Relationship of expression of aquaglyceroporin 9 with arsenic uptake and sensitivity in leukemia cells

Arsenic trioxide (As₂O₃) is a standard treatment for acute promyelocytic leukemia (APL)^1-3  and is showing promise in multiple myeloma and other hematologic malignancies.^4,5  The intracellular actions of As₂O₃ include interaction with proteins possessing high cysteine and histidine contents,⁶  generation of superoxides and reactive oxygen species,^7,8  disruption of the mitochondrial transmembrane potential with release of cytochrome c and caspases activation, and blockade of cell cycle at the G₁/S and G₂/M phases.⁹ 

As₂O₃ is detoxified by the glutathione redox system, comprising thiol-rich glutathione (GSH) carrier proteins, glutathione peroxidase, and glutathione S-transferase (GST-π).^9-12  GSH forms a transient As(GSH)₃ complex with As³⁺, a process catalyzed by GST-π.¹¹  As(GSH)₃ binds the multi-drug–resistant protein 1 (MRP1), inducing a conformational change.^13,14  This in turn leads to an ATP-driven transmembrane transport of As(GSH)₃ to the extracellular space.^13,14  Free reduced GSH carrier protein is then replenished by glutathione peroxidase.

The arsenic GSH redox system might be one of the mechanisms that account for the varying sensitivities of different cell types to As₂O₃. The sensitivities to As₂O₃ had been shown to be inversely proportional to the cellular GSH content.¹⁵  Furthermore, cell lines treated with ascorbic acid and buthionine sulfoxide (BSO), which decreased intracellular GSH levels, had increased sensitivity to As₂O₃.¹⁵  The role of MRP1 in arsenic sensitivity is more controversial. Overexpression of MRP1 had been shown to confer resistance to As₂O₃ in lung cancer cells and the HeLa cell line.¹⁶  However, an HL-60 variant, HL-60/AR, that overexpressed MRP1 did not show increased arsenic sensitivity.¹⁷ 

The transmembrane protein aquaglyceroporin 9 (AQP9) was first identified in adipose tissue and leukocytes.^18,19  It is also expressed in the liver, lung, and spleen.²⁰  AQP9 facilitates the passage of water and glycerol, as well as a large variety of small noncharged solutes (caramides, polyols, purine, pyrimidine, and protonated monocarboxylates).²⁰  Recently, AQP9 was shown to restore arsenic sensitivity to a Saccharomyces cerevisiae mutant that lacked FPS1, the homologue of AQP9.²¹  Furthermore, Xenopus laevis oocytes microinjected with AQP9 cDNA exhibited increased uptake of As³⁺ ions.²¹  Although these results showed that AQP9 facilitated arsenic transport in yeast and Xenopus, further investigations are required to determine the function of AQP9 in arsenic transport in mammalian cells.

In this study, we tested the hypothesis that AQP9 mediated arsenic uptake in leukemic cells and might be a determining factor in arsenic sensitivity.

One APL (NB4), 2 myeloid leukemia (HL-60, ML-1), 1 chronic myeloid leukemia (K562), 1 B-cell acute lymphoblastic leukemia (ALL; NALM-20), 1 non-B non-T ALL (NALM-16), and 5 T-cell leukemia (HPB-ALL, JM, JURKAT, KE-37, MOLT-4) cell lines were studied. They were cultured in RPMI-1640 supplemented with 10% fetal bovine serum (FBS) in 5% CO₂ at 37°C. Cells were treated with As₂O₃ (Sigma-Aldrich, St Louis, MO) and with all-trans retinoic acid (ATRA; Sigma-Aldrich) in selected experiments.

Cells were washed twice with phosphate-buffered saline (PBS) and resuspended in fresh RPMI-1640 medium supplemented with 10% FBS and various concentration of As₂O₃ (0, 0.5, 1.0, 2.0, and 4 μM) at 2.5 × 10⁵ cells/mL. Quadruplicates of 100 μL of each arsenic/cell suspension were incubated for 48 hours. MTT reagent (100 μL) was then added and incubated for 4 hours, followed by the addition of 100 μL of solubilizing buffer and incubation overnight. The absorbance at 560 nm was then measured.

Leukemic cells were washed once in PBS supplemented with Complete protease inhibitor (Roche, Manheim, Germany). Washed pellets (4 × 10⁷ to 5 × 10⁷ cells) were resuspended in 5 mL lysis buffer (7.5 mM sodium phosphate, pH 7.0; 0.25 M sucrose; 5 mM EDTA with Complete protease inhibitor; and 1 mM phenylmethylsulphonylfluoride), sonicated on ice for 60 J, and centrifuged at 1000g for 10 minutes at 4°C. The crude membrane protein fraction was then pelleted by ultracentrifugation (200 000g for 90 minutes at 4°C), resuspended in 150 μL buffer (5.0% SDS; 20 mM Tris, pH 8.0; and 5 mM EDTA with Complete protease inhibitor and 1 mM PMSF), sonicated, and incubated at 37°C for 30 minutes. Protein concentrations of each sample were determined (DC protein assay kit II; Bio-Rad, Hercules, CA) and adjusted to 2 μg/μL in lysis buffer and 5 × sample buffer (15% SDS; 50 mM Tris, pH 6.8; 30% glycerol; 0.05% bromophenol blue; and 0.5 M dithiothreitol). Protein samples were boiled at 95°C for 5 minutes and kept at 37°C, and 30 μg was resolved in 12% acrylamide gel with 4 M urea, electrophoretically transferred to PVDF membrane (Amersham Biosciences, Piscataway, NJ), blocked with 5% low-fat milk for 1 hour, incubated with rabbit anti-AQP9 antibody (3.5 μg/mL) or rabbit anti–enhanced green fluorescent protein (EGFP) antibody (1:10 000; Chemicon, Temecula, CA), and probed with an antirabbit secondary antibody (1:1000; Dako Cytomation, Glostrup, Denmark). Immunoblotting was evaluated after chemiluminescence development (Western Lightning Chemiluminescence Reagent; PE Biosystems, Foster City, CA), followed by PointSau staining (Sigma-Aldrich) for normalization of protein loading.

AQP9 expression was quantified by quantitative polymerase chain reaction (Q-PCR). Briefly, the AQP9 primers (forward, 5′ AGT TGT TGG GAG CCT TTG TG 3′; reverse, 5′ GTT CGC CAG AGA TAG ATA CGG AG 3′) and the TaqMan probe (5′ CAA CAG CAC ACA TTT TTG CAA CAT ACC 3′) detecting a 155-bp amplicon were designed by Primer Express software (PE Biosystems). The primers (forward, 5′ GAA GGT GAA GGT CGG AGT CA 3′; reverse, 5′ CTT CTA CCA CTA CCC TAA AG 3′) and TaqMan probe (5′ GTT CGA AGG GCA AGA GTC GG 3′) detecting a 155-bp amplicon for the internal control gene GAPDH were similarly designed. For Q-PCR, 100 ng of RNA was added to 45 μL of Taqman EZ reverse transcription–PCR (RT-PCR) reaction mixture (1 × Taqman EZ buffer, 3 mM MnOAc, 300 μM of dNTP mixture, 200 nM of primers and probe, 0.1U rTth DNA polymerase, 0.01U uracil N-glycosylase). PCR conditions were 50°C for 2 minutes, 60°C for 30 minutes, 95°C for 5 minutes, and 40 cycles of 94°C for 20 seconds and 60°C (AQP9)/62°C (GAPDH) for 1 minute. Q-PCR data were collected continuously and analyzed by an automated sequence analyzer (ABI PRISM 7700; PE Biosystems). Expression of the multi-drug–resistance 1 (MDR1) gene was quantified as previously reported.²²  The expression levels of AQP9 among the 11 cell lines were analyzed using the comparative C_T (ΔC_T) method (User Bulletin no. 2; PE Biosystems),²³  with GAPDH as the internal reference gene to control for RNA quantity and integrity and NB4 cells as a calibrator to control for interexperimental variations. All samples were quantified in triplicates. Quality assurance of Q-PCR was as previously described.²³ 

To construct an EGFP-tagged AQP9 expression plasmid, the complete coding sequence of AQP9 (888 bp) was cloned by PCR using NB4 cell line cDNA as a template with primers (forward, 5′ CCGCCG AGA TCT CCA TGC AGC CTG AGG GAG CA 3′; reverse, 5′ CAGCAG GGA TCC CTA CAT GAT GAC A 3′) containing BglII and BamHI restriction sites (underlined), respectively. The PCR product was cloned into the pEGFP-C2 vector (BD Biosciences Clontech, Mountain View, CA) to give the plasmid pEGFP-AQP9. The nucleotide sequence of AQP9 in pEGFP-AQP9 was analyzed by DNA sequencing to confirm correct nucleotide sequence and in-frame insertion of AQP9 with EGFP. The plasmid pSV-βgal, containing the β-galactosidase gene, was purchased from Promega (Madison, WI).

Hep3B cells in MEM cultured to 80% confluence were transfected with pEGFP-C2/pSV-βgal or pEGFP-AQP9/pSV-βgal in Lipofectamine 2000 (Invitrogen, Carlsbad, CA). In order to ensure that cells transfected with pSV-βgal would also express AQP9, a 1:5 ratio of pSV-βgal to pEGFP-AQP9 was used. Hence, β-galactosidase activity could be used as an indicator of viability of cells expressing both β-gal and EGFP-C2 or EGFP-AQP9. The transfectants, designated as Hep3B_{EGFP-AQP9/β-gal} and HEP3B_{EGFP-C2/β-gal}, were further incubated for 48 hours. The cells were then treated with As₂O₃ at various concentrations (0, 0.5, 1.0, and 2.0 μM) for 24 hours. For assaying of β-galactosidase activity, transfected Hep3B cells were washed with PBS, lysed with reporter lysis buffer (Promega) for 15 minutes at room temperature, transferred to Eppendorf tubes on ice, vortexed for 20 seconds, and centrifuged at 1000g at 4°C for 1 minute. Cell lysates were mixed with an equal volume of 2 × assay buffer and incubated at 37°C for 1 to 2 hours, after which the reaction was stopped by 1 M sodium bicarbonate. Absorbance at 420 nm was then recorded. Untransfected cells not treated with As₂O₃ served as 100% viability control. All samples were tested in triplicates.

K562 cells at mid-log phase were transfected with pEGFP-AQP9 or pEGFP-C2 as described for Hep3B cells. The K562 transfectants, designated as K562_EGFP-AQP9 and K562_EGFP-C2, were washed once with PBS and then resuspended in RPMI-1640 medium supplemented with 500 μg/mL of G418 (Calbiochem, San Diego, CA) at 1 × 10⁶ cells/mL. Cells resistant to G418 were selected for 3 weeks. To further enrich for transfected cells, sorting for green fluorescent cells (K562_EGFP-AQP9 and K562_EGFP-C2) was performed by an EPICS ALTRA flow cytometer (Beckman Coulter, Fullerton, CA) until a purity of 80% or higher was achieved.

K562_EGFP-AQP9 and K562_EGFP cells at 1 × 10⁶ cells/mL were washed twice with PBS; resuspended in fresh RPMI-1640 medium supplemented with 10% FBS and 1 μM As₂O₃; and incubated in 5% CO₂ at 37°C for 0, 30, 60, 90, and 120 minutes. Cells were then harvested into 1 mL ice-cold PBS and pelleted. After 2 washes in ice-cold PBS, the cell pellets were lysed in 0.9 mL double-distilled water by sonication for 10 minutes. HNO₃ (2%) with 10 parts per billion (ppb) of yttrium (Chem Service, West Chester, PA) was then added as an internal standard. Samples were vortexed and centrifuged at 3000g for 10 minutes, and the supernatants were collected and assayed for arsenic concentration by inductively coupled plasma–mass spectrometry (ICP-MS) as previously described.²⁴ 

Archival bone marrow aspirate samples from patients with APL and other subtypes of acute myeloid leukemia (AML) were quantified for AQP9 expression. Leukemic diagnoses were based on the World Health Organization classification system.²⁵  The blast cells were isolated by centrifugation, snap-frozen, and stored at −80°C. For quantification of AQP9 expression, leukemic samples were thawed, total RNA was extracted, and Q-PCR for AQP9 was performed, with the cell line NB4 used as the calibrator in the ΔC_T method, using GAPDH as the internal reference. The procurement of materials was approved by the institutional review board at Queen Mary Hospital (Hong Kong, China). Patients gave informed consent and the study was conducted in accordance with the Declaration of Helsinki.

Continuous variables were expressed as mean ± standard error of the mean. The correlation between cell viability and AQP9 expression was tested by Pearson correlation test (SPSS software, version 10.0; Chicago, IL). Comparison between groups was performed by Student t test. P values of less than .05 were considered to be statistically significant.

MTT assays were performed on the 11 cell lines tested. The sensitivities of the cell lines differed, although all showed a dose-dependent increase in As₂O₃ cytotoxicity. Of the 11 cell lines, Jurkat, HL-60, and NALM-20 were most resistant, whereas NB4 was most sensitive (Figure 1; Figure S1, available on the Blood website; see the Supplemental Materials link at the top of the online article).

In the 11 leukemia cell lines tested, the expression of AQP9 as determined by Q-PCR was found to be positively correlated with As₂O₃-induced cytotoxicity as assessed by MTT (P = .039; Figure 2A). The results were confirmed by Western-blot analysis (Figure 2B), showing a direct relationship between AQP9 protein expression and As₂O₃ sensitivity (P = .001; Figure 2C). However, the cell line ML-1 was an obvious outlier in both experiments. Finally, there was no relationship between AQP9 and MDR1 gene expression (P = .09, Pearson correlation; Table S1).

As ML-1 showed an apparent high level of AQP9 expression but a low As₂O₃ sensitivity, the entire coding region of AQP9 in ML-1 was amplified by PCR and sequenced. The results showed a homozygous single base substitution, 835G>A, which resulted in an amino-acid change from alanine to serine at codon 279. Whether this might account for the discordant AQP9 expression and As₂O₃ sensitivity is under further investigations.

Fluorescence microscopic examination of the Hep3B and K562 AQP9 transfectants showed a difference in distribution of fluorescence. In Hep3B_EGFP-C2 and K562_EGFP-C2, the green fluorescence was evenly distributed in the cytosol. However, in Hep3B_EGFP-AQP9 and K562_EGFP-AQP9, the fluorescence was mainly distributed on the cytoplasmic membrane (Figure 3A). The results confirmed selective localization of AQP9 to the plasma membrane. Western blotting was performed to confirm protein overexpression in the transfectants. In K562_EGFP-C2 and Hep3B_{EGFP-C2/β-gal}, an EGFP protein band of around 30 kDa was detected with an anti-GFP antibody (Figure 3B-C). In Hep3B_{EGFP-AQP9/β-gal} and K562_EGFP-AQP9 lysates, the EGFP-AQP9 fusion protein was detected as a 61-kDa band with antibodies specific to GFP or AQP9, representing a summation of the 30-kDa EGFP and the 31-kDa AQP9 protein (Figure 3B-C). An additional band of approximately 30 kDa was also detected in both the anti-GFP and anti-AQP9 blotting of the K562_EGFP-AQP9 lysate, which might be due to the degradation of the EGFP-AQP9 fusion protein to EGFP (30 kDa) and AQP9 peptides (31 kDa).

The Hep3B hepatoma line did not express any detectable amount of AQP9 (Figure 3B) and was insensitive to As₂O₃. In Hep3B_EGFP/β-gal and Hep3B_{EGFP-AQP9/β-gal} transfectants, the β-galactosidase activity was used as an indicator of cell viability after treatment with various concentrations of As₂O₃. As shown in Figure 4, Hep3B_EGFP/β-gal cells, similar to the parental Hep3B cells, were resistant to As₂O₃. On the other hand, As₂O₃-treated Hep3B_{EGFP-AQP9/β-gal} cells demonstrated a significantly lower β-galactosidase activity compared with Hep3B_EGFP/β-gal (Figure 4), showing that overexpression of AQP9 increased As₂O₃-induced cytotoxicity. Interestingly, the difference of viability of Hep3_BEGFP/β-gal and Hep3B_{EGFP-AQP9/β-gal} remained constant with increasing concentrations of As₂O₃.

The K562 cells expressed very low levels of AQP9 (Figure 3B) and were insensitive to As₂O₃. The K562_EGFP-C2 transfectant was also insensitive to As₂O₃ (Figure 5A). However, the K562_EGFP-AQP9 transfectants showed a significantly increased sensitivity to As₂O₃ (Figure 5A). Furthermore, in contrast to Hep3B AQP9 transfectants, the difference of viability of K562_EGFP-C2 and K562_EGFP-AQP9 widened with increasing concentrations of As₂O₃. The increase in As₂O₃ sensitivity could be explained by an increased As³⁺ uptake mediated by the overexpressed AQP9. As shown in Figure 5B, the K562_EGFP cells accumulated minimal amounts of arsenic after incubation in 1 μM As₂O₃ for 2 hours. On the other hand, the K562_EGFP-AQP9 showed a time-dependent increase in arsenic accumulation, reaching a significantly higher amount of arsenic (median, 65 pmol/10⁶ cells) after 2 hours.

ATRA acts synergistically with As₂O₃ in vitro and in vivo. The HL-60 cells expressed very low levels of AQP9 (Figure 2A-C) and were insensitive to As₂O₃. However, treatment with ATRA led to a dose-dependent increase in AQP9 gene transcription and protein expression (Figure 6A). To assess the biologic significance of AQP9 up-regulation, HL-60 cells were pretreated with ATRA (100 nM for 48 hours), washed twice in RPMI to clear the ATRA, and treated with As₂O₃. ATRA-pretreated HL-60 cells showed an increased sensitivity to As₂O₃ compared with untreated HL-60 cells (Figure 6B). The increase in As₂O₃ sensitivity was due to increased arsenic uptake into the cells. Incubation of ATRA-pretreated HL-60 cells in 1 μM As₂O₃ led to a time-dependent increase in arsenic accumulation, which was significantly higher than in untreated HL-60 cells (Figure 6C).

Leukemic samples from 80 patients were studied. These included 19 patients with APL (presentation, 5; relapse, 14), 42 patients with AML (minimally differentiated, 3; without maturation, 11; with maturation, 10; myelomonocytic, 5; monoblastic, 1; erythroid, 1; megakaryoblastic, 1; multilineage dysplasia, 10), and 19 patients with ALL. The mean level of AQP9 was significantly higher in APL compared with all the other subtypes of AML (APL 3.07 × 10⁵ ± 2.08 × 10⁵ versus AML 8.10 × 10² ± 2.26 × 10², P = .027; Figure 7A). Interestingly, in subgroup analysis, when APL was compared with AML with maturation (or AML M2 in the French-American-British classification system²⁶ ; n = 10), the AQP9 expression was similar (APL 3.07 × 10⁵ ± 2.08 × 10⁵ versus AML M2 2.14 × 10³ ± 0.72 × 10³, P = .75; Figure 7B). Hence, when APL was compared with AML subtypes other than M2 (n = 32), AQP9 expression was even more significantly different (3.07 × 10⁵ ± 2.08 × 10⁵ versus 3.93 × 10² ± 1.37 × 10², P = .009; Figure 7B). However, AQP9 expression in ALL was comparable with AML (ALL 2.05 × 10⁴ ± 1.94 × 10⁴ versus AML 9.62 × 10⁴ ± 6.63 × 10⁴, P = .6; Figure 7C).

Our findings showed that AQP9 played a critical role in the transmembrane transport of As₂O₃. In neutral pH, As³⁺ exists primarily as As(OH)₃, which might be recognized by AQP9 as a small neutral noncharged solute similar to glycerol for facilitated uptake.²⁷  Accordingly, we showed that in leukemia cells of different lineages, in vitro sensitivity to As₂O₃ was directly proportional to AQP9 expression. The results were confirmed both at the mRNA and protein level. The ML-1 cell line was an exception in that although AQP9 was expressed apparently at high levels, the line was relatively resistant to As₂O₃. Analysis of the AQP9 coding sequence showed a point mutation resulting in a base substitution. Whether this base substitution is of functional significance is being investigated. An alternative explanation for the poor As₂O₃ sensitivity would be different intracellular constituents or mechanisms that are distinctive to this cell line. As the MDR1 gene encoding P-glycoprotein also controlled transmembrane drug trafficking, we investigated if the expression of AQP9 and MDR1 genes might be related. Q-PCR showed that their expressions were unrelated. This was consistent with previous observations that although the MDR1 gene might be up-regulated in relapsed APL, the response to As₂O₃ was unaffected, suggesting that MDR1 did not contribute to As₂O₃ sensitivity.²²  Lastly, for leukemia cell lines of different lineages, the APL cell line NB4 showed the highest expression level of AQP9 and the most exquisite sensitivity to As₂O₃. Interestingly, APL is also clinically the leukemic subtype that responds best to As₂O₃ therapy.

To confirm the role of AQP9 in As₂O₃ entry and sensitivity, a pertinent investigation will be to express AQP9 in an AQP9-negative cell line and to study the response to As₂O₃ before and after AQP9 expression. Furthermore, transient transfection may be preferable to avoid inadvertent introduction of additional properties acquired unexpectedly during the selection of permanent transfectants. However, a leukemic line with undetectable AQP9 expression was not available. Moreover, the very low transfection-efficiency characteristic of leukemia cells in suspension would make transient transfection and subsequent experiments difficult. We therefore had elected to choose the Hep3B line, which were adherent cells and did not express any detectable AQP9, for this proof-of-principle experiment. In Hep3B cells, the transfected EGFP-AQP9 fusion protein showed the appropriate membrane localization. The selective cell-membrane partition of AQP9 may be due to the hydrophobicity of the AQP9 moiety or the presence of a membrane-localizing motif in AQP9. Besides accurate localization, the function of AQP9 in facilitating As₂O₃ uptake was also preserved. This was reflected in an increase in sensitivity to As₂O₃ in the Hep3B_EGFP-AQP9 transfectant. The intracellular arsenic was not measured because harvesting the Hep3B cells required trypsinization, which partly permeabilized the cell membrane and resulted in a substantial loss of arsenic.

To further substantiate these observations in leukemia cells, the K562 line was chosen, as it had very low endogenous AQP9 expression and hence was insensitive to As₂O₃. Due to the low transfection efficiency, stable transfection coupled with flow sorting of fluorescent cells was needed to enrich for AQP9-transfected cells. The stable K562_EGFP-AQP9 transfectants showed results similar to those observed in transient transfection of Hep3B, in that K562_EGFP-AQP9 also demonstrated an increased As₂O₃ sensitivity. Moreover, we were able to show that the higher As₂O₃ sensitivity was due to increased intracellular arsenic accumulation. These observations confirmed that AQP9 facilitated transmembrane transport of arsenic.

Both Hep3B_{EGFP-AQP9/β-gal} and K562_EGFP-AQP9 exhibited a higher arsenic sensitivity than Hep3B_EGFP/β-gal and K562_EGFP-C2. However, the difference in viability of K562_EGFP-AQP9 and K562_EGFP cells increased progressively with higher concentrations of As₂O₃, whereas the difference of viability of Hep3B_{EGFP-AQP9/β-gal} and Hep3B_EGFP/β-gal remained comparable at different concentrations of As₂O₃. Therefore, it might be possible that AQP9-mediated arsenic transmembrane transport could be saturable, depending on the amount of AQP9 expressed. Alternatively, in addition to the amount of intracellular arsenic, other intrinsic properties of the cell may also determine its sensitivity to As₂O₃.

An important prerequisite for putting these observations into therapeutic applications is finding clinically feasible methods for up-regulation of AQP9 in leukemia cells. A synergistic interaction between ATRA and As₂O₃ has been observed in vitro.²⁸  Furthermore, patients with APL at presentation showed a more efficient leukemic load reduction if treated concomitantly with ATRA and As₂O₃.²⁹  Patients with APL at presentation or relapse also responded better to combined ATRA and As₂O₃ treatment.^30-34  In the myeloid leukemia line HL-60 we showed that ATRA pretreatment up-regulated AQP9 expression, which was translated into an increase in arsenic uptake and hence sensitivity to As₂O₃ treatment. As the concentration of ATRA used was as low as 100 nM, which would not induce any significant apoptotic effect on HL-60, the increased cytotoxicity after As₂O₃ treatment could be attributable to the demonstrable increase in arsenic uptake. These observations provided another mechanistic explanation for the synergism between ATRA and As₂O₃. We had also conducted similar experiments with the APL cell line NB4. Unfortunately, NB4 was not a good in vitro model. This is because NB4 cells already express very high levels of AQP9 and are extremely sensitive to ATRA and As₂O₃, even at very low concentrations. Therefore, the exogenous overexpression of AQP9 did not give consistent results of further increases in arsenic uptake or cytotoxicity (data not shown).

To verify the relevance of AQP9 expression in primary leukemia samples, we quantified AQP9 in a series of AML samples. We showed that APL expressed AQP9 significantly (about 2-3 log) more than other subtypes of AML. This observation was consistent with the high sensitivity of APL to As₂O₃ compared with other AML subtypes. Interestingly, in AML with maturation (AML M2), where the leukemic cells show at least 10% granulocytic maturation, the AQP9 expression was comparable to APL. These results suggested that AQP9 expression might be related to myeloid maturation, with promyelocytes and myelocytes/granulocytes showing the highest expression of AQP9. However, not all APL cases expressed high levels of AQP9. As these cases still responded to As₂O₃ much more than AML with a similar AQP9 expression, intracellular mechanisms other than membrane AQP9 expression may also determine As₂O₃ sensitivity. Finally, ALL cells expressed AQP9 levels comparable to AML, implying that the role of As₂O₃ in ALL treatment might be explored.

Our observations provide useful leads in the research of the therapeutic use of As₂O₃ in leukemic treatment. The expression level of AQP9 may be one of the molecular determinants predicting the response to As₂O₃ therapy, although expectedly other cellular mechanisms are also important. Whether AQP9 might also affect other chemotherapeutic agents remains to be determined by experiments measuring drug entry/efflux and AQP9 expression. Secondly, investigations of the transcriptional regulation of the AQP9 gene will be clinically relevant, particularly in AML without granulocytic maturation. Previous studies have shown that steroid hormones and vitamin D₃ may induce AQP9 expression.^35-38  As the promoter of AQP9 does not contain hormone-responsive elements, the actions of these lipophilic agents may be due to up-regulation of other transcription factors.²⁰  In fact, the AQP9 promoter contains binding sites for C/EBP, NF-κB, and AP-1.²⁰  It is intriguing to note that C/EBP is inducible by retinoids such as ATRA.^39,40  Therefore, the definition of agents able to induce AQP9 expression in different cancer cell types may hold promise for increasing the uptake of As₂O₃ and hence its efficacy in these malignancies. Moreover, the potential contribution of AQP9 dysregulation or mutation to As₂O₃ resistance needs further investigation. Finally, the organ toxicity of arsenic may also be related to AQP9 expression. In fact, AQP9 is expressed normally at high concentrations in the liver and brain, organs that are most affected by arsenic therapy.⁶  Agents that may selectively down-regulate AQP9 in these organs may ameliorate arsenic toxicity, which is an important limiting factor in the therapeutic use of As₂O₃. These propositions will require future validations.

This work was supported in part by the S. K. Yee Foundation.

The authors thank W. I. Wong for valuable technical assistance and M. F. Yuen for advice on statistical analysis.