Disease-stabilizing treatment based on all-trans retinoic acid and valproic acid in acute myeloid leukemia – identification of responders by gene expression profiling of pretreatment leukemic cells

The 28 patients included in the microarray studies represent the subset of patients with high enough peripheral blood blast counts to allow sampling from the peripheral blood of sufficient cells for microarray studies. Leukemic peripheral blood mononuclear cells (PBMCs) were isolated by density gradient separation (Ficoll-Hypaque; NycoMed, Oslo, Norway; specific density 1.077) from peripheral blood of patients with at least 80% of the leukocytes being AML cells. Cells were stored frozen in liquid nitrogen. The percentage of AML blasts among leukemia PBMC exceeded 95% [26].

Submicroscopic mutation profiling of 54 genes frequently mutated in myeloid leukemias was done by the Illuminas TruSight Myeloid Gene Panel and sequenced using the MiSeq system and reagent kit v3 (all from Illumina, San Diego, CA, USA) (Additional file 1: Table S4). Amplicon sequencing library was prepared from 50 ng DNA according to the manufacturer’s instructions with the exception of normalization being done manually. 8–16 samples were sequenced each time and the total DNA input on the flow cell was 15 picomolar. Secondary analysis was performed using MiSeqReporter version 2.4.60.8 (Illumina) mapping to the human genome reference hg19. Sequence alignment of selected variants was manually examined with the Integrative Genomics Viewer (IGV) [27]. Annotation was done by snpSIFT og snpEFF v 4.1. As no matching normal DNA was available variants with >1% minor allele frequency in the 1000 genomes data were presumed to be germline and removed from further interpretations. Synonymous substitutions, intronic variants not in the splice site and variant interpreted as benign or most likely benign are not included. The variant allele frequency (VAF) was calculated for each mutation as number of variant reads divided by total reads. Cut-off for reported variants for VAF was 8% and read depth 100. Only variants interpreted as pathogenic, probably pathogenic and variants of unknown significance are reported. The nomenclature is according to Human Genome Structural Variation consortium.

Fragment analysis of FLT3 exon 14–15 and NPM1 exon 12 were done as described in [28] and CEBPA mutation analysis as described previously [28].

All microarray experiments were performed using the Illumina iScan Reader, which is based upon fluorescence detection of biotin-labelled cRNA. Three hundred ng of total RNA from each sample was reversely transcribed, amplified and Biotin-16-UTP-labelled using the Illumina TotalPrep RNA Amplification Kit (Applied Biosystems/Ambion, USA). The amount and quality of the Biotin-labelled cRNA was controlled both by the NanoDrop spectrophotometer and Agilent 2100 Bioanalyzer. Biotin- labelled cRNA (750 ng) was hybridized to the HumanHT-12 V4 Expression BeadChip according to the manufacturer’s instructions. The HumanHT-12 V4 BeadChip targets 47,231 probes that are mainly derived from genes in the NCBI RefSeq database (Release 38).

Data from the array scanning were investigated in GenomeStudio and J-Express 2012 for quality control measures [29]. All arrays within each experiment were quantile normalized to be comparable before being compiled into an expression profile data matrix. The probe with the highest fluorescence was used in the analyses if the expression of the same gene was examined by different probes. In all our analyses of gene expression profiles we used significant analyses of microarray (SAM) [30], and gene set enrichment analysis (GSEA) [31], to compare different patient subsets or samples. The genes encoding proteins with a known function were classified by using the PANTHER (protein annotation through evolutionary relationship) classification system [32].

All patients in the present study were included in two previous clinical studies. These studies included 60 patients (20 responders); the characteristics of all patients are summarized in Additional file 1: Table S1 and the characteristics of individual patients included in the present study are presented in Additional file 1: Table S2. Additional analysis of the mutational profiles was possible for 12 responders and 29 non-responders to the treatment (Additional file 1: Table S3; patients 1–12 and 15–43). The effects of antileukemic treatment on gene expression profile were analyzed for eight patients from the study by Ryningen et al. [13]. A high frequency of patients with high-risk disease according to conventional prognostic criteria (i.e. AML relapse, secondary AML, high-risk cytogenetic abnormalities) was seen both for the whole group of 60 patients, the 41 patients included in the study of mutational profiles (Additional file 1: Table S2; patients 7, 8, 17, 23, 26, 31, 38 and 42). None of the patients had low-risk cytogenetic abnormalities.

A detailed submicroscopic mutational profile was examined for 41 patients (Additional file 1: Table S2; 12 responders and 29 non-responders); the profile included genes frequently mutated in myeloid malignancies (Additional file 1: Table S4, Fig. 1). Thirty-two of these genes were mutated in at least one of the patients, and according to previous studies [33, 34] these mutations were classified as (i) NPM1 mutations, (ii) mutations causing activation of intracellular signaling; (iii) mutated tumor suppressor genes; (iv) mutations in genes involved in DNA methylation or (v) chromatin modification; (vi) mutations in genes encoding myeloid transcription factors; (vii) mutated genes important for the spliceosome or (viii) encoding cohesion protein; and (ix) others.

The number of detected mutations differed between patients, the median number being three mutations (range 0–7). Responders and non-responders did not differ with regard to the number of mutations. FLT3 mutations were most frequent (15 out of 41 patients, 37%) followed by NPM1 mutations (14/41, 34%). Most patients had at least one mutation causing activation of intracellular signaling (25/41, 61%), and all six patients with TP53 mutations had an adverse karyotype. Even though NPM1, FLT3, TP53 and DNMT3A mutations showed higher frequencies for non-responders than for responders, these differences did not reach statistical significance.

We compared the global gene expression profiles for pretreatment AML cell samples derived from 28 patients who later could be classified as responders or non-responders to the leukemia-stabilizing treatment. All these AML cell samples were derived from patients with high peripheral blood blast counts, and highly enriched AML cell populations could thereby be prepared by a simple and highly standardized method based on gradient separation. Nineteen of these patients were classified as non-responders and nine as responders according to the criteria previously described in detail by Fredly et al. [13]. We used SAM to compare the global gene expression profiles for responders and non-responders. When setting the d-score to ±2.5 we identified 243 probes that differed significantly between the two groups, and these probes represent 179 different genes (Additional file 1: Table S5). We then used the Panther database to classify the encoded proteins (Fig. 2), and 159 of these genes encoded proteins that could be classified. Genes encoding for proteins that belonged to the “nucleic acid binding” class were overrepresented (38 out of 159 genes), and these genes encoded both DNA binding protein (16 proteins), RNA binding proteins (20 proteins) and nucleases (2 proteins).

All the genes showing at least a 2-fold difference between responders and non-responders are listed in Additional file 1: Table S6. The proteins encoded by these genes included oncogenes (RGL4, LMO4) as well as regulators of protein degradation/activation/modulation (QPCT, ELANE), transcription (HOXA3, HOXA5, PBX3; the only three with decreased expression), iron metabolism (HP, LTF), energy metabolism (MOSC1, CYP4F3, OLR1, SNCA), apoptosis/proliferation (OLFM4, PGLYRP1) and communication (COL17A1, TACSTD2). The Hox genes may be of particular importance, and HOXA4 expression was also significantly lower although the difference was less than two-fold.

We also performed a GSEA. This alternative analysis also showed that the differentially expressed genes are important for a wide range of cellular function, but several of the identified GO-terms with p-value <0.05 describe binding/function/regulation of nucleic acids and showed increased expression for the responders (Table 1).

Transcriptomic profiling of primary AML cells during in vivo ATRA treatment was only possible for eight patients included in the first study [9]. We first compared AML cell samples derived (i) before start of treatment on day 1, and (ii) after two days of oral ATRA monotherapy (day 3). These eight patients were two responders and six non-responders, and due to this low number of available patients it was not possible to compare the effects of ATRA in responders and non-responders. Thus, by this comparison we identified alterations in global gene expression profiles that are common for responders and non-responders. By this approach we could not identify quantitative differences in the expression of these identified genes and probably not effects of ATRA that are specific for responders/non-responders either. However, differences between responders and non-responders to ATRA may be caused by different downstream responses to common ATRA-induced alterations, and only such mechanisms are likely to be identified by our strategy for analysis.

We performed a SAM analysis that identified the top rankled differently expressed genes with d-score of ±2.0 (400 permutations) when comparing AML cells sampled on day 1 before ATRA treatment and after two days of treatment on day 3. All differentially expressed genes are listed in Additional file 1: Table S5. The Panther classification analysis based on the function of the encoded proteins (Fig. 3); for these analyses we only included the 70 genes (36 upregulated and 34 downregulated on day 3) that were annotated; these genes can be identified from Additional file 1: Table S5). ATRA altered the expression of genes with a wide range of functions, but a major effect of this in vivo therapy was altered expression of genes encoding proteins that show nucleic acid binding and/or being involved in transcriptional regulation. Additional effects were altered expression of receptor-associated genes, whereas the pretreatment differences between responders and non-responders with regard to hydrolases and enzyme modulation (see above) seem to be maintained during ATRA. However, relatively few genes were altered during ATRA therapy for the 5 terms Nucleic acid binding/transcription factors/enzyme modulation/hydrolases/receptors (see Tables 2 and 3), but the genes encode proteins that are involved both in regulation of DNMT3A (SALL3) and retinol metabolism (RBP1). Finally, there was only a small overlap between those genes showing differential expression when comparing responders and non-responders and those genes being altered by ATRA (Tables 2 and 3). This last observation suggests that the pretreatment differences between responders and non-responders with regard to these 5 terms are maintained during treatment.

We compared the gene expression profiles for AML cells derived from the same 8 patients before addition of valproic acid and theophyllamine to the ATRA therapy (days 3) and during treatment with the triple combination (day 8). We then did a SAM analyses identifying the top ranked differently expressed genes with d-score of ±2.0 (400 permutations), and thereafter a Panther protein classification only based on those 81 genes (42 upregulated and 39 downregulated on day 8) that were annotated. The identity of these genes can be seen from Additional file 1: Table S5. The showed that addition of valproic acid/theophyllamine altered the expression of genes that are involved in a wide range of cellular functions, again including altered transcriptional regulation (Fig. 3, Table 3, Additional file 1: Table S7). Altered expression after addition of valproic acid/theophyllamine was observed for relatively few genes, but including genes encoding proteins important for epigenetic regulation (SIRT6) and for the kallikrein system (ITIH4, KLK) (Additional file 1: Table S7). Very few of these genes differed significantly when comparing pretreatment levels for responders and non-responders (see Tables 2 and 3), i.e. this is similar to the ATRA treatment and suggests that the pretreatment differences between these two patient subsets are maintained during this treatment.

We finally compared the effects of the triple drug combination by comparing the global gene expression profiles for primary AML cells derived from the 8 patients before start of treatment and after triple therapy on day 8. The therapeutic serum level for theophyllamine was 55–110 μmol/L. The daily valproic acid dose was increased to the maximal tolerated dose. The therapeutic serum level of valproic acid was 300–600 μmol/L, but the mean valproic acid level during the 5 days of triple treatment varied for individual patients between 178 and 717 μmol/L (median value 407 μmol/L) and did not reach the lower therapeutic limit for 2 of the patients.

We did a SAM analysis identifying the top ranked differentially expressed genes with d-score of ±2.0 (400 permutations), and thereafter we did a Panther protein classification only based on the 76 annotated genes (39 upregulated and 37 downregulated in day 8 samples). The genes included in the terms transcriptional regulation/nucleic acid binding represent only a minority among the genes with altered expression during the triple drug therapy, and the same was true for the term hydrolases (Figure 3, Table 3, Additional file 1: Table S7). Thus, a major part of the pre-therapy differences between responders and non-responders seem to be maintained during the triple treatment and this was also seen for the separate analyses of ATRA and valproic acid/theophyllamine treatment (see above). The triple therapy altered the expression of genes included in several annotations, but major effects seem to be altered receptor expression/function (especially G-protein coupled receptors for neuromediators, i.e. AVPR1B, GALR2, HCRTR1, GPR151) together with altered expression of transcriptional regulators (Additional file 1: Table S7). Finally, altered expression after valproic acid/theophyllaminee was observed for relatively few genes (Table 2) and very few of these genes differed significantly when comparing pre-treatment levels for responders and non-responders (Additional file 1: Table S5), i.e. pre-treatment differences are maintained during treatment.