Modulation of phospho-proteins by interferon-alpha and valproic acid in acute myeloid leukemia

Purpose Valproic acid (VPA) is suggested to be therapeutically beneficial in combination with interferon-alpha (IFNα) in various cancers. Therefore, we examined IFNα and VPA alone and in combinations in selected AML models, examining immune regulators and intracellular signaling mechanisms involved in phospho-proteomics. Methods The anti-leukemic effects of IFNα and VPA were examined in vitro and in vivo. We mapped the in vitro phosphoprotein modulation by IFNα-2b and human IFNα-Le in MOLM-13 cells by IMAC/2D DIGE/MS analysis and phospho-flow cytometry, and in primary healthy and AML patient-derived PBMCs by CyTOF. In vivo, IFNα-Le and VPA efficacy were investigated in the immunodeficient NOD/Scid IL2γ−/− MOLM-13Luc+ mouse model and the syngeneic immunocompetent BNML rat model. Results IFNα-2b and IFNα-Le differed in the modulation of phospho-proteins involved in protein folding, cell stress, cell death and p-STAT6 Y641, whereas VPA and IFNα-Le shared signaling pathways involving phosphorylation of Akt (T308), ERK1/2 (T202/T204), p38 (T180/Y182), and p53 (S15). Both IFNα compounds induced apoptosis synergistically with VPA in vitro. However, in vivo, VPA monotherapy increased survival, but no benefit was observed by IFNα-Le treatment. CyTOF analysis of primary human PBMCs indicated that lack of immune-cell activation could be a reason for the absence of response to IFNα in the animal models investigated. Conclusions IFNα-2b and IFNα-Le showed potent and synergistic anti-leukemic effects with VPA in vitro but not in leukemic mouse and rat models in vivo. The absence of IFNα immune activation in lymphocyte subsets may potentially explain the limited in vivo anti-leukemic effect of IFNα-monotherapy in AML. Electronic supplementary material The online version of this article (10.1007/s00432-019-02931-1) contains supplementary material, which is available to authorized users.


Background
Acute myeloid leukemia (AML) is a heterogeneous aggressive blood cancer characterized by a block in differentiation, elevated threshold for undergoing apoptosis and excessive proliferation of myeloid progenitor cells (Dohner et al. 2017). Median age of diagnosis is approximately 70 years (Juliusson et al. 2009), and 5 year overall survival is only 5% for patients older than 65 years (Visser et al. 2012), underscoring the need of more effective therapy with acceptable toxicity. IFNα has been tested in AML as induction therapy (Berneman et al. 2010), as a post-remission strategy to prevent recurrence after chemotherapy (Goldstone et al. 2001), in consolidation with allogeneic hematopoietic stem cell transplantation (HSCT) (Klingemann et al. 1991), and as a salvage therapy for patients relapsing upon allogeneic HSCT (Arellano et al. 2007). Case reports showing complete remission after IFNα monotherapy in secondary AML following essential thrombocytosis and/or myelofibrosis may indicate that subsets of patients are particularly sensitive to IFNα (Berneman et al. 2010;Dagorne et al. 2013). IFNα also seems to be effective to prevent relapse in minimal residual disease (MRD) positive patients after HSCT (Mo et al. 2015), while no effect has been reported in children's relapsed/refractory leukemia (Ochs et al. 1986). Several formulations of therapeutic IFNα have been available for clinical use. In addition to the most used recombinant IFNα-2b, a human purified preparation of IFNα consisting of six different subtypes (IFNα-Le) has been shown beneficial in melanoma (Stadler et al. 2006). IFNα-Le consists of IFNα1, -α2, -α8, -α10, -α14 and -α21, whereof IFNα2 and IFNα14 are glycosylated. Intriguingly, the IFNα-induced molecular phospho-signaling response has not systematically been characterized in cancer cells, and the anti-leukemic effect of IFNα-Le has previously never been compared with recombinant IFNα-2b.
The combination of IFNα with the histone deacetylase inhibitor valproic acid (VPA) has been reported to be synergistic in several solid cancer models (Jones et al. 2009;Iwahashi et al. 2011;Hudak et al. 2012), suggesting that this combination could represent a valuable novel therapeutic strategy in AML. VPA is an anticonvulsant also used in bipolar disease with well-characterized side effects. Its anti-leukemic effect has been examined in combination with all-trans retinoic acid (ATRA) (Trus et al. 2005), 5-azacytidine or low dose cytarabine with responses in up to 20% of the AML patients (Kuendgen et al. 2006;Raffoux et al. 2010;Corsetti et al. 2011;Fredly et al. 2013).
In this study, we compared recombinant and purified human IFNα formulations and found specific regulation of signaling pathways. The combination of IFNα with VPA was synergistic in vitro, but even though in vivo experiments supported the anti-leukemic effect of VPA, we did not find a beneficial effect of IFNα or the combination of IFNα and VPA in vivo.

Staining of primary cells for mass cytometry (CyTOF)
To assure comparability across samples from different donors, the samples were barcoded using a commercially available metal barcoding kit (Fluidigm), according to the manufacturer's instructions. Twenty samples were multiplexed, making a total of four pooled samples, each containing AML patient and healthy donor samples.      The samples were subsequently stained with the antibody panels (Online Resource Tables 1 and 2) following the Max-Par phospho-protein staining protocol (Fluidigm), with minor adjustments. Briefly, amendments to the protocol include; Fc receptors blocking was done using Human IgG (Octagam ® , Octapharma, Lachen, Switzerland) for 20 min at RT. To block nonspecific antibody binding to eosinophils (Rahman et al. 2016), samples were pre-incubated with 100 IU heparin sodium (Wockhardt, Wrexham, UK) for 20 min, and subsequently stained with antibody cocktails (Online Resource Tables 1 and 2) in the presence of 100iU heparin. DNA intercalation stain (iridium, Fluidigm) was diluted at 1:1250 in 4% PFA in Maxpar PBS, and samples were incubated at 4 °C overnight. After staining, samples were resuspended in a 1:8 solution of Maxpar cell acquisition solution (Fluidigm) and EQ™ four element calibration beads (Fluidigm).
Acquisition of samples was done using a Helios mass cytometer (Fluidigm). After acquisition, the collected data was normalized to EQ bead standard (Finck et al. 2013) and exported to FCS3 files. Data was subsequently uploaded to Cytobank Cellmass software (Cytobank Inc, Santa Clara, CA, USA) and evaluated using established methods (Diggins et al. 2015;Levine et al. 2015). Phenograph was run with the Cyt interphase in Matlab (Mathworks, Natick, MA, USA), see Online Resource Fig. 1 for gating strategy. The non-parametric Kruskal-Wallis H test was used to determine statistical significance (p < 0.05) using R software (R Core Team (2017), R Foundation for Statistical Computing, Vienna, Austria).

Phospho-flow cytometry staining and analysis
Treated MOLM-13 cells were washed in 0.9% NaCl, fixated in 1.6% PFA for 15 min at room temperature (RT), added ice-cold methanol and stored at − 80 °C prior to analysis. Samples were fluorescently barcoded using Pacific Blue and Pacific Orange (Molecular Probe, Eugene, OR, USA), as described previously (Krutzik and Nolan 2006). The Student's unpaired, two-tailed t test (GraphPad, GraphPad Software, Inc., La Jolla, CA, USA) was used to determine statistical significance (p < 0.05). Primary antibodies are described in Online Resource Table 3.

IMAC phosphoprotein purification, two-dimensional differential gel electrophoresis, gel analysis and protein identification by mass spectrometry (IMAC/2D DIGE/MS)
Phosphoproteins were enriched using the PhosphoProtein Purification Kit (Qiagen, Hilden, Germany) as recommended by the manufacturer. In short, 1 × 10 7 cells were lysed after IFNα treatment (48 h) as previously described ). Subsequently, phosphoprotein samples were covalently labeled with fluorescent CyDyes (GE Healthcare, Chicago, Illinois, US) in a minimal labeling reaction (400 pmol dye:50 µg protein) and isoelectrically focused on pH 3-11 DryStrip Immobiline gel strips (GE Healthcare) prior to second dimension gel electrophoresis and mass spectrometry identification as described in the Online Resources.

Phosphoproteome analysis of IFNα-Le and IFNα-2b
We investigated the difference in phosphoprotein regulation between the two IFNα compounds IFNα-Le and IFNα-2b by immobilized affinity chromatography (IMAC) and 2D DIGE in the human AML cell line MOLM-13 (48 h treatment). 2D DIGE showed a total of 47 proteins with higher than 1.3 fold change and a significance level of p ≤0.05 between the compounds (Fig. 1a, Tables 2 and 3). Only nascent polypeptide-associated complex subunit alpha (NACA) and 40S ribosomal protein SA (RPSA) were modulated at both 250 and 2000 IU/mL IFNα-2b. For IFNα-Le only F-actin-capping protein subunit beta (CAPZB) and actin cytoplasmic 2 (ACTG1) were modulated at 250 and 2000 IU/mL (
The expression differences induced by IFNα-2b and IFNα-Le demonstrated no overlap between proteins regulated at low and high dose (Table 3). At 250 IU/mL, 6 of 7 proteins had lower expression after IFNα-Le treatment, whilst only PSMC2 was regulated by IFNα-2b (Online Resource Table 4). This effect was reversed at 2000 IU/ mL where 16 of 18 proteins showed higher expression after IFNα-Le treatment compared to IFNα-2b, exemplified by up-regulation by IFNα-Le for YWHAE and ANP32A, or down-regulation by IFNα-2b for alpha-enolase (ENO1), heat shock protein beta-1 (HSPB1) and T-complex protein 1 subunit alpha (TCP1).

IFNα-Le induces cell death more efficiently than recombinant IFNα-2b
Since IFNα-2b and IFNα-Le differed in the regulation of both known and previously unknown IFNα-regulated proteins, we investigated the difference in cell death induction by the two drugs. VPA and IFNα have been reported to act synergistically in several cancer models (Jones et al. 2009;Iwahashi et al. 2011;Hudak et al. 2012), and we, therefore, combined the two drugs with the aim of increasing the modest apoptotic effects of IFNα. MOLM-13 cells were treated for 48 h and analyzed by Hoechst (Online Resource Fig. 3) and Annexin-V/PI staining (Fig. 2). We found that both IFNα-2b and IFNα-Le induced a low but significant increase in apoptosis compared to the control. Whilst increasing the concentration of IFNα-2b from 250 to 2000 IU/mL did not result in significantly increased levels of cell death, 2000 IU/ mL IFNα-Le caused elevated levels of cell death (p =0.04) (Fig. 2a). Combining 2000 IU/mL IFNα-2b with 1 mM VPA increased the levels of cell death synergistically (46.0%, p =0.01) (Fig. 2b), whilst the combination of 2000 IU/ml IFNα-Le with 1 mM VPA was more efficient at inducing cell death (55.1%, p = 0.009) (Fig. 2c). For the rat IPC-81 cell line, VPA significantly induced cell death compared to the control (Online Resource Fig. 4A). However, no effect was seen on apoptosis by either IFNα drugs, even though 2000 IU/mL IFNα-Le induced STAT1 (pY701) phosphorylation (Online Resource Fig. 4B).

Phosphoprotein signaling by the valproic acid/ IFNα-Le combination
To unravel the reason for the synergistic effect seen by IFNα-Le and VPA in MOLM-13 cells, we performed phospho-flow exploring the same proteins as described above for IFNα mono-therapy. Altered phosphorylation that could account for the observed synergistic effect was not found for any of the analyzed proteins. Treatment with 1 mM VPA for 15 min resulted in increased acetylation of p53 (acK382), whereas no significant change was induced by IFNα-Le (Fig. 3a, Online Resource Fig. 2C). VPA also induced a slight increase in phospho-ERK1/2 (pT202/pT204) and phospho-p38 (pT180/pY182), similar to the response seen after IFNα-Le treatment, indicating p38 and ERK1/2 as common downstream targets for VPA and IFNα-Le. After 48 h, acetylation of p53 remained to be the main cellular response to VPA treatment, whereas both drugs induced phosphorylation of ERK1/2, p38, p53 (pS15) and Akt (pT308) (Fig. 3b, Online Resource Fig. 2D). The increase in S15 phosphorylation of p53 induced by IFNα-Le indicates that the previously reported induction of p53 by IFNα (Takaoka et al. 2003) Fig. 5 Signaling pathways altered by IFNα-2b and VPA in healthy PBMCs. PBMCs from healthy donors treated with IFNα-2b and VPA and combination IFNα-2b/VPA for 48 h ex vivo were evaluated by CyTOF to investigate alterations in intracellular signaling pathways in defined cell subsets. Data are presented as arcsinh ratio relative to control. Statistics are based on treated cells compared to control. Kruskal-Wallis H test *p ≤ 0.05, **p ≤ 0.01 ◂ may be caused by p53 S15 phosphorylation and not acetylation. A STRING analysis of proteins found to be regulated by IFNα in this study and proteins regulated by VPA in our previous study  showed that several proteins were connected, and also showed that proteins found by 2D DIGE interacted with proteins known to be regulated by IFNα (YWHAE and MAPK3(ERK1)/MAPK1(ERK2)/ AKT1) (Online Resource Fig. 5).

Interferon-α gives no survival benefit in MOLM-13 Luc+ NOD/Scid IL2 γ−/− xenograft mouse
To further explore the observed in vitro synergistic effects of VPA and IFNα, we used the MOLM-13 Luc+ NOD/Scid IL2 γ−/− xenograft mouse model. Tumor load evaluation by bioluminescent imaging showed that control mice and mice treated with IFNα-Le (1 × 10 6 IU/kg) developed tumors in femurs and lymph nodes after 21 days, whilst animals treated with VPA showed detectable tumors 7 days later (Fig. 4a). At day 32, mice treated with VPA (350 mg/kg) showed the lowest tumor burden. Control mice had higher tumor burden compared to IFNα-Le-treated mice (Fig. 4b), but IFNα-Letreated mice developed hind limb paralysis earlier than other treatment groups. They did, however, not have significantly reduced survival compared to control mice (p = 0.118). VPA-treated and VPA/IFNα-Le combination treated mice had significantly longer survival compared to mice treated with IFNα-Le as monotherapy (p = 0.0008 and 0.0294, respectively) (Fig. 4c). Necropsy revealed tumor infiltration in lymph nodes and ovaries but no signs of splenomegaly.

VPA treatment significantly increases survival in immune-competent BN myeloid leukemia rats
The anti-leukemic effects of IFN-α are attributed both to a direct action on AML cells and an indirect effect through immune activation (Anguille et al. 2011). As the MOLM-13 Luc+ NOD/Scid IL2 γ−/− model is lacking a functional immune system, we further investigated whether the synergistic apoptotic effect observed in MOLM-13 cells could be reproduced in vivo using the immune-competent BNML rat model. Control rats and rats treated with IFNα-Le (0.8 × 10 6 IU/kg) mono-therapy rapidly presented with hunched posture and paralysis of hind limbs due to the accumulation of leukemic blasts in the bone marrow, and showed median survival of 21 and 22 days, respectively (Fig. 4d). Animals treated with VPA (400 mg/kg), both as mono-therapy and in combination with IFNα-Le, showed no signs of disease during the treatment period and had consistently lower spleen size, white blood cell counts and higher number of platelets (Fig. 4e) compared to control animals. Rats treated with VPA alone and in combination with IFNα-Le also showed significantly longer survival compared to control rats (p = 0.01) and compared to IFNα-Le monotherapy (p = 0.007). Combining IFNα-Le and VPA (median survival 53 days) gave a slight but non-significant prolonged survival (p = 0.07) compared to VPA alone (median survival 50 days), whereas no survival benefit was seen for IFNα-Le monotherapy. The dose of IFNα-Le used in both the rat and mouse model was slightly higher than the dose chosen for a cutaneous melanoma study (Stadler et al. 2006), but is in line with the current practice for IFNα-2b treatment of chronic myeloid leukemia and chronic hepatitis B (Online Resources).
VPA was found to significantly induce acetylation of p53 (K382) and IFNα-2b to significantly induce phosphorylation of STAT1 (Y701) and STAT3 (Y705) both in healthy donor and AML PBMCs (Figs. 5 and 6), validating the finding Fig. 6 Signaling pathways altered by IFNα-2b and VPA in AML patient-derived PBMCs. PBMCs from AML patients treated with IFNα-2b and VPA and combination IFNα-2b/VPA for 48 h ex vivo were evaluated by CyTOF to investigate alterations in intracellular signaling pathways in defined cell subsets. Data are presented as arcsinh ratio relative to control. Statistics is based on treated cells compared to control. Kruskal-Wallis H test *p ≤ 0.05, **p ≤ 0.01 made by flow cytometry in the MOLM-13 cell line (Fig. 3). NPM1 wild type patients had higher levels of pSTAT1 after IFNα-2b treatment compared to mutated patients (DN T cells; p = 0.049) (Online Resource Fig. 6).

Discussion
The phosphoproteome analysis identified the acetyl transferase protein NAA10 as selectively down regulated by IFNα-Le and a potential overlapping signal pathway with the histone deacetylase inhibitor VPA. Knock-down of NAA10 has been found to increase apoptosis and increase the sensitivity to daunorubicin in vitro (Arnesen et al. 2006). Furthermore, mutations in the auto-acetylation site of NAA10 inhibit lung tumor xenograft growth in vivo (Seo et al. 2010). YWHAE and PKM2 expression were also induced by IFNα-Le, and knockdown of these proteins has been shown to result in increased invasion, migration and proliferation in gastric cancer cell lines (Leal et al. 2016), and inhibition of drug-induced differentiation in leukemic K562 cells (Chaman et al. 2015). Additionally, mRNA expression of YWHAE, ARMC6, RAB11B, P4HB, SNRPA and ANP32A is down-regulated and CAPZB up-regulated by VPA in AML cell lines (Rucker et al. 2016), further supporting the existence of overlapping signaling pathways for IFNα and VPA. STRING pathway analysis of proteins regulated by IFNα and VPA ) also showed several interactions (Online Resource Fig. 5) determined experimentally.
Previous case reports indicate that secondary AML transformed from essential thrombocytosis or myelofibrosis particularly benefit from IFNα therapy (Berneman et al. 2010;Dagorne et al. 2013). In the light of the biological and molecular heterogeneity in AML (Dohner et al. 2017), this may suggest that a particular AML subset is sensitive for IFNα. This was, however, not the case for our patient cohort, although the number of individuals was limited. Importantly, immediate and rapid progressing disease has been described in acute lymphoblastic leukemia patients treated with lower doses of IFNα (Ochs et al. 1986) and in vitro testing of AML cells has indicated that approximately a third of the patient samples responded with increased clonogenicity when treated with lower doses IFNα (Ludwig et al. 1983). Our study showed increased in vitro activation of UPR by IFNαinduced phosphorylation of AK2 and HSPH1 in MOLM-13 cells. Increased UPR has been shown to promote faster tumor growth and resistance to common anticancer drugs in xenograft mouse models (Bi et al. 2005;Spiotto et al. 2010). Therefore, moderate or low doses IFNα should be used with care in AML until we know predictive markers for therapy response, like tumor burden and molecularly defined IFNαsensitive subtypes of AML.
The anti-tumor effect of IFNα in combination with VPA has been suggested experimentally in other cancers (Stadler et al. 2006;Iwahashi et al. 2011;Hudak et al. 2012 (Eggermont et al. 2012), and we cannot exclude that the mouse and rat models used in this study may have exceeded the tumor burden accessible for beneficial IFNα therapy. Whereas the lack of anti-leukemic effect of IFNα in the mouse model used in our study could also be due to the absence of important immune cells needed for an effective DC effect against AML cells (Ito et al. 2002), the BNML rat model has an intact immune system. The lack of in vivo potency of the VPA and IFNα-Le combination in this model could be a result of reduced activity of human IFNα-Le in rats. However, we did find 2000 IU/mL IFNα-Le to increase pSTAT1 (Y701) in BNML derived IPC-81 cells (Online Resource Fig. 4), suggesting that human IFNα could be reactive also in BNML rats. Furthermore, activation of rat IFNα receptors by human IFNα is supported by reports of reduced rat endometriosis by human IFNα (Altintas et al. 2008) and in vivo interferon-induced metallothionein (Guevara-Ortiz et al. 2005).
It is well established that IFNα activates DCs, T cells and NK cells, and thus contributes to the generation of a potent anti-leukemic immune response (Zhang et al. 2005;Watanabe et al. 2006;Korthals et al. 2007;Willemen et al. 2015). Investigating immune regulators in healthy and AML patient-derived PBMCs revealed that all cellular subsets apart from CD8 + and DN T cells responded by one or more markers after IFNα-2b treatment in healthy donor PBMCs. For AML-derived PBMCs, however, no response was seen in pDCs, NK or NK T cells. We also observed that PBMCs from AML patients had different immune-associated responses to IFNα-2b compared to healthy donor PBMCs (Online Resource Fig. 8). Particularly, lack of activation of the differentiation markers CD141, CD45RO and CD86 in AML patient monocytes and pDCs could indicate that patients with AML have an inaccessible immune system where cell subsets stay unresponsive to activating stimuli. The absence of pSTAT1 (Y701) induction in AML-derived pDCs, and not healthy pDCs, further supports the inability of these cells to respond to IFNα. The up-regulation of CD141 in response to IFNα treatment in healthy pDCs was importantly not observed in AML-derived DCs. CD141 + DCs are known to induce differentiation of IL-4-and IL-13-producing CD4 + T cells, thereby guiding the adaptive immune response (Yu et al. 2014). Thus, the lack of DC activation in AML samples could explain the lack of response to IFNα monotherapy in AML patients. Increased CD86 promotes myeloid differentiation and suppresses cell proliferation (Fang et al. 2017). AML patients positive for CD86 have been suggested to be candidates for immunotherapy (Re et al. 2002), however, this marker was not activated in AML patients. Neither was CD45RO, whose presentation on lymphocytes in adult T cell leukemia patients is correlated with improved prognosis (Suzuki et al. 1998). Thus, our overall results could indicate that there is a combined loss of the differentiation potential and lack of immune activation in the investigated AML patients, suggesting that these patients would not benefit from IFNα-monotherapy.

Conclusion
IFNα-2b and IFNα-Le have different effects on the regulation of phospho-protein expression as discovered by 2D DIGE proteomic analysis and phospho-flow, and IFNα combined with VPA induced cell death synergism in vitro. The absence of monocyte and pDC activation by IFNα ex vivo could explain the lack of an in vivo anti-leukemic effect, and the therapeutic effect of IFNα may potentially be enhanced by removing this inherent block of activation in healthy immune subsets in AML patients. This needs to be addressed in future studies that take into consideration the complex tumor-host interactions in AML.