to suppress checkpoint immunotherapy response (42). An interesting possibility that requires careful examination is whether suppression of β-catenin signaling is an additional mechanism by which romidepsin promotes the anti-PD-1 response.

While the current study focused on pan-HDACi, we also tested HDACi specific for HDAC1+2, HDAC3, HDAC4, HDAC6, and HDAC8. However, we did not observe induction of Cxcl9/10 T-cell chemokine expression as was evident after romidepsin or vorinostat treatment. Additional studies will be required to determine whether these genes are specifically regulated by an HDAC other than the above or whether inhibition of multiple HDACs is required for induction of expression. In an important recent study, it was shown that enhancer of zeste homologue 2 (EZH2)-mediated histone H3 trimethylation (H3K27me3) and DNA methyltransferase 1 (DNMT1)-mediated DNA methylation repress ovarian tumor cell expression of Cxcl9 and Cxcl10 (43). Importantly, inhibitors of EZH2, such as DZNep or GSK126, and DNMT1 inhibitor 5-AZA-dC enhance chemokine expression to promote effector T-cell trafficking to the tumor and the response to PD-L1 blockade immunotherapy (43). This study, together with our findings, indicate that regulation of chemokine expression is subject to control by multiple and distinct epigenetic mechanisms. Modulation of tumor immunogenicity by drugs that target epigenetic mechanisms may therefore represent a promising area for translational research and clinical intervention (44).

HDACi are approved for treatment of certain hematologic malignancies, but have failed to demonstrate efficacy in solid tumors. Interestingly, the combination of the HDACi entinostat with 5-AZA-dC has shown benefit in lung cancer treatment, highlighting the potential for HDACi to augment response to other epigenetic therapies (45). Importantly, supportive of our findings, a recent study showed that the HDACi vorinostat increased CD8 T-cell infiltration in human lung tumors (46). In the preclinical setting, there is also evidence of immune modulation by HDACi (47–50). On the basis of the novel findings reported here, we strongly believe that the combination of HDACi with immunotherapy represents a powerful approach for cancer treatment. Notably, several clinical trials employing this combinatory approach are now underway, including one at our institution in patients with NSCLC.

No potential conflicts of interest were disclosed.

Conception and design: H. Zheng, X. Ren, S.J. Antonia, A.A. Beg

Development of methodology: H. Zheng, W. Zhao, C. Yan, M. Xie, G.V. Martinez, A.A. Beg

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Zheng, C. Yan, C.C. Watson, M. Messengill, C. Messengill, D. Noyes, G.V. Martinez, R. Afzal

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Zheng, C. Yan, M. Messengill, M. Xie, C. Messengill, G.V. Martinez, R. Afzal, Z. Chen, S.J. Antonia, E.B. Haura, A.A. Beg

Writing, review, and/or revision of the manuscript: H. Zheng, C.C. Watson, M. Messengill, C. Messengill, G.V. Martinez, S.J. Antonia, E.B. Haura, B. Ruffell, A.A. Beg

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Zheng, M. Xie, R. Afzal, X. Ren, E.B. Haura

Study supervision: H. Zheng, X. Ren, A.A. Beg

The authors thank Dr. Kenyon Daniel and Nan Sun (Chemical Biology core) for help setting up chemotherapy agent screening and Dr. Lu Chen for help with statistical analysis, and Dr. J. Kurie (M.D. Anderson Cancer Center, Houston, TX) for providing 393P and 344SQ cells. The authors also acknowledge the help provided by Flow Cytometry, Molecular Genomics, Pathology and Comparative Medicine core facilities at the Moffitt Cancer Center.

This work was supported by lung cancer SPORE P50 CA119997, funds from the Moffitt Cancer Center Lung Cancer Center of Excellence, National Natural Science Foundation of China (grant no. 81472183), and NIH R01 AI0802685.

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.