Preapoptotic protease calpain-2 is frequently suppressed in adult T-cell leukemia

Key Points

• Proteome-wide analysis of HTLV-1–infected T cells identified 17 biomarker proteins for the diagnosis of ATL or HAM/TSP patients.

Abstract

Adult T-cell leukemia (ATL) is one of the most aggressive hematologic malignancies caused by human T-lymphotropic virus type 1 (HTLV-1) infection. The prognosis of ATL is extremely poor; however, effective strategies for diagnosis and treatment have not been established. To identify novel therapeutic targets and diagnostic markers for ATL, we employed focused proteomic profiling of the CD4⁺CD25⁺CCR4⁺ T-cell subpopulation in which HTLV-1–infected cells were enriched. Comprehensive quantification of 14 064 peptides and subsequent 2-step statistical analysis using 29 cases (6 uninfected controls, 5 asymptomatic carriers, 9 HTLV-1–associated myelopathy/tropical spastic paraparesis patients, 9 ATL patients) identified 91 peptide determinants that statistically classified 4 clinical groups with an accuracy rate of 92.2% by cross-validation test. Among the identified 17 classifier proteins, α-II spectrin was drastically accumulated in infected T cells derived from ATL patients, whereas its digestive protease calpain-2 (CAN2) was significantly downregulated. Further cell cycle analysis and cell growth assay revealed that rescue of CAN2 activity by overexpressing constitutively active CAN2 (Δ₁₉CAN2) could induce remarkable cell death on ATL cells accompanied by reduction of α-II spectrin. These results support that proteomic profiling of HTLV-1–infected T cells could provide potential diagnostic biomarkers and an attractive resource of therapeutic targets for ATL.

Introduction

Human T-lymphotropic virus type 1 (HTLV-1) is a human retrovirus that is the pathogenic agent of HTLV-1–associated diseases, such as adult T-cell leukemia (ATL) and HTLV-1–associated myelopathy/tropical spastic paraparesis (HAM/TSP). Recent epidemiological studies revealed that HTLV-1 is endemic mainly in Japan, the Caribbean basin, Iran, Africa, South America, and the Melanesian islands.¹ Other estimates have shown that 20 million to 30 million people worldwide are infected with HTLV-1.² The infection is followed by a prolonged asymptomatic phase of 20 to 30 years, and 2% to 5% of the infected individuals develop ATL during their lifetime.³ ATL is one of the most aggressive hematologic malignancies characterized by increased numbers of lymphocytes with multilobulated nuclei, so-called flower cells, in blood circulation. The prognosis is severe with the median overall survival period and 5-year survival rate of ATL patients of 7 months and 20%, respectively.⁴ Recently, humanized anti-CCR4 (KW-0761) therapeutic antibody achieved a great improvement in ATL treatment in a phase 3 study. However, the disease control rate was restricted to 50%, and long-term prognosis has yet to be known.⁵ For future improvements in the management of ATL, novel biomarkers for early diagnosis are urgently needed for early therapeutic intervention.

To date, comprehensive genomic or proteomic studies using CD4⁺ T cells have been performed for this purpose,^6-9 but reproducibility and reliability of quantification results in the discovery phase were uncertain due to the diverse individual variety of HTLV-1–infected cell contents in CD4⁺ T cells. To overcome the etiologic variety of samples, we focused on the CD4⁺CD25⁺CCR4⁺ T-cell subpopulation since Yamano et al¹⁰ recently revealed that HTLV-1 preferentially infected CD4⁺CD25⁺CCR4⁺ T cells in both ATL and HAM/TSP patients. By targeting CD4⁺CD25⁺CCR4⁺ T cells, we here provide the first quantitative proteome map illustrating molecular disorders in pathogenic human T cells directly associated with the onset or progression of ATL. The comprehensive and comparative interpretation of total proteome in infected cells, especially between asymptomatic HTLV-1 carriers and ATL patients, could immediately lead to specific candidates for biomarkers and drugs.

Another challenge to emphasize in this study is our recently established proteomic profiling technologies. It is indisputable that the greater the number of clinical samples analyzed, the more confidently statistical analysis can be undertaken in order to identify diagnostic markers and druggable targets. Despite this fact, previous proteomics reports could not provide high-throughput quantitative methodologies that were sufficient for dealing with even more than 10 clinical samples, excepting a study utilizing a surface enhanced laser desorption/ionization time of flight mass spectrometer. Although the surface enhanced laser desorption/ionization time of flight method drastically improved the performance in both quantification and throughput, allowing relative quantification analysis for 96 samples in several hours, at most only 250 unidentified protein peaks were detectable. In the present study, we integrated the proteomics server for the huge data set “Expressionist” (Genedata A.G., Basel, Switzerland) with high-end mass spectrometers to maximize the quality and quantity of protein catalogs transferred from mass spectrometers. We first describe the discovery phase providing a panel of novel diagnostic molecules from quantification of 14 064 peptides and identification of 4763 proteins. As the functional validation phase, we further examined the physiological potential of an identified diagnostic marker candidate, calpain-2 (CAN2), particularly concerning the association of its activity with survival or progression of ATL cells.

Materials and methods

PBMCs and cell lines

Peripheral blood mononuclear cells (PBMCs) from 6 normal donors, 5 asymptomatic carriers, and 9 HAM/TSP patients used in the screening analysis were collected in the St. Marianna University School of Medicine. Those from 9 ATL patients were collected in the Imamura Bun-in Hospital. PBMCs from 4 ATL patients used for the validation experiments were provided by the Joint Study on Predisposing Factors of ATL Development. The others from 4 HAM/TSP patients were collected in the St. Marianna University School of Medicine. The use of these human specimens in this study was approved by individual institutional ethical committees: the Ethical Committee of Yokohama Institute, RIKEN (approval code Yokohama H22-3); the Ethical Committee of St. Marianna University School of Medicine; the Institutional Review Board of Imamura Bun-in Hospital; and the Ethical Committee of the University of Tokyo (approval code 10-50). This study was conducted in accordance with the Declaration of Helsinki.

SO-4, KOB, and KK1 cells were kindly provided by Dr Yasuaki Yamada, cultured in RPMI 1640 supplemented with 10% fetal bovine serum (Cell Culture Bioscience, Tokyo, Japan), 100 kU/L interleukin 2 (Cell Science & Technology Institute Inc., Tokyo, Japan), and 1 × antibiotic-antimycotic solution (Sigma-Aldrich, MO). Jurkat, SUP-T1, CCRF-CEM, and MOLT-3 cells were cultured in RPMI 1640 supplemented with 10% fetal bovine serum and 1 × antibiotic-antimycotic solution. All cell lines were grown at 37°C in 5% CO₂. CD3⁺CD4⁺CD25⁺CCR4⁺ T cells were isolated with anti-CD3-FITC (eBioscience, San Diego, CA), anti-CCR4-PE (Becton Dickinson, CA), anti-CD4-Cy7 (eBioscience), and anti-CD25-APC (eBioscience) on a Cell Sorter JSAN (Bay Bioscience, Hyogo, Japan).

Sample preparation for mass spectrometric analysis

The CD4⁺CD25⁺CCR4⁺ T cells were washed with phosphate-buffered saline 3 times and lysed in denaturation buffer (8 M urea in 50 mM ammonium bicarbonate). After sonication, reduction with 5 mM tris(2-carboxyethyl)phosphine (Sigma-Aldrich) at 37°C for 30 minutes, and alkylation with 25 mM iodacetamide (Sigma-Aldrich) at room temperature for 45 minutes, lysates were digested with Trypsin GOLD (Promega, WI) with protein/enzyme ratio of 25:1 at 37°C for 12 hours. The digested peptides were desalted with Oasis HLB µElution plate (Waters, MA). The collected samples were dried up with a Vacuum Spin Drier (TAITEC Co. Ltd., Saitama, Japan) and subjected to mass spectrometric analyses.

Liquid chromatography tandem mass spectrometry (LC/MS/MS)

The digested peptides were separated on a 0.1 × 200 mm homemade C₁₈ column using a 2-step linear gradient, 2% to 35% acetonitrile for 95 minutes and 35% to 95% acetonitrile for 15 minutes in 0.1% formic acid with a flow rate of 200 nL/min. The eluting peptides were analyzed with a QSTAR-Elite mass spectrometer (AB Sciex, CA) in the smart information-dependent acquisition mode of Analyst QS software 2.0 (AB Sciex). The other parameters on QSTAR-Elite were shown as follows: DP = 60, FP = 265, DP2 = 15, CAD = 5, IRD = 6, IRW = 5, curtain gas = 20, and ion spray voltage = 2000 V.

Two-dimensional (2D) LC/MS/MS

Tryptic digests of CD4⁺CD25⁺CCR4⁺ T cells were dissolved in 10 mM ammonium formate in 25% acetonitrile and fractionated by a 0.2 × 250 mm monolith strong cation exchange column (GL Science, Tokyo, Japan). Peptides were eluted with an ammonium formate gradient from 10 mM to 1 M in curve = 3 mode for 70 minutes using a Prominence high-performance liquid chromatography (HPLC) system (Shimadzu Corporation, Kyoto, Japan). The eluate was fractionated into 20 fractions and analyzed individually by LTQ-Orbitrap-Velos mass spectrometer (Thermo Scientific, Bremen, Germany) accompanied with the Ultimate 3000 nano-HPLC system. The fractionated peptide samples were separated with the same gradient used in the QSTAR-Elite system described previously and analyzed by LTQ-Orbitrap-Velos acquiring a full MS scan on Fourier-transition mode with MS resolution = 60 000 and simultaneously MS/MS scans for the 20 most intense precursor ions in each MS spectrum on ion-trap mode with regular resolution. Other important parameters for LTQ-Orbitrap-Velos were as follows: capillary temp = 250, source voltage = 2 kV, MS scan range = mass-to-charge ratio (m/z) 400 to 1600, acquire data dependent CID MS/MS for top-20 intense precursors, and dynamic exclusion enabled during 30 seconds. For protein identification, all MS/MS spectra were searched against SwissProt database version 2012_06 (20 232 human protein sequences) using SEQUEST algorithm on ProteomeDiscoverer 1.3 software (Thermo Scientific) with the following parameters: MS tolerance = 3 ppm, MS/MS tolerance = 0.8 Da, maximum missed cleavages = 2, enzyme = trypsin, taxonomy = Homo sapiens, fixed modification = carbamidomethylation on cysteine, and variable modification = oxidation on methionine. We accepted the protein identification satisfying the false discovery rate <1% by Percolator false discovery rate estimation algorithm on ProteomeDiscoverer.

Label-free quantification analysis

The LC/MS/MS raw data were imported into the Expressionist RefinerMS module and subjected to the following data processing and relative quantification steps. The total work flow on the RefinerMS module is shown in supplemental Figure 1 (see the Blood Web site). The LC/MS/MS raw data set from 29 clinical samples was displayed in 2D planes (m/z vs retention time [RT]). The chromatogram grid was applied to all planes: scan counts = 10, polynom order = 3, and RT smoothing = 0. The planes were simplified by subtracting background noises using chromatogram chemical noise subtraction: RT window = 50 scans, quantile subtraction = 50%, and RT smoothing = 3 scans. After the noise subtraction, data points with intensity <10 were clipped to zero. The RT variety among 29 planes was adjusted by chromatogram RT alignment: RT transformation window = 0.2 minutes, RT search interval = 5 minutes, m/z window = 0.1 Da, and gap penalty = 1. Peaks were detected by chromatogram summed peak detection: summation window = 5 scans, overlap = 50, minimum peak size = 4 scans, maximum merge distance = 10 points, peak RT splitting = true, intensity profiling = max, gap/peak ratio = 1%, refinement threshold = 5, consistency threshold = 0.8, and signal/noise threshold = 1. The detected peaks were grouped into isotopic clusters derived from each molecule using 2-step chromatogram isotopic peak clustering. The first parameters were as follows: minimum charge = 1, maximum charge = 10, maximum missing peaks = 0, first allowed gap position = 3, RT window = 0.1 minute, m/z tolerance = 0.05 Da, isotope shape tolerance = 10, and minimum cluster size ration = 1.2. The second parameters were as follows: minimum charge = 1, maximum charge = 10, maximum missing peaks = 0, first allowed gap position = 3, RT window = 0.1 minute, m/z tolerance = 0.05 Da, and minimum cluster size ration = 0.6.

Expression vectors and siRNA

For the Δ₁₉CAN2 construct, the CAPN2 fragment was amplified with primers 5′-CATGTCGACTCCCACGAGAGGGCCATCAAGT-3′ and 5′-CATTCTAGATCAAAGTACTGAGAAACAGAGCC-3′ from pBlueBacIII CAPN2 and cloned into pEFBOS-Myc. Prior to the overexpression experiments, we confirmed that the sequence of the inserted CAPN2 fragment was identical to the Mammalian Gene Collection sequence (accession number BC021303). The 5-μg vector DNA was transfected to 1 × 10⁶ cells. The small interfering RNAs (siRNAs) against SPTAN1, PTMS, HSPE1, and SHMT2 and siRNA universal negative control were purchased from Sigma-Aldrich. The 500-pmol siRNA oligo was transfected into 1 × 10⁶ cells. The vectors and siRNAs were transfected into all cell lines except CCRF-CEM by Amaxa Nucleoportator transfection Kit V (Lonza, Cologne, Germany) and CCRF-CEM by Kit C (Lonza).

Cell cycle analysis and proliferation assay

For the cell cycle analysis, 1 × 10⁵ to 2 × 10⁵ cells were washed and agitated in 0.1% Triton-X (Sigma-Aldrich) with 100 ng/mL of ribonuclease (Sigma-Aldrich). Following addition of 1 μg/mL propidium iodide, the flow cytometric analysis was performed on FACScalibur (Becton Dickinson). The data analysis was performed using FlowJo software (Tree Star Inc., OR). Doublet events were eliminated from analyses by proper gating on FL2-W/FL2-A primary plots before histogram analysis of DNA content. Cell proliferation was estimated by measuring cell metabolic activity using Cell Counting Kit-8 (Dojindo, Kumamoto, Japan) following the manufacturer’s recommendation.

Western blotting

Cells were lysed in lysis buffer [1% NP-40, 2 mM EGTA, 2 mM MgCl₂, 150 mM NaCl, 20 mM tris(hydroxymethyl)aminomethane–HCl (pH 7.5), 10% glycerol, containing the protease inhibitor cocktail Complete (Roche, IN)] and subjected to sodium dodecyl sulfate–polyacrylamide gel electrophoresis and transferred onto PVDF membranes. Following blocking with 4% Block Ace (Yukijirushi Nyugyo Inc., Tokyo, Japan), membranes were incubated with anti-myc (9E10; Sigma-Aldrich) or anti–α-II spectrin (Abcam, Cambridge, UK) antibodies. Membranes were then incubated with horseradish peroxidase–conjugated anti-mouse IgG (GE Healthcare, NJ) or anti-rabbit IgG (GE Healthcare), respectively, and visualized with Western Lightning kit (Perkin Elmer, MA).

Multiple reaction monitoring (MRM)

CD4⁺ T cells were isolated from PBMCs using flow cytometry. The tryptic digests of the isolated cells were analyzed by 4000 Q-TRAP mass spectrometer (AB Sciex) accompanied with Ultimate 3000 nano-HPLC system. The LC gradient was as follows: 2% to 30% acetonitrile for 10 minutes and 30% to 95% acetonitrile for 5 minutes in 0.1% formic acid with a flow rate of 300 nL/min. The MRM transitions monitored were m/z 409.7/375.2 for α-II spectrin (SPTA2); m/z 538.3/889.5 for parathymosin (PTMS); m/z 507.3/147.1 for heat shock 10-kDa protein, mitochondrial (CH10); m/z 490.3/147.1 for serine hydroxymethyltransferase, mitochondrial (GLYM); and m/z 581.3/919.5 for β-actin, respectively. Individual peak areas were normalized by the peak area of β-actin. Data acquisition was performed with ion spray voltage = 2300 V, curtain gas = 10 psi, nebulizer gas = 10 psi, and an interface heating temperature = 150°C. The parameters were set as follows: declustering potential = 60, entrance potential = 10, collision cell exit potential = 10, and dwell time for each transition = 10 seconds. Collision energy was optimized to achieve maximum intensity for each MRM transition as follows: 34.03 V for m/z 409.7/175.1, 24.68 eV for m/z 538.3/889.5, 23.32 eV for m/z 507.3/147.1, 37.57 eV for m/z 490.3/147.1, and 31.58 eV for m/z 581.3/919.5.

Results

Quantitative proteome profiling of CD4⁺CD25⁺CCR4⁺ T cells

A schematic overview of the screening approach is shown in Figure 1. To identify diagnostic markers expressed in HTLV-1–infected T cells, a CD4⁺CD25⁺CCR4⁺ subset of PBMCs from 6 uninfected volunteers, 5 asymptomatic carriers, 9 HAM/TSP patients, and 9 ATL patients was isolated by flow cytometry (Figure 2). The averaged proportion of CD4⁺CD25⁺CCR4⁺ cells in CD4⁺ T cells from 4 clinical groups was 6.48 ± 2.46%, 13.17 ± 13.06%, 20.55 ± 10.73%, and 55.83 ± 22.40%, respectively, indicating that the occupancy of viral reservoir cells varied drastically among both pathological groups and even individuals within a group. Enrichment of the infected cells was confirmed by viral load measurement of the used samples (supplemental Figure 2). As reported previously,¹⁰ the viral load of CD4⁺CD25⁺CCR4⁺ cells (37.91 copies/100 cells on average) was ∼10 times higher than that of CD4⁺CD25^–CCR4^– cells (4.12 copies/100 cells on average), indicating that the former cells were evidently the HTLV-1–enriched fraction. This fact strongly supports the importance of enriching pathogenic cells for rigorous quantitative biomarker discovery.

Figure 1.

Schematic overview of proteomic profiling for CD4⁺CD25⁺CCR4⁺ cells. PBMCs were collected from 6 normal donors, 5 asymptomatic carriers, 9 ATL patients, and 9 HAM/TSP patients, followed by isolation of the CD4⁺CD25⁺CCR4⁺ subset using the cell-sorting system. The statistical candidate selection steps, including LC/MS/MS data processing, label-free quantification, and statistical analysis, were performed on the Expressionist proteome server. The protein identification database was separately established based on 2D LC/MS/MS analysis. ND, normal donors; AC, asymptomatic carriers.

Figure 2.

Representative sorting results of CD4⁺CD25⁺CCR4⁺ cells. After labeling with anti-CD3-FITC, anti-CD4-Cy7, anti-CD25-APC, and anti-CCR4-PE, the CD3⁺CD4⁺CD25⁺CCR4⁺ fraction was isolated. The averaged content ± standard deviation (%) of CD25⁺CCR4⁺ cells out of CD3⁺CD4⁺ cells was calculated for each clinical group and is displayed in the upper right section of the panels.

An accurately adjusted number of CD4⁺CD25⁺CCR4⁺ cells from 29 cases were digested with trypsin and subjected to LC/MS/MS analysis individually. Because recent mass spectrometers often deal with data on the order of hundreds of megabytes per sample, it has been considered almost impossible to calculate a data set larger than a gigabyte from large-scale clinical samples on desktop computers. Hence, we constructed a proteomics server equipped with a 12-core central processing unit, 36 SAS hard disks, and 192-GB physical memories driving the Expressionist, which was designed to combine the database module, the data processing module, and the statistical analysis module into a single integrative platform for genomics, proteomics, and metabolomics. The detailed work flow for data processing and quantification for 29 LC/MS/MS raw data was described in the “Materials and methods” and is illustrated in supplemental Figure 1. Finally, 68 454 nonredundant peaks were detected and grouped into 37 143 isotopic clusters, or molecules. As tryptic peptides should appear as multivalent ions in electrospray ionization mass spectra, 23 079 singly charged ions were removed, resulting in utilization of 14 064 peptide signals for further statistical selection of diagnostic markers.

Statistical identification of candidate diagnostic markers for ATL

A stepwise statistical extraction was employed for the effective identification of proteins, which demonstrated specific up- or downregulation in the ATL group. In the first stage, a 4-group Kruskal-Wallis test was performed to roughly extract the candidates showing a significantly distinct expression level among 4 clinical groups. Here we set the cutoff line at P < .01 and obtained 1170 first candidate peptides simply because the isolated peptide set using this criterion showed the best performance in the following prediction model.

Next, we selected the final candidates by the support vector machine–recursive feature elimination algorithm in the Expressionist Analyst module. Support vector machine–recursive feature elimination is a candidate elimination method based on SVM, which enabled us to improve the classification outputs by selecting the best-performing peptide set among initially provided candidates.¹¹ As a result, a combination of 91 peptides showed the lowest misclassification rate (7.78%) in a leave-one-out cross-validation test (Figure 3A-B). To evaluate the classification efficiency of 91 selected candidates, the principal component analysis was performed. Figure 3C shows the three-dimensional plot of 29 clinical samples based on the 3 best-explainable components, which illustrated statistically clear segregation among the 4 clinical groups. These assessments indicated that the 91 peptides should be a sufficient set of classifiers that closely associated with the pathological characteristics of the 4 clinical groups.

Figure 3.

Statistical extraction of candidate therapeutic targets. The 14 064 nonredundant peptides detected were subjected to a 4-group Kruskal-Wallis test (ND, AC, ATL, and HAM), resulting in identification of 1170 first candidates (P < .01). ND, normal donors; AC, asymptomatic carriers. (A) Next, the Expressionist ranking method further narrowed down the candidates to 91 peptides based on SVM-REF so that the misclassification rate in the cross-validation test became minimum, 7.8%. (B) The predicted classification result by leave-one-out cross-validation test. The 27 out of 29 cases were successfully classified into the true classes. (C) The three-dimensional plot shows the additional assessment for the classification power of 91 classifiers by principal component analysis. Comp. 1 to 3 indicate principal components 1 to 3.

Based on an independently constructed 6279-protein identification database for CD4⁺CD25⁺CCR4⁺ cells using 2D LC/MS/MS (see details in “Materials and methods”), 19 peptides among the 91 candidate peptides were successfully assigned to 17 proteins listed in Table 1. The mass spectrometric quantification profiles for the 19 peptides are also shown in Figure 4 (box plots).

Table 1.

List of 17 protein classifiers for categorization of normal donors, asymptomatic carriers, HAM/TSP, and ATL

Accession	Protein name	P value (Kruskal-Wallis test)	m/z	RT	Charge	Peptide score	Identity or homology threshold	Sequence
LPPL	Eosinophil lysophospholipase	2.3.E-03	409.722	47.4	2	36.3	27	MVQVWR
CH10	Heat shock 10-kDa protein, mitochondrial	2.5.E-03	430.721	40.6	2	26.2	21	GGIMLPEK
PRG2	Bone marrow proteoglycan	2.4.E-03	528.271	64.6	2	31.6	28	RLPFICSY
MOES	Moesin	8.1.E-04	532.253	26.8	2	46.2	29	EKEELMER
MNDA	Myeloid cell nuclear differentiation antigen	9.4.E-03	647.863	69.1	2	67.3	24	SLLAYDLGLTTK
GLYM	Serine hydroxymethyltransferase, mitochondrial	8.7.E-04	408.551	21.6	3	31.1	18	HADIVTTTTHK
PTMS	Parathymosin	9.7.E-04	453.875	17.8	3	41.2	25	AAEEEDEADPKR
TPIS	Triosephosphate isomerase	9.1.E-03	472.266	71.0	3	54.0	28	QSLGELIGTLNAAK
HSP71	Heat shock 70-kDa protein 1A/1B	9.7.E-03	563.307	65.5	3	93.8	21	IINEPTAAAIAYGLDR
CD6	T-cell differentiation antigen CD6	7.7.E-03	592.306	37.8	3	62.7	22	VLCQSLGCGTAVERPK
ANXA1	Annexin A1	4.4.E-04	612.347	61.5	3	57.0	17	RKGTDVNVFNTILTTR
ANXA6	Annexin A6	2.3.E-03	669.017	70.9	3	54.7	16	AMEGAGTDEKALIEILATR
SPTA2	Spectrin α chain, brain	5.4.E-03	409.718	28.8	2	42.7	30	EAGSVSLR
GLYM	Serine hydroxymethyltransferase, mitochondrial	1.1.E-03	428.240	57.0	2	42.8	27	SGLIFYR
DRB1s	HLA class II histocompatibility antigen, DRB1-1, 4, 10, 11, 13, 15, 16 β chain	1.0.E-02	478.216	25.8	2	55.9	25	AAVDTYCR
CAN2	Calpain-2 catalytic subunit	2.4.E-03	483.253	54.0	2	66.6	29	SDTFINLR
STAT1	Signal transducer and activator of transcription 1-α/β	7.3.E-03	486.290	21.7	2	39.1	29	KILENAQR
PRG2	Bone marrow proteoglycan	9.4.E-04	497.742	49.2	2	31.6	27	FQWVDGSR
CXCL7	Platelet basic protein	1.3.E-03	528.761	43.1	2	51.7	28	ICLDPDAPR

Figure 4.

Summary of quantitative features for the 17 protein classifiers identified. The 19 box plots (see Table 1 for protein names) show the results of mass spectrometric quantification and protein identification. We finally identified 19 peptides out of 91 candidates in Figure 3, which were assigned to 17 proteins. Proteins identified from 2 distinct peptides were shown as GLYM (peptides 1 and 2) or PRG2 (peptides 1 and 2). The y-axis indicates normalized relative intensity of peptides in mass spectrometric data. ND, normal donors; AC, asymptomatic carriers.

Recovering CAN2 activity induced cell death in ATL cells

Our diagnostic marker discovery for ATL identified an enzyme-substrate pair, CAN2 and SPTA2, which demonstrated significantly aberrant expression level in ATL patients (Figure 4). Interestingly, the intensities of the 2 proteins in 27 screening cases (without 2 statistical outliers in Figure 4) showed a clearly inverse correlation (R² = 0.395, Figure 5A). To examine whether CAN2 downregulation and/or SPTA2 upregulation might be essential for the growth of ATL cells, the enzymatic activity of CAN2 was rescued by overexpressing the constitutively active form of CAN2 (Δ₁₉CAN2) in 3 ATL cell lines, SO-4, KOB, and KK1. After 36 hours of transfection, significant inhibition of cell proliferation (Figure 5B) and induction of sub-G1 transition was observed by activation of CAN2 in 3 ATL cells, but not in 4 non-ATL leukemia cell lines (Figure 5C). Furthermore, overexpression of Δ₁₉CAN2 drastically attenuated the expression level of SPTA2 in the ATL cell line SO-4 (Figure 5D), but not in the non-ATL leukemia cell line Jurkat (Figure 5E). On the other hand, an additional cell proliferation assay using siRNA against SPTAN1 revealed that reduction of SPTA2 was not sufficient for the induction of cell death for ATL cells (supplemental Figures 3 and 4).

Figure 5.

Rescue of CAN2 activity induced cell death in ATL cells. (A) Correlation between CAN2 and SPTA2 expression level in 27 cases. (B) Cell proliferation was measured by MTT assay on SO-4, KOB, and KK1 cells 36 hours after transfection of mock vector or Δ₁₉CAN2. *P < .05; **P < .01 by Student t test. (C) Overexpression of Δ₁₉CAN2 significantly accelerated cell death in 3 ATL (SO-4, KOB, and KK1) and 4 non-ATL (Jurkat, SUP-T1, CCRF-CEM, and MOLT-3) cell lines. **P < .05 by Student t test. The drastic attenuation of SPTA2 expression was observed after transfection of Δ₁₉CAN2 in SO-4 cells (D), but not in Jurkat cells (E). The immunoblot of anti-myc tag confirmed the expression of exogenous Δ₁₉CAN2.

In addition, 3 proteins (PTMS, CH10, and GLYM) were also found to be upregulated in ATL cells. To address the roles of these proteins, a cell proliferation assay was conducted using 3 ATL cell lines treated with siRNAs against PTMS, HSPE1 (gene symbol of CH10), or SHMT2 (gene symbol of GLYM) (supplemental Figure 4). As a result, suppression of the SHMT2 gene induced significant growth inhibition for all 3 ATL cell lines. Although siHSPE1-treated KOB cells showed a statistically significant decrease in cell growth rate, siHSPE1 and siPTMS had only partial or no effects on proliferation of ATL cell lines. To further confirm whether the overexpression of SPTA2, PTMS, CH10, or GLYM protein would be an ATL-specific molecular signature, the expression levels of these proteins in 8 clinical samples were evaluated by the mass spectrometric quantification technology MRM (supplemental Figures 5 and 6). Expression of SPTA2, GLYM, and CH10 in cells derived from ATL patients was significantly higher than that in cells derived from HAM/TSP patients. The level of PTMS also showed a clearly increasing tendency in the ATL patient group. Taken together, these results suggested that the deprivation of CAN2 activity and upregulation of GLYM in HTLV-1–infected T cells might have a key role at the onset or progression of ATL.

Discussion

In the past decade, proteomics technologies have developed dramatically for the purpose of obtaining more and more comprehensive and sensitive proteome maps in cells or clinical specimens. The performance of mass spectrometers in particular has exhibited remarkable progress; however, as for sensitivity and throughput, it has still been difficult to identify biomarkers from crude samples including body fluids or total cell lysate. A major reason could be that the range of protein concentration in the analyte is indeed much larger than the dynamic range of recent mass spectrometers.¹² The other essential factor to be improved for clinical proteomics is the capacity of the bioinformatics platform to allow analysis of a sufficient number of clinical samples in order to statistically overcome the significant individual variability.¹³

Concerning the first issue, we previously developed and applied various focused proteomic applications targeting molecular biochemical features including glycan structure biomarkers^14-16 and low-molecular-weight peptide biomarkers.¹⁷ The preenrichment of subproteome fractions effectively reduces the complexity of crude samples and allowed us to identify potential serum cancer biomarkers successfully. Through our previous knowledge, we provide an approach for investigating infectious diseases by employing virus-infected cell-focused proteomics. In addition to HTLV-1, for instance, isolation of HIV-infected cells is highly desired because the frequency of these cells in AIDS patients’ PBMCs is ∼1 out of 10⁴ to 10⁵ cells.¹⁸ Actually, we successfully demonstrated the effect of HTLV-1–infected cell isolation on the elimination of individual variability (Figure 2, supplemental Figure 2) and reliable identification of disease state–associated proteins (Figures 4 and 5). We further showed the potential of the next-generation bioinformatics platform Expressionist to remove the constraint on the capacity of data size acquired from high-end mass spectrometers. Expressionist covered whole discovery steps from processing of raw mass spectrometer data to statistical analyses (Figures 1 and 3, and supplemental Figure 1) and, importantly, could perform quantification analysis using a basically unlimited number of clinical samples. Hence, in parallel with the development of mass spectrometers, high-specification and inexpensive OMICS server systems are necessary for future diagnostic marker and therapeutic target discoveries using hundreds or thousands of clinical specimens.

In this study, we focused on the CD4⁺CD25⁺CCR4⁺ T-cell subpopulation in which T helper 2, T helper 17, and regulatory T (Treg) cells were mainly involved.¹⁰ The purpose for which we used this subset was to technically enrich the preferential viral reservoir cells and to strengthen reliability of screening results. However, investigating proteome behaviors of these subtypes in HTLV-1–associated diseases is also important physiologically because it has been frequently reported that deregulated Treg plays significant roles in pathogenesis of ATL and HAM/TSP. Indeed, aberrant proliferation of Treg cells is considered the main cause of immunodeficiency in ATL patients because of their innate immunosuppressive functions,¹⁹ whereas abnormal production of interferon γ from infected Treg cells might induce chronic spinal inflammation in HAM/TSP patients.²⁰ Given the list of our 17 classifier proteins, activation of signal transducer and activator of transcription 1-α/β is the well-known key factor for HAM/TSP,²¹ whereas upregulation of heat shock 70-kDa protein 1A/1B, CH10, and PTMS were reported in many other types of tumors.^22-24 The association of these 4 proteins with the etiology of HAM/TSP and ATL would be evident according to the previous work, supporting that our other candidates might similarly have a direct impact on the transformation of Treg cells after infection of HTLV-1. Particularly, the specific upregulation of GLYM in ATL cells represents the first evidence that excessive folate metabolism might be essential for the progression or survival of ATL cells because GLYM is a fundamental enzyme catalyzing the supply of glycine accompanying the conversion of tetrahydrofolate to 5,10-methylenetetrahydrofolate.²⁵ Indeed, the suppression of GLYM expression, which was confirmed to be upregulated in ATL patients, resulted in significant reduction of cell growth. This observation suggests that diminishing GLYM expression or enzyme activity could be a promising strategy for molecular-targeting treatment of ATL. Together with the downregulation of CAN2 in the ATL cells shown in Figure 5, the proteins listed in Table 1 could provide the molecular basis for not only interpretation of physiological mechanisms in ATL or HAM/TSP but also development of novel therapeutic agents for HTLV-1–associated diseases.

CAN2 belongs to a Ca²⁺-regulated cytosolic cysteine protease family, which includes 14 calpain isoforms.²⁶ The enzymatic activity of calpain is implicated in diverse physiological processes, such as cytoskeletal remodeling, cellular signaling, and apoptosis.²⁶ As an example of a spectrin-mediated apoptosis pathway, it was reported that CAN2 produced SPTA2 breakdown products following traumatic brain injury.²⁷ Because SPTA2 interacts with calmodulin and constructs the membrane cytoskeletons, its breakdown is considered a process of membrane structural changes during cell death.^28,29 This fact is concordant with our finding in ATL, suggesting that accumulation of SPTA2 in ATL cells can be attributed to the suppression of CAN2 expression and contribute to circumvent apoptosis. In the analysis of basal levels of CAN2 and SPTA2 in 7 cell lines (supplemental Figure 7), 3 ATL cell lines showed endogenous expression of CAN2 and moderate levels of SPTA2. On the other hand, 4 non-ATL leukemia cells demonstrated very high expression of SPTA2 and undetectable levels of CAN2. Although we found the downregulation of CAN2 and accumulation of SPTA2 in ATL cells, this tendency might be more distinctive in HTLV-1 (–) leukemia cells. Taken together, even though the expression level of CAN2 was indeed suppressed in ATL cells, the CAN2-SPTA2 apoptotic pathway itself might remain normal. In contrast, this pathway was considered to be impaired at multiple stages in HTLV-1 (–) leukemia cells because CAN2 expression was completely diminished (supplemental Figure 7) and overexpression of CAN2 could not reactivate the CAN2-SPTA2 apoptotic pathway (Figure 5B-E). In these cells, not only genetic downregulation of CAN2 but also inhibition of CAN2 enzymatic activity might be involved in the carcinogenesis.

In conclusion, comprehensive proteomic profiling of HTLV-1–infected T cells provided 17 disease-associated signature proteins, which have great potential for future clinical use as diagnostic biomarkers. As we described regarding the relationship between the CAN2-SPTA2 pathway and ATL phenotypes, further individual functional analyses will contribute to understanding the detailed molecular mechanisms involved in the onset or progression of HAM/TSP and ATL.

Authorship

Contribution: M.I. and K.U. designed the study, performed experiments, analyzed results, and wrote the manuscript; A.T. and N.S. performed experiments; N.A., T.S., A.U., and Y.Y. collected the clinical samples and performed flow cytometric experiments; Y.N. and H.N. revised the manuscript; and all authors discussed the results and commented on the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Koji Ueda, Laboratory for Biomarker Development, Center for Genomic Medicine, RIKEN, General Research Building 6F, Institute of Medical Science, 4-6-1 Shirokanedai, Minato-ku, Tokyo, Japan, 1088639; e-mail: k-ueda@riken.jp.