Systems analysis of transcriptome and proteome in retinoic acid/arsenic trioxide-induced cell differentiation/apoptosis of promyelocytic leukemia

Results and Discussion

RA-Induced Differentiation of NB4 Cells. Gene clustering of microarray data. At the transcriptome level, NB4 cells were treated with RA at a series of time points and profiled by using cDNA microarrays. Hybridization for each sample was repeated at least twice, and only data with a high correlation coefficient (≥0.95) were further analyzed. For data mining, we used SOM, which exerts distinct advantages in both gene clustering and its visualization (8). As shown in Fig. 1A , the SOM outputs were first visualized by bar-graph display, offering a global view of gene clustering, particularly with respect to the expression patterns of clustered genes. It can be seen that genes mapped to corner/edge areas of the map are mostly regulated, showing evidence of four major functional categories, with two down-regulated categories in the upper two corner/edge areas and two up-regulated categories in the bottom two corner/edge areas (Fig. 1 A and B ). A total of 1,113 genes were found modulated over RA-induced differentiation. To validate the reliability of the data, 30 randomly selected genes were analyzed by real-time RT-PCR, and the results were highly correlated with those of the array data. Representative results of 8 genes are shown in Fig. 1C . To address the clinical relevance of the results obtained in the NB4 cell line, a side-by-side comparison study was conducted through microarray, and a high correlation was revealed between NB4 cells and patient samples (data not shown).

Transcriptional remodeling and molecular pathways coordinating the RA-induced differentiation. A large number of transcription factors/cofactors (TFs/CoFs), particularly those related to hematopoiesis, were found to be up-regulated at an early stage (within 6 h) of the differentiation, clustered to the bottom-left-corner areas of the map (Fig. 1B ). Among these factors are members of the CCAAT/enhancer-binding protein (C/EBP) family and the helix–loop–helix (HLH) family (Fig. 2A ). Although modulations of C/EBPs, including C/EBPα, -β, and -ε in the RA-induced granulocyte differentiation have been reported (9, 10), there was a lack of coherent data in terms of the complex roles played by C/EBPs. It has been shown that dimerization of different C/EBP proteins precisely regulates the transcriptional activity of target genes (11). Accordingly, modulations of multiple C/EBPs (i.e., C/EBPβ, -δ, and -ζ) strongly suggest that C/EBPs may undergo specific homo- and heterodimerization. Because C/EBP homologous protein (CHOP) is a dominant-negative regulator preventing heterodimer binding of C/EBPα/β to C/EBP enhancer sequences (12), CHOP's induction may represent a balanced mechanism in C/EBP action. Members of the HLH family represent another group of TFs coordinating the differentiation. Most HLH proteins have a basic amino acid motif, and these bHLH proteins behave as transactivators or repressors. However, a small number of HLH proteins lacking the basic motif, known as ID proteins, act as dominant-negative regulators of bHLH proteins (13). Although ID proteins have been previously viewed as inhibitors of cell differentiation (14), markedly increased expression of ID1 and ID2 genes in our setting strongly suggests their engagement in APL cell differentiation, probably through coordinating lineage-specific gene expression. Furthermore, other TFs/CoFs might also be involved in early-stage gene regulation, as implicated by modulation of IRF1 (confirmed by real-time RT-PCR), HHEX, SMARCD2, and TADA3L. Thus, it can be deduced that global remodeling in transcription regulation occurs in APL cells upon RA treatment.

An obvious consequence of transcriptional remodeling is the modulation of genes involved in calcium signaling and effector cascades (Figs. 2A and 3), which was one of the landmarks of the second wave of regulation (at 12–24 h). Inositol trisphosphate receptors and other factors constitute major calcium-releasing channels on the endoplasmic reticulum (ER), whereas NUCB2 represents a key calcium-storage protein inside the ER. Accordingly, markedly increased expression of ITPR2 and decreased expression of NUCB2 suggest a remodeling event of calcium-regulatory cascades, which may consequently result in a net increase of cytosolic influx from the internal stores during RA-induced differentiation. An increased level of cytosolic calcium influx may stimulate various calcium-dependent activities that can be important for neutrophil function. For instance, calcium-binding proteins S100A8 and S100A9 are known to be involved in various neutrophil-specific activities, such as migration, chemotaxis, degranulation, and phagocytosis (15). PLSCR1 (phospholipid scramblase 1), an endofacial plasma-membrane protein that mediates accelerated transbilayer migration of phospholipids upon binding to calcium, was indeed found critical to RA- and phorbol 12-myristate 13-acetate-triggered leukemic-cell differentiation during the preparation of this article (16). The versatile impact of enhanced calcium signaling on this differentiation can also be deduced from up-regulation of a number of other signal-pathway- and gene-regulation-related factors, such as calcium-sensitive PTK2β, CAMK1, and CGREF1, a calcium-dependent growth inhibitor. Moreover, the release of calcium from the ER can be paralleled by the increase of mitochondrial calcium, which may activate calcium-sensitive matrix dehydrogenases, resulting in the increase of ATP production and, eventually, the induction of apoptosis in mature neutrophils.

Members of the IFN signaling cascade and its effector cascades represent another prominent group that is up-regulated during differentiation (Fig. 2A ). Here, we have gathered emerging evidence that a spectrum of IFN-responsive genes is induced through multiple transcriptional events. Induction of IRF1 is an early event, whereas the group induced after that (at 12–24 h) consists mostly of effectors of IFN-γ signaling, as represented by genes encoding components of the immunoproteasome, such as PSMB8, PSMB9, and PSMB10, and subunits of the NB, such as SP100 and SP110. The IFN-responsive genes, such as IRF7, up-regulated at late stages of the differentiation are highlighted by TFs of IFN signaling. Accordingly, regulation of key members of IFN signaling and its downstream effectors implicates a fundamentally important role in antiproliferation, immune responses, restoration of the NB, and, probably, degradation of the PML–RARα protein during RA-induced differentiation.

Cell growth rate seems to be little impacted until late stages of the differentiation (at 48–72 h), probably because of a proper balance between survival and death signals essential for cell differentiation. Indeed, a number of survival factors, such as BCL2A1 and PDCD6IP, are up-regulated during differentiation. Restoration of the apoptotic potential seems to parallel the progression of differentiation. In addition to the recovery of the NB, up-regulated caspases (CASP1 and -7) and MADD can be of particular value. Postmaturation apoptosis in RA-treated APL cells is probably triggered by different stimuli, such as overproduction of ATP inside mitochondria through activated calcium-sensitive matrix dehydrogenases.

Integration of proteomic data from RA-induced differentiation. Of the total protein spots separated by 2D gel electrophoresis, 793 appeared to be significantly changed in percentage volumes during differentiation. Of these changed protein spots, 63 were randomly chosen and further characterized by MS. Among them are TFs, signal transduction and cytoskeleton molecules, and proteins involved in calcium signaling and cell-cycle control (Figs. 1 D and E and 2B ). The use of current technologies to establish a direct correlation between proteomic and transcriptome data is extremely challenging because of multiple layers of discrepancies, such as the distinct sensitivities of cDNA array hybridization and 2D gel electrophoresis, differences in the translational efficiency of different genes, inability of the current cDNA array platform to distinguish mRNA isoforms, posttranslational modifications of proteins that are essential for protein stability, and timing discordance of modulation at mRNA and protein levels. Proteomic data appear to be complementary to rather than duplicative of transcriptome data (see Table 1, which is published as supporting information on the PNAS web site). For example, modulation of several key members of calcium signaling was revealed only by proteome analysis, as highlighted by ryanodine receptor 3 (RYR3), an intracellular calcium-ion-release channel (Fig. 2B ). Equally notable is AKAP9, a kinase anchor protein involved in various signal-transduction pathways, including PKA signaling (17). AKAP9 promotes the phosphorylation of inositol trisphosphate receptors through PKA and, thus, contributes to calcium regulation. The detection of at least eight spots of AKAP9 protein with distinct regulatory patterns (Fig. 2B ) provides evidence supporting the notion that AKAP9 is actively involved in multifunctional pathways probably essential for RA-induced differentiation. To this end, it may be worth pointing out that a number of cytoskeleton proteins, including calcium-sensitive/-dependent proteins, were also modulated at the proteomic level (Fig. 2B ).

ATO Effects on NB4 Cells at the Transcriptome and Proteome Levels. ATO effects at the transcriptome level. ATO-treated NB4 cells revealed a much smaller number of regulated genes (487) than did RA-treated cells (1,113), an observation that is supported by regulatory patterns of the CPP-SOM as well (Fig. 1B ). This finding appears to be verified when we look further into details of the functionally classified genes. For example, a total of 74 TFs/CoFs were regulated by RA, whereas only 29 were regulated by ATO. It should be noted that many ATO-regulated genes (316 of 487, 65%) were also regulated by RA but in a more profound manner (Fig. 1B ). Hence, the effects of both RA and ATO on these genes may be additive. Transcriptional modulation of many of these 316 genes could be related to essential aspects of cell-activity control, such as the induction of differentiation antigens (PECAM1 and SELPLG), the modulation of apoptosis regulators (BAK1, BCL2, and MADD), and the regulation of genes involved in cell-cycle and growth control (CDC7L1 and PLK3). Based on these observations, we assume that ATO may have some RA-like effects on APL cells, probably because of the ATO-targeted degradation of PML–RARα. Unlike RA, ATO mainly induces the degradation rather than the activation of the hormone-inducible TF PML–RARα. This assumption probably also explains a less-wide spectrum of regulated genes and a less-profound degree of gene regulation in cells treated by ATO than in those treated by RA. ATO may also target or interact with many other proteins, resulting in the regulation of specific genes different from those regulated by RA, which may underlie the synergistic effect with RA, as highlighted by a set of genes involved in protein degradation, cell defense, stress response, and protein processing (Fig. 2A ). For example, several U/P genes appear to be specifically regulated by ATO, which may consequently contribute to a more effective and efficient protein-degradation system in RA plus ATO-treated cells than in RA-alone- or ATO-alone-treated cells. Another interesting phenomenon was that a total of five HLA-class I genes were specifically down-regulated in ATO-treated cells but up-regulated in RA-treated cells. Because ATO is known to be a strong inducer of oxidative stress, it is not surprising to find that several stress-response-related genes including HSPA8, HSPCA, and AHSAI were modulated by ATO. ATO also appears to be effective in modulating genes involved in protein modification and synthesis, such as ICMT and PPIF.

ATO effects at the proteome level. The number of modulated protein spots in ATO-treated cells (982) appears to be much larger than that in RA-treated cells (793), as highlighted by regulatory profiles of ATO-treated cells at the proteomic level (Fig. 1E ). By comparing regulatory patterns of profiles between treatment series, it can be seen that a time-dependent pattern of regulation in ATO-treated NB4 cells is evident. At 12 h of treatment, regulatory patterns of ATO-treated cells were quite similar to those of RA-treated cells. However, after 48 h, ATO-treated cells exhibited a profound reduction of protein levels at many map units on CPP-SOM (Figs. 1E and 2B ). After protein characterization through MS, a group of enzymes involved in biochemical metabolism were found to be significantly down-regulated, e.g., HMOX1, MTR, and ALDOA, implicating the suppression of related biochemical cascades in ATO-treated cells. A strong reduction of cytoskeleton proteins, e.g., MYOM1, MYO9B, and FBN2, appeared to be another prominent feature in ATO-treated cells, implying a considerable reorganization of the cell nucleus and cytoplasmic structures. Interestingly, down-regulation of some translational initiation and elongation factors seems to be important to RA-induced differentiation, as suggested by a recent proteomic report (18), whereas in our data, elongation factors EEF1A1 and EEF1G were significantly down-regulated by ATO. Compared with relatively minor changes of corresponding genes at the transcriptome level, significant changes observed at the proteomic level in ATO-treated cells suggest that ATO may particularly enhance mechanisms of posttranscriptional/translational modifications.

Synergistic/Additive Effects Between RA and ATO on APL Cells. Transcriptome of RA/ATO-treated NB4 cells. Based on the above data, the existence of overlapping but distinct target properties between RA and ATO may contribute both synergistic and addictive effects on APL after combined treatment. As shown in Fig. 1, profiles of the cotreatment series are highly similar to those of the RA-treatment series, suggesting that many features are shared by the two, whereas some differentially displayed patterns (including 220 genes), particularly highlighted by those at the upper-left-corner/edge areas of the map (Fig. 1B ), may suggest a synergistic impact of RA and ATO on transcriptional regulation in APL cells. Among synergistically/additively up-regulated genes were members of the U/P system, including UBE2L6, PSMC2, and PSMD13 (Fig. 2A ). Based on the observation that genes encoding components of typical immunoproteasome were specifically up-regulated by RA, whereas those encoding subunits of the conventional U/P system (e.g., PSMD11) were significantly induced by ATO, it can be deduced that the protein-degradation system is probably much more enhanced in degrading futile proteins, including the PML–RARα, in cells cotreated with RA and ATO than in cells treated with either of them alone. This finding seems to be in accordance with previous data suggesting that the more degradation of the PML–RARα occurs, the better is recovery from the disease (4). Interestingly, synergistically/additively down-regulated genes include a group of genes known to be involved in various chromosome translocations in human malignances (Fig. 2A ). For instance, ARHGAP26, SH3GL1, and MLLT4 are translocation partners of the MLL gene in acute leukemia. Synergistic/additive down-regulation of these genes may represent a more effective manner of eliminating oncogenic properties or reducing cell-survival potentials in APL cells treated by RA and ATO than in those treated by RA alone or ATO alone. Moreover, a group of genes related to cell proliferation, e.g., IFI16b and GAS7, and apoptosis, e.g., STK24, were also found among the synergistically/additively modulated genes. On the other hand, an antagonistic effect between RA and ATO on gene regulation was also detected, although impacted genes appeared to be only a small percentage of the total. One such example is CYP51, a gene of the cytochrome P450 family involved in the catabolism of RA (19). CYP51 seems to be induced by RA, whereas its induction was abolished under the joint effect of RA/ATO, possibly leading to an increased sensitivity of APL cells to RA treatment.

Proteomic features of RA/ATO-treated NB4 cells. Because RA and ATO exert preferential effects on transcriptome and proteome, respectively, it would be interesting to further investigate protein expression profiles of RA/ATO-treated APL cells. Indeed, we found that a total of 414 protein spots were synergistically/additively modified by RA/ATO on 2D gels, and some of the protein spots were characterized further, as shown in Fig. 2B . For example, translation-initiation factor EIF4A1 and translation-elongation factors EEF1B2, EEF1A1, and EEF1D were down-regulated in a much deeper way under cotreatment than those in either of the single treatments. It has been suggested that the regulation of translational factors may play a role in coordinating proliferation and tumor growth. Accordingly, synergistic/additive inhibition of these factors by RA and ATO probably favors growth arrest. Another factor belonging to this category was TRRAP, a subunit of histone acetyltransferase complexes and an essential cofactor for the c-myc and E1A/E2F-involved transcriptional regulation. RAD23B, an evolutionarily conserved factor involved in DNA-damage recognition as well as ubiquitin-mediated proteolysis (20), was one of the few synergistically/additively up-regulated proteins.

Systems View of Pathways Underlying RA/ATO Therapy of APL. Information obtained from transcriptome and proteome levels complemented by literature-mining allowed the identification of a number of pathways with temporospatial relationships, which might contribute to the mechanisms of RA/ATO-induced differentiation/apoptosis of APL cells (Fig. 3). The early modulated molecules (within 6 h) are mostly TFs/CoFs associated with myeloid-specific gene expression, nuclear receptor signaling molecules, IFN pathway members, and factors involved in some other cascades. A second wave of gene/protein regulation (at 12–24 h) appears to be associated with several prominent features. First, modulation of genes/proteins seems to be an amplification of RA signaling, as highlighted by molecules involving IFN, calcium, cAMP/PKA (21), G-CSF, MAPK/JNK/p38, and TNF pathways (22). These pathways are regulated, not in an isolated situation, but in concert and harmony, thereby forming a network (Fig. 3). Second, there is a strong activation of the U/P system. Among the proteins under the control of this system is PML–RARα. Degradation of this oncoprotein exerts a dual effect. On one hand, the transrepression caused by PML–RARα is relieved. On the other hand, the PML released from the PML–RARα/PML complex may lead to the restoration of the NB. Third, cell survival seems to be secured by the induction of an antiapoptotic program, which may ensure the terminal differentiation of granulocytes in RA-treated cells. After 48–72 h of treatment with RA/ATO, a third wave of gene/protein regulation with some interesting characteristics is visible. The expression of differentiation markers SELL, SELPLG, and CD44 as well as functional molecules DEFA3, DEFA4, and CCL2 reaches a maximum. Meanwhile, genes/proteins promoting cell cycle or enhancing cell proliferation, such as those involved in the MAPK/JNK/p38 pathway, are significantly repressed (Fig. 3). As the cells approach terminal differentiation, the expression of apoptosis agonists increases gradually. Obviously, RA exerts its effects on APL cells mainly through nuclear-receptor-mediated transcriptional regulation, whereas ATO exercises its impact through targeting multiple pathways/cascades at the levels of proteome, transcriptome, and, probably, metabolome. Although understanding the complexity and dynamics of the in vivo synergy between RA and ATO in APL can be more difficult than is currently recognized, this study has clearly provided a framework to reach that goal.