Discovery of a Novel CIP2A Variant (NOCIVA) with Clinical Relevance in Predicting TKI Resistance in Myeloid Leukemias

Cancerous inhibitor of protein phosphatase 2A (CIP2A) functions as an oncoprotein by directly binding to the tumor suppressor PP2A-B56α (1, 2). CIP2A is overexpressed in a vast variety of human cancers, and high CIP2A expression has been shown to correlate with poor patient survival in a broad spectrum of human malignancies (3–6). Furthermore, CIP2A is required for malignant cellular growth in vitro and for tumor formation in vivo in a number of cancers, and its overexpression broadly promotes cancer cell drug resistance (6–11). Prior to the advancement of this potential cancer therapy target in drug development, there should be a comprehensive understanding of CIP2A protein and/or mRNA variants. In regard to the CIP2A (KIAA1524) gene, there are no genetic homologs in the human genome, and virtually no information exists about the variant forms of CIP2A at either the mRNA or protein level.

Current evidence suggests a pivotal role of alternative splicing abnormalities in leukemia pathogenesis (12, 13). Particularly in acute myeloid leukemia (AML), a prominent component of the disease is recurrent mutations in spliceosome machinery and genome-wide aberrant splicing events (14–17). Despite therapeutic progress, the outlook for AML remains unsatisfactory (18), and up to 50% of patients with AML will experience relapse (18, 19). In contrast, chronic myeloid leukemia (CML) treatment was revolutionized by the use of targeted tyrosine kinase inhibitors (TKI; refs. 20, 21). However, despite their remarkable clinical efficiency in most patients with CML, a significant portion of patients develop resistance to first-generation TKI, imatinib. The identification of such a patient population at the diagnostic phase is important so that these patients could be treated with first-line second-generation TKI therapy instead. However, there are no molecular markers that predict treatment failure at the time of diagnosis or for selection of first-line TKI therapy for optimal outcomes (22). Interestingly, both AML and CML are among very few human malignancies in which CIP2A mRNA is not overexpressed; although presumably due to posttranscriptional stabilization, CIP2A is overexpressed at the protein level, and this correlates with more aggressive disease (4, 23).

Here, we identified a novel CIP2A variant, novel CIP2A variant (NOCIVA), which is produced via alternative splicing. NOCIVA translates to a unique human protein that can heterodimerize with CIP2A, and bind to the PP2A B56α subunit. In AML and CML, high NOCIVA expression is found as a marker of poor clinical outcome. Of particular clinical relevance in CML, high NOCIVA expression is associated with resistance to the first-generation TKI, imatinib, but this effect is not seen in patients treated with second-generation TKIs, such as dasatinib or nilotinib.

For both 3′ and 5′ rapid amplification of cDNA ends, Invitrogen 3′ RACE (catalog No. 183743-019) and 5′ RACE (catalog No. 18374-058) Kits were used according to the manufacturer's protocols. Primer sequences are listed in Supplementary Table S1.

The primer and probe sequences used in this study for quantitative real-time PCR (RQ-PCR) analysis are listed in Supplementary Table S1. NOCIVA RQ-PCR No. 1, No. 2, and No. 3 assays were designed to amplify the NOCIVA-specific mRNA sequence (Supplementary Fig. S1A). The CIP2A RQ-PCR assays were designed to amplify the exon 13 and 14 (CIP2A e13 assay) or exon 20 and 21 (CIP2A e20 assay) branch site. The primer concentration in each reaction was 300 nmol/L and probe concentration 200 nmol/L.

The standard curve analysis for amplification efficiency and the melting curve analysis for NOCIVA No. 1 and NOCIVA No. 2 RQ-PCRs are shown in Supplementary Fig. S1B–S1J. The amplification efficiency of all used assays, including control genes β-actin and GAPDH, was 90% to 100%. Agarose gel electrophoresis of RQ-PCR products also revealed single band for all main assays used in this study (Supplementary Fig. S1K). Minor groove binding–based NOCIVA No. 3 RQ-PCR assay was used for the analysis of the CML study cohort 2 (Supplementary Fig. S1A).

Amplification of target cDNAs was performed using KAPA PROBE FAST RQ-PCR Kit (Kapa Biosystems) and 7900 HT Fast Real-Time PCR System (Thermo Fisher Scientific) according to the manufacturer's instructions. RQ-PCR was executed under the following conditions: 95°C for 10 minutes, followed by 45 cycles of 95°C for 15 seconds and 60°C for 1 minute. Relative gene expression data were normalized to the expression level of endogenous housekeeping genes, GAPDH and β-actin, using the 2^−ΔΔC_t method with SDS Software (version 2.4.1, Applied Biosystems) or with the Thermo Fisher Cloud Real-time qPCR Relative Quantification Application. To estimate the degree of overexpression in the AML and CML cohort 1, the expression of each gene was normalized to the expression level in a commercial normal bone marrow (BM) control sample (pooled from 56 males and females, 636591, lot 1002008, Clontech Laboratories). In CML cohort 2, a pool of cDNA from four normal individuals was used as calibrator and all the samples were normalized to GAPDH as an endogenous control. Results were derived from the average of at least two independent experiments and two technical replicates.

The study was conducted in accordance with the ethical guidelines of the Declaration of Helsinki.

Basic molecular and clinical characteristics for the 80 patients with AML are listed in Supplementary Table S2. Detailed information for AML study cohort can be found in (24). All 80 patients received regimens comprising anthracycline and high-dose cytarabine as induction therapy. Their median age was 50 years (Q₁ = 38.8 and Q₃ = 58), median overall survival (OS) was 5.4 years [95% confidence interval (CI), 2.8–7.9], and median follow-up time was 5.4 years (range, 6 days–16 years). The European LeukemiaNet (ELN) 2010 genetic risk group classification (25) was used for risk stratification (Supplementary Table S3). All patients gave informed written consent which was signed, and the local Ethical Review Board of Turku University Hospital (Turku, Finland) approved the study protocol.

This cohort comprised 35 patients with newly diagnosed chronic phase CML from the University of Liverpool CML biobank. One patient lacked follow-up data. Twenty patients received imatinib as a first-line therapy and 14 received a second-generation TKI, either dasatinib or nilotinib. Their median age was 53.5 years (Q₁ = 42.3 and Q₃ = 62), the median follow-up time was 32.5 months (range, 9–75), and median event-free survival (EFS) was 30.9 months (95% CI, 24.1–39.4). All CML cohort 1 patients gave informed written consent which was signed.

This cohort consisted of 159 patients with newly diagnosed CML from the United Kingdom–wide SPIRIT2 clinical trial (26). The samples were the first 141 biobanked samples plus 18 additional patients whose disease progressed. A total of 81 patients received imatinib and 78 dasatinib as their first-line treatment. Their median age was 53 years (Q₁ = 43 and Q₃ = 63) and median follow-up time was 60 months (range, 1–60).

The vast majority of SPIRIT2 entrants gave informed consent to donate samples to the SPIRIT2 biobank housed at Molecular Pathology Laboratory at the Imperial College Healthcare NHS Trust (London, England, United Kingdom), in addition to the informed written consent which was signed, required to enter the trial. This project was approved by the National Cancer Research Institute CML Subgroup, which has ownership of this biobank, and ethical approval was given by the Liverpool East Committee of the UK National Research Ethics Committee.

Statistical analysis was performed using SAS Software (version 9.3, SAS Institute Inc.) or GraphPad Prism (version 8.3., GraphPad Software). Normal distribution of the data was tested and, if needed, transformations were performed. All statistical tests were two-sided and declared significant at a P value of less than 0.05.

Continuous variables were summarized by descriptive statistics (median, interquartile range, and range), while frequencies and percentages were calculated for categorical data. Kruskal–Wallis test, Mann–Whitney U test, one-sample t test, and Student t test were used for analyzing continuous variables. For categorical variables, frequency tables were analyzed using Fisher exact test. A Pearson pairwise correlation analysis was performed in a gene-to-gene manner.

Univariable survival analysis was based on the Kaplan–Meier method, where stratum-specific outcomes were compared using log-rank statistics or on Cox proportional hazards regression model. To adjust for the explanatory variables (diagnosis age and expression levels of NOCIVA, CIP2A, SET, EVI1, WT1, ARPP19, TIPRL, and PME1), a Cox proportional hazards regression model was used for multivariable analysis. In multivariable analysis, covariates were entered in a stepwise backward manner.

OS was defined for all patients, measured from the date of diagnosis to the date of death from any cause. EFS was defined as the time from the date of diagnosis to the first occurrence of any of the following: death from any cause during treatment, progression to the accelerated phase or blast crisis, or loss of a cytogenetic response. Time to complete molecular response (CMR) was defined from the date of diagnosis to the date of no detectable BCR-ABL1 transcripts in two consecutive samples with good quality control values (BCR-ABL1/ABL1 ratio of ≤0.0032%, in the presence of at least 31,623 control ABL1 transcripts). Freedom from progression (FFP) was defined from the date of diagnosis of chronic phase to the date of accelerated phase or blast crisis.

Eight AML patient BM samples were used for deep RNA sequencing study. Samples with RNA quality numbers ≥9 (Fragment Analyzer, Advanced Analytical Technologies) were selected for RNA library preparation with TruSeq RNA Library Preparation Kit v2 (Illumina). Paired-end sequencing with a read length of 100 or 150 bp was performed on an Illumina HiSeq 2500 sequencer, yielding approximately 114 million read pairs per sample. Image analysis, base calling, and quality check were performed with Illumina data analysis pipelines. The sequencing reads were further trimmed of Illumina adapters using Trimmomatic (version 0.39; ref. 27). The trimmed reads were then aligned to human reference genome (hg38), allowing for novel junctions using STAR-2.6.1b (28) in two-pass mode. The aligned reads were visualized against the known CIP2A gene models and the predicted novel CIP2A isoform (NOCIVA) using the Integrative Genomics Viewer (IGV; ref. 29). The IGV sashimi plots were drawn to verify the support for the novel junction site between CIP2A exon 13 and 14.

Two NOCIVA-specific antibodies were generated by immunizing rabbits against NOCIVA-specific peptide NNKNTQEAFQVTS by BioGenes GmbH. Antibodies for immunofluorescence imaging were anti-CIP2A (sc-80659, 1:5,000, Santa Cruz Biotechnology) and Alexa Fluor–conjugated anti-mouse or rabbit secondary antibodies (488, 555, 1:300, Thermo Fisher Scientific). Antibodies used in NOCIVA binding assays included anti-B56α (sc-136045, 1:5,000, Santa Cruz Biotechnology), anti-V5 (E10/V4RR, 1:5,000, Thermo Fisher Scientific), and anti-GST (CAB4169, 1:10,000, Thermo Fisher Scientific).

CIP2A, NOCIVA, and B56α recombinant protein expression, purification, and interaction assays were performed as described in reference 1.

To identify potential mRNA variants of CIP2A (gene alias KIAA1524), rapid amplification of cDNA ends by PCR assays (3′ RACE and 5′ RACE) was employed in human cell line mRNA samples (PNT2, MDA-MB-231, and HeLa). As a result, a novel CIP2A mRNA splice variant (named here as NOCIVA) with alternative exon inclusion was identified (Fig. 1A; Supplementary Fig. S2A). NOCIVA composed of exons 1 to 13 of CIP2A fused C-terminally to a part of the intron between exons 13 and 14 (Fig. 1A). This 349-nucleotide intronic region (Fig. 1A; Supplementary Fig. S2B) is normally located within intron 13 of the CIP2A gene, more precisely ranging from 108,561,721 to 108,562,069 in Homo sapiens chromosome 3 (GRCh38.p13 reference, annotation release 109.20200228). As clear evidence that NOCIVA constitutes a functional mRNA transcript, NOCIVA mRNA contains a stop codon followed by a 330-nucleotide 3′ untranslated region (UTR) with a polyadenylation signal (PAS) AATAAA and a poly(A) tail (Fig. 1B; Supplementary Fig. S2B).

As an indication that NOCIVA mRNA is created by alternative splicing, the spliced intron region was found to be flanked by GT and AG dinucleotides (Supplementary Fig. S2B, yellow, GU-AG intron) and the junction site between CIP2A and NOCIVA contains exonic splicing silencer (ESS) matrices, especially Fas ESS and PESS-octamers (Fig. 1C). Furthermore, binding sites for many splice factors, including YB-1, SRp20, Sam68, SLM2, SRp40, and multiple hnRNPs (including hnRNP K), were predicted in the near vicinity of the junction site by SpliceAid 2 (30) and SFmap (version 1.8; ref. 31; Fig. 1C).

Validation PCR for full-length NOCIVA mRNA expression was conducted in the HeLa cell line, with forward primers targeting CIP2A exon 1 and reverse primers targeting the NOCIVA-specific 3′ end of the mRNA (Fig. 2A; Supplementary Fig. S3A for PCR assay design). In addition, validation PCR for NOCIVA expression was conducted in multiple cancer cell lines with primers specific to the mRNA sequence that codes for the unique C-terminal portion of NOCIVA (Fig. 2B; Supplementary Fig. S3B for PCR assay design). The correctly sized PCR products were subsequently sequenced to confirm the NOCIVA transcript.

To provide methodology-independent validation of the NOCIVA transcript, we analyzed the deep RNA sequencing data of eight AML patient samples. The sequencing reads were aligned to the human reference genome (hg38), allowing for novel junctions using STAR-2.6.1b in two-pass mode (28). The aligned reads were visualized against the known CIP2A mRNA sequence and the NOCIVA transcript using the IGV (32). Two of eight analyzed samples had detectable expression of reads aligning with the NOCIVA-specific sequence (Fig. 2C; Supplementary Fig. S3C). Both IGV analysis and Sashimi plots verified the existence of a novel junction site resulting in a fusion between CIP2A exon 13 and the NOCIVA-specific sequence within intron 13 of the KIAA1524 gene (Fig. 2C; Supplementary Fig. S3D).

Together, these results identify NOCIVA as a novel, alternatively spliced CIP2A variant that is expressed in multiple cancer cell lines. The NOCIVA transcript has been assigned a GenBank identifier, 2402366.

Interestingly, in NOCIVA mRNA, the 5′ end of the NOCIVA-specific intronic sequence is fused in the coding frame with the preceding 3′ end of the CIP2A mRNA sequence. After 39 nucleotides, corresponding to 13 amino acids (red text in Fig. 1B), the C-terminal tail is followed by a classical stop codon TAA. Therefore, the potential NOCIVA protein consists of 545 amino acids that are shared with CIP2A, followed by the NOCIVA-specific peptide sequence NNKNTQEAFQVTS (Fig. 1B). The novel 13-amino acid peptide sequence in NOCIVA did not match any known protein sequence in the human proteome based on a BLAST homology search (ref. 33; Supplementary Fig. S4A, BLASTP 2.8.1+, Database: nonredundant protein sequences). We used the recombinant NOCIVA peptide to generate two affinity chromatography-purified NOCIVA-specific antibodies. The specificity of the antibodies was tested by using bacterially produced NOCIVA and CIP2A proteins. Anti-NOCIVA antibodies specifically recognized NOCIVA, but did not recognize CIP2A protein fragments (Fig. 2D; Supplementary Fig. S4B for NOCIVA antibody No. 2 data). Importantly, the NOCIVA signal could be abolished by using a blocking peptide (Fig. 2D). In addition, the C-terminal CIP2A antibody did not recognize NOCIVA (Supplementary Fig. S4C).

Interestingly, whereas CIP2A resided predominantly in the cytoplasm of MDA-MB-231 breast cancer cells as expected (3), endogenous NOCIVA positivity was clearly nuclear (Fig. 2E; Supplementary Fig. S4D). A similar conclusion could be drawn from GFP fusion overexpression studies in both the MDB-MB-231 (Fig. 2F) and HeLa (Supplementary Fig. S4E) cell lines. Hence, NOCIVA expresses a novel immunogenic peptide sequence, and constitutes a CIP2A variant protein that is abundantly localized to the nucleus.

To address NOCIVA protein functions, recombinant GST-NOCIVA (CIP2A 1–545+13-amino acid peptide) and GST-CIP2A 1–560 were compared (Supplementary Fig. S4F for Coomassie staining) in terms of two functions critical for CIP2A-mediated PP2A modulation: protein homodimerization and direct binding to the B56α subunit of PP2A (1). Consistent with the location of B56α binding regions in the N-terminal part of CIP2A 1–560 (1), which is identical between NOCIVA and CIP2A, both proteins coimmunoprecipitated B56α with equal efficiency in vitro (Fig. 3A). NOCIVA could also competently heterodimerize with CIP2A 1–560, albeit with lower affinity than that seen with CIP2A 1–560 homodimers (Fig. 3B). This can be explained by the partial overlap of the CIP2A-NOCIVA fusion site with the amino acid region mediating CIP2A homodimerization (ref. 1; Fig. 3C and D), and when compared with CIP2A homodimers, in NOCIVA-CIP2A heterodimers, some of the stabilizing interactions were lost (Fig. 3E).

To compare expression levels of NOCIVA and CIP2A mRNA in biological samples, we designed and validated two RQ-PCR assays for both NOCIVA (NOCIVA No. 1 and No. 2 assays) and CIP2A (CIP2A e13 and e20 assays; Supplementary Fig. S5A). The primer sequences are listed in Supplementary Table S1, and if not otherwise indicated, NOCIVA No. 1 and CIP2A e20 were the mainstay assays. Notably, CIP2A and NOCIVA RQ-PCR assays were optimized to yield similar amplification efficiencies, allowing a direct comparison between their respective expression levels.

NOCIVA showed overall low levels of expression across normal human tissues (Fig. 4A), but consistent with its regulation from the same promoter region as CIP2A, the expression profile across different tissues, including high expression in testis, was comparable with that of CIP2A (Fig. 4A; Supplementary Fig. S5B). Although the absolute expression of NOCIVA was below 7% of CIP2A in all tissues, leukocytes, the kidney, and the pancreas had the highest NOCIVA/CIP2A mRNA ratio (Fig. 4B). To address the potential overexpression of NOCIVA in cancer, we assessed NOCIVA mRNA expression between normal epidermal keratinocytes (Ker, NHEK) and patient-derived cutaneous and head and neck squamous cell carcinoma (SCC) UT-SCC cells (34, 35) in which CIP2A was overexpressed (Supplementary Fig. S5C). NOCIVA mRNA also showed significantly elevated expression in SCC samples compared with NHEKs (Fig. 4C; P = 0.0001 by Student t test).

Considering that the highest NOCIVA/CIP2A ratio was found in lymphoid cells (Fig. 4B), we examined NOCIVA expression in lymphoid cancer cells. Relatively higher expression of NOCIVA than CIP2A was observed in most AML (F36P, Eol-1, Kasumi-1, KG-1, and MOLM-13) and CML (K562, Ku812, and Meg01) cell lines (Fig. 4D). We confirmed the expression of the NOCIVA protein in Kasumi-1 and KG-1 cells by Western blotting. Consistent with the preferential expression of NOCIVA mRNA in lymphoid cells, the expression of NOCIVA protein in AML cells exceeded that of the glioblastoma cell line, T98G (Fig. 4E). Encouraged by these results, we validated NOCIVA gene expression from a panel of 80 clinical AML (BM) and 35 CML (peripheral blood) samples. Consistent with earlier results (4, 24), 96% of patients with AML and 94% of patients with CML expressed lower levels of CIP2A than normal BM controls pooled from 56 males and females (Fig. 4F and G). However, fully supporting active alternative splicing from CIP2A to NOCIVA in myeloid cancers, 77% of AML and 65% CML samples displayed overexpression of NOCIVA compared with BM controls (Fig. 4F and G). On the basis of Pearson pairwise correlation analysis, we found that NOCIVA expression levels in AML samples were significantly correlated among the PP2A inhibitor proteins with PME1 (r = 0.43; P = 0.0002), ARPP19 (r = 0.37; P = 0.0014), and SET (r = 0.30; P = 0.0104), but not with established AML markers Wilms' tumor 1 (WT1; ref. 36) and ectopic viral integration site-1 (EVI1; ref. 37; Fig. 4H).

These results show that NOCIVA mRNA is robustly overexpressed in patient AML and CML cells compared with normal BM.

To understand the potential clinical significance of NOCIVA alternative splicing, we next analyzed the prognostic significance of NOCIVA mRNA expression in 80 patients with AML treated with intensive chemotherapy (AML study cohort; ref. 24; Supplementary Table S2). After dividing NOCIVA expression into high- and low-expression groups according to the median (2.18, Q₁ = 1.14 and Q₃ = 6.65), Kaplan–Meier estimates revealed that high NOCIVA mRNA expression was a strong indicator of shorter OS (Fig. 5A; P = 0.022 by log-rank test). Interestingly, low CIP2A e20 (Fig. 5B; P = 0.073 by log-rank test) expression was instead a borderline significant predictor of longer OS, indicating that active alternative splicing from CIP2A to NOCIVA is oncogenic in AML.

An additional analysis of the prognostic role of the studied genes for OS was performed by a multivariable Cox proportional hazards model, which included age at diagnosis and diagnostic mRNA expression levels of CIP2A e13, CIP2A e20, SET, ARPP19, TIPRL, PME1, EVI1, WT1, and NOCIVA. All the covariates were included in the model as independent predictors. After excluding the nonsignificant markers, age at diagnosis (Fig. 5C; P = 0.0013; HR, 1.07), EVI1 (P = 0.0004; HR, 1.27), and NOCIVA gene expression (P = 0.0205; HR, 1.51) were found to be independent prognostic factors for OS. It was notable that the HR for NOCIVA mRNA expression was even higher than that for EVI1 expression or diagnosis age, both of which are considered strong predictors of AML outcome in current clinical practice (37, 38).

We also analyzed the association of the studied markers with clinical characteristics and risk groups. The expression of NOCIVA did not show correlations with any of the clinical characteristics: age, sex, leukocyte or BM blast count, secondary leukemia, or presence/absence of a normal karyotype (P > 0.05 in all the analyses; Supplementary Tables S4 and S5). In regard to genetic risk group associations, neither NOCIVA nor CIP2A expression levels showed an association with the ELN 2010 risk group (Fig. 5D; P > 0.05 by Kruskal–Wallis test). On the other hand, and as expected, EVI1 mRNA expression at diagnosis was significantly different among the three risk groups, and its expression increased in accordance with the risk group (P = 0.005 by Kruskal–Wallis test).

As CIP2A and NOCIVA mRNA expression levels had opposite roles in predicting AML patient OS (Fig. 5A and B), we further tested the predictive role of the ratio of their expression in the same AML cohort. Importantly, among the four evaluated expression ratio pairs, high NOCIVA/low CIP2A selectively predicted poor OS (Fig. 5E). When pooling the other three ratios together, the high NOCIVA/low CIP2A ratio was a significant predictor of inferior OS (Fig. 5F; P = 0.042 by log-rank test). These data strongly indicate that rather than the absolute expression of these genes, the ratio between spliced and nonspliced CIP2A mRNA is the determining factor for AML survival. This finding could help develop a clinically translatable assay including the measurement of CIP2A mRNA as an internal control.

Together, these data demonstrate a risk group–independent association between a high NOCIVA/CIP2A expression ratio and a poor clinical outcome among patients with AML treated with intensive chemotherapy.

Next, we evaluated the prognostic significance of NOCIVA mRNA expression in 34 patients with newly diagnosed chronic phase CML (CML study cohort 1). Twenty patients received imatinib (first-generation TKI), and 14 received dasatinib or nilotinib (second-generation TKI) as first-line therapy. As calculation of OS was not reasonable in this cohort due to only one death at 60 months, Kaplan–Meier estimates were used to analyze the EFS. Importantly, after dividing NOCIVA expression into high- and low-expression groups according to the median (5.5, Q₁ = 0.20 and Q₃ = 20.0), analysis revealed that high NOCIVA mRNA expression was associated with significantly shorter EFS (Fig. 6A; P = 0.024 by log-rank test).

Although TKIs have revolutionized CML therapy, the identification of patients likely to have resistance to first-line first-generation TKI, imatinib, at diagnosis is still a significant unmet clinical challenge (22). Interestingly, EFS was significantly shorter in the high NOCIVA patient group treated with imatinib (Fig. 6B; P = 0.004 by log-rank test), but this was not seen among patients treated with second-generation TKIs (Fig. 6C; P = 0.429 by log-rank test). An analysis of the time to CMR was used to assess the depth of a patient's response, with CMR being the deepest form of response. Patients with high NOCIVA expression had a significantly inferior time to CMR (Fig. 6D; P = 0.039 by log-rank test). Critically, none of the patients with high levels of NOCIVA mRNA at diagnosis achieved CMR. Again, among patients treated with second-generation TKIs, no association was found between NOCIVA expression and CMR, indicating that second-generation TKI therapy may overcome the adverse effect of high NOCIVA mRNA expression.

These findings were then validated in an independent cohort of 159 patients (CML study cohort 2) from the SPIRIT2 clinical trial (26). In this cohort, 81 patients had received imatinib and 78 had received dasatinib as first-line therapy. Similar to CML study cohort 1, high NOCIVA expression at diagnosis was associated with disease progression exclusively among imatinib-treated patients. Imatinib-treated patients who subsequently progressed to blast crisis had higher expression of NOCIVA at diagnosis than patients who did not progress (Fig. 6E; P = 0.04 by Mann–Whitney U test). No significant difference was observed for patients treated with dasatinib (Fig. 6E). Interestingly, imatinib-treated patients with the highest quartile NOCIVA expression at diagnosis had significantly inferior FFP compared with patients with lower NOCIVA expression (Fig. 6F; P = 0.039 by log-rank test). Consistent with the results from CML study cohort 1, no association between NOCIVA expression and FFP was observed among the dasatinib-treated patients (Fig. 6G). Finally, as a further indication of its independent role as a leukemia oncoprotein, we did not detect a significant association between NOCIVA expression and any of the BRC-ABL transcript variants from the CML cohort 2 (Fig. 6H).

Regardless of the thorough profiling of CIP2A expression across human cancers (2, 5), the mRNA or protein variants of CIP2A remain uncharacterized. Here, we describe a novel clinically relevant CIP2A variant, NOCIVA. One of the most interesting features of NOCIVA is that it codes for a unique immunogenic C-terminal 13-amino acid peptide tail (Fig. 1B). As no homologous sequences could be identified to the 13-amino acid NOCIVA tail in the human proteome, NOCIVA can be considered a novel human protein. Strongly indicative of the alternative cellular functions of CIP2A and NOCIVA, the NOCIVA protein was found to be predominantly nuclear, whereas CIP2A mainly resides in the cytoplasm. However, similar to CIP2A, NOCIVA retains the capability to dimerize and bind to B56α, indicating that it functions similarly to CIP2A as a PP2A inhibitor protein (1). Further studies on the differential functional roles of NOCIVA and CIP2A are warranted. In particular, the functional importance of the C-terminal truncation of CIP2A in NOCIVA is an important future topic to be addressed, as currently no molecular functions have been assigned to the C-terminal CIP2A tail. Unfortunately, during the project, we failed to yet develop siRNA or CRISPR/Cas9 tools selectively suppressing NOCIVA, but not CIP2A.

AML and CML patient samples clearly displayed higher expression of NOCIVA mRNA than CIP2A, suggesting that during myeloid leukemogenesis, a splicing switch that creates NOCIVA from CIP2A is activated (Fig. 6I). This is interesting, as AML and CML are the only cancer types in which CIP2A seems to be underexpressed at the mRNA level compared with corresponding normal tissue (Fig. 6I; refs. 4, 23). At the NOCIVA junction site, ESS sequences, as well as binding sites for hnRNPs and various splice factors were found (Fig. 1C). A recent study reported expression changes in 13 hnRNPs affecting mRNA processing in AML (39), among which hnRNP A1, A2B1, and C were predicted to bind to the NOCIVA junction site. In addition, the expression of hnRNP K (40), SRSF3 (SRp20) (41), and YB-1 (42) has been shown to be altered in AML, but also to contribute to leukemia progression. Interestingly, SRSF3 (43) and YB-1 (44) have also been shown to specifically promote exon inclusion during alternative splicing. Detailed analysis of the role of these splicing factors in the alternative splicing of CIP2A to NOCIVA will be needed in the future to better understand the regulation of NOCIVA in myeloid cancers.

Our data firmly indicate that high NOCIVA mRNA expression is associated with poor clinical outcomes in both AML and CML. The data suggest that if NOCIVA is highly expressed in AML cells at the diagnosis phase, treatment of those patients with standard cytotoxic chemotherapy is not sufficient to kill these cells. Thus, we cautiously propose that NOCIVA contributes to cytotoxic chemotherapy resistance in AML. These novel findings are particularly interesting because NOCIVA expression was independent of the current genetic risk classification in AML, or other risk factors, suggesting that the evaluation of NOCIVA expression at diagnosis could provide clinically relevant predictive value. Furthermore, mechanistic follow-up studies of these findings could reveal novel mechanisms mediating AML chemotherapy resistance. In CML, high NOCIVA mRNA expression was associated with inferior EFS and FFP, as well as lower rates of CMR in patients with CML treated with imatinib. Hence, the data suggest that second-generation TKI therapy is needed to overcome the adverse effects of high NOCIVA expression. Currently, a significant number of patients with CML still receive first-generation TKI as first-line therapy due to the higher costs and worse side effect profiles of second-generation and third-generation TKIs. We propose that together with other diagnostic biomarkers, such as BCR-ABL transcript variants, the detection of NOCIVA/CIP2A ratio at the CML diagnostic phase might help in treatment decision-making between imatinib and second-generation TKIs. The urgent clinical utility of such an assay for CML patient first-line TKI therapy selection was recently highlighted by an expert group (22).

Although previously debated, mRNA expression–based diagnostics have entered clinical CML diagnostics (45). As an example, BCR-ABL1/ABL1 mRNA ratios on the international scale were shown to predict the success of treatment-free remission attempts for patients with de novo chronic phase CML (46). The technical platforms allowing a standardized clinical assessment of mRNA expression levels for CML samples include droplet digital PCR (ddPCR; ref. 46) and the Cepheid GeneXpert qPCR cartridge system (47). However, currently, there are no patient stratification markers to guide clinicians in the selection of first-line TKI therapy to achieve optimal outcomes in patients with CML (22). On the basis of our nearly identical results from two independent clinical CML cohorts, and independence of NOCIVA from other candidate markers, the determination of NOCIVA mRNA levels from patients with de novo chronic phase CML by ddPCR assay or the Cepheid qPCR cartridge system could provide significant support for clinicians in recognizing patients in need of first-line second-generation TKI therapy. Similarly, in AML, high NOCIVA expression at diagnosis could indicate a need for first-line therapy intensification.

In summary, this work describes the discovery of a novel human gene and protein product with the characteristics of a clinically relevant PP2A inhibitor in myeloid malignancies. Further studies to validate the clinical diagnostic value of NOCIVA mRNA in the identification of patients with CML resistant to imatinib, and the mechanistic basis for imatinib resistance are clearly warranted.

E. Mäkelä reports grants from The Päivikki and Sakari Sohlbergin Foundation, The Cancer Foundation Väre, The Finnish Concordia Fund, The Finnish Cultural Foundation, The Faculty of Medicine at the University of Turku, and personal fees from Turku Doctoral Program of Molecular Medicine during the conduct of the study, as well as has a patent for “A novel cip2a variant and uses thereof” (PCT/FI2018/050844) WO/2019/097122 pending and “Method for predicting response to treatment with tyrosine kinase inhibitors and related methods” (PCT/FI2020/050257) WO/2020/212650 pending. R.E. Clark reports grants from Bristol Myers Squibb during the conduct of the study, personal fees from Pfizer and grants and personal fees from Novartis outside the submitted work. L.L. Elo reports grants from Academy of Finland during the conduct of the study. C.M. Lucas reports grants from University of Turku during the conduct of the study and Bristol Myers Squibb outside the submitted work. J. Westermarck reports grants from The Sigrid Juselius Foundation, Turku University Hospital ERVA, and Business Finland during the conduct of the study, as well as has a patent for “A novel cip2a variant and uses thereof” (WO/2019/097122) pending and “Method for predicting response to treatment with tyrosine kinase inhibitors and related methods” (WO/2020/212650) pending. No disclosures were reported by the other authors.

E. Mäkelä: Conceptualization, formal analysis, investigation, visualization, methodology, writing–original draft. K. Pavic: Resources, formal analysis, investigation. T. Varila: Formal analysis, investigation. U. Salmenniemi: Resources, data curation. E. Löyttyniemi: Data curation, formal analysis. S.G. Nagelli: Investigation. T. Ammunét: Formal analysis, visualization. V.-M. Kähäri: Resources. R.E. Clark: Resources, writing–review and editing. L.L. Elo: Resources, supervision. V.K. Bachanaboyina: Formal analysis, investigation. C.M. Lucas: Resources, formal analysis, investigation, writing–review and editing. M. Itälä-Remes: Conceptualization, resources. J. Westermarck: Conceptualization, resources, supervision, funding acquisition, writing–review and editing.

Taina Kalevo-Mattila is acknowledged for superior technical assistance. Dr. Veli Kairisto (Turku University Hospital) is acknowledged for his valuable contribution to AML mRNA samples. We gratefully acknowledge the CML subgroup of the United Kingdom National Cancer Research Institute, especially Prof. Jane Apperley, and Sandra Loaiza for access to the SPIRIT2 CML samples and Newcastle University for supplying data from the SPIRIT2 trial. We thank Prof. Maria D. Odero, Dr. Otto Kauko, Dr. Juha Okkeri, Dr. and Mikko Frilander for constructive discussions and advice. This work was supported by funding from The Sigrid Juselius Foundation (to J. Westermarck), Turku Doctoral Program of Molecular Medicine (to E. Mäkelä), University of Turku faculty of medicine (to J. Westermarck and E. Mäkelä), Turku University Hospital ERVA (13283 and 13336 to J. Westermarck and M. Itälä-Remes), The Päivikki and Sakari Sohlbergin Foundation (to E. Mäkelä), The Cancer Foundation Väre (to E. Mäkelä), The Finnish Cultural Foundation (to E. Mäkelä), The Finnish Concordia Fund (to E. Mäkelä), and Business Finland TUTL (2445/31/2017 to J. Westermarck).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.