Harmine, a dual-specificity tyrosine phosphorylation-regulated kinase (DYRK) inhibitor induces caspase-mediated apoptosis in neuroblastoma

Harmine (Fig. 1), LDN-192960, and INDY were purchased from Cayman Chemical. The compound solids were stored at − 20 °C. Stock solutions were prepared by dissolving harmine (100 mM), LDN-192960 (40 mM) and INDY (40 mM) into sterile DMSO (VWR). Stock concentrations were filtered prior to being added to the cell cultures.

The human NB cell lines KELLY (also known as N206; #92110411 from Sigma), SK-N-AS (called SKNAS in this study; CRL-213 from ATCC), SK-N-BE (clone SKNBE(2)C, called SKNBE in this study; CRL-2268 from ATCC), and SK-N-FI (called SKNFI in this study; from the Children’s Oncology Group) were cultured in RPMI 1640 (VWR), supplemented with 10% heat-inactivated fetal bovine serum (Invitrogen), Penicillin (100 IU/mL) and Streptomycin (100 μg/mL) (30-002-CI, Corning). All cells were purchased from their respective suppliers within the last 2 years. The cells were maintained at 37 °C in a humidified atmosphere containing 5% CO₂.

Cell viability and IC₅₀ was determined using the RealTime-Glo MT cell viability assay (G9712, Promega). This reagent allows continuous measurement of cell viability in the same well. Cells were plated at a density of 4000 cells/well into white-walled, opaque assay plates. After the plated cells had been given 24 h to adhere, they were treated with harmine concentrations ranging from 0 to 1 mM. The MT Cell Viability Substrate and NanoLuc Enzyme were equilibrated to 37 °C, 2× RealTime-Glo reagent was prepared, and an equal volume was added to each well. For time zero measurements, cells were incubated with reagent for 20 min at 37 °C, and luminescence was measured on a Biotek Synergy microplate reader. Luminescence was measured at 24, 48, and 72 h after addition of harmine.

The colorimetric SRB assay was used to measure cytotoxicity as previously described, following the treatment with DYRK2 inhibitor LDN-192960 or DYRK1A/B inhibitor INDY [15‐17]. Briefly, NB cells were plated in transparent flat 96-well plates and allowed to attach overnight. At the initiation of each experiment (t = 0) and after drug treatments, cells were fixed with 10% TCA at 4 °C for 1 h, washed with deionized water, and dried at room temperature. Cells were then stained with 100 μl of 0.4% SRB in 1% acetic acid for 20 min at room temperature, rinsed five times with 1% acetic acid and allowed to dry at room temperature. One hundred µl of 10 mM Tris–HCl pH 7.0 was added to each well, shaken for 10 min at room temperature and read at 540 nm using a Molecular Devices Flexstation 3 microplate reader.

The quantification of the relative caspase activity in harmine-treated cells was carried out using the Caspase-Glo 3/7 and Caspase-Glo 9 Assay kits (G8091 and G8211, respectively, from Promega). Cell lines were plated at a density of 16,000 cells/well into white-walled, opaque 96-well plates. Twenty-four hours after plating, cells were treated with 0, 25, 50 and 100 µM harmine for 24 h. Caspase Glo reagents were added to the cells and incubated at room temperature. Luminescence was measured using a Biotek Synergy microplate reader every 20 min for 3 h.

Whole cell lysates were prepared using radioimmunoprecipitation assay (RIPA) buffer, (20 mM Tris–HCl [pH 7.5], 135 mM NaCl, 2 mM EDTA, 0.1% (w/v) sodium lauryl sulfate, 10% (v/v) glycerol, 0.5% (w/v) sodium deoxycholate, and 1% (v/v) Triton X-100). The RIPA buffer was supplemented with cOmplete™ Protease Inhibitor Cocktail (Roche), and 0.27 mM Na₃VO₄ and 20 mM NaF as phosphatase inhibitors. Protein concentration was determined using the Bradford dye reagent protein assay (Bio-Rad). Equal amounts of protein were resolved using 12% SDS-PAGE, and transferred to 0.45 µM polyvinylidene difluoride Immobilon-P membrane (Millipore). The transferred proteins were incubated with the following primary antibodies: PARP rabbit polyclonal antibody (#9542) from Cell Signaling Technologies at 1:1000 dilution and GAPDH mouse monoclonal antibody (#SC-47724) from Santa Cruz Biotechnology at 1:1000 dilution. After incubating the blots with the primary antibody at 4 °C for at least 8 h, they were washed and incubated with secondary antibodies at 1:10,000 dilution for 1 h at room temperature. The secondary antibodies used were LiCor IRDye^® 680RD Goat anti-Mouse IgG, (925-68070), and LiCor IRDye^® 680RD Goat anti-Rabbit IgG, (925-68710).

To quantify the percentage (%) of apoptosis induction in harmine-treated NB cells, the Annexin V Detection Kit APC (88-8007, Invitrogen) assay was used. Samples were prepared according to the manufacturer’s protocol and analyzed by flow cytometry. In brief, the cells were plated overnight and then treated with 0 μM or 100 μM harmine for 24 h at 37 °C. Cells were collected by centrifugation and stained with Annexin V APC/DAPI staining solution at room temperature for 20 min in the dark. Next, cells were collected by centrifugation suspended in PBS (pH 7.4) and analyzed immediately using flow cytometry Excitation/emission for Annexin APC and DAPI were 633/700 and 350/450 nm respectively.

The molecular docking model for DYRK2 in complex with harmine was constructed by identifying the target structures from the Protein Data Bank (PDB) [18]. The crystal structure of human DYRK1A (PDB ID: 3ANR) in complex with harmine was previously published [13, 18]. To predict if DYRK2 also forms a complex with harmine, we generated a molecular docking model with coordinates from an existing crystal structure of DYRK2 in complex with Leucettine (PDB:4AZF) using the online docking web server SwissDock (http://​www.​swissdock.​ch/​) [18‐21]. Chain A of 4AZF was isolated from the original crystal structure, and the Leucettine L41 ligand was removed. The modified structure was then entered into SwissDock, along with the chemical structure of harmine (ZINC ID: 27646846) taken from the ZINC molecule database (http://​zinc.​docking.​org/​) [19, 20]. The conformational ΔG was calculated and deemed best fit by the parameters set by SwissDock. The resulting molecular interaction model was visualized using visual molecular dynamics (VMD) [22].

Alignment of the DYRK family sequences was carried out using the Blastp tool provided by the National Center for Biotechnology Information (NCBI) [23, 24]. The amino acid sequences for human DYRK1A and human DYRK2 used for the alignment were taken from the Basic Local Alignment Search Tool (BLAST) database, (Accession No. Q13627 and AAH06375, respectively) (https://​blast.​ncbi.​nlm.​nih.​gov/​Blast.​cgi) [18, 24]. The query cover reported herein was given as part of the output of the Blastp program, as was the percentage of identity and the corresponding E-value.

For analysis of DYRK gene expression in human NB patients, the largest public NB cohort for which genome-wide tumor RNA-sequencing has been performed, SEQC-498, (n = 498; GSE62564) was analyzed using the R2 genomics analysis and visualization platform developed in the Department of Oncogenomics at the Academic Medical Center—University of Amsterdam (http://​r2.​amc.​nl). Expression data for the datasets were retrieved from the public Gene Expression Omnibus (GEO) dataset on the NCBI website (http://​www.​ncbi.​nlm.​nih.​gov/​geo/​) and analyzed as previously described [25].

The statistical significance for the cell viability measurements (Fig. 2b) was calculated using the GraphPad Prism 7 package (https://​www.​graphpad.​com/​) and an unpaired student’s t-test assuming the null hypothesis was performed. The change in caspase activity was quantified by calculating the average fold change, after normalizing experimental samples to their respective controls. The significance of both caspase activation and Annexin V apoptotic measurements (Figs. 3, 4) was calculated using an unpaired student’s t-test, assuming the null hypothesis. DYRK gene NB tumor mRNA expression correlation with survival probability (Fig. 7) was evaluated by Kaplan–Meier analysis using the log-rank test as described [26]. To determine the optimal value of gene expression to set as cutoff value, all tumor samples were first sorted according to gene mRNA expression and subsequently divided into two groups. Analyses were performed on groups separated by median or average tumor mRNA expression values. DYRK mRNA expression correlation with tumor MYCN gene amplification was determined using the non-parametric, rank-based Kruskal–Wallis test. For all tests, a P-value < 0.05 was considered to be statistically significant.

To study the effect of harmine on the morphology of NB cells, four NB cell lines were exposed to 100 μM harmine. As early as 24 h after treatment, the plated cells began to show signs of morphological changes associated with apoptosis (Additional file 1: Fig. S1). In comparison to control cells, treated cells were smaller in size, rounded in shape, and more detached from the plate surface. After 72 h, the majority of treated cells had become spherical in shape and had detached (Fig. 2a). To determine if harmine induces dose- and time-dependent cell death, the four NB cell lines were treated with increasing drug concentrations (0, 6, 12.5, 25, 50, and 100 μM) and cell viability was measured after 24, 48 or 72 h of treatment (Fig. 2b). Increased treatment length with harmine caused decreased IC₅₀ in all four cell lines. The IC₅₀ values for SKNBE, KELLY, and SKNFI after a 72 h treatment of harmine were 169.6 ± 0.10, 170.8 ± 0.10, and 791.7 ± 0.77 μM, respectively. The IC₅₀ value of SKNAS after 72 h could not be calculated (for all IC₅₀ values including 24 and 48 h time points; see Additional file 1: Table S1). The IC₅₀ decrease was consistent with the morphological changes observed in the treated cells (Fig. 2a). Moreover, the IC₅₀ values suggest that harmine is more toxic to MYCN-amplified NB cell lines (SKNBE and KELLY), than to NB cell lines with a normal MYCN gene copy number (SKNAS and SKNFI).

The dose- and time-dependent cell death and associated morphological changes in response to harmine treatment prompted us to determine if harmine induces caspases, known to be activated during apoptosis. NB cells were treated with 0, 50, and 100 μM of harmine for 24 h, after which caspase activity was measured using the Promega Caspase-Glo 3/7 and Caspase-Glo 9 Assay kits. Caspase-3/7 activity increased with increasing harmine concentration (Fig. 3a) and this was significant for all four cell lines at the highest concentration (100 μM). The average fold changes for the caspase-3/7 activity were: SKNBE = 4.32 (P < 1.0 × 10⁻⁴), KELLY = 8.26 (P < 1.0 × 10⁻⁴), SKNAS = 6.98 (P < 1.0 × 10⁻⁴), and SKNFI = 6.77 (P < 1.0 × 10⁻⁴). Under identical cell treatment conditions, a significant increase in caspase-9 activation was observed in all four cell lines (Fig. 3b). The average fold changes for caspase-9 activity were: SKNBE = 2.51 (P < 1.0 × 10⁻⁴), KELLY = 2.96 (P < 1.0 × 10⁻⁴), SKNAS = 4.18 (P < 1.0 × 10⁻⁴), and SKNFI = 3.86 (P < 1.0 × 10⁻⁴). The results suggest that harmine triggers apoptotic cell death in NB cells.

To confirm that caspase activation leads to progressive apoptosis, harmine-treated NB cells were analyzed for caspase-mediated PARP cleavage. As shown in Fig. 4, cleaved PARP appeared in whole cell lysates of harmine-treated SKNBE and KELLY cells, but not SKNAS or SKNFI cells. These results confirm that harmine induces apoptosis, with a profound effect on MYCN-amplified NB cells (SKNBE and KELLY); both cell lines representing the most aggressive sub-types of NB tumors.

To further confirm actual apoptosis in the harmine-treated NB cells, the amount of apoptotic cells in four NB cell lines was quantified using Annexin V staining and flow cytometry (Fig. 5a). The total amount of apoptotic cells in the SKNBE cells significantly increased from 12.5% ± 0.60 (control) to 36.8% ± 14.5 in cells that had been treated with harmine (100 μM) (Fig. 5b). Similarly, harmine significantly increased the total amount of apoptotic cells in KELLY cells from 11.3% ± 2.31 (control) to 45.3% ± 6.96 in the presence of harmine (100 μM). The change in apoptotic cells in SKNAS cells was less dramatic and increased from 6.20% ± 2.24 (control) to 16.6% ± 5.31 (100 μM) and there was no change in the percentage of apoptotic cells in SKNFI cells (Fig. 5b). The results further confirm our observation that harmine appears more toxic to MYCN-amplified NB cells (SKNBE, KELLY) than to NB cells with a normal MYCN gene copy number (SKNAS, SKNFI).

Based on literature supporting the potential interaction between harmine and DYRK2, we performed a docking simulation using SwissDock [19, 20]. To accomplish this, we utilized the crystal structures of human DYRK1A in complex with harmine (PDB ID: 3ANR) (Fig. 6a) and human DYRK2 in complex with Leucettine (PDB:4AZF) [13, 21]. Chain A of human DYRK2 was isolated and was entered into the SwissDock portal with harmine as the ligand, after removing Leucettine. The structure of harmine was provided by the ZINC molecular database (ZINC ID: 27646846). The most favorable conformation of the DYRK2/harmine complex had a ΔG value of − 6.89 kJ mol⁻¹ (Fig. 6a). The figure displays the highest affinity binding between human DYRK2 and harmine. The crystal structure of chain A of DYRK1A in complex with harmine is provided for comparison (Fig. 6b). The two proteins were also compared by performing a sequence alignment between human DYRK1A (Accession No. Q13627) and human DYRK2 (Accession No. AAH06375) using the NCBI Blastp online tool (Fig. 6c) [24]. The query identity was 41%, with an E-value of 4 × 10⁻⁹².

Our in vitro data in NB cell lines were supported by crystal structure-derived computational docking models and suggest that targeting DYRK family kinases through harmine treatment may be a potential new route of therapy for patients with NB. This notion prompted us to investigate DYRK family kinases in human NB tumors. To accomplish this, we analyzed the mRNA expression of the DYRK gene family in SEQC-498, the largest RNASeq dataset on human NB samples in the public domain. Analyses were performed using the R2 website (see Materials and Methods). We first determined whether DYRK genes are expressed in human NB tumors (Fig. 7a). All five DYRK genes show robust expression, especially DYRK1A and DYRK2.

We next determined whether high DYRK gene expression is beneficial to NB tumorigenesis and translates into poor patient outcome (Fig. 7b–f). Kaplan–Meier graphs representing the overall survival prognosis of NB patients in the SEQC-498 cohort based on grouping of the patients according to median DYRK gene tumor mRNA expression showed that high-level tumor mRNA expression of DYRK1B, DYRK2, and DYRK3, but not of DYRK1A or DYRK4, is significantly predictive of poor patient outcome (Fig. 7b–f). Although DYRK1A and DYRK2 are the most highly expressed, only DYRK2 is prognostic of survival outcome, suggesting that in NB the deleterious effect of harmine treatment acts mostly through DYRK2. Harmine toxicity tests and apoptosis analyses on NB cell lines suggest that MYCN-amplified cell lines are more sensitive to harmine than NB cell lines with two MYCN gene copies. We were interested whether this was reflected in the correlation between DYRK gene mRNA expression and MYCN tumor amplification in human NB tumors. We found that DYRK2 and DYRK3 mRNA expression in NB tumors was significantly higher in samples with MYCN gene amplification (Fig. 8). In contrast, DYRK1B and DYRK4 tumor expression was not correlated to MYCN amplification status, and DYRK1A was actually lower in tumors with MYCN amplification. These results might indicate that the most aggressive, MYCN-amplified NB tumors that have the highest DYRK2 and DYRK3 expression could be more sensitive to harmine treatment, similar to the results we found in NB cell lines. These expression profiles along with the harmine-DYRK2 interaction model would suggest that the cytotoxic effect of harmine in NB is primarily through DYRK2 inhibition.