Stabilized Peptide HDAC Inhibitors Derived from HDAC1 Substrate H3K56 for the Treatment of Cancer Stem–Like Cells In Vivo

Two of the most fundamental hallmarks of malignant tumors are their heterogeneous nature and limitless proliferation potential (1). The cancer stem cell (CSC) theory states that malignant proliferation can be fueled by a small subset of stem-like tumor cells, which possess similar properties as typical stem cells and can contribute to drug resistance and cancer relapse (2). Cancer stem–like cells have been found in many common cancer types such as breast cancer (3), colorectal cancer (4), ovarian cancer (5), and lung cancer (6). Increasing evidence has revealed that epigenetic mediators, especially HDACs, can contribute to the emergence and maintenance of CSCs (7–9). HDACs control cancer cell fortune through the removal of the acetyl group of histone and nonhistone proteins. The aberrant acetylation of histones has been identified in oncogenesis, for example, the loss of acetylation of H4K16 (H4K16ac) is found in various cancer types such as lymphoma, and cancers of the breast, colon, and lung (10). Additionally, HDACs can interact directly with some transcription factors, such as E2F, Stat3, p53, Rb, and NF-κB, to regulate important signaling pathways (11). Thus, the aberration of HDACs in cancer cells has fueled intense interest in the development of HDAC inhibitors for cancer therapy (12).

HDACs are grouped into four classes based on their homology: class I (HDACs 1, 2, 3, and 8), class II (HDACs 4–7, 9, and 10), and class IV (HDAC 11) are Zn²⁺ dependent, and class III, referred to as sirtuins, are NAD⁺ dependent (13). Different HDACs are thought to have their own substrate specificity, including histone substrates and nonhistone substrates. However, the identification of their histone substrates is difficult as different HDACs or HDAC complexes may have synergistic effects (14). Nowadays, HDAC1 is reported to have preference for H3K56ac and H4K16ac, while HDAC3 preferentially deacetylates H4K5 and H4K12 (14–16). For nonhistone substrates, HDAC1 and Sirtuin 1 are reported to deacetylate P53 (17, 18), and HDAC6 is reported to deacetylate lysine 40 of α-tubulin (19).

Due to the high similarity of the Zn²⁺-dependent enzyme pockets of HDACs, the four FDA-approved drugs are all pan-HDAC inhibitors and usually share a common dose-limiting toxicity, impeding their further clinical application (20). The intrinsic biocompatibility and complex chemical structure of peptides have attracted chemists to develop peptide inhibitors with improved HDAC isoform selectivity. The most famous peptide-based HDAC inhibitors are cyclic tetrapeptides equipped with active Zn²⁺ binding groups, such as the approved drug FK228 (21). These macrocyclic peptides usually display class I selectivity, and the selectivity is hypothesized to be attributed to peptide complex cap groups that interact with a large area of the enzyme surface (22). Despite its preference for class I HDACs, FK228 still shows the common side effects of other approved HDACIs. Thus, it is still unclear whether or not an isoform-selective HDAC inhibitor would be advantageous over pan-HDAC inhibitors in cancer therapy (20). Therefore, taking the peptides’ biocompatibility and affinity for target binding into consideration, we hypothesized that a suitable HDAC-interacting peptide bearing an active Zn²⁺ binding group might provide a suitable scaffold for novel HDACIs, which would result in a significantly increased safety window and better selectivity toward malignant cells.

Peptidomimetics are efficient tools to inhibit protein–protein interactions and have emerged as a new modality for cancer therapy. Recently, our group has developed several efficient helical stabilization methodologies including the N-terminal helix nucleating template (TD strategy) and the in-tether chiral center induced peptide helicity (CIH strategy) based on the concept that a precisely tuned in-tether chiral center could dominate the peptide's secondary structure and its biological functions (23, 24). Peptides constructed with these methods are successfully utilized to target ERα–coactivator interaction and p53–MDM2/MDMX interaction (25, 26). Inspired by the previous success of FK228 and our own experience in stabilized peptides, we hypothesized that peptide substrate analogues with large binding surfaces could achieve enhanced efficiency and reduced toxicity through effectively disrupting the interaction between HDACs and their substrates. The HDAC substrates acetylated H3K56 and H4K16 have been reported to have a preference for HDAC1 and were previously examined (15, 16). Thus, we chose to modify the helical region (comprising residues 45–56) of H3 and stabilized the peptide structure using our developed method—the TD strategy. We envisioned that these stabilized peptide-based HDAC inhibitors with large interacting interfaces could improve the molecules’ selectivity for HDACs over other nonspecific interacting proteins and ultimately reduce nonspecific toxicity. In this report, we found the designed peptide inhibitors showed potent pan-HDAC inhibition but selective cytotoxicity toward cancer stem–like cells (PA-1 cells and NTERA-2 cells) along with negligible toxicity toward normal cells even at much higher concentrations. The superior tumor-inhibiting effect displayed in the PA-1 and NTERA-2 xenograft animal models resulted in negligible organ lesions at the high dosage (50 mg/kg, every other day), further supporting the efficiency and safety of our designed peptides. To our knowledge, this study was the first to showcase the stabilized peptide as a cap group used to decrease the nonspecific toxicity of HDAC inhibitors, demonstrating the wide potential for combined application of stabilized peptides with clinical approved therapeutics.

All solvents and reagents used for solid-phase peptide synthesis were purchased from Shanghai Hanhong Chemical Co., J&K Co. Ltd., Shenzhen Tenglong Logistics Co., or Energy Chemical Co. and were used without further purification unless otherwise stated. Antibodies against HDAC6, HDAC8, P21 Sox2, Ac-tubulin, GAPDH, and H3 were obtained from Proteintech; antibodies against histone H4K16ac were obtained from Millipore; antibodies against HDAC1, HDAC4, HDAC5, SIRT6, LSD1, H3K4m1, H3K4me2, H4, and active caspase-3 were obtained from Abcam; antibodies against HDAC2 and HDAC3 were obtained from BETHYL; antibody against H3K56ac was obtained from Millipore and Active motif. The purified recombinant human HDAC 3, 6, 8 and their corresponding substrates were purchased from BioVision Inc. The purified recombinant human HDAC1 was obtained from Caymanchem Inc. Annexin V: FITC Apoptosis Detection Kit I was obtained from BD Pharmingen. Reagents used for biological assays were purchased from Sigma-Aldrich and Thermo Fisher. Cells were purchased through ATCC and cultured according to ATCC guidelines.

Human ovarian cancer cell lines, PA-1 cells (CRL-1572, ATCC, 2018), were cultured in MEM with 10% (v/v) fetal bovine serum (FBS) and penicillin/streptomycin (100 g/mL). Human malignant pluripotent embryonal carcinoma, NTERA-2 cells (CRL-1973, ATCC, 2018), were cultured in DMEM with 10% (v/v) FBS and penicillin/streptomycin (100 g/mL). Human lung adenocarcinoma, A549 cells (CCL-185, ATCC, 2018), were cultured in RPMI-1640 supplemented with 10% (v/v) FBS and penicillin/streptomycin (100 g/mL). Human cervix adenocarcinoma, HeLa cells (CCL-2, ATCC, 2018), and human kidney cells, 293T (CRL-11268, ATCC, 2018), were cultured in DMEM with 10% (v/v) FBS and penicillin/streptomycin (100 g/mL). Chang liver cells (gifts from the South China University of Technology) were cultured in DMEM with 10% (v/v) FBS and penicillin/streptomycin (100 g/mL). All reagents were purchased from Gibco. All these cells were maintained in a humidified incubator containing 5% CO₂ at 37°C. All cells were authenticated at the Nevada Cancer Institute by specific markers such as Oct4, Sox2, p53, and p16^Ink4a and by their cell morphology and used within 3 months.

Details of the syntheses of peptide inhibitors and their structure analysis were given in the Supplementary Methods.

The purified recombinant human HDAC 3, 6, 8 and their corresponding substrates were purchased from BioVision Inc. The purified recombinant human HDAC1 was obtained from Caymanchem Inc. as the unavailable source from BioVision Inc. HeLa nuclear extract was obtained according to the Nuclear Extraction Kit (Solarbio). Assays were carried out in 384-well format using the fluorometric histone deacetylase kit according to the manufacturer's protocol (BioVision and Caymanchem). HDAC activity was detected at the excitation wavelength of 360 nm and the emission wavelength of 460 nm using Envision (PerkinElmer). The analytical software, GraphPad Prism 5.0 (GraphPad Software, Inc.) was used to generate IC₅₀ value via nonlinear regression analysis.

Cell viability for different cell lines was measured by MTT (3-(4, 5-dimethylthiazol-2-yl)-2, 5-diphenylt-etrazolium bromide, from Sigma) assays. The cells were incubated on a 96-well plate for 24 hours in growth medium prior to drug treatment. Then the media were removed followed by adding peptides or suberoylanilide hydroxamic acid (SAHA) in medium with 5% FBS (v/v) for 24-hour incubation, replaced by 10% FBS medium for another 24 hours, if necessary. MTT (5 mg/mL, 20 μL) in phosphate-buffered saline (PBS) was added, and the cells were incubated for 4 hours at 37°C with 5% CO₂. DMSO (150 μL, Sigma) was then added to solubilize the precipitate with 10 minutes of gentle shaking. Absorbance was measured with a microplate reader (Bio-Rad) at a wavelength of 490 nm.

For Western blot analysis, different cancer cells were seeded in 12-well plates with 24-hour incubation, then treated with peptides or SAHA for another 24 or 48 hours. The cells were washed with PBS and harvested using the lysis buffer (50 mmol/L Tris-Cl, PH = 6.8, 2% SDS, 6% glycerol, 1% β-mercapitalethanol, 0.004% bromophenol blue). The extracted protein concentrations were measured by a spectrophotometer (Nano-Drop ND-2000). An equal amount of protein was loaded onto a SDS-PAGE gel and resolved by electrophoresis. Protein bands were then transferred to Nitrocellulose Blotting membranes followed by incubation with an appropriate primary antibody overnight at 4°C. Antibody dilutions were used as follows: 1:500 for rabbit monoclonal anti-HDAC1 (Abcam), rabbit monoclonal anti-HDAC6 (Proteintech), rabbit monoclonal anti-H4K16ac (Millipore), and mouse monoclonal anti–Ac-tubulin (Proteintech) and 1:1,000 for rabbit polyclonal anti-H3 (Proteintech), rabbit monoclonal anti-H3K56ac (Millipore), and mouse monoclonal anti-GAPDH (Proteintech). Then, the membranes were washed with buffer (1× TBS, 0.05% Tween 20). Secondary antibodies of goat anti-rabbit or anti-mouse were used for 1-hour incubation at room temperature. Proteins were then visualized with chemiluminescent substrates.

Cells were pretreated as indicated in the Western blot experiment before the RT-PCR experiment. Forty-eight hours after incubation with peptides and SAHA, total RNA was extracted from PA-1 or NTERA-2 cells using TRIzol reagent (Invitrogen) according to the manufacturer's protocol. The isolated mRNA was quantified by a spectrophotometer (Nano-Drop ND-2000). Total RNA (2 μg) was reverse transcribed to cDNA using the reverse transcriptase kit (Takara) according to the manufacturer's instructions. The mRNA levels of the target genes were detected by real-time PCR using SYBR green (Takara) in an ABI Prism 7500 real-time PCR system (Applied Biosystems). Data represent two independent experiments with three technical replicates per experiment. Relative changes in transcript levels for each sample were determined by normalizing to β-actin housekeeping gene mRNA levels. Primers were shown in Supplementary Table S1.

To detect cell permeability by flow cytometry, all cells were incubated with 10 μmol/L FITC-labeled peptides for 4 hours in medium with 5% FBS at 37°C. After washing with media, the cells were harvested by trypsinization and washed twice with PBS. To quench extracellular fluorescence, the cells were incubated with 0.05% trypan blue for 3 minutes prior to fluorescence activated cell sorting (FACS) analysis.

The apoptosis assay was performed according to the manufacturer's instructions using an Annexin V:FITC Apoptosis Detection Kit I (BD Pharmingen). Briefly, PA-1 cells were seeded in a 12-well plate and allowed to grow for 24 hours in medium with 10% FBS. Then, the cells were treated with different peptides in medium with 5% FBS for 24-hour incubation. The cells were harvested by trypsinization and washed twice with PBS and suspended in 1 × binding buffer. The suspended cells were treated with FITC-labeled Annexin V and propidium iodide (PI) as the protocol indicated. The suspended cells were stained with PI and Annexin V and then analyzed by flow cytometry to determine apoptotic cells. Cells with positive fluorescence intensity signals for both FITC and PI were used for apoptotic cell count.

As for the cell-cycle arrest experiments, PA-1 cells were seeded in a 12-well plate and allowed to grow for 24 hours in medium with 10% FBS. The cells were treated with 20 μmol/L peptides in medium of 5% FBS for 24-hour incubation. Then, the cells were washed twice with PBS and harvested by trypsinization. The cells were fixed with cold 70% ethanol for 4 hours and isolated by centrifugation at 2,000 rpm for 5 minutes. Then, the precipitant cells were suspended in PBS with 1% Triton-100, 5 mg/mL PI, 1 mg/mL RNase and stained at 37°C for 30 minutes. The samples were detected by flow cytometer and the percentages of cells in G₀–G₁, S, and G₂–M phases were analyzed by FlowJo software.

PA-1 cells (or other cancer cells) were seeded in 24-well culture plate on coverslips. The cells were grown in DMEM with 10% FBS (v/v) in a 37°C, 5% CO₂ incubator for 24 hours. The cells were incubated with 10 μmol/L peptides for 4 hours at 37°C. Then, the cells were washed 3 times with PBS and then fixed with 4% (wt/vol) formaldehyde (Alfa Aesar, MA) in PBS for 30 minutes. They were then washed 3 times with PBS and stained with 6-diamidino-2-phenylindole (DAPI; Invitrogen) for 10 minutes. Images of peptide localization in cells were taken via confocal laser scanning microscope (FV1000, Olympus). Image processing was done using the Volocity software package (Zeiss Imaging).

The lactate dehydrogenase (LDH) release assay was used to detect the cell membrane damage and performed by the LDH release kit (Dojindo). PA-1 cells were seeded in a 96-well plate at a density of 5 × 10³ cells/well. Then, the cells were treated with 100 μL peptides with 2-fold serial dilution starting at 80 μmol/L for 12 hours at 37°C. Lysis buffer was added as positive control and incubated at 37°C for 30 minutes. Then, 100 μL working solution was added to each well and incubated for 30 minutes at room temperature. The reaction was stopped by 50 μL stop solutions and detected immediately at 490 nm absorption by a microplate reader (PerkinElmer, Envision). LDH release activity was calculated by [(ODsamples − ODblank)/(ODpositive − ODblank)] × 100.

Fresh mouse red blood cells were collected and centrifuged at 1,500 rpm for 10 minutes. Erythrocytes were washed 4 times and then resuspended in 0.9% NaCl to a final density of 10⁸/mL. A serial dilution of peptides was added staring at 200 μmol/L and incubated at 37°C for 1.5 hours. Then, rythrocytes were centrifuged at 6,000 rpm for 10 minutes and the release of hemoglobin was monitored by measuring the absorbance of supernatant at 570 nm by a microplate reader (PerkinElmer, Envision). Triton X-100 (0.1%) and 0.9% NaCl were used as positive and negative controls, respectively. The percentage of hemolysis was calculated according to the following equation: %Hemolysis = [(A_{576 nm} of sample − A_{576 nm} of negative control)/(A_{576 nm} of positive control − A_{576 nm} of negative control)] × 100.

PA-1 cells were maintained in MEM containing 10% FBS and penicillin/streptomycin for 24 hours. Whole-cell extracts were generated using lysis buffer [50 mmol/L Tris (pH 8.0), 150 mmol/L NaCl, 1 mmol/L EDTA, 0.5% Nonidet P-40]. Protein concentration of the soluble fraction was measured by a spectrophotometer (Nano-Drop ND-2000). Biotinylated peptides (40 μmol/L) were added to protein extracts (500 μg) and incubated at 4°C overnight. Then, the mixture was incubated with Streptavidin Magnetic Beads (Thermo) at room temperature for 2 hours. The beads were collected and washed 3 times with lysis buffer, eluted by boiling in SDS buffer, and subjected to Western analysis with HDAC antibody.

Exponentially growing PA-1 cells (10 cm dish) were treated with 10 μmol/L peptide 16cyc-HxA or equivalent volume of DMSO for 16 hours. Whole-cell extracts were generated using lysis buffer [50 mmol/L Tris (pH 8.0), 150 mmol/L NaCl, 1 mmol/L EDTA, 0.5% Nonidet P-40]. Protein extracts (500 μg) were precleared for 1 hour with 40 μL protein A sepharose beads (50%, Sigma) before the addition of the indicated antibodies. Then, pretreated cell extracts were incubated with 2 μg rabbit antibody anti-HDAC1(Abcam) or anti-LSD1(Abcam) overnight at 4°C. The mixture was incubated with protein A sepharose beads at 4°C for 2 hours. The beads were washed five times with cold lysis buffer. Precipitated proteins were subjected to Western blotting analysis.

For chromatin immunoprecipitation (ChIP) assays, 1 × 10⁷ to 5 × 10⁷ PA-1 cells were used for each sample. Proteins were crosslinked to DNA by addition of formaldehyde to a final concentration of 1.0% for 10 minutes at room temperature. After incubating with 125 mmol/L glycine for 5 minutes, cells were harvested using cold PBS and resuspended in FA lysis buffer (50 mmol/L HEPES-K⁺, pH 7.5, 140 mmol/L NaCl, 1 mmol/L EDTA pH 8.0, 1% Triton X-100, 0.1% sodium deoxycholate, 0.1% SDS, protease inhibitors), and sonicated to generate DNA fragments of 500 to 1,000 bp in average length. Soluble chromatin fragments were incubated with 2 μg primary antibodies or IgG overnight. Immunocomplexes were incubated with protein A sepharose beads for 2 hours, briefly centrifuged, and washed sequentially with the wash buffer (0.1% SDS, 1% Triton X-100, 2 mmol/L EDTA, 150 mmol/L NaCl, 20 mmol/L Tris-HCl, pH 8.0) three times and the final wash buffer (0.1% SDS, 1% Triton X-100, 2 mmol/L EDTA, pH 8.0, 500 mmol/L NaCl, 20 mmol/L Tris-HCl, pH 8.0) one time. Immunocomplexes were eluted, and the crosslinks were reversed in the elution buffer (1% SDS, 0.1 mol/L NaHCO3) at 65°C. Purified DNA was quantified by real-time qPCR. Sequences of the primers used for ChIP assays are listed in Supplementary Table S2.

Athymic nude mice (BALB/c ASlac-nu) were obtained from Vital River Laboratory Animal Technology Co. Ltd. of Beijing, People's Republic of China and allowed an acclimation period of 1 week. Mice were maintained in an isolated biosafety facility for specific pathogen-free animals with bedding, food, and water. All operations were carried out in accordance with the National Standard of Animal Care and Use Procedures at the Laboratory Animal Center of Shenzhen University, Guangdong Province, People's Republic of China (the permit number is SZU-HC-2014-02).

For tumor suppression assay, we first harvested PA-1 cells or NTERA-2 cells and resuspended in DMEM at a density of 1 × 10⁸/mL. Then, athymic nude mice (female; 6 weeks old) were inoculated with 1 × 10⁷ PA-1 cells or NTERA-2 cells (100 μL volume of each) propagated in vitro subcutaneously in the lower flank of mice. After 15–20 days, mice with tumors exceeding 100–200 mm³ in volume were randomly divided into 5 groups of 5–6 mice per treatment group. Mice bearing PA-1 or NTERA-2 tumors were intraperitoneally injected with peptides (50 mg/kg) or SAHA (50 mg/kg) using PBS as a negative control. Mice were injected every other day starting on day 0. Tumor volumes were measured by calipers (accuracy of 0.02 mm) every other day and calculated using the following formula: V = L × W²/2 (W being the shortest dimension and L the longest dimension). Each tumor was independently measured and calculated by changes in volume (folds) relative to day 0. Statistical significances between groups were tested by one-way analysis of variance.

When the tumor reached an appropriate volume of 100–200 mm³, mice were injected with Cy5-labeled 16cyc-HxA and 16lin-HxA peptides by intratumoral injection at 50 mg/kg. After injection, mice were anesthetized with 100 μL 1% pentobarbital sodium. Once the mice were properly anesthetized, they were imaged at indicated time points to monitor the accumulation of cyc5-labeled peptides in tumors using the CRI Maestro ex/in vivo imaging system (ex: 605 nm; filter: 700 nm).

The mice were sacrificed after 21 days treatment, and organ and tumor tissues were collected and fixed in 4% buffered formalin-saline at room temperature for 24 hours for histologic experiments. Then, the tissues were embedded in paraffin blocks, and paraffin sections of 4-mm thickness were mounted on a glass slide for hematoxylin and eosin (H&E) staining. The H&E staining slices were examined under a light microscopy (Olympus BX51). For IHC assay, the ovarian cancer tissue microarray slides were immersed in 3% H₂O₂ for 5 minutes to inactivate the endogenous peroxidase. BSA (5%) was used to block the nonspecific binding sites for 15 minutes. Antibodies against HDAC1, HDAC6, H3K56ac, H4K16ac, and caspase-3 were diluted as the primary antibodies and incubated with slides at 4°C overnight. The samples were washed and then incubated with Rabbit-Probe or Mouse-Probe MACH3 HRP-polymer detection system according to the supplier's instructions. Slides were developed with 3,3′-diaminobenzidine substrate using the ImmPACT DAB Peroxidase Substrate Kit (Vector Laboratories) for 1–5 minutes, counterstained with hematoxylin. (All reagents were obtained from Biocare Medical.)

BALB/c mice (female; 6 weeks old) were obtained from Vital River Laboratory Animal Technology Co. Ltd. of Beijing, People's Republic of China and allowed an acclimation period of 1 week at 22 ± 2°C with a 12-hour light–dark cycle (lights on 8 am, lights off 8 pm). Subsequently, BALB/c mice were randomly divided into 5 groups (4–5 mice per group) and were subcutaneously treated with PBS, peptides or SAHA (50 mg/kg), respectively, via intraperitoneal injection every other day. A voluntary running system consisting of six separated chambers (Chengdu TME Technology Co., Ltd) was used in the animal performance study. During the training session, mice were placed on the motorized rod (30 mm in diameter) in the chamber. Rotation speed gradually increased from 0 to 100 rpm over the course of 100s. The rotation speed was recorded when the animal fell off from the rod. Performance was measured as the average rotation speed animals achieved during the training session. The different groups were all trained in five same continuous time points (days 0, 2, 4, 8, 10, 12, 14, 16, 18, and 20). No significant differences were found between these five groups.

In 2016, Watson and colleagues developed a substrate-like peptide inhibitor based on the HDAC1-specific substrate H4K16, with K16 replaced with a hydroxamic acid functionality, which showed nanomolar HDAC1 inhibition (27). The crystal structure of the HDAC1–MTA1 complex with this peptide revealed that the peptide alone can make complementary contact with the rim of the HDAC active site. This is a new attempt to develop a substrate mimic for HDAC inhibition. However, calls for further study on the cellular functions of this peptide inhibitor remain unanswered. Herein, we chose to mimic another HDAC1-specific substrate, H3K56, based on the preferential deacetylation of HDAC1 on H3K56. The crystal structure of H3 indicates that H3K56 is on the C terminus of the α-helical region of H3 (comprising residues 45–56), suitable for incorporation of hydroxamic acid at K56 and N-terminal stabilization using the TD strategy (23, 28). Thus, we chose to modify the helical region (comprising residues 45–56), and a detailed design is shown in Fig. 1A and Supplementary Table S3. The three additional arginines on the N-terminus of peptides were used to enhance their cellular penetration. To study the structure and activity relationship, enzyme inhibition assays were performed on peptides with different lengths or sequences toward HDACs from HeLa nuclear extract shown in Fig. 1B (HeLa nuclear extract is reported to mainly contain HDAC1/2 protein, which was validated in Supplementary Fig. S1A and S1B; refs. 29–31). The peptide 16cyc-HxA showed better HDAC1/2 enzyme inhibition, emphasizing the importance of proper peptide sequences (Fig. 1B). Similar to TSA, the peptides 16cyc-HxA showed pan-HDAC inhibition shown in Fig. 1C, but minimal effect on the NAD⁺-dependent SIRT1 enzyme (Supplementary Fig. S2A and S2B). The pan-HDAC inhibition may be due to the high homology of enzyme pocket in HDAC isoforms including HDAC1, which can tightly bind to hydroximic acid. Although HDAC1 preferably binds to H3K56, the strong binding of hydroximic acid and the flexibility of peptide H3K56 still somehow rendered bindings with other isoforms. The 16cyc-HxA was more potent than 16lin-HxA, suggesting that helical stabilization can enhance target-binding affinity (Supplementary Figs. S3A–S3B and S4A–S4D). Despite their limited HDAC isoform selectivity, we believe with further understanding of the interactions between HDACs with their partners, isoform-selective HDAC inhibitors will emerge through precise adjustment of peptide structure.

To better investigate the cellular functions of these peptide inhibitors in different cell lines, cellular uptake efficiency and intracellular distribution were tested by cell flow cytometry and confocal microscopy shown in Fig. 2A and B and Supplementary Figs. S5A–S5B to and S8A–S8B. FITC-labeled TAT is a commonly used cell-penetrating peptide derived from HIV integrase (32). The cyclic peptide 16cyc-HxA showed better cellular uptake than the other tested peptides. In addition, all of the peptides had better cell-penetrating ability in PA-1 cells over any of the other cell lines. The mechanisms of peptide cellular uptake are a focused research area but still far from been fully elucidated (33). Synergetic contribution from peptides’ secondary structure, amphiphilicity, and charges correlates with peptides’ cellular uptake (34). Different types of cells with different membrane compositions and different expressed receptors made the question more elusive. Intracellular fluorescence at different periods of time indicated the peptide 16cyc-HxA had good cellular stability shown in Supplementary Fig. S9A–S9C.

As cell penetration is a prerequisite for cellular functions, we chose the cell-penetrating peptides 16cyc-HxA, 16lin-HxA, and 16KA to further study their cellular functions. We chose pluripotent human ovarian teratocarcinoma PA-1 cells and testicular embryonic carcinoma NTERA-2 cells as they are well-accepted models for studying cancer cell stemness and are highly sensitive to HDAC1 (15, 35–37). The approved drug SAHA was chosen as a positive control. As shown in Fig. 2C, in cancer stem–like cells, cyclic peptide 16cyc-HxA showed a significant antitumor effect, with an IC₅₀ of 14.65 μmol/L in PA-1 cells and an IC₅₀ of 62 μmol/L in NTERA-2 cells, better than 16lin-HxA. Notably, peptide 16KA did not have a significant effect on these cells. Compared with the two scrambled peptides Scr-1 and Scr-2, 16cyc-HxA showed better antitumor effects to PA-1 cells; nevertheless, this peptide showed weaker cellular uptake than Scr-2, ruling out the possibility that the peptide's high efficiency was simply caused by increased cellular uptakes as shown in Supplementary Fig. S10A–S10C. In the other two cancer cell lines, A549 cells and HeLa cells, peptide 16cyc-HxA displayed a decreased effect, partly because they were reported to be less sensitive to HDACs (15, 36). The negligible effect on the two normal cell lines indicated the low nonspecific toxicity of these peptides. Notably, SAHA exhibited potent toxicity across all cell lines, including the normal ones (Fig. 2D). To further study the HDAC isoform-related cytotoxicity, the class I selective small inhibitors MS275 and HDAC6 and selective small inhibitors Tubacin were selected as positive controls shown in Supplementary Fig. S11A–S11C. In PA-1 and NTERA-2 cells, both SAHA and MS275 had potent antitumor effect, while Tubacin had much weaker antitumor effects, indicating the importance of class I HDACs in cancer stem–like cells. Meanwhile, these small inhibitors all showed similar toxicity to normal cells. The nonspecific toxicity of these peptide inhibitors was further excluded using the LDH release assay in PA-1 cells (Supplementary Fig. S12A and S12B) and the hemolysis assay (Supplementary Fig. S13A and S13B). In summary, peptide 16cyc-HxA exhibited preferential inhibition on cancer stem–like cells, PA-1 cells or NTERA-2 cells, which was sensitive to HDAC1 inactivation. Importantly, the nonspecific toxicity of peptide inhibitors suggested their superiority based on a wider safety window over SAHA.

To investigate the association between HDAC inactivation and antiproliferation in cancer stem–like cells, we used an in vitro streptavidin pulldown assay to test whether our designed peptides could bind to HDACs endogenously expressed in PA-1 cells (Fig. 3A; Supplementary Fig. S14A). We first tested the expression levels of these HDAC isoforms in PA-1 cells as shown in Supplementary Fig. S14A. HDAC1 showed much higher expression than other isoforms. Streptavidin pulldown assays indicated that peptide 16cyc-HxA and 16lin-HxA could bind to class I HDACs (HDAC1, -2, -3, -8) and class II HDACs (HDAC4, -5, -6), but minimal binding to class III (SIRT 1 and SIRT 6) shown in Supplementary Fig. S14B, which was also supported by SIRT1 inhibition experiments shown in Supplementary Fig. S2A and S2B. The trapping of HDAC1 and HDAC6 depended on the concentration of the peptides shown in Supplementary Fig. S14C and S14D.

We subsequently detected representative acetylation changes of HDAC1-specific substrates (H3K56 and H4K16) and an HDAC6-specific substrate (α-tubulin) in PA-1, NTERA-2, and A549 cells (Fig. 3B; Supplementary Fig. S15A and S15B). In these cell lines, peptides 16cyc-HxA and 16lin-HxA significantly increased the acetylation of H3K56, H4K16, and α-tubulin in a dose-dependent way, indicating the cellular inactivation of HDAC1 and HDAC6 without downregulating their expression indicated in Supplementary Fig. S15A and S15B. Meanwhile, peptides 16cyc-HxA and 16lin-HxA showed irreversible HDAC inhibition in PA-1 cells for 48-hour treatment (Supplementary Fig. S16A and S16B). The positive controls SAHA and MS275 showed more obvious effect on the acetylation of H3K56 and H4K16, and Tubacin showed preference for the acetylation of tubulin shown in Supplementary Fig. S17A and S17B, which further suggested the inactivation of HDAC1 and HDAC6 of 16cyc-HxA.

Most of HDACs performed their distinct functions by forming multiple protein complexes, among which, CoREST complex containing HDAC1, LSD1, and CoREST are reported to play a key role in cancer stem–like cells and have attracted special interests (15, 38). In this study, we found that 16cyc-HxA could upregulate the expression of P21 and P53, significantly stimulate mono- and dimethylation of H3K4 in PA-1 cells, suggesting the inactivation of LSD1 (Fig. 3C; Supplementary Fig. S18). To further study the peptide influence on the LSD1-CoREST-HDAC1 complex, CO-IP assays were performed and demonstrated that peptide 16cyc-HxA could disrupt their interactions as shown in Fig. 3D. Meanwhile, the peptide inhibitors could alter the recruitment of HDAC1 or LSD1 to Sox2 promoters by the ChIP experiment shown in Supplementary Fig. S19A and S19B. We also found that inactivation of HDACs could downregulate the expression of the pluripotent stem cell marker Sox2 (Fig. 3C and E), verifying that HDAC inactivation could induce cell differentiation of cancer stem–like cells.

Loss of HDACs can cause cell differentiation and induce the expression of genes associated with cell cycle, such as P21 (39). Significant downregulation of the stemness gene Sox2 and upregulation of the differentiation gene FoxA2 are convincing evidence of CSC differentiation (15). To further investigate whether our designed peptides could influence related gene expression, we monitored the expression of the genes Sox2, FoxA2, and P21 by quantitative RT-PCR (Fig. 3E). Compared with SAHA and 16lin-HxA, peptide 16cyc-HxA could significantly activate the expression of the FoxA2 and P21 genes and downregulate the expression of the Sox2 gene in both PA-1 cells and NTERA-2 cells. Together, these data indicated that deactivation of HDACs by our designed peptides could interrupt the LSD1-CoREST-HDAC1 complex and might induce the cancer stem–like cells to differentiation.

The aberrance of HDACs is reported to be associated with cellular functions such as cell apoptosis and cell-cycle arrest (40). To verify if our designed peptides could induce cell apoptosis, Annexin V/PI assay was performed as shown in Fig. 4A and Supplementary Fig. S20A and S20B. The peptide 16cyc-HxA had greater apoptotic effect than 16lin-HxA in PA-1 and NTERA-2 cells. To investigate whether the apoptotic effect was mediated by a caspase-3-dependent pathway, PA-1 cells or NTERA-2 cells were treated with peptide inhibitors and finally analyzed using a caspase-3 assay kit (Fig. 4B) or Western blot assays (Supplementary Fig. S21). The activation of caspase-3 demonstrated that peptide inhibitors could induce the caspase-3-dependent apoptosis pathway.

HDAC inhibition has been shown to have antiproliferative effects through the induction of cell-cycle arrest in G₁ or G₂ (40). To study the alteration of cell-cycle distribution of our designed peptides, FACS assays were performed as shown in Fig. 4C and Supplementary Fig. S22. The peptide 16cyc-HxA displayed a more appreciable arrest of cell-cycle progression in G₂–M phase than peptide 16lin-HxA. The 16KA-treated group did not show obvious cell-cycle arrest compared with the control group. The cell-cycle results were further confirmed by Western blot assays against cell-cycle marker cyclin A, cyclin B1, P21, and H3PS10 shown in Supplementary Fig. S23A and S23B. Our designed peptides could upregulate the expression of P21, cyclin B1, and H3PS10, but had little effect on the expression of cyclin A, which validated the G₂–M cell-cycle arrest (41, 42).

To attain an overview of the genes directly affected by the inactivation of HDACs, genome-wide mRNA microarrays were performed, as shown in Fig. 5 and Supplementary Figs. S24 and S25A–S25C. Compared with DMSO-treated cells, treatment with 16cyc-HxA yielded detection of a total of 2,255 upregulated and 786 downregulated genes (fold change >2), which affected more genes than treatment with peptide 16lin-HxA or SAHA (Fig. 5A and B). The gene ontology analysis revealed that the affected genes are associated with many important signaling pathways relative to widely concerning diseases, such as metabolic diseases, cancer, immune diseases, neurodegenerative diseases, and so on (Fig. 5C; Supplementary Figs. S24 and S25A–S25C). In addition, there are 330 cancer-related genes affected by HDAC inhibition, such as CDKN1A, CDKN1B, DNMT3A, EGFR, NFKB1, NFKBIA, PIK3, WNT, BMP4, TP53, MAPK10, NOTCH4, and so on (Supplementary Table S4). These results were further confirmed by RT-PCR assays and Western blot assays shown in Supplementary Figs. S26 and S27. In summary, all these results demonstrated that inactivation of HDACs had a broad effect on the expression of most genes at the same time, some of which were associated with tumorigenesis.

To investigate the potential tumor inhibition effect of peptide inhibitors in the PA-1 and NTERA-2 xenograft animal models, we first assessed peptide biocompatibility using the voluntary cage-wheel exercise assay. Over a period of 21 days following injection, voluntary running cycles increased steadily with no significant differences between these five groups, indicating no obvious effects on the motor learning ability of mice (Supplementary Fig. S28A and S28B). Additionally, the body weight of these five groups was not influenced during the 21-day duration of treatment. These findings revealed that the peptides were well tolerated in mice.

Subsequently, we examined the tumor inhibition of these peptides in the PA-1 and NTERA-2 xenograft animal models in nude mice (Fig. 6A–D; Supplementary Fig. S29A–S29D). Tumor-bearing mice were treated with peptides 16cyc-HxA, 16lin-HxA, 16KA, SAHA, or vehicle (5 mice/group, 50 mg/kg, every other day, intraperitoneal injection) for a duration of 3 weeks. We chose this route of administration as it is a common practice for peptide therapeutics referred to previous reports (25, 43). In the PA-1 and NTERA-2 xenograft animal model, in contrast to the vehicle-treated group, peptide 16cyc-HxA resulted in significant inhibition of tumor, at a rate of over 80% (Fig. 6A–C; Supplementary Fig. S29A–S29D). Although the linear peptide was not as stable as the cyclic peptides in the serum stability assay (Supplementary Fig. S30), it also showed an 80% reduction in tumor volume. This phenomenon could most likely be explained by the high dosage of administration (50 mg/kg). The negligible antitumor effect of 16KA excluded the nonspecific toxicity of peptides. The weaker in vivo effect of SAHA may be associated with its poor aqueous solubility and low bioavailability by intraperitoneal injection, which may also be the cause of failure to the clinical treatment of solid tumor (44, 45). These peptide inhibitors were also well tolerated in mice based on data for stable body weight (Fig. 6D) accompanied by the lack of any signs of organ lesions by H&E staining (Supplementary Fig. S31A–S31E). The two animal models of cancer stem–like cells provide strong evidence that our peptide inhibitors had great potential for cancer therapy.

To further investigate the peptide distribution in vivo, tumor-bearing mice were intratumoral injected with Cy5-labeled 16cyc-HxA and 16lin-HxA followed by imaging at different time points using an ex/in vivo imaging system (Supplementary Fig. S30A–S30G). We chose this route of administration as the peptide inhibitors were not labeled with tumor targeting ligands. The optical signal intensity of the Cy5-labeled 16cyc-HxA did not show any diminishment for 24 hours, while the Cy5-labeled 16lin-HxA displayed decreasing signal intensity after 24-hour treatment, suggesting the cyclic peptide was more stable than the linear peptide for in vivo tumor therapy.

To detect in vivo HDAC inactivation after the 21-day treatment, mice were sacrificed, and tumors were carefully removed for IHC assay (Supplementary Fig. S32A–S32E). When compared with the SAHA-treated group, peptide inhibitors showed more obvious acetylation levels of H3K56 and H4K16, and higher activation of caspase-3, which is consistent with cellular results. Meanwhile, the tumors from mice treated with the two potent peptide inhibitors displayed less cell density than those of the other groups, further confirming the potent anticancer effect. To further check the proliferative index of these tumor tissues, the Ki-67 staining assay was performed as shown in Supplementary Fig. S33, and the peptides (16cyc-HxA and 16lin-HxA) treated tumor tissue had a lower proliferative index than the PBS-treated group, indicating their potent antitumor effects. In summary, our designed peptides showed superior antitumor activities and good biocompatibility in the PA-1 and NTERA-2 xenograft animal model, which is consistent with the cellular results.

Developing HDAC inhibitors to efficiently target cancer stem–like cells in combination with other drugs could have great potential to exterminate cancer (46). Recently, HDAC inhibitors have been shown to enhance the sensitivity of pancreatic ductal adenocarcinoma (PDAC) cells to gemcitabine and have the potential to overcome the tumor relapse (47). These studies demonstrate a place for HDACs as promising targets for cancer resistance and elimination.

In the current study, we have developed a new class of stabilized peptide HDAC inhibitors to reduce the cellular nonspecific toxicity of conventional HDAC inhibitors. The helical stabilization method is facile and improves the stability and cellular uptake of these peptides. Unlike the traditional small HDAC inhibitors, the designed peptides showed preferential inhibition of cancer stem–like cells (PA-1 and NTERA-2 cells) and malignant tumor (A549 cells) cells with negligible effects on normal cell lines (293T and Chang liver cells; Fig. 2). HDACs especially HDAC1 were reported to play an important role in regulating stem cells and cancer stem–like cells, and the inactivation of HDACs could decrease the expression of stemness gene Sox2, upregulate differential genes and induce cell death. Although cancer cells and normal cells, such as HeLa cells and 293T cells, were not sensitive to HDAC1 inactivation, peptide inhibitors showed little toxicity toward HeLa cells. Although peptide 16cyc-HxA showed higher IC₅₀ than SAHA itself, it significantly attenuates the nonspecific toxicity of SAHA and increases the therapeutic safety window. This study clearly showed the promising potential of developing peptide hybrid for HDAC inhibition. We believe the peptide hybrid efficacy could be further enhanced by the discovery of more potent HDAC-binding peptides. Our findings also indicated that the cyclic peptides could inactivate HDACs and disrupt the LSD1-CoREST-HDAC1 interactions (Fig. 3; Supplementary Figs. S16A–S16B and S19A–S19B), which are reported to play a key role in cancer stem–like cells. Moreover, these peptides can also alter the properties of cancer stem–like cells by downregulating the stemness gene Sox2, upregulating the differential gene FOXA2 and inducing caspase-3-dependent apoptosis and cell-cycle arrest (Figs. 3 and 4). Besides, our designed peptides could influence the proliferation of cancer stem–like cells in vivo with negligible organ lesions (Fig. 6). All these results indicated the great potential of our designed peptides for cancer therapy.

This proof-of-principle study has clearly shown the tremendous opportunity for the development of peptide/small-molecule hybrids to improve the safety window of these therapies, an application that could be applied for other important targets with similar needs. We also believe that with further understanding of the structurally based interaction between HDACs and their specific substrates, establishment of isoform-selective HDAC inhibitors could be realized.

No potential conflicts of interest were disclosed.

Conception and design: D. Wang, W. Liang, F. Yin, Z. Li

Development of methodology: D. Wang, W. Liang, F. Yin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Wang, W. Li, R. Zhao, L. Chen, N. Liu, M. Xie, Q. Fang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Wang, Y. Tian, H. Zhao, F. Lu, W. Liang, F. Yin, Z. Li

Writing, review, and/or revision of the manuscript: D. Wang, W. Liang, F. Yin, Z. Li

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Wang, W. Liang, F. Yin, Z. Li

Study supervision: W. Liang, F. Yin, Z. Li

We acknowledge financial support from the Natural Science Foundation of China grants 21778009 (Z. Li), 81701818 (F. Yin), and 81572198 (Z. Li COPled); the Shenzhen Science and Technology Innovation Committee, JCYJ20170412150609690 (F. Yin), KQJSCX20170728101942700 (Z. Li), and JCYJ20170807144449135 (F. Yin COPled). This work is supported by the High-Performance Computing Platform of Peking University.

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.