A Noncanonical Flt3ITD/NF-κB Signaling Pathway Represses DAPK1 in Acute Myeloid Leukemia

Recent evidence indicates that attenuation of the unfolded protein response (UPR)—an endoplasmic reticulum (ER)–dependent stress response—may explain therapeutic failure of acute myeloid leukemia (AML; refs. 1–3). Downstream transcriptional mediators of the Flt3 internal tandem duplication (ITD) in AML may impose such a status by regulating expression of effectors that control stress-dependent apoptosis. For example, the ratio of ER levels of bcl-2 versus death-associated protein kinase 1 (DAPK1) as well as other effectors may determine this output at a defined set-point (4–6).

DAPK1 is a calcium-calmodulin–dependent serine–threonine protein kinase, which suppresses tumor cell survival and metastasis via autophagy and apoptosis. It plays a central role in ER stress–dependent apoptosis (7). DAPK1 expression is affected by a variety of oncogenic signals (7, 8). We previously showed the existence of a Flt3–JNK1–c-Jun pathway in Flt3ITD⁺ AML (9). c-Jun is known to drive the expression of not only bcl-2, but also DAPK1 (10, 11). However, the latter circumstance would be antagonistic to the progression of poor-prognosis Flt3ITD⁺ AML. On the other hand, NF-κB and CRE/c-Jun regulatory sites coexist on the promoters of certain tumor suppressor or cytokine genes including DAPK1, and the importance of NF-κB signaling in AML is known (12–15). We hypothesized that a resistance to apoptosis in certain AMLs occurs via severe repression of DAPK1, through recruitment of p52NF-κB to the putative NF-κB site at −134bp of its promoter (12, 13). Indeed, expression of DAPK1 is lost in a number of human cancers, including leukemias (16). Although epigenetic suppression of DAPK1 is well reported, the upstream mechanisms that contribute to tumor promotion via the recruitment of epigenetic apparatus to the DAPK1 promoter are not defined (16–19).

We hypothesized that tandem activation of both c-Jun N-terminal kinase 1 (JNK1) and IκB kinase (IKK)/NF-κB may be involved in concerted regulation of antiapoptotic as well as proapoptotic genes to achieve an antiapoptotic/proapoptotic effector balance (e.g., bcl-2/DAPK1) to permit higher aggressiveness in Flt3ITD⁺ AML (4, 6, 14, 20). This postulation was tested in context of prior functional and cohort analyses of AML blasts carried out in our laboratory that linked Flt3 phosphorylation/activation to JNK1 phosphorylation (9) and with regard to the known role for c-Jun1/AP-1 in DAPK1 and bcl-2 expression (10, 11). In addition, we inferred that a conserved dual-activation mechanism, which relies upon TAK1, for JNK1–c-Jun and IKK–NF-κB may exist (21) to promote the optimal antiapoptotic/proapoptotic effector balance. This hypothesis was given emphasis by the recognition that TAK1 is among the most highly expressed genes in a leukemic stem cell (LSC) signature of poor-risk AML in which DAPK1 repression coexists (22, 23).

In this study, we showed that TAK1 activated p52NF-κB, binds at the tandem NF-κB and CRE sites of DAPK1, and recruits certain transcriptional repressors, belonging to the histone deacetylase (HDAC) family (12). Because p52NF-κB is a downstream target of Flt3 signaling, we hypothesized that interruption of this signaling arm of Flt3 would result in a de-repression of DAPK1 to contribute toward enhanced apoptosis in Flt3 ITD⁺ AMLs. Finally, we propose a therapeutic model for the rational combination of Flt3- and HDAC-inhibitors for suppressing AML growth.

The human leukemic cell lines, HL-60 and MV-4-11 (derived from a biphenotypic leukemia), were obtained from the American Type Culture Collection (ATCC) in McCoy's and RPMI media, respectively, supplemented with 10% fetal calf serum. The MV-4-11 cell line overexpresses Flt3-ITD and harbors the 11q23 translocation [t(4)] involving the MLL gene (9). Blast cells from the bone marrow of patients with AML were obtained at the time of diagnosis, after informed consent. The buoyant fraction was isolated over Ficoll–Hypaque, and then washed with PBS before processing. The cohort of AMLs subjected to gene expression analysis showed mean 88 ± 10% blasts (Supplementary Table S1). Cells were lysed and fractionated into nuclear and cytoplasmic fractions using the NE-PER Extraction Kit (Pierce Biotechnology). For Western blot analysis, bone marrow samples with ≥70% blast cells in the purified aspirate were used.

MV-4-11 cells were electroporated using the Amaxa system and then placed in McCoy's medium supplemented with 10% FBS. Cells were transfected with a luciferase reporter (13) driven by the DAPK1 promoter (1.2 κB) harboring either wild-type or mutated CRE site (−177 bp). Wild-type c-Jun or vector control was co-transfected in some experiments with the DAPK1 reporter. Renilla luciferase was cotransfected to normalize for variations in transfection efficiency.

Cytosolic or nuclear proteins were subjected to Western blotting with indicated antibodies as described previously (9). Densitometry was carried out to quantify specific bands, and data were normalized to either cytoplasmic glyceraldehyde-3-phosphate dehydrogenase (GAPDH) or nuclear (Sp1) internal controls, depending on the case.

Chromatin immunoprecipitation (ChIP) assays were conducted as previously described (13). In brief, DNA–protein complexes were cross-linked by incubating 5 × 10⁷ cells with 1% formaldehyde. After washing, cells were lysed and chromatin was sheared to yield DNA fragments (∼800 bp) using a Branson Digital Sonifier. The lysate was then centrifuged, and soluble chromatin was immunoprecipitated with 5 μg of the indicated antibody. The protein–DNA complexes were collected after incubating with protein-A magnetic beads, washed, and reverse cross-linked by heating at 65°C overnight using 1% SDS and 0.1 mol/L NaHCO3. DNA released from the complexes was purified using the QIAquick Spin Kit (Qiagen, Inc) and collected. The purified DNA samples from the input, immunoprecipitated with either a nonspecific immunoglobulin G (IgG; control) or specific IgG (experimental), were subjected to either PCR or quantitative PCR (qPCR) analyses with DAPK1 promoter specific primers: 5′-AGTCCTCAGAAATCTC ATGCAAG-3′ and 5′-CATTAGAGTCCAAGACAGTA-3′. DNA extracted from soluble chromatin was used as an input control for each reaction. Each experiment was repeated at least 3 independent times with multiple samples (n = 4) for ensuring the consistency of the results.

Total RNAs from blood/bone marrow specimens were isolated using the RNeasy Kit (Qiagen, MD) according to the manufacturer's instructions. Total RNA (500 ng) from each specimen was converted to cDNA using the Superscript III First strand cDNA Synthesis Kit (Invitrogen). Relative gene expression was quantified using TaqMan Gene Expression Assays (Applied Biosystems) and ABI PRISM 7900 real-time PCR machine. Quantitative reverse transcriptase (qRT-PCR) for 31 AML associated genes and 1 housekeeping gene was carried out at low-density array (LDA) format according to the manufacturer's protocol (TaqMan Gene Expression Micro Fluidic card, 4346799; Applied Biosystems). The amount of cDNA per card was 200 ng. 18S rRNA was chosen as an internal control. Relative expression was calculated using RQ manager Ver 1.2 (Applied Biosystems) using a 1-patient volunteer sample (CBF⁺, very low c-Jun and negative for Flt3ITD and Meis-1 expression) as a calibrator. Copy number or fold-change in expression was calculated using the 2^−ΔΔCt method (24).

Individual patient gene expression data within the defined cohorts was subjected to statistical analysis using the Mann–Whitney/Wilcox test and a P value <0.05 was considered significant.

Bcl-2 and DAPK1 are targets of transcriptionally active p-Jun, and are potentially relevant to the pathogenesis of Flt3ITD AML (10, 11). However, optimal progression of Flt3ITD-driven AML would be served by repression of DAPK1 with simultaneously overexpressed bcl-2, whose transcription is critically dependent on a CRE/c-Jun site (25).

We postulated that the poor prognosis of Flt3ITD AMLs may relate to a lack of DAPK1 expression, dependent on a transcriptional milieu affecting the tandem CRE and κB elements of its promoter, thus distinguishing DAPK1 regulation in Flt3ITD⁺ versus Flt3ITD⁻ AML. Expression of DAPK1 in Flt3ITD⁺ AML may be selectively blocked by a p52NF-κB/HDAC-associated complex at the κB/CRE site (12). Both NF-κB and c-Jun can be generated by activation mechanisms (IKK and JNK1) regulated by the MAP3K, TAK1, respectively (21).

Moreover, analysis of AML blasts separated by their Flt3ITD status, in chronologic succession, showed an apparent relationship with expression of phospho-c-Jun and TAK1 activation, through phosphorylation (Fig. 1A and B; and data not shown). Interestingly, in an Flt3ITD⁺ sample (#2797) with very high pTAK1/p-Jun activity, the lowest expression of DAPK1 was observed, when compared with 2 other AMLs lacking the Flt3ITD mutation (Fig. 1A). This is contrary to the expectation based on the known role for c-Jun and its transcriptional partner canonical NF-κB on DAPK1 expression (refs. 11, 13, 15; and see below), and expression levels of p-Jun were quite similar in Flt3ITD⁻ AML #2993 (normal karyotype) with high DAPK1, compared with NKFlt3ITD⁺ 2797 (Fig. 1A).

To further explore the role of Flt3ITD/TAK1 in promoting cooperation between p-Jun and NF-κB species, a series of AML blasts characterized by normal karyotype with Flt3-ITD (NK-Flt3ITD) was studied for c-Jun phosphorylation and TAK1 activation. We found that, like the positive control cell line MV-4-11, the normal karyotype Flt3ITD⁺ patient blasts express significant levels of Flt3 protein, active/total TAK1, and p-Jun (Fig. 1). In one of the AML blasts (#2857), treatment with Flt3-ligand ex vivo led to downregulation of Flt3, followed by transient upregulation of phospho-TAK1 and DAPK1 levels, which returned to baseline (Fig. 1B). Indeed, DAPK1-Luc reporter analyses in MV-4-11 revealed abundant activity, and exogenous c-Jun could significantly augment luciferase expression by 2-fold (P < 0.004). However, a mutant promoter lacking the CRE site had blunted activity. RNA interference (RNAi) with Flt3 and JNK1 showed that interference with the Flt3–JNK1 axis inhibited DAPK1 and bcl-2 expression (Fig. 1C).

On the other hand, Western blot analyses for DAPK1 expression in relation to p-Jun, showed widely varied expression of DAPK1 between blast samples, contrary to expectation if only p-Jun were the regulatory determinant (Fig. 1B). All Flt3ITD⁺ samples, except #2841, had an extremely low (#2857) or barely detectable DAPK1 expression (#2797, 2854), similar to that observed in MV-4-11—where DAPK1 was undetectable (Fig. 1). This was noted in spite of high and largely invariant levels of p-Jun (Fig. 1). Furthermore, the occurrence of highest bcl-2 levels in #2797 primary AML blasts with lowest DAPK1 was contrary to expectation and quite similar to the situation in MV-4-11 (Fig. 1).

We next pursued the clinical/biologic significance of these findings in MV-4-11 and in a larger series of NKFlt3ITD⁺ samples. Genome-wide gene expression profiling coupled with real-time RT-PCR was carried out using AML blasts from patients with defined cytogenetic and molecular Flt3ITD status from 3 distinct cohorts: (i) normal karyotype Flt3ITD; (ii) tMLL-AML; and (iii) other genotypic/cytogenetic alterations, including those with monosomy 5/7 and a complex karyotype. These analyses included the cohort of blasts from the normal karyotype of patients with Flt3ITD AML (Fig. 1). We studied the expression of DAPK1, because it is responsive to c-Jun/AP-1 and (p65) canonical NF-κB, to examine the basis for gene repression guided by p52NF-κB in the absence of p65NF-κB. In such circumstance, p52NF-κB heterodimers with relB could recruit transcriptional co-repressors (12,14, 26, 27). In addition to c-jun and relB (26), expression of another c-Jun target gene meis1 (11) was analyzed as a positive control in real-time RT-PCR.

Compared with CBF⁺ AML or PML-RAR⁺ AML (which express low c-Jun and which fail to express hoxA9/meis1; data not shown; unpublished data), the NK Flt3ITD⁺ samples had very high meis1 and c-jun expression (Fig. 2). However, there was a significant suppression of DAPK1 transcripts (Fig. 2A). In most of the examples noted above, DAPK1 levels were 10- to 100-fold lower than in the controls, when normalized to c-Jun (Fig. 2A). In fact, a statistically significant difference in DAPK1/c-jun expression was noted when we compared the populations with NKFlt3ITD [or tMLL; as a positive-control for DAPK1 repression; ref. 19] to AMLs having other cytogenetic/genotypic profiles, where differences in Meis1/c-jun levels were not significantly different (Fig. 2B; DAPK1/c-jun: Mann–Whitney/Wilcoxon one-sided test: NKFlt3ITD or tMLL vs. other cytogenetic/genotype groups: P < 0.041 or P < 0.0006, respectively; Meis1/c-jun = not statistically significant). RelB, which is partially c-Jun-responsive (26), was uniformly highly expressed (Fig. 2A).

Therefore, Flt3ITD⁺ MV-4-11 cell line was used as a model to further explore the relationship between Flt3, TAK1, and JNK1/phospho-c-jun versus IKK/NF-κB (Fig. 3). We first knocked down Flt3 using RNAi. Flt3 knockdown (78%) resulted in a significant loss of TAK1 phosphorylation (53%; Fig. 3A). [We previously showed that Flt3 knockdown leads to a loss of JNK activity and c-Jun phosphorylation (Fig. 1C; ref. 9)]. Indeed, RNAi-mediated knockdown of either JNK1 or TAK1 also suppressed the expression of phosphorylated c-Jun, by 100% or 80%, respectively (Fig. 3B). In addition, we found that TAK1-inactivation essentially led to total loss/destabilization of NIK (Fig. 3A). NIK is a MAP3K required for noncanonical NF-κB activation but sensitive to TAK1 activation-dependent stabilization (28). Thus, loss of Flt3 would be expected to affect activation of either canonical NF-κB (solely involving TAK1) or noncanonical NF-κB (TAK1/NIK; Fig. 3A; and see below).

To study the activation status of NF-κB species, cellular lysates were fractionated into cytoplasmic and nuclear components because nuclear retention of these proteins determines their transcriptional activity. We and others have shown that phosphorylated c-Jun (at S⁶³, S⁷³) is always predominantly present in the nucleus as a heterodimer with other proteins (refs. 9, 29). Surprisingly, we found that, in MV-4-11 cells, p65 NF-κB was largely absent in the nuclear fraction, although an abundant quantity of Sp1 (a constitutively nuclear transcription factor), but not GAPDH (a cytoplasmic marker), was found (Fig. 3C). Furthermore, the RNAi-mediated knockdown of Flt3, TAK1, NIK, and JNK1 failed to increase nuclear levels of p65NF-κB (Fig. 3C).

In contrast, control MV-4-11 (Fig. 3C) had ∼50% fraction of p52NF-κB in the nucleus, and a lesser amount of relB, (a primary heterodimeric partner of p52NF-κB; Fig. 3C). Knockdown of either TAK1 or NIK strongly decreased nuclear p52NF-κB levels by 73% or 76%, respectively, with a corresponding increase in cytoplasmic levels (Fig. 3C, panel 1; and see below). Knockdown of JNK1 or Flt3 appeared to diminish overall cytoplasmic and nuclear content of both relB and p52NF-κB by 76% and 50%, respectively (Fig. 3C, panels 1 and 2). This is consistent with stability of this heterodimeric complex relying on partner relB, whose expression is induced by c-Jun (Fig. 3C; ref. 26). Taken together, our data support the existence of an Flt3–p52NF-κB pathway, which may negatively regulate DAPK1 expression (Figs. 1–3).

As three fourths of primary Flt3ITD⁺ AML cases presented in Fig. 1B were refractory to primary treatment (all except #2841) and had low DAPK1 expression, we hypothesized that this pathway might have a biologic/prognostic significance. Indeed, sample #2841 with high DAPK1 expression differed greatly from 7 other Flt3ITD⁺ samples and a control tMLL AML because it had trivial amounts of nuclear p52NF-κB and abundant and predominantly nuclear p65NF-κB (Fig. 3D; and see below). These NKFlt3ITD⁺ AMLs with little or no DAPK1 expression, as in the control tMLL AML, were distinguished by predominantly nuclear p52NF-κB (mean% nuclear translocation p52NF-κB, 72.8 ± 9.9%; vs. mean% nuclear translocation p65NF-κB, 19.6 ± 6.6%). In fact, qPCR analysis showed very low DAPK1 transcripts among samples with no nuclear p65, for example, 2797, 2874, and 2857 (compare Fig. 3D with Fig. 2A). On the other hand, the higher and dominant content of nuclear p65NF-κB in #2841 was more typical of the above-mentioned Flt3ITD⁻ patients—where DAPK1 levels were high, and had little or no nuclear p52NF-κB (#2930, 2993 in Figs. 1, 2, and 3D). In addition, the dominant nuclear presence for p52NF-κB compared with p65NF-κB in the different cohorts was confirmed by gel mobility-shift assays in addition to immunoblotting (Supplementary Fig. S1). We ascertained that the isolation procedure/washing had not created artifactual deficiency of nuclear p65NF-κB (on any potential autocrine cytokine washout) by carrying out timed re-additions to cultured AML blasts of Flt3 ligand (ref. 30; data not shown).

We next investigated a possible Flt3ITD-driven mechanism that linked p52NF-κB to DAPK1 repression in MV-4-11. Because HDACs are known to cause gene repression, we next determined whether p52 NF-κB, c-Jun, and HDAC would bind to the DAPK1 promoter. ChIP assays revealed a strong binding of both p52NF-κB and HDAC2 to the DAPK1 promoter (Fig. 4). In contrast, a relatively weaker binding of c-Jun and HDAC1 to DAPK1 promoter was observed, when compared with HDAC2 (Fig. 4A). In another series of ChIP experiments, we found that, among HDACs, strong binding of the HDAC2 (class I) to the promoter accompanied a corresponding absence of HDAC-5 (class II) and class IV HDAC11, which have been found to be predominantly localized to cytoplasm in certain hematopoietic cells and are expressed low in MV-4-11 (Fig. 4B). On the other hand, a moderate binding of the class II HDAC6 occurred. The latter has a nuclear localization signal and has been found to shuttle between nucleus (e.g., in association with c-Jun/CREB/RUNX2/steroid receptors on chromatin) and cytoplasm for functions. In the cytoplasm, a dominant role for HDAC6 in regulating hsp90 chaperone function has been noted in AML cell lines as well (Fig. 4B; refs. 31–36).

To provide additional evidence for a role of p52NF-κB in repressing DAPK1, knockdown experiments in MV-4-11 cells were undertaken (Fig. 4C–E). RNAi-mediated knockdown of p52NF-κB, or its upstream activator NIK, caused an increase in DAPK1 protein expression by 2-fold (Fig. 4C and D). As expected, knockdown of either JNK1 or Flt3 in MV-4-11 downregulated DAPK1 levels by 40% or 60%, respectively (Fig. 4E), similar to the data shown in Fig. 1C. Therefore, DAPK1 was de-repressed following the knockdown of p52NF-κB when JNK1 arm of Flt3 signaling was sustained (Fig. 4). In addition, the identification of severe DAPK1 repression (Figs. 2 and 3D) is consistent with the relationship between p52NF-κB and DAPK1 in MV-4-11.

Genome-wide sequencing studies to identify gene mutations among the majority of solid tumors can be compared with similar screens carried out with AML. Such comparison indicates that the mutational complexity of adult AML is modest and may involve 10- to 300-fold lower numbers of mutated genes (37–42). In the case of normal karyotype AMLs with Flt3ITD mutation, only 10 to 14 genes have been found mutated and the important founder mutations appear to involve not only the Flt3 and NPM1 genes, but also epigenetically active enzymes (39–42). Perhaps because of the involvement of intrinsic epigenetic mechanisms in many examples of the heterogenous disease process, clinical use of single-agent Flt3-selective tyrosine kinase inhibitors has not showed truly significant disease-free survival in most patients with Flt3ITD⁺ AML.

As the epigenetic signature of AMLs is robust, particularly in the setting of Flt3ITD mutation (41–43), we postulated that the unique signaling pathway used by Flt3ITD may contribute to those “stress-induced” steps involved in epigenetic reprogramming in Flt3ITD⁺ AMLs (12, 17, 18). We further postulated that TAK1, which is overexpressed in a poor-risk AML LSC signature (22), may contribute to p52NF-κB–induced epigenetic silencing, similar to the ability for Evi-1 to attract a cis-localized epigenetic apparatus to distinct genes in an AML subgroup (44). Indeed, our data point to the existence of an instructive mechanism (as opposed to a “stochastic” origin) for transcriptional repression of certain tumor suppressor genes, as shown here for DAPK1. DAPK1 is regulated by a binary cis-element consisting of CRE/Jun and κB sites (12, 13). These sites are subject to dominant-negative regulation by p52NF-κB and HDAC2/HDAC6 under the Flt3-/TAK1 pathway (Figs. 3–5). Inhibition of p52NF-κB caused a de-repression of DAPK1.

Very little is known about the role of p52NF-κB as a regulator of gene expression in AML (45). In one study, epigenetic repression of another tumor suppressor gene, DIF2 was indirectly linked to the activity of p52NF-κB. In addition, there has been inadequate explanation for the severe reduction of DAPK1 expression in poor-risk AML. In fact, previous studies have identified much lower levels of DAPK1 promoter hypermethylation than would have been expected of the observed extent of transcriptional repression in AML (16, 41–43, 46, 47). This is in contrast to MLL-r infant ALL, where the DAPK1 repression was associated with its promoter hypermethylation (19). In this study, we showed that, for the subset of AMLs characterized by Flt3ITD, often with additional NPM1 and/or IDH1/2 or DNMT3 mutation, upstream signals may provide the stimulus for p52NF-κB, and its partner relB, to attract HDACs and possibly other repressors to specific gene promoters to enforce a repression (12, 47).

The Flt3-/TAK1-dependent activation of IKK's/NF-κB and JNK1 that we identified in this study is pertinent to activation of these latter 2 crucial downstream tumor effectors. In addition, TAK1 may bear upon other antiapoptotic participants, including AMP kinase, which is activated by TAK1 for additional antiapoptotic function in the TRAIL pathway (48, 49). Although TAK1 has more frequently been associated with the canonical NF-κB pathway rather than the NIK-dependent noncanonical pathway we identified, there is ample precedent that TAK1 can activate NIK, and that TAK1/TRAF6-ubiquitin conjugates activate IKKα as well as IKKβ (28, 50).

Furthermore, the data we report has implications for the design of a mechanistically driven therapeutic regimen-to inhibit Flt3ITD⁺ AML. DAPK1 suppression in context of Flt3ITD activation of p52NF-κB implies that 1 criterion for selection of tyrosine kinase inhibitors, classified as Flt3 inhibitors, is in their ability to inhibit p52NF-κB activation in this subgroup of AMLs. In turn, HDAC inhibitors contribute independently toward DAPK1 de-repression for its participation in ER stress apoptosis. Bortezomib is also known to inhibit steps required for proteasomal activation of p52NF-κB (51) This indicates a synergistic potential for combining these drug classes in “targeted” therapy of Flt3ITD⁺ AML (Fig. 5). Indeed, we have both in vitro and in vivo evidence that such an approach may enforce apoptosis via enhanced ER stress following DAPK1 de-repression in association with p52NF-κB depletion by the combination of a Flt3 inhibitor and an HDAC inhibitor in Flt3ITD⁺ AML (H Sayar et al., manuscript in preparation).

Finally, the enzymatic activity of the co-oncogene peptidyl prolylisomerase 1 (PIN1), required for tyrosine kinase-mediated JNK/c-Jun signaling in breast cancer and AML, is abrogated by DAPK1-mediated phosphorylation on serine 71 of PIN1 (52–54). Thus, the pathophysiologic significance of DAPK1 repression in these malignancies is further emphasized, along with the therapeutic strategy for combining tyrosine kinase and HDAC inhibitors.

No potential conflicts of interest were disclosed.

R. Shanmugam and P. Gade planned and carried out experiments, analyzed results, and wrote the manuscript. A. Wilson-Weekes planned and carried out experiments and analyzed results. S. Gupta planned and carried out experiments. H. Sayar screened patients and edited the manuscript. A. Suvannasankha analyzed results and edited the manuscript. C. Goswami, L. Li, A.A. Cardoso, and T. Al Baghdadi analyzed results and reviewed the manuscript. K.J. Sargent and L.D. Cripe screened patients and reviewed the manuscript. D.V. Kalvakolanu planned experiments, analyzed results, and wrote the manuscript. H.S. Boswell designed the study, planned and carried out experiments, analyzed results, and wrote the manuscript.

This work was supported in part by the Department of Veterans Affairs Merit Review Award (H.S. Boswell), and stipend in memory of Dr. Gary D. Tollefson. D.V. Kalvakolanu is supported by NIH grant CA78282.

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.