NK Cell Genotype and Phenotype at Diagnosis of Acute Lymphoblastic Leukemia Correlate with Postinduction Residual Disease

The natural killer (NK) cell is a lymphocyte important in cancer control (1). A diverse repertoire of cell surface inhibitory and activating receptors dictates the response of the NK cell to both normal and abnormal cells. Once activated, the NK cell can lyse the target directly or secrete cytokines and chemokines that indirectly stimulate the adaptive immune response of the host (2).

NK cell cytotoxicity has been studied in hematologic malignancies, including acute myelogenous leukemia (AML; refs. 3, 4) and acute lymphoblastic leukemia (ALL; refs. 5, 6). In addition, many have examined the best use of NK cells to achieve maximum benefit in hematopoietic stem cell transplantation for AML or ALL (7, 8). Because of differences in gene content and variable surface receptor expression, not all NK cells are alike, however (7). We aimed to understand which NK cell–related factors are most relevant in childhood ALL therapy.

Early response to treatment is the most predictive marker for the risk of ALL relapse (9). It is because early response assessment by minimal residual disease (MRD) testing provides a precise and objective measurement of drug sensitivity of leukemia cells and the efficacy of treatment as well as host pharmacogenetics and immune surveillance (10). In this study, we prospectively investigated the relationship between features of the NK cells at the time of ALL diagnosis and MRD at the end of induction chemotherapy.

Patients, 18-years-old or younger, with newly diagnosed B or T lymphoblastic leukemia were enrolled on the institutional protocol Total Therapy Study XVI (TotalXVI). A total of 244 patients were enrolled from October, 2007 to October, 2011 at the time this laboratory research was performed. These studies were approved by the Institutional Review Board. Written informed consent was obtained from the guardians, and assent from the patient, as appropriate.

All patients received intrathecal chemotherapy with methotrexate, cytarabine, and hydrocortisone on days 1 and 15. Patients with Philadelphia chromosome, MLL rearrangement, hypodiploidy (<44 chromosomes), or WBC count ≥100 × 10⁹/L at presentation received additional intrathecal treatment on days 8 and 22 and those with T-cell ALL, t(1;19)/TCF3-PBX1, CNS 2 status (<5 WBC/μL CSF with blasts), CNS 3 status (≥5 WBC/μL of CSF with blasts or cranial nerve palsy), or traumatic lumbar puncture (>10 RBC with blasts) received additional intrathecal treatment on days 4 and 11. Induction treatment began with prednisone, vincristine, daunorubicin, and polyethylene glycol (PEG) asparaginase, followed by cyclophosphamide plus cytarabine and thioguanine. Dexamethasone was substituted for prednisone in patients with early T-cell precursor immunophenotype and mercaptopurine for thiopurine in those with thiopurine methyltransferase deficiency. An extra dose of PEG asparaginase was given if the day 15 MRD was ≥1% and fractionated cyclophosphamide at 300 mg/m² per dose every 12 hours for four doses was used instead of 1,000 mg/m² for one dose if the day 15 MRD was ≥5%.

Bone marrow aspirate for MRD determination was performed in all patients between days 38 and 42 of remission induction, when the absolute neutrophil count recovered to >300/μL, WBC >1,000/μL, and platelets >50,000/μL. MRD was measured with flow cytometry or PCR of immunoglobulin or T-cell receptor gene rearrangement, as described previously (10). MRD <0.01% was considered negative.

Peripheral whole blood (3 mL) was obtained at diagnosis for immunophenotyping by flow cytometry of NK cells and leukemic blasts. The following antibodies were used to enumerate NK cells and their surface expression: anti-KIR2DL1 (143211), anti–NTB-A (292811), anti-ULBP1 (170818), anti-ULBP2 (165903), anti-ULBP3 (166510), anti-MICA (159227), and anti-MICB (236511) were obtained from R&D Systems; anti-KIR3DL1 (DX9), anti-KIR2DL2/3 (CH-L), anti-CD11a (HI111), anti-NKG2D (1D11), anti-TRAIL (RIK-2), anti-DNAM (DX11), anti-2B4 (2-69), and anti-granzyme B (GB11) from BD Biosciences; anti-NKG2A (Z199), anti-NKp30 (Z25), anti-NKp44 (Z231), anti-NKp46 (BAB281), and anti-FAS (CIB2) from Beckman Coulter; and anti-FASL (14C2) from AbD Serotec. Blasts obtained at diagnosis were evaluated by anti–NTB-A (292811) from R&D Systems; anti-ICAM (84H10), anti-FAS (CIB2), and anti-CD48 (J4-57) from Beckman Coulter; anti-Nectin (R2.525) from BD Biosciences, anti–TRAIL-R1 (DJR1) and anti–TRAIL-R2 (DJR2-4) from BioLegend; anti-PVR (TX21) from MBL International; anti–HLA-ABCEFG (W6/32) from Dako; and anti–PI-9 (7D8) from AbD Serotec.

PCR amplification was performed with the Olerup SSP KIR Genotyping Kit (Qiagen). The following gene alleles were tested: 2DL1*001-010, 2DL2*001-005, 2DL3*001-007, 2DL5A*0010101-0010102 and 0050101-005010102, 2DL5B*00201010-004 and 00601-009, 2DS1*001-004, 2DS2*0010101-005, 2DS3*00101-004, 2DS4*00101010-00103, 003, 004, 006, 007 and 009, 2DS5*001-008, 3DL1*00101-002, 00401-009, 01501-044, 056 and 057, 3DL2*00101-021, 3DL3*00101-031, 3DS1*010-014, 045-049N and 055, 2DP1*00101-003, and 3DP1*001-006, 00301-00301 and 004-006. KIR2DL1 functional allele typing and KIR (killer-cell immunoglobulin-like receptor) ligand typing were performed using an SNP assay on the 7900 HT Sequence Detection System (Applied Biosystems) as described previously (11).

Quantification of NK cell transcripts was performed for the following: KIR2DS1-5, KIR2DL1-4, KIR3DL1-2, KIR3DS1, NKp30, NKp44, NKp46, and NKG2D. Real-time quantitative PCR (RQ-PCR) was performed by the 7900 HT Fast Real-Time PCR system. cDNA was obtained by performing reverse transcription on RNA using the Invitrogen Vilo Kit (Invitrogen).

PCR were performed with either 2.5 μL cDNA (from 12.5-ng RNA), 2.5 μL of standards, or 2.5 μL of RNAse-free water for the negative control. The master mix for all receptors, except KIR2DS3, included 12.5 μL of SYBR Green PCR master mix (Applied Biosystems), 2 μL of forward and reverse primers at a concentration of 5 μmol/L, and 6 μL of RNAse-free water. Standards were diluted in EB buffer from 10⁵ copies down to one copy per 2.5μL. Cycling parameters were as follows: 50°C for 2 minutes, 95°C for 10 minutes, 40 cycles at 94°C for 30 seconds, 58°C for 30 seconds, and 72°C for 45 seconds, and one cycle at 72°C for 3 minutes. The dissociation steps were 95°C for 15 seconds, 60°C for 20 seconds, and 95°C for 15 seconds.

The PCR for KIR2DS3 was performed with master mix containing 10 μL of Fast SYBR Green and 3.5 μL of RNAse- and DNAse-free water. Cycling parameters were as follows: 95°C for 20 seconds, 40 cycles at 95°C for 1 second, and 60°C for 20 seconds, and a hold at 72°C for 3 minutes. The dissociation steps were the same as the setup for all other receptors.

Forward and reverse primers for receptors 2DL1-4, 3DL2, 2DS1-5, and 3DS1 are described previously (12). Forward and reverse primers for the remaining receptors were as follows: KIR3DL1 5′-CAAGCTCCAAATCTGGTAACCC-3′ and 5′-CCAACTGTGCGTATGTCACC-3′, NKp30 5′-CCCACTTGCTTCTTCCCGTTTCC-3′ and 5′-CACCACCAGCCGAGTCCCATTCC-3′, NKp44 5′-TCTCTAAGTCCGTCAGATTC-3′ and 5′-GATGGTAGATGGAGACTCAG-3′, NKp46 5′-ACGGGACTCCAGAAAGACCAT-3′ and 5′-CAGGCCCATCCGAAGGA-3′, and NKG2D 5′-GGCTCCATTCTCTCACCCA-3′ and 5′-TAAAGCTCGAGGCATAGAGTGC-3′.

We used the Fisher exact test and Wilcoxon rank sum test to compare categorical and continuous baseline variables between patients with positive or negative MRD at the end of induction chemotherapy, respectively. Univariate logistic regression was used to test associations between KIR genotype, NK cell receptor surface expression, leukemia blast characteristics, KIR haplotypes, and MRD status, respectively. The Wilcoxon rank sum test was used for comparison of the NK cell receptor mRNA transcript level between patients with positive or negative MRD. ROC curves, AUC of the ROC, sensitivity, and specificity were calculated to determine levels of PI-9, FasL, granzyme B, and NKp46 that best differentiate positive MRD versus negative MRD using the maximum Youden index method implemented in R package pROC (13, 14). The smooth ROC curves were obtained using the method of maximum-likelihood fitting of univariate distributions (method = “fitdistr” in pROC). An MRD risk system based on the cutoffs of these four variables and the presence of KIR2DL5A was developed using the logistic regression model. All the reported P values are two-sided and are considered significant if <0.05 because of exploratory nature of the study. Statistical analyses were performed with R-2.15.1 (15).

Table 1 shows the presenting clinical and biologic features of the 244 patients studied and the distribution of these features according to the MRD status at the end of induction. Not surprisingly, the MRD-negative group was younger than the MRD-positive group (P = 0.0087).

Table 2 shows the proportion of patients positive for each KIR gene according to their MRD status. The frequency distribution of positivity in the entire cohort was no different than the general United States population (16). In a univariate logistic regression analysis, KIR2DL5A, KIR2DS1, and KIR2DS3 were statistically significant and were associated with increased odds of MRD by 3.05- to 4.5-fold. Notably, these three genes are found exclusively in the B haplotypes (Fig. 1).

The patients' KIR genotypes were then categorized on the basis of the presence of centromeric (Cen) or telomeric (Tel) A and B haplotype motifs as described previously (7). The odds of being MRD positive in patients with Tel-A/B were 2.85 times than those with Tel-A/A [95% confidence interval (CI), 1.075–7.73; P = 0.035]. No statistically significant difference was found between Cen-A/A and Cen-A/B. The effect of Cen-B/B and Tel-B/B was not examined because of the small number of patients (<15).

Receptor transcripts and surface expression were quantified by RQ-PCR and flow cytometry, respectively. As there was no statistically significant difference between the MRD-positive group and MRD-negative group in the number of NK cells in the blood at diagnosis (P = 0.12), the quantification of receptor transcripts was adjusted on the basis of the amount of RNA (expressed as per μg of RNA, Table 3), rather than the amount of NK cells. Patients with positive MRD had a higher mRNA level of Tel-B–associated KIR2DS1 (P = 0.022) and a lower level of Tel-A–associated KIR3DL1 (P = 0.007) and KIR2DL1 transcripts (P = 0.011) than did those with negative MRD (Table 3).

The association of MRD with surface expression of KIR2DL1, KIR2DL2/3, and KIR3DL1 as enumerated by flow cytometry was analyzed in the context of the presence of their corresponding ligand. The odds of being MRD positive in a patient with HLAC2 ligand increased by 2.01-fold (95% CI, 1.05–3.85) for every percentage increase in NK cells expressing KIR2DL1 on their surface (P = 0.034). There were no significant association between MRD and receptor–ligand pairs for KIR2DL2, KIR2DL3, and KIR3DL1.

The proportion of NK cells expressing NKG2A, NKp46, and FAS receptor were significantly associated with the MRD status. The NK cells from the MRD-positive patients had lower frequency of expression of NKG2A (mean ± SD, 27.1% ± 16.1% vs. 38.3 ± 14.3%, P = 0.0012) and NKp46 (37.2% ± 37% vs. 56.6% ± 29%, P = 0.0074), but higher expression of FAS (44.8% ± 43.2% vs. 23.8% ± 31.8%, P = 0.032) than NK cells from MRD-negative patients. In terms of surface density, the mean fluorescence intensity (MFI) of FASL was lower in the MRD-positive group than the MRD-negative group (348 ± 201.3 vs. 543.1 ± 358.7, P = 0.029).

Among the 13 biomarkers, ICAM, PI-9, and NTBA were statistically associated with MRD status. The leukemia blasts from the MRD-positive patients had a greater frequency of expression of ICAM (mean ± SD, 34.9% ± 30.2% vs. 22.5% ± 24.5%, P = 0.035) and PI-9 (mean ± SD, 63.3% ± 35% vs. 45.6% ± 35.2%, P = 0.027) than blasts in the MRD-negative group. Furthermore, the amount of PI-9 in the MRD-positive group was greater than that in the MRD-negative group (MFI mean ± SD, 996 ± 677.8 vs. 691.6 ± 527.5, P = 0.038). The amount of NTB-A expressed on the blasts was also greater in the MRD-positive group (MFI mean ± SD, 235.9 ± 186.5 vs. 184.1 ± 79.4, P = 0.04). There was no statistically significant difference between the MRD-positive group and MRD-negative group in the expression of HLA-Class I on the leukemia blasts (MFI mean ± SD, 20.5 ± 19.7 vs. 18.8 ± 17.4 × 10³, P = 0.64).

Among the significant results of immunophenotyping and KIR genotyping, five NK-related parameters were revealed by ROC analyses to be highly correlated collectively with the risk of MRD, including PI-9, FasL, granzyme B, NKp46, and KIR2DL5A. The probability of positive MRD ranged from as low as 3.32 × 10⁻⁹% to as high as 81.6% (Table 4), using cutoffs that were established statistically from individual ROC (>650, 415, 4,070, and 30 for the first four predictors, respectively, and the presence of KIR2DL5A). The five-parameter composite model had a sensitivity of 100% and specificity of 80% (Fig. 2); the AUC of its ROC curve improved to 0.93 when compared with those of individual factors (all AUC < 0.67).

The activity of NK cells is regulated by the engagement of their inhibitory and activating surface receptors with ligands on the leukemia lymphoblasts. Cytotoxicity is achieved through perforin/granzymes or FasL (2, 17). Herein, we presented our comprehensive evaluation of these molecules in NK cells and target leukemia cells, and identified five NK-related parameters that correlated strongly with MRD after induction chemotherapy for childhood ALL.

KIR diversity is generated through haplotype gene content on chromosome 19q13.4, allele polymorphism, and stochastic surface expression (18, 19). In our study, the presence of genes KIR2DL5A, KIR2DS1, and KIR2DS3 was significantly associated with a positive MRD. Interestingly, these three genes are exclusively found in group B KIR haplotypes, suggesting that B haplotypes are associated with a poor induction response in pediatric ALL. When we compared the haplotypes A/A versus A/B directly, the Tel rather than the Cen-B motifs appeared to be the key determinant, as the odds of being MRD positive with Tel-A/B genotypes were significantly higher than that of Tel-A/A, whereas no difference was observed between Cen-A/B and Cen-A/A. In contrast with our study, some investigators have found that the group B KIR haplotypes were associated with improved overall survival and event-free survival, as well as lower transplant-related mortality (20, 21). However, others corroborate our findings of an A haplotype benefit and a B haplotype disadvantage (22, 23). The A haplotypes were associated with a complete molecular response in chronic myelogenous leukemia (24) and an improved disease-free survival in patients receiving a haploidentical T-cell–depleted stem cell transplant for leukemia. The risk of relapse in the latter study increased with the number of activating donor KIR genes present (25). Because B haplotypes typically contain larger numbers of activating KIR genes than the A haplotypes (which contain only two activating genes KIR2DL4 and KIR2DS4 that are often disrupted by mutations rendering them nonfunctional; refs. 26–28), these data collectively suggest that the risk of relapse might be dependent on the activating gene content in the B haplotypes. In this regard, NK cells possessing Tel-B–associated KIR2DS1 have been shown to be hyporesponsive because of tolerance induced by homozygous C2 ligands (29, 30).

To further dissect the role of activating KIRs from that of inhibitory KIRs in MRD, we directly measured the patients' KIR repertoire by RQ-PCR. This approach had two advantages. First, the RQ-PCR allowed us to analyze all 15 family members of KIRs individually, as monoclonal antibodies for flow cytometry are either not available or cannot discern the activating versus inhibitory KIRs. Second, previous studies have shown that genotypes do not always correlate with phenotypes (31, 32). In our cohort, the MRD-positive group had more mRNA transcripts of the B haplotype–associated KIR2DS1-activating receptor than the MRD-negative group, collaborating with our genotype findings of detrimental Tel-B effect. In contrast, the MRD-negative group expressed more mRNA of inhibitory receptors KIR2DL1 and A haplotypes associated inhibitory KIR3DL1, supporting the hypothesis of the importance of inhibitory KIR acquisition during NK cell licensing for missing-self recognition (33–35). The licensing or arming model proposes that inhibitory receptor and MHC ligand interactions educate the NK cells to kill when it encounters a target cell missing the cognate ligand (36–39). The greater the number of inhibitory receptors on NK cells, the stronger the cytotoxic responsiveness (40, 41). The absence of these inhibitory receptors, even in the presence of activating receptors, leaves the NK cell hyporesponsive. Alternatively, our Tel-B genotype and RQ-PCR data also support the disarming model, which states that NK cells exhibiting more stimulatory signal become dampened overtime (42, 43). In our patients, those with a poor response to induction chemotherapy are typified by the Tel-B genotypes with high level of expression of activating KIR2DS1 and low level of expression of inhibitory KIR2DL1 and KIR3DL1, a phenotype found in immature NK cells and unlicensed NK cells (44, 45). In fact, KIR2DL5A is the most significant risk factor for MRD. Although the function and ligand of KIR2DL5A are unknown, this KIR gene is almost in complete linkage disequilibrium with the absence of Tel–KIR3DL1 and the presence of Tel–KIR2DS1, KIR2DS3, KIR2DS5, and KIR3DS1; NK cells of these genotypes (more activating and less inhibitory gene content) would be predictably hyporesponsive based on the arming and disarming model (42, 43).

In addition to KIRs, other NK cell receptors and effector molecules are crucial determinants of antileukemia response (2, 7, 17, 46). An NCR^dull phenotype, for example, has been associated with poor leukemia control in adult AML (3, 4), in line with our data showing that MRD was associated with low expression of NKp46. Furthermore, the amount of FasL in the NK cells of the MRD-positive group was smaller than that of the MRD-negative group; FasL is essential to induce the death pathway in the target cells (47).

The leukemia cell-intrinsic factors may also play a role in NK cell escape. In our patients, MRD positivity was associated with the presence of C2 ligand and the abundance of intracellular PI-9. Whereas C2 may inhibit KIR2DL1, PI-9 can resist granzyme B (48) or FasL-mediated cytotoxicity, rendering both cytotoxic pathways ineffective (49).

Postinduction MRD has been established to be one of the most important prognostic markers for childhood ALL and MRD-based risk-adaptive continuation regimens have been shown to improve patient outcomes (50). Herein, we identified five NK-related factors that collectively correlated strongly with postinduction MRD. For instance, patients in groups 1 to 16 as listed in Table 4 had a <1 × 10⁻⁵% risk of positive MRD; these patients would be expected to have an excellent outcome after treatment with TotalXVI regimen. In contrast, patients in groups 17 to 25 had an intermediate risk ranging from 0.2% to 10%, whereas those in the high-risk groups had a risk ranging from 13.3% to 81.6%. The composite model had favorable operating characteristics (100% sensitivity and 80% specificity, corresponding with a 100% negative predictive value), thus providing strong support to our biologic hypothesis that NK cells with a strong effector phenotype in the setting of decreased leukemia resistance are associated with better leukemia control.

No potential conflicts of interest were disclosed.

Conception and design: E.M. Sullivan, R. Bari, C.-H. Pui, W. Leung

Development of methodology: E.M. Sullivan, R. Bari

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Jeha, B. Rooney, M. Holladay, C.-H. Pui, W. Leung

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.M. Sullivan, G. Kang, C. Cheng, B. Rooney, W. Leung

Writing, review, and/or revision of the manuscript: E.M. Sullivan, S. Jeha, G. Kang, C. Cheng, R. Bari, C.-H. Pui, W. Leung

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E.M. Sullivan, S. Jeha, S. Schell, M. Tuggle, C.-H. Pui

Study supervision: W. Leung

The study was supported in part by research grant no. CA-21765 from the NIH and by the Assisi Foundation of Memphis. Research funding from NIH and Alpha Tau Pharmaceuticals (to C. Cheng). Research funding from NCI CA36401 and GM92666 (C.-H. Pui).

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.