Significance of Expression of ITGA5 and its Splice Variants in Acute Myeloid Leukemia: A Report from the Children’s Oncology Group

Abstract

Acute myeloid leukemia (AML) encompasses a heterogeneous group of diseases, and novel biomarkers for risk refinement and stratification are needed to optimize patient care. To identify novel risk factors, we performed transcriptome sequencing on 68 diagnostic AML samples and identified 2 transcript variants (−E2 and −E2/3) of the α-subunit (ITGA5) of the very late antigen-5 integrin. We then quantified expression of ITGA5 and these splice variants in specimens from participants of the AAML03P1 trial. We found no association between ITGA5 expression and clinical outcome. In contrast, patients with the highest relative expression (Q4) of the −E2/3 ITGA5 splice variant less likely had low-risk disease than Q1–3 patients (21% vs. 38%, P=0.027). Q4 patients had worse response to chemotherapy with a higher proportion having persistent minimal residual disease (50% vs. 23%, P=0.003) and inferior overall survival (at 5 years: 48% vs. 67%, P=0.015); the latter association was limited to low-risk patients (Q4 vs. Q1–3: 56% vs. 85%, P=0.043) and was not seen in standard-risk (51% vs. 60%, P=0.340) or high-risk (33% vs. 38%, P=0.952) patients. Our exploratory studies indicate that transcriptome sequencing is useful for biomarker discovery, as exemplified by the identification of ITGA5 −E2/3 splice variant as potential novel adverse prognostic marker for low-risk AML that, if confirmed, could serve to further risk-stratify this patient subset.

INTRODUCTION

Acute myeloid leukemia (AML) remains difficult to treat. As the efficacy of chemotherapy and chance of cure vary widely [1–3], there is a long-standing interest in determining the predictors of therapeutic response. Of the numerous recognized disease-related factors, cytogenetic abnormalities and somatic mutations are the most important ones and provide the framework for risk-stratification schemes [3, 4]. Nevertheless, recent data from >2,000 patients indicate that even complex models cannot accurately predict therapeutic failure, the major cause of death in AML [5]. Thus, there is ongoing need for novel, critical biological features that can be used as biomarkers to refine prognostication and allow more tailored, risk-stratified treatment strategies.

The development of second-generation sequencing technologies has enabled comprehensive, unbiased, genome-wide characterization of cancers to gain further insight into the mechanisms of cancer pathogenesis and improve diagnostics and treatment selection [6]. Over the last several years, such DNA-based approaches have provided further evidence of the complexity of the leukemic genome in AML [7, 8]. More recently, transcriptome sequencing has been introduced as efficient, relatively cost-effective strategy for the discovery of novel disease-associated somatic mutations in patients with acute leukemias [9, 10]. This method can also be utilized to identify novel transcripts and alternative splice forms of genes [6, 11]. We have now employed genome-wide transcriptome sequencing to find new biomarkers for response to chemotherapy and outcome/prognosis in AML. Given our interest in adhesion molecules [12], we herein present our detailed findings on splice variants in the α subunit (ITGA5) of the α5β1 integrin, very late antigen-5 (VLA-5; CD49e/CD29), alongside studies exploring the clinical relevance of ITGA5 expression levels.

PATIENTS AND METHODS

Patient Samples for Genome-Wide Transcriptome Sequencing Studies

Whole-transcriptome sequencing (RNA sequencing) was performed on diagnostic bone marrow or peripheral blood specimens from selected patients with untreated AML enrolled on COG protocols AAML03P1 and AAML0531 (n=68). These studies aimed to identify novel biomarkers associated with relapse risk in patients that were not considered at high risk for relapsed based on known cytogenetic/molecular markers; therefore, samples were chosen from patients who lacked known high-risk cytogenetic features but relapsed, maximizing those with normal karyotype and core-binding factor [CBF, t(8;21) or inv(16)/t(16;16)] translocations.

Patient Samples for Determination of Clinical Significant of ITGA5 and its Splice Variants

The clinical significance of total ITGA5 expression levels and ITGA5 splice variants were determined in participants of AAML03P1, a phase 3 pilot study that determined the safety and feasibility of adding gemtuzumab ozogamicin to intensive chemotherapy in previously untreated childhood AML [13]. Briefly, from December 2003 to November 2005, AAML03P1 enrolled 340 eligible children (aged 1 month to 21 years) with newly diagnosed de novo AML, excluding those with acute promyelocytic leukemia, bone marrow failure syndromes, juvenile myelomonocytic leukemia, or Down syndrome. Treatment consisted of a remission induction phase (Induction I and Induction II) followed by an intensification phase with either chemotherapy alone (Intensification I, II, and III) or chemotherapy (Intensification I) plus allogeneic hematopoietic cell transplantation (HCT), depending on the availability of a 5/6 or 6/6 matched family donor. Cycle regimens were cytarabine/daunorubicin/etoposide (ADE) plus GO (Induction I), ADE (Induction II), high-dose cytarabine and etoposide (Intensification I), mitoxantrone/cytarabine plus GO (Intensification II), and sequential high-dose cytarabine and asparaginase (Intensification III). Pretreatment (“diagnostic”) bone marrow specimens from all patients enrolled on AAML03P1 who consented to participation in biology research studies (as those described in this report) and for whom marrow specimens were available were used for this study. The patient and disease (cytogenetic/molecular) characteristics of these subsets of AAML03P1 studied in this analysis were relatively comparable to the entire AAML03P1 study cohort [13]. Specifically, while there were small differences with regard to some disease characteristics (i.e., higher median WBC [P<0.001] and lower proportion of standard-risk disease [54% vs. 67%, P=0.021]) and short-term outcomes (i.e., lower rate of CR after 1 course of therapy [80% vs. 89%, P=0.036] with a lower rate of patients in complete remission [CR] with minimal residual disease [MRD; 17% vs. 31%, P=0.027]) for patients with available diagnostic marrow specimens, there were no significant differences between patients with and those without available specimens with regard to overall survival (OS), event-free survival, relapse-free survival (RFS), risk of relapse (RR), or treatment-related mortality. Informed consent was obtained in accordance with the Declaration of Helsinki. The institutional review boards (IRBs) of all participating institutions approved the clinical protocol, while the Fred Hutchinson Cancer Research Center IRB and the COG Myeloid Disease Biology Committee approved this research study.

Risk Stratification

Following the current COG approach, a combination of cytogenetic and molecular information was used for risk stratification [12, 14–17]. Patients were considered low-risk if a chromosomal abnormality/mutation was present in CBFs, nucleophosmin [NPM1], or CEBPA (n=68). Patients were classified as high-risk if they had −5/5q-, monosomy 7, or FLT3/ITD with high allelic ratio (n=25). All other patients with data sufficient for classification were considered standard-risk (n=107). Patients were classified as unknown-risk if cytogenetic/molecular data were insufficient for classification (n=16).

Transcriptome Sequencing of Human AML Specimens and Normal Bone Marrow

Genetic material from AML specimens and 4 normal bone marrow samples was extracted using AllPrep DNA/RNA Mini Kits via QIAcube automated system (Qiagen, Valencia, CA). Then, 1 µg of high-quality total RNA was used for conversion of mRNA into a cDNA library of template molecules based on mRNA capture with poly(T) magnetic beads, fragmentation, and reverse transcription to first-strand cDNA with reverse transcriptase and random primers, using Illumina’s TruSeq RNA Sample Prep Kit (Illumina, San Diego, CA) according to the manufacturer’s instructions. After adaptor ligation, each cDNA library was purified and enriched by PCR amplification; the final average fragment size, including adaptors, was 280 bases. Each library was then subjected to 50-cycle paired-end sequencing on the Illumina HiSeq, with 4 samples multiplexed into each flow cell lane.

Bioinformatics Approach to Identify Novel Splice Variants

To identify novel splice variants, we developed a largely automated analysis pipeline based on publicly available tools and data resources. Beginning with a single de-multiplexed Fastq file for each sample, overall quality assessments with FastQC were first performed to check for various sequence-composition biases and base-call quality trends. Then, each library was screened for abundant sequences (i.e., ribosomal RNA), common contaminants, and adaptor sequences, and adaptors were trimmed when sufficient informative sequence remained. Filtered Fastq files were processed with TopHat [18], an efficient read-mapping algorithm designed to align reads to a reference genome without relying on known splice sites, to identify novel splice variants of genes. Data from normal tissue specimens served as controls and to assess whether a newly identified splicing event was AML specific: in addition to normal bone marrow, data from the Illumina Body Map 2.0 study were used, which comprise very deep sequencing of polyadenylated mRNA from 16 normal tissues (kidney, heart, liver, lung, lymph, prostate, skeletal, white blood cells, ovary, testes, thyroid, adipose, adrenal, brain, breast, and colon) for this purpose. Inferred splice junctions were merged across all samples (AML and normal tissues) denoting each junction by the chromosomal locations of its 5’ and 3’ splice sites. Each splice junction was annotated by whether it is contained within a known transcript model (RefSeq, UCSC KnownGenes, or Ensembl), with the number of supporting expressed sequence tags, and with the gene symbol for the containing locus. The abundance of a splice junction in each sample, represented by the number of supporting reads, was also recorded to form a “splice junction array” with 1 row for each junction and 1 column for each sample.

To characterize the complete catalog of splicing events observed in the AML samples, we applied the following criteria to first remove observations that were likely artifacts of the alignment and splice-junction inference process: 1) removal of any junction with less than 15 nucleotides sequenced from either flanking exon across all samples. This may remove a small number of true microexon-spanning reads, but our experience with manual curation has shown that junctions that do not meet this criterion are far more likely to be alignment artifacts. 2) removal of any junction seen in fewer than 3 AML samples. And 3) removal of any junction not supported by at least 5 reads in at least one AML sample. These criteria filter for quality, prevalence, and abundance of the observed splice junctions. We found that only 15% of the junctions excluded by this process are part of known annotated gene models (from RefSeq, UCSC KnownGenes, or Ensembl), suggesting that this process is effective at selectively excluding spurious un-annotated splice junctions from further consideration.

Verification of ITGA5 Splice Variants and Determination of Splice Variant Abundance

To verify the ITGA5 splice variants identified by bioinformatics, ITGA5 transcripts were amplified by reverse-transcriptase polymerase chain reaction (RT-PCR) with primers annealing to exon 1 (5’-CTCCTTCTTCGGATTCTCAGTGGAG-3’) and exon 4 (5’-GCAGGGTGCATACTCCAGAATTCG-3’). Subsequently, amplicons were electrophoretically separated and subjected to Applied Biosystems (ABI) BigDye™ Terminator sequencing. To determine the abundance of ITGA5 splice variants in AML specimens, individual amplicons were quantified by optical DNA fragment length determination, using an ABI 3730xl DNA Analyzer in diagnostic bone marrow specimens of 216 participants of AAML03P1. Then, the ratio of splice variant transcript abundance to abundance of wild-type ITGA5 (ITGA5 splice variant ratio) was calculated. In a subset of 20 AML specimens, splice variants were also determined in CD34⁺/CD33⁻ and CD34⁺/CD33⁺ immature cell subsets that were isolated by fluorescence-activated cell sorting (FACS).

Quantification of Total ITGA5 Expression

From diagnostic bone marrow specimens of 113 participants of AAML03P1, total RNA was extracted with the AllPrep DNA/RNA Mini Kit via QIAcube automated system and quantified with a micro-volume spectrophotometer (NanoDrop™; Thermo Scientific, Wilmington, DE). To allow multianalyte profiling directly from purified RNA preparations, total ITGA5 mRNA expression was quantified from 125 ng of RNA using a multi-analyte sandwich nucleic acid hybridization method employing branched DNA molecules to amplify the signal from captured target RNA [19–21] (QuantiGene™ Plex 2.0; Panomics/Affymetrix, Santa Clara, CA) in combination with the xMAP system fluorescent-dyed microspheres (xMAP™ beads; Luminex, Austin, TX), according to the manufacturer’s instructions. The ITGA5 probe set covered nucleotides 792–1,244 of ITGA5, spanning exons 7–13 based on Ensembl (http://www.ensembl.org; transcript ID ENST00000293379, ITGA5-001). ITGA5 mRNA expression levels were normalized using beta glucuronidase (GUSB). Two technical replicates obtained from each specimen were analyzed with the xPonent software (Life Technologies, Grand Island, NY).

Statistical Analyses

The Kaplan-Meier method [22] was used to estimate OS (defined as time from study entry to death) and RFS (defined as time from end of induction I for patients in CR until relapse or death due to progressive disease, censoring patients who died from non-progressive disease). Patients who withdrew from therapy due to relapse, persistent CNS disease, or refractory disease with >20% bone marrow blasts by the end of induction I were defined as induction I failures. RR was calculated by cumulative incidence defined as time from the end of induction I for patients in CR to relapse or death due to progressive disease where deaths from non-progressive disease were considered competing events [23]. The significance of predictor variables was tested with the log-rank statistic for OS and RFS, and with Gray’s statistic for RR. Children lost to follow-up were censored at their date of last known contact or at a cutoff 6 months before the date of analyses, which was June 30, 2011. Cox proportional hazards models [24] were used to estimate the hazard ratio (HR) for defined groups of patients in univariate and multivariate analyses. The chi-squared test was used to test the significance of observed differences in proportions, and Fisher’s exact test was used when data were sparse. Differences in medians were compared by the Mann-Whitney test. Linear associations between 2 variables were measured with the Pearson product-moment correlation coefficient. All reported P-values were 2-sided; P-values <0.05 were considered statistically significant.

RESULTS

Initial Identification of ITGA5 Splice Variants in AML

In order to identify novel biomarkers associated with relapse risk in patients that were not considered at high risk for relapsed based on known cytogenetic/molecular markers, unbiased transcriptome sequencing was performed on 61 bone marrow and 7 peripheral blood specimens that were chosen from patients who lacked known high-risk cytogenetic features but relapsed, maximizing those with normal karyotype and CBF translocations. After a series of filtering steps as described in Patients and Methods, a total of 184,572 splice junctions were identified. Approximately 77% of these were contained in some annotated gene model. Of the remaining 42,957 un-annotated splice junctions, 9,634 were exon-skipping events with respect to some annotated gene model, with single-exon skips dominating (85% of the exons skips omitted a single exon, while only 3% omitted 3 exons). 19% of un-annotated junctions were consistent with at least one spliced EST in the public dbEST database, and an additional 27% were supported by at least one tissue from the Illumina BodyMap 2.0 data. This finding suggested that at least half of the filtered un-annotated junctions were supported by an orthogonal technology (EST sequencing) or additional tissues. We selected the top 30 candidate junctions that were both un-annotated and implicated in relapse. We manually inspected each of these candidates using the Integrative Genomics Viewer (http://www.broadinstitute.org/igv/). On inspection, 12 of the 30 were found to be questionable alignments (PCR duplication artifacts, misalignment to repetitive regions or pseudogenes, etc.) and were dropped from further consideration. The remaining 18 candidates comprise two intergenic splice sites and 16 representing novel splicing events in known genes: ANPEP, C17orf85, DCTN4, DCUN1D3, EHD1, GYG1, ITGA5, ITGAE, KLHDC2, PRTN3, RANBP10, SPNS2, STK24, and WAC.

Given our interest in adhesion molecules [12], we initially focused our efforts on validating the splicing events identified in ITGA5, the gene encoding the α subunit of a major integrin found on AML cells, α5β1 (VLA-5) [25]. Overall, transcriptome sequencing discovered the presence of splice variants of ITGA5 in 10 of 68 samples. Specifically, 2 dominant splice variants were found: one lacking all of exon 2 (“−E2”) and the other lacking the entirety of exons 2 and 3 (“−E2/3”; Figure 1). Of note, the −E2 variant has been reported previously (ITGA5-002 [ENST00000435631] in Ensembl), but, to our knowledge, the −E2/3 variant has not yet been described.

Figure 1. Verification of ITGA5 splice variants in AML patient samples.

(A) Genomic organization of ITGA5 located at chromosome 12q13.13. Wild-type (WT) ITGA5 contains an amino-terminus signal peptide (exon 1), an α-light chain cleavage site (exon 26), and a transmembrane domain (exon 29). The indicated ITGA5 splice variants were identified by RNA sequencing, including both a previously reported splice variant (−E2) and a novel variant (−E2/3). Shown are UTRs (green boxes), skipped exons (open boxes), translation stop signal (red line), and approximate location of primers used for RT-PCR (red arrows). (B) RT-PCR on RNA extracted from 5 representative AML patient samples and a human AML cell line (ML-1 cells), using primers recognizing ITGA5 sequences on exon 1 and exon 4. Major amplicons are indicated (arrows); also shown are water control and a sizing ladder. (C) Amplicon sizes as determined by optical DNA fragment length determination on one representative AML patient sample. (D) Exon/exon boundaries were confirmed through gel extraction of PCR products shown in panel B, subsequent purification, and sequence analysis. Representative sequences for ITGA5-WT, ITGA5-E2 and ITGA5-E2/3 are shown. (E) Amino acid sequence of ITGA5 exons 1–5, ITGA5-E2 (predicted), and ITGA5-E2/3 (predicted) splice variants. Aberrant amino acids due to frame shift are indicated in red; *denotes early translational stop signal. (F) Functional domains of WT ITGA5 (1049 amino acids) as well as structures of ITGA5 −E2 (predicted, 135 amino acids) and −E2/3 (predicted, 85 amino acids). Full length ITGA5 protein includes a signal peptide (membrane localization) within the first 42 amino acids, a 7-bladed β-propeller domain for ligand binding, an integrin α-2 domain for interaction with the integrin β-subunit, as well as a transmembrane and cytoplasmic domain. ITGA5 −E2 and −E2/3 are predicted to be severely truncated because of an early termination signal in exon 4, containing only the signal peptide as well as a relatively short aberrant amino acid sequence due to frame shift (red) but lacking all functional domains characteristic of ITGA5.

Leftover mRNA was available from 5 of these 10 samples for additional studies to verify the findings from transcriptome sequencing. Specifically, standard Sanger (i.e. dideoxy chain termination) sequencing confirmed the existence of both the−E2 and−E2/3 variant in each of these 5 specimens (Figure 1). Moreover, the presence of amplicons of the predicted length after RT-PCR–based targeted amplification was confirmed by optical DNA fragment length analysis and gel electrophoresis in all tested specimens. The predicted protein sequences of the −E2 and −E2/E3 variants indicate that frame shift mutations introduced by the alternate splicing event will lead to translation stop signals in exon 4 and severely truncated ITGA5 proteins in both isoforms, with only little more than the signal peptide retained from the endogenous ITGA5 sequence (see Figure 1).

Association of Total ITGA5 Expression with Clinical Characteristics and Treatment Outcome

Having identified ITGA5 as a gene of interest in AML, we first assessed total ITGA5 expression levels in diagnostic (i.e. pre-treatment) bone marrow specimens from 113 participants of AAML03P1 selected based on RNA availability, using a probe set that spans nucleotides 796–1,244 (exons 7–13). Total ITGA5 mRNA expression levels varied more than 60-fold relative to GUSB in 110 samples (from 0.033 to 2.089 [median, 0.248]; Figure 2A). Given the distribution of total ITGA5 expression across our study cohort, we compared the characteristics of the 25% patients with the highest total ITGA5 expression (“Q4” patients) with the 75% of patients with lower total ITGA5 expression (“Q1–3” patients), with a cut point of 0.38 (relative expression to GUSB mRNA expression) separating Q4 from Q1–3. As summarized in Table 1, baseline characteristics were relatively comparable in these 2 patient subsets; however, of note, patients with high total expression of ITGA5 were significantly more likely to present with hepatomegaly (48% vs. 21%, P=0.006) or splenomegaly (52% vs. 26%, P=0.020). Although patients with high (Q4) expression of ITGA5 had a slightly worse RFS (49±22% vs. 67±13%) and higher RR (46±21% vs. 28±12%) than those with lower (Q1–3) expression, these differences were not statistically significant (P=0.199 and P=0.180, respectively), and OS was relatively similar (61±18% vs. 69±10%, P=0.594).

Figure 2.

(A) Quantitative expression of ITGA5 relative to beta glucuronidase (GUSB) in the 113 diagnostic bone marrow specimens from patients enrolled in AAML03P1. (B, C) Ratios of abundance of exon 2 (A; “−E2”) and exon 2/3 (B; “−E2/3”) ITGA5 splice variants relative to abundance of other ITGA5 transcripts containing exons 2 and 3 across the 216 diagnostic specimens from patients enrolled in AAML03P1.

TABLE 1.

Comparison of Baseline Characteristics of Patients with Low (Patient Quartile 1–3) vs. High (Patient Quartile 4) ITGA5 Expression

Patient Characteristics	ITGA5 Expression (Patient Quartile)		P-value
	Q1-Q3 n = 84	Q4 n = 29
Median Age (range), years	9.9 (0.11–18.3)	11.7 (0.4–20.8)	0.664
Male Gender, n (%)	47 (56%)	18 (62%)	0.566
Median WBC % (range)	28.6 (1.6–302)	19.5 (2.2–405)	0.344
Median Bone Marrow Blasts, %	70 (2–100)	56 (5–94)	0.257
Cytogenetics, n (%)
Normal	23 (27%)	4 (14%)	0.139
t(8;21)(q22;q22)	11 (13%)	3 (10%)	1.000
inv(16)/t(16;16)(p13.1;q22)	12 (14%)	7 (24%)	0.253
t(9;11)(p22;q23) or other abn 11q23	12 (14%)	2 (7%)	0.513
t(6;9)(p23;q34)	3 (4%)	1 (3%)	1.000
Monosomy 7	1 (1%)	2 (7%)	0.161
Del7q	1 (1%)	0 (0%)	1.000
−5/5q-	0 (0%)	2 (7%)	0.064
Trisomy 8	9 (11%)	2 (7%)	0.726
Other	12 (14%)	6 (21%)	0.396
Risk Group, n (%)
Standard	43 (51%)	11 (38%)	0.218
Low	32 (38%)	11 (38%)	1.000
High	9 (11%)	7 (24%)	0.118
Molecular Alterations, n (%)
FLT3/ITD	12 (15%)	3 (10%)	0.755
NPM1 Mutation	2 (3%)	2 (8%)	0.302
CEBPA Mutation	7 (9%)	0 (0%)	0.185
Hepatomegaly, n (%)	18 (21%)	14 (48%)	0.006
Splenomegaly, n (%)	22 (26%)	15 (52%)	0.020
CNS Involvement, n (%)	5 (6%)	1 (3%)	1.000
Extramedullary Disease (non-CNS), n (%)	3 (4%)	1 (3%)	1.000

Quantification of ITGA5 Splice Variants in Specimens from Participants of AAML03P1

We were next interested in studying ITGA5 splice variants in human AML in more detail. RNA sequencing can have several technical variations, including widely varying read coverage for each sample and confounding dynamic range issues, which may lead to false negatives for low-abundance transcripts that more sensitive assays such as PCR (or significantly increased sequencing read depth) may not have. We therefore screened diagnostic bone marrow specimens from 216 patients enrolled on AAML03P1 for the presence of these splice variants, and used optical DNA fragment analysis to quantify the abundance of splice variant transcripts relative to transcripts that contain both exons 2 and 3 (Figure 1) by calculating splice variant ratios. While, in AML, the usefulness of ratios is perhaps best established for the prognostic assessment of FLT3/ITD mutations (where a high burden of mutated relative to wild-type alleles [“high allelic ratio”] has been demonstrated to bear a very poor prognosis [26]), splice variant ratios have been commonly utilized in the evaluation of potential cancer biomarkers in several human cancers, including AML, and found to be predictive of outcome [27–33]. Unlike human AML cell lines (ML-1, Kasumi), which did not express detectable amounts of −E2 or −E2/3, all primary AML specimens expressed variable amounts of these splice variants, although their relative abundances greatly differed among individual specimens when compared with those of ITGA5 transcripts containing exons 2 and 3, with ratios varying more than 4,400-fold (range, 0.026 to 113.7; median, 1.14) for the −E2 variant and more than 2,300-fold (range, 0.017 to 39.5; median, 0.74) for the −E2/3 variant, respectively (Figure 2B and 2C); there was only a relatively weak correlation between total ITGA5 expression levels and the −E2 variant ratio (r=0.256, P=0.0078, n=107) or the −E2/3 variant ratio (r=0.290, P=0.0025, n=107). Of note, we did not find any evidence that ITGA5 splice variant expression depended on the differentiation stage of the cell: specifically, among a subset of 20 diagnostic specimens, the expression of the −E2 and −E2/3 was very similar in corresponding less mature CD34⁺/CD33⁻ and more mature CD34⁺/CD33⁺ cell subsets (Figure 3).

Figure 3. ITGA5 splice variants in immature AML cell subsets.

RT-PCR on RNA extracted from equal numbers of less mature CD34⁺/CD33⁻ and more mature CD34⁺/CD33⁺ cells isolated from diagnostic bone marrow specimens from 6 AML patients via FACS as well as a human AML cell line (Kasumi cells), using primers recognizing ITGA5 sequences on exon 1 and exon 4. Major amplicons are indicated (arrows); also shown are water control and a sizing ladder. Data is representative of diagnostic bone marrow specimens from 20 AML patients.

Given the distribution of the ITGA5 splice variant ratios across our study cohort, we again compared the 25% of patients with the highest variant ratios with the remaining 75% of patients having lower ratios (“Q4” vs. “Q1–3”); in our cohort, a transcript ratio of 1.8 separated −E2 Q4 from Q1–3, whereas a transcript ratio of 1.3676 separated −E2/3 Q4 from Q1–3. Demographics, baseline laboratory findings, and pretreatment characteristics are summarized in Table 2. Of note, high −E2/3 splice variant ratio was associated with lower prevalence of favorable-risk disease (21% vs. 38%, P=0.027), in particular t(8;21) (2% vs. 16%, P=0.019) but, conversely, a higher prevalence of t(6;9) (9% vs. 1%, P=0.024) and FLT3/ITD (21% vs. 15%, P=0.034).

TABLE 2.

Comparison of Baseline Characteristics of Patients with Low (Patient Quartile 1–3) vs. High (Patient Quartile 4) −E2 and −E2/3 Splice Variant Ratios

Patient Characteristics	ITGA5 −E2 Variant Ratio (Patient Quartile)		P- value	ITGA5 −E2/3 Variant Ratio (Patient Quartile)		P- value
	Q1-Q3 n = 162	Q4 n = 54		Q1-Q3 n = 162	Q4 n = 54
Median Age (range), years	9.6 (0.11–20.8)	10.5 (0.2–20.8)	0.491	9.3 (0.11–19.0)	10.4 (0.2–20.8)	0.887
Male Gender, n (%)	99 (61%)	23 (43%)	0.017	94 (58%)	28 (52%)	0.428
Median WBC % (range)	31.1 (0.7–409)	24.6 (0.8–495)	0.816	29.2 (1.4–409)	28.2 (0.7–495)	0.944
Median Bone Marrow Blasts, %	68 (2–100)	74.6 (24–98)	0.032	69 (2–100)	72 (10–98)	0.259
Cytogenetics, n (%)
Normal	32 (22%)	7 (15%)	0.289	31 (20%)	8 (18%)	0.746
t(8;21)(q22;q22)	20 (14%)	5 (10%)	0.576	24 (16%)	1 (2%)	0.019
inv(16)/t(16;16)(p13.1;q22)	25 (17%)	4 (8%)	0.147	24 (16%)	5 (11%)	0.467
t(9;11)(p22;q23) or other abn 11q23	26 (18%)	13 (27%)	0.151	31 (20%)	8 (18%)	0.746
t(6;9)(p23;q34)	3 (2%)	3 (6%)	0.158	2 (1%)	4 (9%)	0.024
Monosomy 7	2 (1%)	2 (4%)	0.252	2 (1%)	2 (5%)	0.218
Del7q	1 (1%)	3 (6%)	0.046	3 (2%)	1 (2%)	1.000
−5/5q-	2 (1%)	0 (0%)	1.000	2 (1%)	0 (0%)	1.000
Trisomy 8	15 (10%)	4 (8%)	1.000	14 (9%)	5 (11%)	0.772
Other	22 (15%)	7 (15%)	1.000	19 (13%)	10 (23%)	0.092
Unknown	14	6		10	10
Risk Group, n (%)
Standard	77 (51%)	30 (61%)	0.212	78 (51%)	29 (60%)	0.270
Low	56 (37%)	12 (24%)	0.106	58 (38%)	10 (21%)	0.027
High	18 (12%)	7 (14%)	0.664	16 (11%)	9 (19%)	0.133
Unknown	11	5		10	6
Molecular Alterations, n (%)
FLT3/ITD	20 (13%)	6 (12%)	1.000	15 (10%)	11 (21%)	0.034
NPM1 Mutation	7 (5%)	1 (2%)	0.681	6 (4%)	2 (6%)	1.000
CEBPA Mutation	6 (4%)	3 (7%)	0.438	6 (4%)	3 (6%)	0.698
CNS Involvement, n (%)	8 (5%)	3 (6%)	1.000	7 (4%)	4 (7%)	0.473
Extramedullary Disease (non-CNS), n (%)	10 (6%)	1 (2%)	0.299	9 (6%)	2 (4%)	0.735

Association between ITGA5 Splice Variants and Clinical Outcome

To investigate the relationship between ITGA5 −E2 and −E2/3 splice variants and treatment effects, we studied both initial responses to chemotherapy as well as long-term outcome and found that the −E2/3 but not −E2 splice variant ratio was associated with clinical outcome (Table 3). Specifically, although the rate of morphologic CR after the 1^st induction course was similar between patients in Q4 and those in Q1–3, patients in Q4 had significantly higher likelihood of MRD as assessed by multidimensional flow cytometry after completion of the 1^st course of induction therapy (P=0.003) and inferior OS than those in Q1–3 (P=0.015; Figure 4A). Accordingly, in a univariate Cox model, the highest quartile of ITGA5 −E2/3 splice variant ratios was associated with a significantly higher hazard of death than those in Q1–3 (hazard ratio [HR]=1.76 [95% confidence interval: 1.11–2.78], P=0.016). Q4 patients also tended to have higher RR (P=0.108) and worse RFS (P=0.096; Figure 4B and Table 3).

TABLE 3.

Comparison of Treatment Responses of Patients with Low (Patient Quartile 1–3) vs. High (Patient Quartile 4) −E2 and −E2/3 Splice Variant Ratios

Outcome	ITGA5 −E2 Variant Ratio (Patient Quartile)		P- value	ITGA5 −E2/3 Variant Ratio (Patient Quartile)		P- value
	Q1-Q3	Q4		Q1-Q3	Q4
All Patients
Number of patients	162	54		162	54
Morphologic CR after 1st Induction	79%	83%	0.550	81%	75%	0.319
MRD after 1st Induction	27%	37%	0.245	23%	50%	0.003
5-year OS	63±8%	57±14%	0.349	67±8%	48±14%	0.015
5-year RFS	65±9%	52±16%	0.296	66±9%	47±17%	0.096
5-year RR	31±8%	44±16%	0.276	30±8%	48±16%	0.108
Low-Risk Patients
Number of patients	56	12		58	10
5-year OS	83±10%	72±28%	0.429	83±10%	56±33%	0.043
5-year RFS	79±12%	67±31%	0.572	81±12%	50±41%	0.146
5-year RR	19±12%	33±31%	0.497	18±11%	50±41%	0.081

Figure 4.

Clinical outcome in patients with high (Q4) and low (Q1–3) abundance of ITGA5 −E2/3 transcript variants: shown are estimates of the probability of (A) overall survival and (B) relapse-free survival for 216 patients enrolled in AAML03P1.

Given the association between cytogenetic risk and high ITGA5 −E2/3 splice variant ratio, the worse outcome for patients in Q4 could be attributable to the lower prevalence of low-risk disease in this subgroup. Consistent with this possibility, a multivariate Cox model showed that, after adjustment for cytogenetic/molecular disease risk (which, as described in Patients and Methods, classifies patients as low/standard/high risk based on cytogenetics and FLT3/ITD, NPM1 and CEBPA mutation data), patients in Q4 were no longer significantly associated with inferior OS when adjusted for disease risk (HR=1.41 [0.87–2.31], P=0.167); an alternative multivariate model using cytogenetic risk (low vs. intermediate vs. high), FLT3/ITD (negative/low allelic ratio vs. high allelic ratio) and CEBPA (mutated vs. not) as separate covariates provided a similar estimate (HR=1.54 [0.93–2.54], P=0.094; information on NPM1 was not included in this alternative model as patients with NPM1 mutations were all alive at the time of analysis). However, when patients were stratified by cytogenetic risk, it became evident that the association between high −E2/3 splice variant ratios and outcome differed in dependence of the specific risk subgroup, pointing to relevant effect modification, with associations limited to low-risk patients. Specifically, when using the same quartile cut-offs as used for the entire study cohort, low-risk patients in the highest quartile of −E2/3 variant ratios had significantly inferior OS than low-risk patients in Q1–3 (56±33% vs. 85±10%, P=0.043) and tended to have a higher RR (50±41% vs. 18±11%, P=0.081) and worse DFS (50±41% vs. 81±12%, P=0.146; Figure 5); there was also a suggestion of worse early response for low-risk patients in Q4 in that they had a lower CR rate after the first (75% vs. 91%) and second (88% vs. 98%) course of induction therapy but these differences did not reach statistical significance in this relatively small set of patients (P=0.214 and P=0.243, respectively). In contrast, the OS for Q4 and Q1–3 patients was relatively similar in the subset of standard-risk (51±19% vs. 60±11%, P=0.340) and high-risk (33±31% vs. 38±24%, P=0.952) patients, respectively.

Figure 5.

Clinical outcome in low-risk patients with high (Q4) and low (Q1–3) abundance of ITGA5 −E2/3 transcript variants: shown are estimates of the probability of (A) overall survival, (B) relapse-risk, and (C) relapse-free survival for 68 patients with low-risk disease enrolled in AAML03P1.

DISCUSSION

Alternative splicing, an important process to amplify genome complexity, has been associated with the development of cancer and the modification of the metastatic potential of malignancies [34]. Using target gene approaches, several previous studies have consistently reported splice variations in AML that may lead to changes in the biology of the leukemia or responsiveness to therapy [35–38] and could serve as predictors of outcome [39]. To the best of our knowledge, our exploratory studies described herein are the first to use transcriptome sequencing as an unbiased, genome-wide approach to identify alternative splice variants that could serve as biomarkers in AML. Two main conclusions can be drawn from our investigations. First, transcriptome sequencing is a useful method for the discovery of novel transcript variants in AML, as demonstrated by the identification of 2 splice variants in ITGA5 in specimens from pediatric patients with AML, one of which has not been described previously. An independent, PCR-based sequencing method was used to confirm the existence of these splice variants to validate the findings obtained with mRNA sequencing. Importantly, although we focused on a single gene, ITGA5, in this study as a paradigm for this strategy, transcriptome sequencing will allow a comprehensive search for splice variants across all other mRNA species and thus provide a powerful tool for the study of human AML. And second, splice variants discovered by mRNA sequencing of AML specimens can directly serve as cancer biomarker candidates. By quantifying individual ITGA5 splice variants with conventional PCR-based methods, we found highly variable abundances across a large number of AML patients. Subsequent correlative studies suggested that high levels of one of the ITGA5 splice variants, −E2/3, may be associated with certain disease characteristics, including a low prevalence of low-risk cytogenetic abnormalities as well as adverse outcome after intensive chemotherapy for newly diagnosed disease in patients classified as having low-risk disease based on cytogenetic/molecular abnormalities, providing the first evidence of prognostic significance in AML that warrants confirmation in future, larger, independent patient cohorts.

Recent research has highlighted the importance of cellular adhesion as prognostically-relevant biological factor in AML, with most studies focusing on CXCR4/CXCL12 and VLA-4 [12, 40–42]. While little is known about the functional or prognostic role of ITGA5/VLA-5 in AML, high expression of VLA-5 has been associated with poor outcome in patients with non-small cell lung, laryngeal, or ovarian cancer [43–46]. In our study, we did not find any association between ITGA5 expression and outcome, although patients within the highest quartile of ITGA5 expression more likely presented with hepatomegaly and splenomegaly than other patients, perhaps suggesting an effect of ITGA5 on tumor localization, although further studies will be required to test this idea.

The mechanisms by which the ITGA5 −E2/3 splice variant could affect outcome in AML is unknown. As depicted in Figure 1F, both −E2 and −E2/3 are severely truncated because of an early termination signal in exon 4, containing only the signal peptide as well as a relatively short aberrant amino acid sequence due to frame shift (red). Due to the lack all functional domains characteristic of ITGA5, functioning as a dominant negative is exceedingly unlikely. Future mechanistic studies will need to determine how variant ITGA5 proteins, if at all expressed as mature proteins to any significant degree, function, and whether AMLs with high levels of ITGA5 variant transcripts are differentially sensitive chemotherapy. Alternatively, high abundances of −E2/3 transcript variants may merely be a surrogate for another factor that affects chemosensitivity and prognosis.

Interestingly, we found an association between high −E2/3 transcript variant ratios and a decreased prevalence of low-risk cytogenetic abnormalities. The reason for this observation is not clear, but it may suggest that certain AML-associated alterations of cellular function (e.g., dysfunctional myeloid transcription factors) could affect the transcription of this splice variant. Most importantly, however, our correlative data suggest that the −E2/3 ITGA5 splice variant may serve as a biomarker for adverse outcome in AML, as evidenced by an increased rate of MRD (as measure of early treatment response) as well as inferior DFS and OS in patients with high −E2/3 variant ratios. Of note, this association was limited to low-risk patients and was not seen in the other cytogenetic/molecular risk groups. Thus, the presence of a high −E2/3 ITGA5 transcript ratio may denote a subset of patients who, despite being classified as “low-risk”, have a high likelihood of disease relapse. By comparison, the remaining patients in the low-risk subgroup may have excellent long-term survival, with very low likelihood of disease recurrence. Importantly, we consider these analyses exploratory, and our findings will need to be independently confirmed in larger patient cohorts before information on the ITGA5 −E2/3 splice variant should be integrated as robust biomarker into the design of risk-adapted treatment strategies, in particular as they are derived from a subgroup analysis in a relatively small number of patients. Therefore, we plan to validate our results in participants of AAML0531, a recently completed phase 3 study on >1,000 children, adolescents, and young adults with newly diagnosed AML once outcome data from that trial becomes available. These future studies may also help to determine the transcript ratio cut points that provide optimal prognostic information. If our findings can be validated in this larger set of patients, the −E2/3 ITGA5 splice variant could serve an important role in outcome prognostication in low-risk AML by virtue of designating the subset of these patients that fares very well with standard chemotherapy and, conversely, identifying those that may benefit from intensified therapy, e.g. allogeneic transplantation.