Cysteine and glycine-rich protein 2 (CSRP2) transcript levels correlate with leukemia relapse and leukemia-free survival in adults with B-cell acute lymphoblastic leukemia and normal cytogenetics

Relapse is the major cause of treatment-failure in adults with B-cell acute lymphoblastic leukemia (ALL) achieving complete remission after induction chemotherapy. Greater precision identifying persons likely to relapse is important. We did bio-informatics analyses of transcriptomic data to identify mRNA transcripts aberrantly-expressed in B-cell ALL. We selected 9 candidate genes for validation 7 of which proved significantly-associated with B-cell ALL. We next focused on function and clinical correlations of the cysteine and glycine-rich protein 2 (CSRP2). Quantitative real-time polymerase chain reaction (RT-qPCR) was used to examine gene transcript levels in bone marrow samples from 236 adults with B-cell ALL compared with samples from normals. CSRP2 was over-expressed in 228 out of 236 adults (97%) with newly-diagnosed B-cell ALL. A prognostic value was assessed in 168 subjects. In subjects with normal cytogenetics those with high CSRP2 transcript levels had a higher 5-year cumulative incidence of relapse (CIR) and worse relapse-free survival (RFS) compared with subjects with low transcript levels (56% [95% confidence interval, 53, 59%] vs. 19% [18, 20%]; P = 0.011 and 41% [17, 65%] vs. 80% [66–95%]; P = 0.007). In multivariate analyses a high CSRP2 transcript level was independently-associated with CIR (HR = 5.32 [1.64–17.28]; P = 0.005) and RFS (HR = 5.56 [1.87, 16.53]; P = 0.002). Functional analyses indicated CSRP2 promoted cell proliferation, cell-cycle progression, in vitro colony formation and cell migration ability. Abnormal CSRP2 expression was associated with resistance to chemotherapy; sensitivity was restored by down-regulating CSRP2 expression.


INTRODUCTION
B-cell acute lymphoblastic leukemia (ALL) is characterized by clonal expansion of developmentallyarrested B-cell precursors [1]. Although survival of adults with B-cell ALL has improved relapse is an important problem. Prognostic models for relapse include age, WBC levels at diagnosis, immune phenotype, cytogenetics, mutational landscape, response to induction therapy and measureable residual disease (MRD) after completing therapy [2]. Adverse cytogenetic and mutations include hypo-diploidy (< 44 chromosomes), MLL/11q23 translocations, complex cytogenetics (≥ 5 abnormalities) and t (9; 22) and/or BCR-ABL1 [2]. However, about onehalf of adults with B-cell ALL have none of the adverse prognostic variables at diagnosis making predicting relapse difficult, especially so in those with normal cytogenetics [3,4]. Identifying a new prognostic variable in these persons is important [5].
Analyzing differential expression of mRNAs is a new approach to predicting outcomes of persons with B-cell ALL. For example, in adults with B-cell ALL increased CTGF (connective tissue growth factor) and LEF1 (lymphoid enhancer binding factor-1) expression are associated with worse RFS [6,7] whereas increased BAALC (brain and acute leukemia, cytoplasmic) expression is associated with an unfavorable response to chemotherapy and worse survival [8].
A bioinformatics-based evaluation of candidate mRNAs improves efficiency compared with random sampling [9]. We used publicly available genome-wide mRNA expression data from patients with B-cell ALL to identify differentially expressed transcripts compared with normals. We identified 9 candidate genes 7 of which we validated and focused our attention on CSRP2 (cysteine and glycine-rich protein 2). CSRP2 is a member of CSRP family encoding a group of short LIM domain proteins (21 kDa) which are critical regulators of development and differentiation [10]. The three CSRPs (CSRP1-3) are preferentially expressed in muscle cells localizing to the nucleus and cytoplasm [11]. In the nucleus, they facilitate smooth muscle differentiation via interactions with transcription factors [12]. In the cytoplasm they decorate filamentous actin structures and participate in cytoskeletal remodeling [13]. CSRP2 maps to 12q21 which is reported abnormal in haematological neoplasms including T-cell ALL and lymphoma [14][15][16]. Increased CSRP2 transcript levels are associated with dedifferentiation in hepatocellular carcinoma [17]. In microarray-based analyses highexpression of CSRP2 is associated with basal-like breast cancer [18,19]. However, there were no reports regarding the role of CSRP2 in hematological neoplasms. Here, we studied levels of CSRP2 transcripts for an association with relapse probability in adults with B-cell ALL. We show increased CSRP2 transcript levels are independentlyassociated with higher cumulative incidence of relapse (CIR) and worse relapse-free survival (RFS) in adults with B-cell ALL and normal cytogenetics.

Validation of new biomarkers for B-cell ALL based on genome-wide mRNA analyses
We studied differentially-expressed genes in normal and B-cell ALL using data from the ImmuSort database (http://immusort.bjmu.edu.cn; Table 1). We focused on the top 20 differentially expressed genes based on the delta values > 45 and average rank scores (ARSs) > 80 in B-cell ALL samples. To increase reliability of our analyses we updated these data with relevant data from the Gene Expression Omnibus (GEO) [9]. The final dataset was based on 400 B-cell samples (GEO samples/GSMs, arrays or measurements) from normals and 690 samples from persons with B-cell ALL and confirmed our target gene selection.
Next we used RT-qPCR to verify differential mRNA levels of these genes in bone marrow cells from 26 adults with newly-diagnosed B-cell ALL compared with cells from 23 normals (Figure 2A). mRNA levels of CSRP2, COL5A1, RASD1 and C5orf62 were significantly increased whereas HBEGF, GNA15, FRMD4B, C19orf77 and CPNE2 were not.
Our re-analysis of cases of B-cell ALL in ImmuSort revealed most samples were from children with B-cell ALL. Consequently, we re-searched the GEO database and found the GSE34861 dataset was from adults with B-cell ALL (191 GSMs) [30]. However, this dataset was not derived from the Affymetrix Human Genome U133 Plus 2.0 Array platform (GEO platform/GPL570) but from the NimbleGen Human Expression Array (GPL15088) despite the fact both platforms include genome-wide transcriptome arrays. We next rank-normalized the dataset to derive ARS values for these genes as described [9]. We found all genes shown in Table 1 except C19orf77 not included in the GPL15088 platform were also upregulated. FRMD4B had the lowest ARS value (61.76) and the remaining genes had ARS values > 75.
Because our RT-qPCR validation studies used ABL1 as an internal control we compared ABL1 expression in children and adults with B-cell ALL. ABL1 mRNA levels were dramatically higher in adults with B-cell ALL with ARS ≥ 93 compared with children B-cell ALL with ARS ≥ 86 (Supplementary Figure 1). Consequently, we switched to a GAPDH (glyceraldehyde-3-phosphate dehydrogenase) internal control and re-studied 9 genes 7 of which were upregulated ( Figure 2B). There were insufficient data to critically-analyze FRMD4B and CPNE2 which had Q3 values which skewed higher than controls ( Figure 2B). CSRP2 was the most differentially expressed gene in our validation studies ( Figure 2).  Table 1).

CSRP2 transcript levels in B-cell ALL
Next we studied bone marrow samples from subjects with B-cell ALL at diagnosis. CSRP2 transcript levels were significantly higher (median 53%; range 0 -1066%) compared with levels in 43 normals (0.44%; 0 -1.78%; P < 0.0001; Figure 3B). Receiver-operator characteristic (ROC) curve analyses identified an area under the curve (AUC) for CSRP2 transcript levels of 0.980 (95% confidence interval [CI], 0.964, 0.996; P < 0.0001) with a maximum CSRP2 transcript level Youden index of 1.83%. Using this cut-off value the rate of CSRP2 over-expression in newly-diagnosed adult B-cell ALL was 97%. Subjects  Figure 3C). These results agree with the data set of GSE34861 [30] which included samples from 191 adults with B-cell ALL ( Figure 3D). The subjects were divided into cohorts with high or low CSRP2 transcript levels at the median CSRP2 transcript value. All subjects with an MLL translocation and more subjects with a WBC > 30 × 10E+9/L were in the high CSRP2 transcript level cohort (Table 2). There was no significant association between CSRP2 transcript levels and age, sex, platelet level, percent bone marrow blasts, immune phenotype, BCR-ABL1, IKZF1 deletion, risk group, MRD-test result at the end of induction therapy and/or post-remission therapy in multi-variate analyses ( Table 2).

CSRP2 transcript level is independently associated with CIR and RFS in subjects with normal cytogenetics
We next analyzed the prognostic impact of CSRP2 mRNA expression in the 56 subjects with normal  Table 4). Female sex and chemotherapy/allotransplant were also associated with a lower CIR and better RFS (Table 4).

CSRP2 promotes cell proliferation in vitro and in vivo
To study the biological role of CSRP2 in B-cell ALL we developed 2 B-cell lines: (1) BV173 in which CSRP2 was stably knocked-down (CSRP2-KD); and (2) a CSRP2overexpressing Ramos cell line (CSRP2-OE; Figure 5A). Proliferation was significantly decreased in CSRP2-KD BV173 cells compared with cells transfected with control lentiviral particles. In contrast, CSRP2-over-expression in CSRP2-OE markedly-increased proliferation ( Figure 5B).
We also tested whether CSRP2 expression promoted colony formation. Silencing of CSRP2 expression significantly decreased numbers of colony-forming units compared with controls. CSRP2 over-expression had the converse effect ( Figure 5C).
To further study the in vivo oncogenic activity of CSRP2 unmodified Ramos cells and CSRP2-OE Ramos cells were injected subcutaneously into the dorsal right flank of nude mice. Tumors induced by CSRP2-OE Ramos cells were significantly larger than tumors induced by control Ramos cells (P < 0.01; Figure 5D). These data indicate CSRP2 increases the oncogenicity of neoplastic B-cell lines in vitro and in vivo.

Subcellular localization of CSRP2 in neoplastic B-cells
We used cell fractionation analyses to assess subcellular localization of CSRP2 in neoplastic B-cells. CSRP2 protein was detected in the cytoplasm (C) and nucleus (N) and was more intense in the latter ( Figure 7A). Silencing CSRP2 slightly decreased nuclear CSRP2 localization in BV173 cells whereas overexpression slightly increased nuclear localization in Ramos cells ( Figure 7B).

CSRP2 knock-down increases drug-sensitivity
Drug resistance is the main reason for treatmentfailure and relapse in B-cell ALL. We studied the relationship between CSRP2 transcript levels and B-cell ALL sensitivity to dexamethasone, methotrexate, daunorubicin, cytarabine and imatinib (in BV173 with BCR-ABL1). CSRP2-KD BV173 cells showed increased sensitivity to dexamethasone, methotrexate, daunorubicin and imatinib compared with controls ( Figure 8A) whereas CSRP2-OE Ramos cells showed increased resistance to these drugs compared with controls ( Figure 8B). These data suggest down-regulation of CSRP2 may improve therapy-outcomes in adult B-cell ALL.

DISCUSSION
We used bioinformatics-based analyses to identify potentially important genes transcribed in B-cell ALL. We focused on CSRP2 because it was the most differentially expressed gene in our studies. CSRP2 is a possible oncogene in hepatocellular carcinoma and breast cancer [17,18]. CSRP1 belongs to the CSRP family and is considered a tumor suppressor gene in hepatocellular carcinoma and colorectal cancer [32,33] but a possible oncogene in gastric cancer [34]. In this study, we found CSRP2 transcripts were uniformly low in bone marrow mononuclear cells from normals whereas transcript levels were high in cells from adults with newly-diagnosed B-cell ALL.
An in silico analyses using publicly available gene expression datasets reported worse survival of women with the basal-like subtype of breast cancer with high expression of CSRP2 [18]. Our analyses of the prognostic impact of CSRP2 transcript levels on CIR and RFS in adults with B-cell ALL was complex because of confounding with other prognostic variables and therapies. We used multivariate analyses to help resolve this complexity. CIR and RFS of subjects receiving transplants from different donors were similar so these data were combined. We found a positive MRD-test at the end of induction therapy, male sex, MLL translocation and chemotherapy-only were independently-associated with a higher CIR and worse RFS whereas other variables including age, WBC, BCR-ABL1, IKZF1 deletion and CSRP2 transcript levels were not. We discuss lack of a significant association between BCR-ABL1 and CIR and RFS previously [35]. Although high CSRP2 mRNA expression was associated with CIR in univariate analyses this association was not significant in multivariate analyses possibly reflecting confounding by high levels  of CSRP2 expression in persons with MLL translocation. Consequently, we re-analyzed an association between CSRP2 transcript levels, CIR and RFS in subjects with normal cytogenetics. In multivariate regression analyses high levels of CSRP2 transcripts were independentlyassociated with a higher CIR and worse RFS regardless of post-remission therapy.
In several experimental models of B-cell ALL increased CSRP2 transcription promotes cell proliferation, migration and cell-cycle progression, a finding concordant with our bio-informatic analyses. Increased CSRP2 transcription also promotes migration of breast cancer cells via an actin bundling factor [18]. In our study CSRP2 was found predominantly in the nucleus. This contrasts with the cytoskeletal localization reported in breast cancer cells [18]. Moreover, knockdown of CSRP2 transcription increased drug-sensitivity whereas increased CSRP2 transcription increased drug resistance. These data suggest down-regulating CSRP2 transcription might decrease drug resistance and thereby decrease CIR and improve RFS. These conclusions are preliminary but warrant consideration.
There are several limitations to our study which was retrospective and susceptible to selection biases. The cohort with normal cytogenetics was not pre-specified and had relatively few subjects. Also, there is the potential for an interaction between CSRP2 transcript levels and type of post-remission therapy. Because of these limitations our conclusion need validation in a larger, independent prospective cohort. If validated, determination of CSRP2 transcript levels in adult B-cell ALL with normal cytogenetics might inform therapy-decisions. Also, consideration could be given to down-regulating CSRP2 expression as a way to reverse drug resistance.

Bioinformatics analyses
To identify possible B-cell ALL relevant genes data from the ImmuSort database dataset related to B-cell ALL and normal B-cell samples was updated and re-analyzed [9]. This database included gene expression profiles from >20,000 genes in human and mouse immune cells based on micro-array platform. Differences (delta values) in average rank score (ARS) were transformed from expressional signal value reflecting gene expression intensity. These data were used for gene expression comparison [9]. When a gene had multiple probe sets the probe with the maximum ARS was used.

Cell lines
The human B-cell ALL cell lines BV173, Sup-B15 and BALL-1 were obtained from Guangzhou Jennio Biotech Co. Ltd (Guangzhou, China). The human Burkitt lymphoma cell line Ramos was a kind gift from Professor H.S. Zhao (Peking University Health Science Center, Beijing, China). The human T-cell ALL cell lines 6T-CEM and MOLT4, the human AML cell lines KG-1, NB4 and HL60, the human chronic myelogenous leukemia cell line K562 and the human lymphoma cell lines MAVER, U937, Raji and MOLP2 were available in our laboratory. Cell lines were cultured in Roswell Park Memorial Institute (RPMI) 1640 medium (Gibco, Billings, MT, USA) containing 10% fetal bovine serum (Gibco), penicillin (100 U/ml, Gibco) and streptomycin (100 μg/ ml, Gibco). Cells were grown at 37°C in a humidified 5% CO2 atmosphere.

Subjects
Bone marrow samples were obtained from adults with B-cell ALL (N = 236) and normals (N = 43) at the Hematology Department of Peking University People's Hospital, Beijing, China and CSRP2 transcript levels assessed. Complete clinical and laboratory data were available for 168 subjects enrolled December, 2008 to June, 2014. Subjects were followed until death, loss to follow-up or June, 2016. The study was approved by the Ethics Committee of Peking University People's Hospital and informed consent was obtained according to the Declaration of Helsinki. Details of treatment regimens are reported [35]. 92 subjects (55%) received an allotransplant, 28 from an HLA-identical sibling, 1 from an HLA-matched-unrelated donor and 63 from HLAhaplotype-matched related donors [36,37]. Complete remission, refractory disease, relapse and risk-stratification were defined as described [2]. Relapse-free survival (RFS) was determined from the date of first complete remission to the date of first relapse. Cumulative incidence of relapse (CIR) was determined from the date of first complete remission to the date of first relapse or death in complete remission.

RNA preparation and RT-qPCR
Mononuclear cells were isolated from bone marrow samples by Ficoll-Hypaque™ density gradient centrifugation and RNA extracted using the TRIzol™ technique (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's instructions and cDNA synthesized as described [43]. mRNA expression levels were analyzed using SYBR ® green (Applied Biosystems, Foster City, CA, USA) to validate differential expression screened out by bioinformatics analyses with ABL1 or GAPDH as internal controls [44]. Gene transcript levels were determined using the 2 -ΔΔCt method. Average gene transcript levels in bone marrow samples from normals were used as calibrator. Other mRNA levels were determined by the TaqMan ® method [45]. CSRP2 transcript levels were normalized to ABL1 expression as recommended by the Europe Against Cancer group [46]. Copy numbers of CSRP2 and ABL1 were calculated from standard curves using the Ct values. Samples were assayed in duplicate to evaluate data reproducibility and average threshold Ct values calculated for expression analyses. Serial dilutions of plasmids expressing ABL1 and CSRP2-positive bone marrow specimens were amplified to construct standard quantification curves [41]. These curves indicated similar amplification efficiency for ABL1 and CSRP2 with slopes of −3.50 and −3.49. Detection sensitivity was approximately 1-10 copies in the plasmid DNA standards and 10E-5 in CSRP2-positive bone marrow samples. For each measurement the curve threshold amplification was set at 0.08 for ABL1 and CSRP2. Primers and probe sequences are shown in Supplementary Table 2.

Cell-cycle analyses
Cells were seeded to 6-well plates and starved by adding serum-free medium for G1 synchronization. After 24 hours, medium containing 10% fetal bovine serum was added for an additional 48 hours. Cells were fixed in 75% ethanol, stained with propidium iodide (BD Pharmingen, San Jose, CA, USA) and analyzed by flow cytometry. Results were analyzed with ModFit LT2.0 software (Coulter Electronics, Hialeah, FL, USA).

Cell proliferation and viability assay
Cell proliferation was determined with the Cell Counting Kit-8 (CCK8, Dojin Laboratories, Kumamoto, Japan) assay. Briefly, 4 × 10E+4 cells were seeded into each well of 96-well plates. 2, 4 or 7 d later 10 μl of the kit reagent was added to each well and 2 h later all plates were scanned by a microplate reader at 450 nm. CCK8 was also used to determine cell viability after drug exposures including daunorubicin, dexamethasone, methotrexate, cytarabine and imatinib (Solarbio, Beijing, China). Cells were seeded and 72 h later 10 μl of the kit reagent was added to each well and 2 h later plates were scanned by a microplate reader at 450 nm. Cell viability was assessed based on the value of fluorescent signal of live cells with no drug treatment. Experiments were performed in triplicate for 3 times independently.

Colony formation assays
Cells were suspended in 1 mL of complete MethoCult™ medium and plated in 6 well plates at a concentration of 4 × 10E+3 /well. Colonies were maintained at 37°C with 5% CO2 and 95% humidity for 7 d and then counted and scored at day 7 after staining with 1% crystal violet (Sigma, St. Louis, MO, USA). Only colonies of ≥ 50 cells were scored. Assays were done in triplicate for 3 times independently.

Cell migration assay
Cells were seeded into the upper chamber of a Transwell insert (pore size, 8 μm) in RPMI-1640 supplemented with 1% FBS. The upper chamber was then placed into the Transwell containing medium with 10% FBS in the lower chamber. After 24 h, cells remaining in the lower surface of the insert were stained with crystal violet. Experiments were conducted in triplicate for 3 times independently.

Tumor xenograft mouse model
Male athymic 6-week-old Balb/c nude mice (Beijing HFK Bioscience Co., Ltd.; Beijing, China) were housed in a controlled environment with a 12 h light/dark cycle at 23°C (± 2°C) and 40-50% relative humidity with free access to chow and water. Animal experiments were approved by the Animal Ethics Committee of Peking University Health Science Center. Mice were pretreated by intraperitoneal injections of cyclophosphamide once daily at a dose of 100 mg/kg for 2 consecutive days. Two days later, Ramos cells (1.5×10E+7 cells in 0.1 mL PBS) transduced with a lentivirus containing CSRP2 lentiviral activation particles or control lentiviral activation particles were injected subcutaneously into the dorsal right flank of 6-week-old male Balb/c nude mice (5 mice/group). Tumor diameters were measured every 2 days until day15. Tumor volume (mmE+3) was estimated by measuring the longest and shortest diameter of the tumor as described [49]. Mice were euthanized on day 15 and tumors surgically removed and photographed.

Statistical analyses
Differences across groups were compared using the Pearson Chi-square analysis or Fisher exact test for categorical data and Mann-Whitney U test or Student t-test for continuous variables. Receiver operating characteristic (ROC) curves were constructed to evaluate the predictive power of transcript levels for diagnosis of B-cell ALL. The Youden Index was used to calculate optimal cutoff points for gene transcript levels in diagnosis of B-cell ALL [50]. Survival functions were estimated by the Kaplan-Meier method and compared by the log-rank test. Cumulative incidences were estimated for relapse to accommodate competing risks. A Cox proportional hazard regression model was used to determine associations between CSRP2 transcript levels and CIR and RFS. Variables with P > 0.1 were sequentially excluded from the model and those with a P < 0.05 considered significant. A two-sided P < 0.05 was considered significant. Analyses were performed by SPSS software version 18.0 (Chicago, IL, USA), Graphpad