Overexpression of P-glycoprotein, encoded by the MDR1 (multidrug resistance 1) gene, is often responsible for multidrug resistance in acute myeloid leukaemia. We have shown previously that MDR1 (P-glycoprotein) mRNA levels in K562 leukaemic cells exposed to cytotoxic drugs are up-regulated but P-glycoprotein expression is translationally blocked. In the present study we show that cytotoxic drugs down-regulate the Akt signalling pathway, leading to hypophosphorylation of the translational repressor 4E-BP [eIF (eukaryotic initiation factor) 4E-binding protein] and decreased eIF4E availability. The 5′-end of MDR1 mRNA adopts a highly-structured fold. Fusion of this structured 5′-region upstream of a reporter gene impeded its efficient translation, specifically under cytotoxic stress, by reducing its competitive ability for the translational machinery. The effect of cytotoxic stress could be mimicked in vivo by blocking the phosphorylation of 4E-BP by mTOR (mammalian target of rapamycin) using rapamycin or eIF4E siRNA (small interfering RNA), and relieved by overexpression of either eIF4E or constitutively-active Akt. Upon drug exposure MDR1 mRNA was up-regulated, apparently stochastically, in a small proportion of cells. Only in these cells could MDR1 mRNA compete successfully for the reduced amounts of eIF4E and translate P-glycoprotein. Consequent drug efflux and restoration of eIF4E availability results in a feed-forward relief from stress-induced translational repression and to the acquisition of drug resistance.
Multidrug resistance, the phenomenon by which cancer cells treated with a single chemotherapeutic agent become cross-resistant to multiple functionally and structurally unrelated compounds, is most frequently associated with up-regulation of P-glycoprotein. This membrane protein, coded by the MDR1 (ABCB1; multidrug resistance 1) gene, is an ATP-dependent efflux pump that maintains intracellular drug concentrations below cytotoxic levels .
Previously we have used a K562 myelogenous leukaemia cell model to determine the mechanisms by which MDR1/P-glycoprotein-mediated multidrug resistance is acquired , with the aim of developing therapeutic strategies that will target up-regulation of P-glycoprotein expression. The K562 cell line is a useful model system as drug-resistant sub-lines can readily be derived by selection with clinically relevant doses of chemotherapeutic agents . Short-term (1–3 days) exposure of naïve K562 cells to a range of cytotoxic agents results in up-regulation of MDR1 mRNA steady-state levels by stabilization of the otherwise short-lived MDR1 mRNA. However, this stabilized mRNA is not associated with translating polyribosomes and P-glycoprotein is not synthesized. Selection for drug resistance, by long-term exposure to drugs, leads to resistant cells in which this translational block is overcome. Translational repression of MDR1 mRNA has also been described in EBV (Epstein–Barr virus)-transformed human B lymphocytes and in leukaemic CEM cells [2,3], and the well-documented lack of correlation between MDR1 mRNA and P-glycoprotein levels in leukaemia  implies that translational regulation represents a general regulatory step in P-glycoprotein expression.
The rate-limiting step in translation is initiation and is the most common target for translational control. For the majority of eukaryotic mRNAs, initiation of protein synthesis occurs via a cap-dependent mechanism . The initial binding of the translation apparatus at the 5′-end of an mRNA begins with the formation of a ternary complex [eIF (eukaryotic initiation factor)2, GTP and a tRNA initiator] which is recruited by the small ribosomal subunit (40S) together with its associated initiation factors to form the 43S pre-initiation complex which recognizes the 5′-cap of an mRNA. This requires the eIF4F protein complex which comprises eIF4E (which recognises and binds to the 5′-cap structure), eIF4A and eIF4G. Since the levels of eIF4E are normally low, assembly of the eIF4F complex and its interaction with the 5′-mRNA cap provides the rate-limiting step in translation initiation. Once bound to the mRNA cap, the 43S pre-initiation complex migrates along the 5′-UTR (untranslated region) in an ATP-dependent process known as scanning, until it reaches the translation initiation codon (AUG) [5,6]. For a proportion of cellular mRNAs, cap-independent translation involves internal ribosomal entry sites .
Several features of an mRNA molecule can influence its translation . Loop structures involving the 5′-UTR of mRNAs with free energy values of less than between −30 and −50 kcal/mol, generally inhibit translation . In vitro RNase mapping has shown that the 5′-terminal region of MDR1 mRNA (nucleotides 1–289) is highly structured . This 289 nucleotide region includes the 140 nucleotide 5′-UTR which folds with the following 149 nucleotides of the MDR1-coding sequence to form a stable structure, with the initiation codon located in a hexanucleotide loop at its centre (Figure 1A). In the present study we show that the highly structured region at the 5′-end of MDR1 mRNA affects translation of reporter transcripts and that this down-regulation is more pronounced under cytotoxic stress, when the Akt pathway is down-regulated and available levels of eIF4E are low. Drug resistance is acquired by up-regulation, by an apparently stochastic mechanism, of MDR1 mRNA in a small subpopulation of cells. This allows effective competition for eIF4E, translation of P-glycoprotein, and consequent removal of cytotoxic drugs from the cell therefore restoring eIF4E availability. This positive feed-forward cycle leads to the development and maintenance of multidrug resistance.
The 5′-end of MDR1 mRNA is highly structured
Human myelogenous leukaemic K562 cells and their colchicine-resistant derivative KC40, have been described previously . KC40 cells were routinely maintained in the presence of 20 ng/ml colchicine. Cells were transfected using an Amaxa Nucleofector, following the manufacturers' protocol. For transient transfections, 4.5 μg of the test plasmid was co-transfected with 0.5 μg phRGTK (expressing Renilla luciferase to normalize for transfection efficiency; Promega). For stable transfections, plasmids (2 μg) were linearized with PvuI, co-transfected as above with 0.1 μg pGK-puro, and transfectants were selected with 2 μg/ml puromycin.
Generation of drug-resistant K562 cells
Cells were seeded at 5×105 cells/ml in 50 ml of RPMI 1640 culture medium containing 20 ng/ml colchicine. At days 3, 10 and 14, 50 ml of fresh culture medium without drug was added. At day 18 cells were spun down (100 g for 10 min at 25 °C) and resuspended in 160 ml of RPMI 1640 culture medium with 20 ng/ml colchicine. From day 25 to 40 the colchicine-containing medium was routinely changed twice per week.
The constitutively active Akt expression plasmid (pCAAKT) was from Upstate Biotechnology. The constitutively active Mnk-1 [MAPK (mitogen-activated protein kinase) interacting kinase 1] expression plasmid (pCAMNK) , and the MDR1 downstream promoter-luciferase reporter plasmid pMDR1(−1202)  have been described previously. The 289 nucleotide MDR1 5′-structure (consisting of the 140 nucleotide 5′-UTR and 149 nucleotides of the proximal coding region) was amplified from reverse-transcribed KC40 mRNA using primers OLEY24bis and OLEY26 (all oligonucleotide sequences are listed in Supplementary Table 1 at http://www.BiochemJ.org/bj/406/bj4060445add.htm). The 140 nucleotide 5′-UTR was amplified with primers OLEY24bis and OLEY25. PCR products were inserted into the pGL3Control vector (Promega) upstream of the luciferase gene, using the NcoI and EcoRI sites, generating pGL-Loop and pGL-UTR respectively. The luciferase translation initiation codon within pGL-UTR was then mutated to that found in the native MDR1 using primers OLEY176 and OLEY177, and the QuikChange® site-directed mutagenesis kit (Stratagene), according to the manufacturer's protocol. The translation start codon and two other out of frame-AUGs in the MDR1-coding region, but upstream from the luciferase gene in pGL-Loop, were sequentially mutated by site-directed mutagenesis as described above with oligonucleotides OLEY27 and OLEY28, OLEY41 and OLEY42, and OLEY43 and OLEY44 respectively, to generate pGL-LoopM. Thus pGL-UTR, pGL-Loop and pGL-LoopM had the same 5′-UTR length and sequence surrounding the translation initiation codon. T7 promoter sequences, produced by annealing the complementary oligonucleotides OLEY203 and OLEY204, were inserted into EcoRI-digested pGL-UTR and pGL-Loop to generate T7pGL-UTR and T7pGL-Loop respectively. eIF4E cDNA was obtained from reverse transcribed K562 mRNA by PCR using the Expand High Fidelity PCR system (Roche Diagnostics) and primers OLEY184 and OLEY185. The resulting 1391 bp DNA fragment was inserted into the pEF6/V5-His-TOPO plasmid vector (Invitrogen) to generate peIF4E.
Genomic DNA and Southern blotting
Genomic DNA was digested with MfeI (for luciferase) or EcoRI (for β-globin), blotted and hybridized to radioactive probes, derived from pGL3control (luciferase-coding sequence) or the β-globin locus as described previously .
A Dual-Luciferase reporter assay (Promega) was used to measure both firefly and Renilla luciferase in lysates of transiently transfected cells 16 h after transfection. Protein content was measured using a DC protein assay (Bio-Rad).
De novo protein synthesis
Incorporation of [35S]methionine into nascent proteins was determined by trichloroacetic acid precipitable counts as described previously .
Luciferase mRNA analysis was performed by semi-quantitative RT (reverse transcriptase)-PCR. Briefly, after reverse transcription , cDNA was amplified by PCR with OLEY218 and OLEY219 for luciferase, and OLEY8 and OLEY70 for GAPDH normalization  for 20–29 cycles. Detection of amplified cDNA was by Southern blotting and hybridization to radioactively end-labelled internal oligonucleotides, essentially as described previously . Quantification was with a PhosphorImager (Molecular Dynamics). MDR1 mRNA analysis was by reverse transcription and real-time Q-PCR (quantitative PCR), as described previously . The mRNA half-life was determined as described previously  except that Id2 mRNA turnover was determined by RT-Q-PCR with primers Id2F and Id2R (PrimerBank ID: 31982933a1)  and normalized to 18S ribosomal RNA (primers 18Sf and 18Sr).
m7GTP Sepharose pulldown
eIF4E-containing protein complexes were isolated by binding to m7GTP Sepharose as described previously .
In vitro transcription, in vitro translation and RNA competition
T7pGL-UTR and T7pGL-Loop plasmids were linearized down-stream from the luciferase stop codon and in vitro transcribed using an mMessage mMachine T7 Ultra kit (Ambion). A cap analogue and a poly(A) tail were added to the 5′- and 3′-ends respectively, following the manufacturers' protocols. In vitro translation was carried out with 0.4 μg of in vitro-transcribed mRNA and 0–20 μg of competitor RNA (isolated either from K562 cells or from yeast) in a rabbit reticulocyte cell-free system (Promega), and luciferase activity was measured after 90 min at 30 °C following the manufacturers' protocols. Where indicated, in vitro transcription and translation was also carried out using the TNT Quick Coupled Transcription and Translation Rabbit Reticulocyte System (Promega). All experiments were performed in triplicate on at least three independent occasions.
siRNA (small-interfering RNA)
siRNA (2.7 μg) targeted to eIF4E (Cell Signalling Technology) or non-specific AllStars negative control (Qiagen), were transfected into cells using an Amaxa Nucleofector Kit V. A decrease in cellular eIF4E protein levels were assayed by Western blotting 24 h after transfection.
Antibodies and flow cytometry
The 5′-MDR1 mRNA structure down-regulates translation of a reporter gene
The structure of the 5′-end of the MDR1 mRNA, nucleotides 1–289,  (Figure 1A), has a calculated free energy value of less than −100 kcal/mol (1 cal≡4.184 J) . We hypothesized that if formed in vivo, such a structure would down-regulate translation from a linked reporter. The 289 bp region and the 140 bp region corresponding to the 5′-UTR were cloned upstream from the luciferase reporter gene in a pGL3control vector, generating pGL-Loop and pGL-UTR respectively (Figure 1B). The luciferase activity of pGL-UTR was approx. 2.5-fold above pGL3control in transiently transfected K562 cells (Figure 1C). This may be due to the longer 5′-UTR sequence in pGL-UTR transcripts (138 nucleotides compared with 35 nucleotides in pGL3control mRNAs), resulting in a greater translation efficiency of the luciferase mRNA . The presence of the 289 bp structured region (pGL-Loop) resulted in a significant reduction in luciferase activity (5- to 7-fold) relative to pGL-UTR (Figure 1C), indicating that this region of MDR1 mRNA is inhibitory to reporter gene expression. Changes in luciferase activity were due to changes at the translational level and not at the level of transcription or mRNA stability since the steadystate levels of luciferase mRNA, as detected by semi-quantitative RT-PCR, were comparable (Figure 1D). Equally, the small fusion protein formed between the first amino acids of P-glycoprotein and firefly luciferase in pGL-Loop did not affect luciferase enzymatic activity (results not shown). Following site-directed mutagenesis to eliminate AUGs in the loop structure (pGL-LoopM), the predicted fold and free energy was unchanged (results not shown). Therefore we reasoned that if the loop at the 5′-end of MDR1 mRNA were formed in vivo, luciferase translation from pGL-LoopM, when compared with pGL-Loop, would be greatly reduced. Transfection of pGL-LoopM into K562 cells showed that luciferase activity was completely absent in pGL-LoopM-expressing cells (Figure 1C). This suggests that the 43S pre-initiation complex cannot scan or shunt through the loop region present in pGL-LoopM, preventing translation of the down-stream luciferase reporter. In contrast, in pGL-Loop-expressing cells, the initiation codon in the hexanucleotide loop at the centre of the structure improves the translatability of these transcripts, presumably because the 43S pre-initiation complex has to scan or shunt through a shorter region before reaching the starting AUG. Thus the 5′-end of MDR1 mRNA is highly structured in vivo and down-regulates translation of the downstream reporter gene.
The loop at the 5′-end of MDR1 mRNA specifically down-regulates translation of a luciferase reporter gene upon treatment with cytotoxic drugs
To test the effects of cytotoxic stress on the translation of transcripts with the 5′-MDR1 structure, we generated stable K562 cell lines with similar numbers of copies of pGL3control, pGL-UTR or pGL-Loop, named K-pGL3, K-pGL-UTR and K-pGL-Loop cells respectively (Figure 2A). Treatment of each of these cell lines with 1 μM colcemid for 3 days resulted in a similar decrease in the incorporation of [35S]methionine into nascent proteins (Figure 2B). Luciferase activity of K-pGL3 and K-pGL-UTR cells decreased by 25%. However, in K-pGL-Loop cells, which include the entire 5′-MDR1 structure upstream of the luciferase reporter, luciferase activity decreased by 75% (Figure 2B). This effect was due to a decrease in translation, and not transcription of the transgenes, since the steady-state levels of the corresponding mRNAs did not change significantly upon drug treatment (Figure 2C). Cytotoxic stress results in a global reduction in translation . Although the observed decrease in luciferase activity in K-pGL3 and K-pGL-UTR cells following colcemid treatment probably reflects this generalized effect, the greater decrease observed in K-pGL-Loop cells indicates a specific effect of the MDR1 5′-mRNA structure on translation of the luciferase reporter under cytotoxic stress.
The structure at the 5′-end of MDR1 mRNA down-regulates translation specifically upon treatment with cytotoxic drugs
Down-regulation of signalling pathways due to cytotoxic stress
Since control of protein synthesis is generally achieved by changes in the phosphorylation state of translation initiation factors, or the regulators that interact with them, we examined several signalling pathways known to affect translation following cytotoxic stress. A number of kinases, activated by various types of stresses, can phosphorylate eIF2α at Ser51 and down-regulate global translation . Following exposure of K562 cells to 1 μM colcemid for 3 days, an increase in the phosphorylation of eIF2α was observed (Figure 3A, left-hand panel), which is probably responsible for the down-regulation of [35S]methionine incorporation into nascent proteins reported previously by our group  and shown in Figure 2(B).
Down-regulation of signalling pathways controlling the phosphorylation of eIF4E and 4E-BP
Extracellular stimuli also modulate the activities of both ERK1 (extracellular-signal-regulated kinase 1) and p38 MAPK, which activate Mnk-1 which phosphorylates eIF4E at Ser209 [21,22]. Colcemid treatment of K562 cells resulted in down-regulation of phosphorylated Mnk-1 and p38 MAPK, but not of ERK, leading to hypophosphorylation of eIF4E (Figure 3A, middle panel).
Activation of the PI3K (phosphoinositide 3-kinase)/Akt signalling pathway and phosphorylation by mTOR (mammalian target of rapamycin) reduces the affinity of 4E-BPs (eIF4E-binding proteins, a family of translational repressors that bind and sequester eIF4E) for eIF4E, allowing binding to eIF4G and translation inititation. Many cellular stresses also inhibit 4E-BP phosphorylation, leading to increased binding of eIF4E, reduced availability of eIF4E and ultimately a decrease in the levels of translation [21,23,24]. Exposure of K562 cells to colcemid led to down-regulation of the PI3K/Akt/mTOR signalling pathway and hypophosphorylation of 4E-BP (Figure 3A, right-hand panel). The effect on the phosphorylation status of eIF4E and 4E-BP was independent of the drug used, since doxorubicin and cytarabine produced a similar down-regulation in phosphorylation (Figure 3B).
To confirm that the drug-induced hypophosphorylation of 4E-BP affected the formation of translation initiation complexes we pulled-down eIF4E and associated proteins from colcemid-treated K562 cell extracts using a 5′-end mRNA cap analogue (m7GTP-Sepharose). Bound proteins were analysed by Western blotting. Treatment with colcemid resulted in a time-dependent increase in the binding of 4E-BP to eIF4E, and a reduction in the binding of eIF4G (Figure 3C). Thus cytotoxic stress in K562 cells leads to a decrease in the ability of eIF4E to associate with eIF4G, resulting in down-regulation of translation.
Down-regulation of available eIF4E following cytotoxic stress decreases the translation efficiency of transcripts with the MDR1 mRNA 5′-structure
To study how the changes in 4E-BP phosphorylation and eIF4E availability affected the translation of luciferase reporter genes in K-pGL-UTR and K-pGL-Loop cells following cytotoxic drug exposure, we transfected these cell lines transiently with plasmids expressing constitutively-active forms of Mnk1 (pCAMNK) and Akt (pCAAKT), singly or together, and measured the luciferase activity following treatment of the cells with 1 μM colcemid for 3 days. Expression of constitutively active Mnk1 increased the phosphorylation of eIF4E after drug exposure in both cell lines, as expected (Figure 4A). Expression of constitutively active Akt also increased 4E-BP phosphorylation after drug exposure, but to a lesser extent. This is probably because, unlike Mnk1 which phosphorylates eIF4E directly, Akt phosphorylates and activates mTOR which then in turn phosphorylates 4E-BP (Figure 4B) . There was no significant difference in the steady-state levels of the reporter mRNAs in each transfected cell line upon drug exposure (results not shown). Since expression of constitutively active Mnk1 did not increase expression of luciferase from pGL-Loop under cytotoxic stress (Figures 4A, 4C and 4D) we can infer that, at least under these conditions, the phosphorylation status of eIF4E does not contribute to the down-regulation of translation of pGL-Loop transcripts. In contrast, translation of pGL-Loop transcripts following cytotoxic stress was enhanced significantly in cells expressing constitutively active Akt (Figure 4D).
The phosphorylation status of 4E-BP controls the translation efficiency of transcripts with the structured MDR1 mRNA 5′-end
We confirmed the importance of Akt by down-regulating mTOR with rapamycin in K-pGL-UTR and K-pGL-Loop cells. The concentrations of rapamycin used in these assays were not toxic to K562 cells, as determined by proliferation analyses (results not shown). Western blot analysis demonstrated that treatment with rapamycin decreased the level of phosphorylated 4E-BP in cells, as expected (Figure 4E). Similarly, steady-state levels of luciferase reporter mRNAs (results not shown) and the translation of pGL-UTR transcripts (Figure 4F) did not change significantly. Nonetheless, rapamycin strongly repressed translation of pGL-Loop transcripts (Figure 4F). Thus translational repression of transcripts containing the 5′-structure of MDR1 mRNA is probably due to a decrease in Akt activity and 4E-BP phosphorylation. In conclusion, inhibition of the PI3K/Akt/mTOR/4E-BP phosphorylation cascade, and a subsequent decrease in the availability of eIF4E in drug-exposed cells, specifically reduces the translation of transcripts including the structured 5′-end of MDR1 mRNA.
The 5′-end MDR1 mRNA loop renders chimaeric luciferase reporter mRNAs poorly competitive for eIF4E
The limiting factor for translation initiation is eIF4E, and transcripts with long and structured 5′-UTRs are generally considered to be poorly competitive [8,25]. We tested whether the 5′-end MDR1 mRNA structure affected mRNA competitiveness in vitro. Luciferase transcripts were synthesized in vitro with either the MDR1 5′-UTR or the intact 5′-MDR1 loop structure from T7pGL-UTR and T7pGL-Loop respectively. Translation of these transcripts in the presence of increasing amounts of total K562 RNA was performed in a rabbit reticulocyte cell-free system. Luciferase mRNA with the 5′-UTR of MDR1 mRNA (T7pGL-UTR) competed efficiently with increasing amounts of RNA up to 7 μg. Conversely, a small increase (0.1 μg) in the amount of competitor RNA negatively affected translation from luciferase mRNA with the 5′-end MDR1 mRNA structure (T7pGL-Loop). A progressive increase in the amount of competitor RNA decreased luciferase translation from T7pGL-Loop (Figure 5A). Similar results were obtained when in vitro transcription and translation was performed in a single reaction using a TNT system (Promega) (results not shown). Equally, yeast RNA could be substituted for K562 RNA as the competitor, thus ruling out the effect of any putative microRNA from K562 cells (results not shown).
The 5′-MDR1 mRNA structure makes a reporter mRNA poorly competitive for eIF4E
We hypothesized that over-expression of eIF4E would compensate for the loss of available endogenous eIF4E through repression by 4E-BP, and so would increase the competitive ability of pGL-Loop transcripts under cytotoxic stress. We therefore transfected K-pGL-UTR and K-pGL-Loop cells with an eIF4E expression plasmid (Figure 5B) and treated them with 1 μM colcemid for 3 days. m7GTP-Sepharose pulldown of translation–initiation complexes confirmed that, upon cytotoxic treatment, more eIF4G was associated with eIF4E in cells transfected with the eIF4E expression plasmid (Figure 5C). As before, luciferase activity in transfected K-pGL-UTR and K-pGL-Loop cells was measured to estimate the relative translation efficiencies of pGL-UTR and pGL-Loop transcripts. Although translation from pGL-UTR was not affected by overexpression of eIF4E, translation from pGL-Loop was increased by approx. 40%. There was no significant difference in the steady-state levels of luciferase mRNA in each transfected cell line upon drug exposure (results not shown).
Further confirmation of the poor competitiveness of pGL-Loop transcripts was obtained by siRNA. Transfection of siRNA targeting eIF4E in K-pGL-UTR and K-pGL-Loop cells decreased eIF4E levels equally in both cells (Figure 5E). Although the eIF4E decrease in unstressed K562 cells did not affect global protein synthesis or translation from pGL-UTR transcripts, it specifically decreased the translation from pGL-Loop transcripts by approx. 25% (Figure 5F).
Thus reporter transcripts with the 5′-structure from MDR1 mRNA compete inefficiently for the limiting amounts of eIF4E in K562 cells, such that, under cytotoxic stress, which decreases the availability of eIF4E in the cells, translation from MDR1 transcripts is diminished.
MDR1 mRNA is translated when global translation is not down-regulated
How do drug-resistant K562 cells arise and overcome the translational block in MDR1 mRNA to translate P-glycoprotein in conditions of cytotoxic stress when eIF4E is downregulated? A mechanism that would allow efficient translation of MDR1 mRNA in drug-resistant K652 cells would be by increasing the levels of phosphorylated 4E-BP, with the consequent increased availability of eIF4E and enhanced translation initiation. However, Western blot analysis showed that the levels of phosphorylated 4E-BP are similar in both naïve and drug-resistant K562 cells (results not shown). Therefore to translate MDR1 mRNA, drug-resistant K562 cells must increase the steady-state levels of this transcript, enabling it to compete effectively for the translation machinery.
If the above competition model is correct, MDR1 mRNA should be translated into P-glycoprotein in cells in which eIF4E availability has not been down-regulated (i.e. in cells not exposed to cytotoxic stress). PMA treatment of K562 cells up-regulates MDR1 mRNA steady-state levels to those levels found in P-glycoprotein-expressing, drug-resistant KC40 cells (see Supplementary Figure 1A at http://www.BiochemJ.org/bj/406/bj4060445add.htm). This up-regulation is due both to an increase in transcription from the MDR1 promoter (see Supplementary Figure 2A at http://www.BiochemJ.org/bj/406/bj4060445add.htm) and to specific stabilization of MDR1 mRNA, increasing its half-life to 20 h (see Supplementary Figure 2B at http://www.BiochemJ.org/bj/406/bj4060445add.htm). PMA treatment did not affect the turnover of other short-lived mRNAs such as Id2  (see Supplementary Figure 2C at http://www.BiochemJ.org/bj/406/bj4060445add.htm) or alter the MDR1 transcriptional start point (results not shown), ensuring that the same structured mRNA was formed as that from P-glycoprotein-expressing drug-resistant K562 cells. Treatment with PMA did not increase the phosphorylation status of eIF2α or alter the phosphorylation status of eIF4E and 4E-BP1, which remained similar to that of drug-naïve cells, as opposed to drug-induced K562 cells (see Supplementary Figure 1B at http://www.BiochemJ.org/bj/406/bj4060445add.htm).
After 16 h of treatment with 32 nM PMA, approx. 30% of K562 cells expressed P-glycoprotein on the plasma membrane, at 20% of the levels of KC40 cells (see Supplementary Figure 1C at http://www.BiochemJ.org/bj/406/bj4060445add.htm). PMA-treated K562 cells were sorted by flow cytometry into P-glycoprotein positive and negative populations. These populations showed differential up-regulation of MDR1 mRNA. Steady-state MDR1 mRNA levels were up-regulated approx. 20-fold in P-glycoprotein-positive cells when compared with P-glycoprotein-negative cells (see Supplementary Figure 1D at http://www.BiochemJ.org/bj/406/bj4060445add.htm). In addition, although MDR1 mRNA levels in P-glycoprotein-negative PMA-treated K562 cells were 10-fold greater than in naïve cells, this degree of up-regulation was insufficient to increase the competitive ability of MDR1 transcripts (see Supplementary Figure 1D at http://www.BiochemJ.org/bj/406/bj4060445add.htm). Thus P-glycoprotein was only translated in the subpopulation of PMA-treated K562 cells that up-regulate levels of MDR1 mRNA sufficiently to increase its competitive ability for eIF4E.
Drug-resistant K562 cells originate from a small subpopulation that differentially up-regulates MDR1 mRNA
We next asked whether differential up-regulation of MDR1 mRNA in a subpopulation of drug-treated K562 cells might occur during selection for drug resistance. We followed cultures of cells for 37 days after treatment with colchicine, in the presence and absence of 10 μM verapamil (a competitive inhibitor of P-glycoprotein). Four parameters were monitored: changes in MDR1 mRNA levels were detected by RT-Q-PCR; and the percentage of viable cells (TOTO-3 negative), P-glycoprotein expression (antibody binding) and P-glycoprotein activity (BODIPY-taxol efflux) were detected by flow cytometry. Treatment with 20 ng/ml colchicine resulted in progressive cell death (Figure 6A), reaching a maximum at 26 days when only approx. 5% of the cells were alive. After longer incubation times, colchicine-resistant clones began to emerge, and after an additional 10 days most of the culture was composed of viable, colchicine-resistant cells. The first clones of colchicine-resistant cells were discernible at day 12, consisting of at least 8 cells. Assuming that the growth rate of these cells was equivalent to that of KC40 (approx. 2 days doubling time), we can estimate that the first colchicine-resistant cells appeared as early as day 6, equivalent to approx. 5 cells per million in the original population. Although addition of verapamil alone (10–25 μM) had no effect on the growth of K562 cells, its inclusion during the selection process abolished the appearance of colchicine-resistant cells, implying that P-glycoprotein function was essential for the development of drug resistance in these cells (Figure 6 and results not shown).
Drug-resistant K562 cells arise from a small subpopulation of cells which differentially up-regulate MDR1 mRNA upon drug treatment
MDR1 mRNA levels increased slightly during the first 20 days of culture (from 0.1% to 3% of the level found in KC40), and dramatically thereafter, reaching similar levels to those of KC40 cells at day 37 (Figure 6B). Similarly, the percentage of cells expressing surface P-glycoprotein (UIC2 positive), and the percentage of cells showing P-glycoprotein-mediated drug efflux (BODIPY-taxol), remained at basal levels for the first 14 days and then increased progressively until day 37 where a population phenotypically similar to KC40 was obtained (Figures 6C and 6D). Verapamil abolished the appearance of P-glycoprotein- and BODIPY-taxol efflux-positive cells.
Using flow cytometry we sorted the small percentage of UIC2 positive cells (3–4%) after 7 days of treatment with 20 ng/ml colchicine and determined their MDR1 mRNA levels by RT-Q-PCR. As observed for PMA-treated K562 cells (see Supplementary Figure 1D at http://www.BiochemJ.org/bj/406/bj4060445add.htm), there was a differential up-regulation of MDR1 mRNA (approx. 5-fold) between P-glycoprotein-positive and -negative cells. Levels of MDR1 mRNA in P-glycoprotein-negative and -positive cells were approx. 25- and 125-fold those of naïve K562 cells respectively (Figure 6E).
We next asked whether the naïve K562 cell line was a heterogeneous cell population (i.e. not clonal) and whether the cells up-regulating MDR1 mRNA were predetermined. For this we isolated, by serial dilution, 20 clones each derived from a single cell, exposed them to 1 μM colcemid for 3 days and determined the percentage of P-glycoprotein-positive cells as described above. The K562 cell line and its 20 single-cell-derived clones all expressed equivalent percentages of P-glycoprotein-positive cells following drug exposure (results not shown). This indicated that the small percentage of cells which differentially up-regulate MDR1 mRNA and synthesize very small amounts of P-glycoprotein is not predetermined and is a stochastic process within a clonal population. Thus differential up-regulation of MDR1 mRNA occurs stochastically in a small subpopulation of K562 cells upon exposure to cytotoxic drugs.
Acquisition of the multidrug resistance phenotype appears to be a multistep process involving many different regulatory steps and may differ between cell types [26–28]. We have previously shown that, upon low-level drug selection, multidrug-resistant K562 leukaemic cells arise due to overexpression of P-glycoprotein. However, this only occurs in a small proportion of cells within the drug-exposed population which can overcome a translational block of P-glycoprotein synthesis. This block prevents MDR1 mRNA associating with polysomes . In the present study we determined how this translational block is overcome and define the mechanism, at least for this cell type, by which drug resistance develops.
Several features of an mRNA can influence its translation. mRNA transcribed from the MDR1 downstream promoter, which is the only promoter active in K562 cells , does not have upstream open reading frames in the 5′-UTR, and other features influencing translation such as the poly(A) tail length and polyadenylation site are unchanged irrespective of the translational status of MDR1 mRNA (results not shown). Secondary structures in the 5′-UTR, such as hairpins, are known to block translation . Although the short 5′-UTR of MDR1 mRNA appears to have no negative effects on translation, it has previously been shown, by RNase mapping, that the 5′UTR can fold with the first 149 nucleotides of coding region into three highly structured hairpins with several internal loops . We have shown that, in vivo, when this entire 5′-MDR1 structure is cloned upstream of a reporter gene it reduces translation, specifically under conditions of cytotoxic stress.
Following cytotoxic stress, we have shown in the present study that eIF2α is phosphorylated and that there is a down-regulation in global cellular translation. In addition, the Akt signalling pathway is down-regulated, leading to hypophosphorylation of 4E-BP. Hypophosphorylation of 4E-BP increases its affinity for, and sequestration of, the cap-binding protein eIF4E . eIF4E is generally considered rate-limiting for translation. Crucially, not all mRNAs are affected equally by down-regulation of the eIF4E levels available. It has been proposed that transcripts with long and structured 5′-UTRs are poor competitors for limiting eIF4E [23,29], and it has long been known that mRNAs containing extensive secondary structure in their 5′-UTR translate efficiently when eIF4E is over-expressed . We demonstrate that, under cytotoxic stress, there is a down-regulation of the available eIF4E levels, and that the highly structured 5′-end MDR1 mRNA competes poorly for eIF4E and its translation is specifically repressed. This poor competitiveness explains the translational block of P-glycoprotein synthesis under conditions of cytotoxic stress. This suggests that compounds that mimic the 5′-end cap of mRNA, such as ribavirin, or which inhibit mTOR or Akt, could decrease the competitive ability of MDR1 mRNA, preventing translation of P-glycoprotein and thereby avoiding drug resistance.
Activation of Akt had a greater effect on the translation of MDR1 5′-end reporter mRNAs than did overexpression of eIF4E. In the same way, inhibition of mTOR activity by rapamycin had a greater effect on the translation of the same transcripts than specifically targeting eIF4E with siRNA. This suggests that other targets of mTOR (such as eIF4G and p70 S6 kinase), whose role in the activation of translational initiation is well-established [31,32], may also play a role in MDR1 mRNA translational control.
The PI3K/Akt signalling pathway is over-activated in a wide range of tumour types, promoting a variety of responses from cell growth and proliferation to survival and motility, leading to tumour progression . Akt activation has also been associated with tumour metastasis and resistance to chemotherapy . It has been suggested that activation of the PI3K/Akt pathway in tumours leads to an increase in the availability of eIF4E and this in turns activates the translation of poorly competitive mRNAs, many of which are important for cell growth [25,29,35]. Our data suggest a similar phenomenon for MDR1 mRNA, implying a mechanistic link between uncontrolled growth and drug resistance.
The poor competitiveness of MDR1 mRNA implies that, in order to develop drug resistance in the presence of drugs which reduce the availability of eIF4E, MDR1 mRNA must be up-regulated above a certain threshold in order to compete effectively for eIF4E and translate P-glycoprotein. An alternative hypothesis would be that an inhibitory factor, such as a microRNA, is outcompeted when the transcripts reach a critical level. We have found different putative microRNAs targeting MDR1 3′-UTR when using the miRBase (http://microrna.sanger.ac.uk) or TargetScanS (http://genes.mit.edu/tscan/) databases, but none of them are shared between the five vertebrate target sequences available (results not shown). Thus although we cannot rule out the possibility that some as yet uncharacterized microRNA negatively regulates MDR1 mRNA translation in K562 cells, the in vitro competition data using rabbit retyculocyte lysates (Figure 5A and results not shown) rule out that the effect of the 5′-end MDR1 loop involves a microRNA.
Under conditions in which the levels of eIF4E are not affected, such as following PMA treatment, we have shown that, for P-glycoprotein to be translated, MDR1 mRNA levels must be increased at least 10-fold above those in naïve K562 cells. The threshold for MDR1 mRNA translation would be expected to be even higher under conditions, such as cytotoxic drug exposure, in which eIF4E is sequestered. On a population basis, the increase in MDR1 mRNA levels, due to increased stability, is 60- to 70-fold. Importantly, in the present study we show that the up-regulation of MDR1 mRNA in response to cytotoxic stress is not homogeneous but that, in a small percentage of cells (less than 5%) there is a large increase in mRNA levels of up to 125-fold. This is stochastic, in that fluctuation tests showed that the small subpopulation of cells in which MDR1 mRNA is stabilized following drug treatment is not predetermined. The mechanism of this stochastic stabilization is unknown. However, the consequence is that this subpopulation of cells produces sufficient MDR1 mRNA to effectively compete for the reduced eIF4E, enabling a small amount of P-glycoprotein to be synthesized. The active P-glycoprotein reduces intracellular concentrations of the cytotoxic drug, leading to increased availability of eIF4E. This, in turn, increases the ability of MDR1 mRNA to compete and translate more P-glycoprotein. This positive feedback loop, which is abolished by inhibiting P-glycoprotein function, leads to the appearance and maintenance of drug resistance.
We thank Dr J. A. Cooper (Fred Hutchinson Cancer Research Centre, Seattle, WA, U.S.A.), Professor A. Willis (University of Nottingham, Nottingham, U.K.) and Dr K. Scotto (Robert Wood Johnson Medical School, Piscataway, New Jersey, U.S.A.) for reagents, and Mr E. O'Connor and Mr E. Ng (both at the MRC Clinical Sciences Centre, London, U.K.) for advice with FACS.
eukaryotic initiation factor
mitogen-activated protein kinase
multidrug resistance 1
mammalian target of rapamycin
small interfering RNA
Present address: UCB-Celltech (NCE Biology), Slough SL1 3WE, U.K.
Present address: Vice-Chancellor's Office, Durham University, The University Offices, Old Elvet, Durham DH1 3HP, U.K.