Multidrug resistance in relapsed acute myeloid leukemia: Evidence of biological heterogeneity


  • Chirayu Patel BA,

    1. Laboratory of Cell Biology, National Cancer Institute, National Institutes of Health, Bethesda, Maryland
    2. Howard Hughes Medical Institute (HHMI)-National Institutes of Health (NIH) Research Scholars Program, Bethesda, Maryland
    Search for more papers by this author
  • Leif Stenke MD,

    1. Department of Medicine, Karolinska Institutet and University Hospital, Stockholm, Sweden
    2. Department of Oncology-Pathology, Karolinska Institutet and University Hospital, Stockholm, Sweden
    Search for more papers by this author
  • Sudhir Varma PhD,

    1. Bioinformatics and Computational Biosciences Branch, Office of Cyber Infrastructure and Computational Biology, Office of Science Management and Operations, National Institute of Allergy and Infectious Diseases, NIH, Bethesda, Maryland
    Search for more papers by this author
  • Marita Lagergren Lindberg MD,

    1. Department of Medicine, Karolinska Institutet and University Hospital, Stockholm, Sweden
    2. Department of Oncology-Pathology, Karolinska Institutet and University Hospital, Stockholm, Sweden
    Search for more papers by this author
  • Magnus Björkholm MD,

    1. Department of Medicine, Karolinska Institutet and University Hospital, Stockholm, Sweden
    Search for more papers by this author
  • Jan Sjöberg MD,

    1. Department of Oncology-Pathology, Karolinska Institutet and University Hospital, Stockholm, Sweden
    Search for more papers by this author
  • Kristina Viktorsson MD,

    1. Department of Oncology-Pathology, Karolinska Institutet and University Hospital, Stockholm, Sweden
    Search for more papers by this author
  • Rolf Lewensohn MD,

    1. Department of Oncology-Pathology, Karolinska Institutet and University Hospital, Stockholm, Sweden
    Search for more papers by this author
  • Ola Landgren MD,

    1. Metabolism Branch, NCI, NIH, Bethesda, Maryland
    Search for more papers by this author
  • Michael M. Gottesman MD,

    Corresponding author
    1. Laboratory of Cell Biology, National Cancer Institute, National Institutes of Health, Bethesda, Maryland
    • Corresponding author: Michael Gottesman, MD, Laboratory of Cell Biology, National Cancer Institute, National Institutes of Health, 37 Convent Drive, Room 2108, Bethesda, MD 20892-4255; Fax: (301) 402-0450;

    Search for more papers by this author
  • Jean-Pierre Gillet PhD

    1. Laboratory of Cell Biology, National Cancer Institute, National Institutes of Health, Bethesda, Maryland
    Search for more papers by this author



Studies of mechanisms mediating resistance to chemotherapy led to the discovery of the multidrug transporter ABCB1 (ATP-binding cassette, subfamily B, member 1), often expressed in leukemic cells of patients with acute myeloid leukemia (AML). Most clinical trials evaluating the strategy of inhibiting efflux-mediated chemotherapeutic resistance have been unsuccessful, clearly indicating the need for a better approach.


This study investigated the clinical relevance of 380 genes whose expression has been shown to affect the response to chemotherapy, mostly through in vitro studies, in 11 paired samples obtained at AML diagnosis and at relapse. The expression profiling of these 380 genes was performed using TaqMan-based quantitative reverse-transcription polymerase chain reaction. Patients had a median age of 58 years at diagnosis, a median duration of complete remission of 284.5 days, and a median overall survival of 563 days. Cytogenetic abnormalities were detected at diagnosis in 4 patients, whereas 5 displayed a normal karyotype and 2 were not investigated.


Hierarchical clustering shows that samples taken at diagnosis and relapse clustered in pairs for 6 patients of the 11 studied, suggesting recurrence of the same leukemic blast, whereas for the other 5 patients, the data indicate their relapse blasts arose from different origins. A patient-by-patient analysis of the paired samples led to the striking observation that each had a unique gene signature representing different mechanisms of resistance.


The data underline the need for personalized molecular analysis to tailor treatment for patients with AML. Cancer 2013;119:3076—3083. © 2013 American Cancer Society.


Acute myeloid leukemia (AML) is a heterogeneous disease characterized by clonal proliferation of malignant precursors of a myeloid lineage with impaired differentiation of normal hematopoietic progenitors.[1] Although the majority of patients with AML achieve remission, most eventually relapse, ultimately succumbing to the disease. Recent studies revealed the clonal architecture of secondary AML, which is a dynamic process shaped by multiple cycles of mutation acquisition and clonal selection.[2] However, the underlying alterations in gene transcription that allow relapse after chemotherapy remain poorly understood.

ABCB1 (ATP-binding cassette, subfamily B, member 1; also known as multidrug resistance 1/P-glycoprotein) exports a wide variety of drugs that are mechanistically and structurally unrelated,[3] and is often found to be highly expressed in leukemic blasts.[4-6] Although inhibition of ABCB1 to increase chemosensitivity has been successful in vitro, translating this strategy to clinical settings has failed.[7, 8] Indeed, trials evaluating ABCB1 inhibitors in more than 2000 patients with AML have failed to yield positive results.[9] Since the discovery of ABCB1 and the emergence of the genomic era, numerous genes have been revealed to mediate drug resistance.[10] Yet the benefits of this research for patients who develop resistance to chemotherapy are minimal.

The current analytical methods for high-throughput gene expression profiling are based on mean values to find differentially expressed genes, with the goal of identifying a common gene signature that defines a trend among patients. However, such a strategy is clearly insufficient, because the variability among patients is totally ignored. For such an analysis to identify rare signatures in individuals that are associated with drug resistance in which the prediction is based on a Boolean rule, the analysis of several hundred samples would be needed to avoid overfitting problems. Nevertheless, in a heterogeneous disease such as AML, one effective way to identify rare resistance mechanisms is to do a patient-by-patient analysis of paired samples taken at diagnosis and after relapse, thereby minimizing false hits due to interpersonal variability.[11]

We conducted a unique study using paired samples to discern the mechanisms involved in the acquisition of multidrug resistance (MDR) in patients with AML, using highly sensitive and specific TaqMan-based quantitative reverse-transcription polymerase chain reaction (qRT-PCR).[12] The 380 genes involved in phase I and II metabolism, signal transduction, DNA repair, stress response, tumor suppressor activity, oncogenic transformation, apoptosis, and drug uptake or efflux were chosen on the basis of their potential role in MDR, as indicated by reports published over the past 3 decades.[13] Here, we extend to the transcriptional level the recent genetic discoveries in AML showing great heterogeneity of the tumor,[2] because the data demonstrate that the mechanisms of resistance are highly heterogeneous.


Patient Samples

The collection of tumors for research and specific molecular analysis was first approved by the ethical review board of the Karolinska Institutet, Stockholm, Sweden (ethical permit number: KI Dnr 03-600) and written, informed consent was obtained from the patients. Peripheral blood was then collected from 11 patients with AML at diagnosis and at relapse after treatment, and separated with Ficoll-Paque (GE Healthcare, Piscataway, NJ) according to the manufacturer's instructions. Blasts were cryopreserved at the biobank at the Department of Hematology, Karolinska University Hospital. Pathological analysis at the Karolinska Institutet confirmed that each sample contained at least 80% leukemic blasts. Patients had undergone treatment with cytarabine combined with daunorubicin, idarubicin, etoposide, mitoxantrone, and/or thioguanine. Data on clinical outcomes were obtained from patient records. Duration of complete remission (CR) was the number of days between CR (determined by bone marrow aspirates that showed less than 5% blasts) and first documented relapse in blood or bone marrow. Patients had a median age of 58 years at diagnosis, ranging from age 28 to 72 years. The median duration of CR was 284.5 days, with a range of 48 to 1166 days and the median overall survival was 563 days, ranging from a low of 193 to a high of 1664 days (Table 1).[14] Cytogenetic aberrations were found in 4 patients at the time of diagnosis, 5 displayed a normal karyotype, whereas 2 were not investigated. Regarding mutations of individual genes, eg, FLT3 or NPM1, no such assessments were made at diagnosis, because the collection of samples took place from 1987 to 2001.

Table 1. Characteristics of 11 Adult Patients With Acute Myeloid Leukemia Who Had Undergone Conventional Chemotherapy and Contributed Paired Samples (at Diagnosis and After Relapse)
PatientFABaSexAgeCR (days)TreatmentbOS (days)Karyotype According to Chromosome Banding AnalysisChrom ScorecWBCdPlateletsSCTeAHDf
  1. Abbreviations: CR, complete remission; F, female; M, male; OS, overall survival.

  2. a

    FAB type: French-American-British classification system

  3. b

    Treatment: A/E/N = Ara-C/etoposide/novantrone, A/D/T = Ara-C/daunorubicin/thioguanine, A/D = Ara-C/daunorubicin; A/E/I = Ara-C/etoposide/idarubicin; A/I = Ara-C/idarubicin

  4. c

    Chrom score: risk groups according to World Health Organization: 9 = unknown, 1 = good, 2 = intermediate, 3 = bad, 4 = very bad

  5. d

    White blood cell count (WBC), at diagnosis, × 109/L

  6. e

    SCT: allogeneic stem cell transplantation performed (y/n= yes/no)

  7. f

    AHD: associated hematological disease (y/n= yes/no).

1M4F6256A/E/N26445,XX,del(5)(q?),der(11)r(11;?) (p15q25?;?)−18319.644nn
2M1M67588A/E/N1252del (7q)(q22) 6 of 8 metafases2965nn
3M1F63190A/E/N44046,XX,Ph+/46,XX,Ph+, 18,mar/46,XX383303nn

Preparation of Total RNA

Total RNA was prepared using the Trizol method (Invitrogen, Carlsbad, Calif). RNA was quantified using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, Del). The integrity of the RNA samples was assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies, Foster City, Calif) and then stored at −80°C. RNA integrity number > 7 for each RNA sample analyzed.

Reverse Transcription

Synthesis of complementary DNA (cDNA) from 1 μg total RNA in a 20 μL reaction volume was carried out using the High Capacity cDNA kit with ribonuclease inhibitor (Applied Biosystems) as per the manufacturer's instructions. The reverse-transcription conditions were as follows: 10 minutes at 25°C, 120 minutes at 37°C, 5 minutes at 85°C. Following reverse transcription, cDNA was stored at 4°C.

TaqMan Low Density Arrays (TLDAs)

Expression levels of 380 MDR-associated genes were measured using custom-made TaqMan Low Density Arrays in addition to the 18S ribosomal RNA being arrayed in 4 replicates (Applied Biosystems).[13, 15] The cDNA was mixed with 2× TaqMan Universal PCR Master Mix (Applied Biosystems), loaded on a TaqMan Low-Density Array (TLDA) card, and run on an ABI-Prism 7900 HT Sequence Detection System (Applied Biosystems) as per the manufacturer's instructions.

Data Analysis

Data from the TLDA cards was collected for each of the 11 paired AML samples (available from the GEO Database, accession number: GSE33787). The normalization of the data was performed in subtracting the median expression of each sample from all gene expression data for that sample. The expressions from replicate probes were averaged together. This analysis was carried out using BRB ArrayTools, a microarray-data statistical analysis tool ([16] In order to remove genes that are not likely to be informative, those expressed by fewer than 50% of the samples were filtered out, resulting in analysis of 331 genes. Hierarchical clustering was done on the normalized, filtered data, using an average linkage algorithm with 1-Pearson correlation as the distance. A Spearman rank test was used to assess the correlation between French-American-British (FAB) classification and the expression of each gene (threshold: Spearman rank P < .05). The Cox proportional hazards test was used to study the correlation of gene expression with the duration of CR. The false discovery rate (FDR) for each gene was calculated using the Benjamini-Hochberg method for both methods. Pairwise comparisons were manually performed using the ΔΔCt method.[17]


Comparison of MDR-Linked Gene Expression Profiles of Paired AML Samples Taken at Diagnosis and After Relapse

Unsupervised hierarchical clustering was performed on all paired samples (Fig. 1). For 6 of 11 patients, the pair of samples (at diagnosis and after relapse) clustered together, indicating that the development of resistance did not involve a major change in pattern of gene expression. This also supports that at least for this set of patients, diagnosis and relapse leukemias had similar origins. For the other 5 pairs of samples that clustered apart, for which large changes in the gene expression profile were observed between the paired samples, no distinguishing trend could be found for age, the subtype (based on the cell type from which the leukemia developed and its degree of maturation, ie, FAB classification[18]), the CR duration, the treatment, or the overall survival time (Table 1).

Figure 1.

Unsupervised hierarchical clustering is shown for 11 paired samples. The x-axis shows clusters of samples, whereas the y-axis shows the expression levels according to the filtering criteria (red and green represent upregulation and downregulation, respectively. The data suggest the role of differentiation in contributing to relapse, because 3 specimens of M5a and M5b subtypes had diagnosis (Diag) and recurrence (Rec) samples that clustered apart, whereas all 3 patients with acute myeloblastic leukemia with minimal maturation (M1 subtype) had diagnosis and relapse samples that clustered together.

Genes that significantly differ in expression between diagnosis and relapse blast samples from the same patient may give insight into what contributes to relapse and why relapsed AML may be more resistant to treatment. Supervised comparison of samples taken at diagnosis and after relapse, paired according to the patient, revealed 27 genes that are differentially expressed, in most cases only to a small extent, between these 2 groups (P < .05; Table 2). None of these genes fulfilled the stringent criteria of FDR < 0.05, but this may be due to the low number of samples analyzed.

Table 2. Differentially Expressed Genes in Samples Taken at Diagnosis Compared With Samples Taken at Relapse
GenesParametric P ValueFDRaGeometric Meanb(Level of Expression at Diagnosis/Level of Expression at Relapse)
  1. a

    False discovery rate (FDR) is measured using the Benjamini-Hochberg method and is an estimate of the proportion of the genes with P < .05 that represent false positives. For example, there is a 50.8% chance that MSH3 is a false positive. One can consider that genes with a FDR > 15% are not statistically significant.

  2. b

    The geometric mean denotes the average of the logarithmic values (base 2), converted back to linear values.


Gene Correlation With FAB Classification Reveals That the Expression Level of BCL2A1 And Glutathione Reductase Increases Along With FAB Subtypes

We expected that some of the MDR genes measured in this study would reflect the state of differentiation of AML cells and would therefore correlate with FAB classification. In fact, 52 genes of the 331 genes that passed the filtering criteria (see “Data Analysis” section above) were found to correlate with FAB classification for samples taken at diagnosis (P < .05). Among these genes, 2 genes, encoding the antiapoptotic bcl2-related protein A1 (BCL2A1) and glutathione reductase (GSR), fulfilled a FDR < 15% and were found to be positively correlated, implying that the expression level of these genes increases along with FAB (from 0 to 5) (Fig. 2, Table 3). More precisely, the probability of finding these genes by chance at P < .05 is 7.1% and 14.3%, respectively. Four additional genes fulfilled a FDR < 15% and were negatively correlated with FAB classes, and were therefore upregulated in immature cells compared with differentiated cells. These genes were POLH, NOLA2, ABCD4, and MNAT1 (Table 3). Fifteen genes were found to be correlated with FAB at P < .05 for samples taken at relapse; however, none of these genes fulfilled a FDR < 15% (data not shown).

Figure 2.

Transcript level of the antiapoptotic BCL2-related protein A1 (BCL2A1) and glutathione reductase (GSR) correlated with FAB classification in samples taken at diagnosis. The transcript level values are presented as −ΔCT. (A) BCL2A1 and (B) GSR were found to correlate with FAB classification for samples taken at diagnosis (P < .05). These genes fulfilled a false discovery rate (FDR) lower than 15% and were found to be positively correlated, implying that the expression level of these genes increases along with FAB (from 0 to 5) (Table 3). More precisely, the probability of finding these genes by chance at P < .05 is 7.1% and 14.3%, respectively.

Table 3. Genes Correlated With FAB for Samples Taken at Diagnosis
Gene SymbolCorrelation CoefficientFDRParametric P Value
MNAT1−0.928<1 × 10−7<1 × 10−7

Gene Correlation With Duration of CR Does Not Highlight Any Trend Among Patients

Genes that correlate with the duration of the first and second CRs may be important predictors of relapse. This would hold true unless a small population of cells present at diagnosis is responsible for relapse or there is a fundamental change in the biology of AML between diagnosis and relapse.[19] We found 38 genes to be correlated with CR in samples taken at diagnosis at P < .05 (data not shown). None fulfilled a FDR < 15%. Seven were found in recurrent samples, but none were statistically significant (data not shown).

Patient-By-Patient Analysis of Paired Diagnosis and Relapse Samples

Given that there were no striking correlations between expression of specific genes and overall resistance to chemotherapy, we hypothesized that each patient may have a unique set of genes within his or her AML blasts that contribute to the observed MDR, which would have been missed by statistical analysis highlighting trends of gene patterns observed “on average” across samples.[20] This issue was addressed by performing a patient-by-patient analysis of paired diagnosis and relapse samples (Table 4), which led to the striking observation that when each of these patients relapsed, their AML cells expressed genes representing different potential mechanisms of resistance. Table 5 shows the expression profile of 9 ABC transporters that have been shown to mediate MDR (data extracted from Table 4). Although no trend was revealed based on FAB classification, as mentioned above, it is observed that the leukemic cells of all patients except patient 1 (FAB: M4) showed increase at relapse of at least one ABC transporter capable of transporting either anthracyclines, vinca-alkaloids, or both. The analysis of samples taken from patient 9 (FAB: M5b) reveals that 4 ABC transporters—ABCB1 (MDR1/P-gp), ABCC1 (MRP1), ABCC5 (MRP5), and ABCG2 (BCRP)—are overexpressed in leukemic cells. This indicates that conventional chemotherapy (anthracyclines/vinca-alkaloids) is not likely to benefit this patient and will probably only have a negative effect by causing deterioration of the patient's condition through the side effects of these compounds.

Table 4. Patient-by-Patient Analysis Showing Genes With Greater Than 4-Fold Up- or Downregulation
Patient 1 (M4)Patient 2 (M1)Patient 3 (M1)Patient 4 (M5b)Patient 5 (M4)Patient 6 (M5a)
  ABCB1−9.4SGPP15.0NR1H35.8AKR1C1 a4.2FN1−12.2
  TNF−13.6  GSTT2−9.8MMP9−9.8  
  IGF1R−14.7  SLC28A3−10.0CCL2−46.4  
  ABCB4−14.9  TGFA−10.5CYP2E1−145.8  
  MYC−17.1  HIF1A−11.6    
  SLC2A5−33.4  NFKBIA−11.6    
  GJA1−90.0  MMP9−14.6    
  CLU−127.3  CLDN7−17.6    
Patient 7 (M1)Patient 8 (M4)Patient 9 (M5b)Patient 9 (continued)Patient 11 (M0)
  1. a

    AKR1C1 assay also detects AKR1C2; CYP2C8 assay also detects CYP2C19; CYP2A6 also detects CYP2A7 and CYP2A13.

KLF1−11.2ABCC4−10.0IGF1R−5.3  ETS15.5
CYP2A6a−11.7SLC2A5−11.9MT1X−5.3  TNFSF105.1
IGF1R−28.1CLDN3−14.2ABCA6−5.7Patient 10 (M2)AKR1C1 a−5.1
Table 5. ABC Transporter Gene Expression Profilesa
GenesP 11P 2P 3P 7P 10P 1P 5P 8P 6P 4P 9
  1. a

    Fold-change of 9 ABC transporters known to be involved in chemotherapy resistance, in samples taken after relapse compared with their counterparts taken at diagnosis.

  2. Abbreviations: NA, not applicable (no detection of the measured genes); P, patient.


ABCB1, which has been evaluated in patients with AML as a potential target for pharmacologic downregulation of efflux-mediated chemotherapy resistance, was found to be overexpressed in the leukemic cells of 5 patients of the 11 (Fig. 3). A similar observation was made for ABCG2, although that transporter may actually be associated with the intrinsic biology of the leukemia rather than with drug efflux mediating resistance per se (Fig. 3).

Figure 3.

ABC transporter-mediated multidrug resistance. Histogram presenting the expression levels of ABCB1 and ABCG2, two ABC transporters intensively studied in patients with AML, in relapse samples compared with their paired samples taken at diagnosis. No bar appears when the measured gene is not expressed.


Genome-wide analyses have revealed the molecular genetic heterogeneity of AML.[21-23] A number of reports have established that AML cases can be classified into various groups based on their chromosomal abnormalities, somatic acquired mutations, gene and microRNA expression profiles, and methylation status.[24] However, our understanding of the mechanisms causing the relapse of patients is limited. Although most patients with AML achieve CR, a large fraction of them will relapse and have a dismal prognosis, demonstrating the dire need to specifically address the reason for relapse. Pairs of samples taken at diagnosis and after relapse, combined with a reliable gene expression profiling assay, are prerequisites to understanding mechanisms mediating MDR.

In this study based on paired samples obtained at AML diagnosis and at relapse, for the first time, we assessed 380 MDR-related genes, through use of TaqMan-based qRT-PCR, which is a state-of-the-art gene expression profiling assay. An unsupervised hierarchical clustering revealed that samples taken at diagnosis and relapse clustered in pairs for 6 patients of the 11 studied, indicating return of the same leukemic blast, whereas the other 5 relapses had different transcription patterns and may have originated from different blasts. None of the gene signatures found to be correlated with either relapse or duration of remission were statistically significant. The observed enormous heterogeneity in gene expression across study patients remained true when we examined AML by subtype (M1, M4, and M5 subtypes are represented by 3 samples each; Table 4).

This investigation extends to the transcriptional level the recent findings highlighting the dynamic clonal evolution during AML relapse.[2] Importantly, these data also demonstrate that systemic chemotherapy has a substantial effect on the increased number of new mutations. Identifying new targets whose expression is altered after chemotherapy is critical to limit the occurrence of relapse. However, a personalized approach presents multiple challenges, such as pinpointing the clinically relevant targets from among numerous candidates in each individual gene signature, and eventually designing multiply-targeted treatment regimens to limit the survival of tumor cells through alternative drug resistance pathways. Although by definition the individual changes we see in this study are not “statistically significant,” each patient acts as his or her own control. By evaluating only genes known to be capable of conferring resistance to drugs such as those used to treat AML, we can generate a specific hypothesis for each patient about how best to treat their relapsed AML. These hypotheses could be tested in short-term ex vivo cultures of their cells, using drugs known not to be subject to resistance mechanisms expressed in their AML cells, or by using inhibitors of demonstrated resistance mechanisms.


The HHMI-NIH Research Scholars Program supported the research training of Chirayu Patel. This work was supported by the Intramural Research Program of the National Institutes of Health, National Cancer Institute, Center for Cancer Research; the Swedish National Board of Health and Welfare; and the Adolf H. Lundin Charitable Foundation.


The authors made no disclosure.