Identification of Immunological Parameters as Predictive Biomarkers of Relapse in Patients with Chronic Myeloid Leukemia on Treatment-Free Remission

BCR-ABL is an aberrant tyrosine kinase responsible for chronic myeloid leukemia (CML). Tyrosine kinase inhibitors (TKIs) induce a potent antileukemic response mostly based on the inhibition of BCR-ABL, but they also increase the activity of Natural Killer (NK) and CD8+ T cells. After several years, patients may interrupt treatment due to sustained, deep molecular response. By unknown reasons, half of the patients relapse during treatment interruption, whereas others maintain a potent control of the residual leukemic cells for several years. In this study, several immunological parameters related to sustained antileukemic control were analyzed. According to our results, the features more related to poor antileukemic control were as follows: low levels of cytotoxic cells such as NK, (Natural Killer T) NKT and CD8±TCRγβ+ T cells; low expression of activating receptors on the surface of NK and NKT cells; impaired synthesis of proinflammatory cytokines or proteases from NK cells; and HLA-E*0103 homozygosis and KIR haplotype BX. A Random Forest algorithm predicted 90% of the accuracy for the classification of CML patients in groups of relapse or non-relapse according to these parameters. Consequently, these features may be useful as biomarkers predictive of CML relapse in patients that are candidates to initiate treatment discontinuation.


Introduction
Chronic myeloid leukemia (CML) is a clonal hematopoietic stem cell (HSC) disorder characterized by the presence of Philadelphia (Ph) chromosome, which is produced by a reciprocal translocation between chromosomes 9 and 22, t(9;22)(q34;q11) [1]. This In summary, many clinical parameters may impact the outcome of CML after treatment interruption and several studies have tried to define biomarkers to predict successful TFR but results are contradictory so far [32][33][34]. Positive outcomes of treatment interruption have also been observed in patients who were on treatment with other immunomodulatory drugs such as IFNα, which highlights the importance of the immune response to control residual leukemic cells [7,35,36]. Consequently, in this study we analyzed several immunological parameters related to the antileukemic cytotoxic response in order to determine the best biomarkers to predict relapse of CML during controlled treatment interruption.

Participants
Blood samples were obtained from 93 patients with CML Ph+ in different stages of evolution (Table 1). Twenty healthy donors with similar age and gender distribution were recruited as controls. Peripheral blood mononuclear cells (PBMCs) were isolated by centrifugation through Ficoll-Hypaque gradient (Pharmacia Corporation, North Peapack, NJ, USA). Due to sample size limitations, not all the analyses were performed with all the samples.
For intracellular staining of interferon gamma (IFNγ), tumor necrosis factor alfa (TNFα) and granzyme B (GZB) from NK cells, PBMCs were treated for 4h at 37 • C with Hsp70 peptide 1 µgr/mL (Abcam, Cambridge, UK) and brefeldin-A (BD Biosciences, San Jose, CA, USA). Cells were then stained with antibodies against CD3, CD56 and CD16. After Data acquisition was performed in a BD LSRFortessa X-20 flow cytometer using FACS Diva software (BD Biosciences). Data analysis was performed with FlowJo_V10 software (TreeStar, Ashland, OR, USA).

KIR Haplotyping
KIR genotyping was performed using LinkSeq KIR Typing kit (One Lambda, Thermo Fisher Scientific), which allows testing all 15 human KIR genes and 2 pseudogenes by StepOne Real-Time PCR System (Thermo Fisher Scientific). Data were exported to Sure-Typer software (One Lambda, Thermo Fisher Scientific) for identification of KIR genotypes. KIR haplotypes AA and BX were assigned according to Allele Frequency Net Database [37]. B haplotypes carry one or more of the following genes: KIR2DL2, -2DL5, -3DS1, -2DS1, -2DS2, -2DS3 and -2DS5; whereas A haplotypes do not carry any of these genes.

Random Forest Algorithm
A Random Forest algorithm [38] was applied to predict the relapse of patients with CML who meet the criteria for safe treatment interruption and evaluate the resulting accuracy. The selection of these parameters was performed according to the existence of significant differences (p < 0.05) in the expression patterns when both groups were compared, and they were as follows: related to NK and NKT cells: CD56+, CD3 ± CD56 + CD16 + IFNγ+, CD3 ± CD56 + CD16 + TNFα+, CD3 ± CD56 + CD16 + GZB +, CD56 + NKG2A +, HLA-E*0101, HLA-E*0103, HLA-E*0101/HLA-E*0101, HLA-E*0103/HLA-E*0103, HLA-E*0101/HLA-E*0103, KIR2DS3, KIR2DL5, KIR haplotype AA, KIR haplotype Bx; related to TCRγδ+ T cells: CD8 ± TCRγδ +. Data from 27 patients on TFR who did not relapse and 15 patients who relapsed within 8 months after treatment discontinuation were included in the analysis. In order to avoid bias in the selection of training, testing and validation sets due to the limited number of samples, we performed a combined feature selection and classification procedure using a Random Forest classifier with a nested K-fold cross validation procedure for each competing algorithm, as previously described [39][40][41][42]. In order to ensure that every fold contains the exact number of samples, the inner loop, which was used to fine tune the hyperparameters of the Random Forest model, applied 5-fold cross validation, whereas the outer loop, which was used to test the error of the best model obtained, applied 6-fold cross validation. The variable importance measure (VIM) allowed the categorization of patients according to the normalized total reduction in the criterion brought by each feature, as determined by Gini importance method [43]. The features that show the highest importance were used to classify polymorphisms regarding their ability to predict the investigated phenotype, which in this case is the highest likelihood of relapse.

Statistical Analysis
Statistical analysis was performed with Graph Pad Prism 8.0 (Graph Pad Software Inc., San Diego, CA, USA) using ordinary one-way ANOVA and Tukey's multiple comparisons test, with a single pooled variance. P values < 0.05 were considered statistically significant and were represented as *, **, ***, or **** for p < 0.05, p < 0.01, p < 0.001, or p < 0.0001, respectively.

Profile and Cytokine Production of NK Cell Populations Induced by Treatment with TKIs
Expression of CD56 was increased 1.8-fold in patients On TKI in comparison with healthy controls (p < 0.01), whereas it was reduced 1.5-and 2.8-fold (p < 0.05) in patients with new CML diagnosis in comparison with healthy donors and patients On TKI, respectively ( Figure 1A). This high expression of CD56 was maintained, although to slightly lower levels, in patients Off TKI with sustained DMR for more than 2 years. However, CD56 was reduced 4.3-fold in patients who relapsed (p < 0.05). The expression of CD16 was analyzed in CD56+ cells in both CD3-and CD3+ populations in order to differentiate between NK ( Figure 1B) and NKT ( Figure 1C) cells, respectively. There were no significant differences among groups but interestingly, although CD3-CD56 + CD16+ cell population was increased in patients Off TKI, it was reduced to basal levels in patients who relapsed ( Figure 1B).

Profile and Cytokine Production of NK Cell Populations Induced by Treatment with TKIs
Expression of CD56 was increased 1.8-fold in patients On TKI in comparison with healthy controls (p < 0.01), whereas it was reduced 1.5-and 2.8-fold (p < 0.05) in patients with new CML diagnosis in comparison with healthy donors and patients On TKI, respectively ( Figure 1A). This high expression of CD56 was maintained, although to slightly lower levels, in patients Off TKI with sustained DMR for more than 2 years. However, CD56 was reduced 4.3-fold in patients who relapsed (p < 0.05). The expression of CD16 was analyzed in CD56+ cells in both CD3-and CD3+ populations in order to differentiate between NK ( Figure 1B) and NKT ( Figure 1C) cells, respectively. There were no significant differences among groups but interestingly, although CD3-CD56 + CD16+ cell population was increased in patients Off TKI, it was reduced to basal levels in patients who relapsed ( Figure 1B). , patients Off TKI who did not relapse (▲), patients Off TKI who relapsed (▼)). Each dot corresponds to one sample and lines represent mean ± standard error of the mean (SEM). Statistical significance was calculated using one-way ANOVA and Tukey's multiple comparisons test. * p < 0.05; *** p < 0.001; **** p < 0.0001.
We also evaluated the functionality of NK and NKT cell populations by measuring the synthesis of IFNγ, TNFα and GZB in response to Hsp70 peptide [44]. Patients Off TKI had CD3-CD56 + CD16+ NK cells in peripheral blood with high capacity to synthesize IFNγ and TNFα, which were increased 2.8-fold (p < 0.01) and 1.7-fold, respectively, in , new CML diagnosis ( ), patients On TKI ( ), patients Off TKI who did not relapse ( ), patients Off TKI who relapsed ( )). Each dot corresponds to one sample and lines represent mean ± standard error of the mean (SEM). Statistical significance was calculated using one-way ANOVA and Tukey's multiple comparisons test. * p < 0.05; *** p < 0.001; **** p < 0.0001.
We also evaluated the functionality of NK and NKT cell populations by measuring the synthesis of IFNγ, TNFα and GZB in response to Hsp70 peptide [44]. Patients Off TKI had CD3-CD56 + CD16+ NK cells in peripheral blood with high capacity to synthesize IFNγ and TNFα, which were increased 2.8-fold (p < 0.01) and 1.7-fold, respectively, in comparison with patients On TKI (Figure 2A, left and middle graphs). However, the capacity to synthesize IFNγ and TNFα of CD3-CD56 + CD16+ NK cells from patients who relapsed was reduced 34-fold (p < 0.01) and 18-fold (p < 0.01), respectively, in comparison with patients Off TKI. Similar results were observed for the induction of the synthesis of GZB from NK cells, which was reduced 1.5-fold (p < 0.01) in patients who relapsed in comparison with patients Off TKI (Figure 2A, right graph). In CD3 + CD56 + CD16+ NKT cells, the synthesis of IFNγ and TNFα was reduced 11.6-(p < 0.01) and 7.4-fold (p < 0.05), respectively, in patients who relapsed in comparison with patients Off TKI ( Figure 2B, left and middle graphs). The synthesis of GZB from NKT cells after stimulation was similar to NK cells, as it was reduced 1.4-fold (p < 0.05) in patients who relapsed in comparison with patients Off TKI ( Figure 2B, right graph).
J. Clin. Med. 2021, 10, x FOR PEER REVIEW 10 of 21 comparison with patients On TKI (Figure 2A, left and middle graphs). However, the capacity to synthesize IFNγ and TNFα of CD3-CD56 + CD16+ NK cells from patients who relapsed was reduced 34-fold (p < 0.01) and 18-fold (p < 0.01), respectively, in comparison with patients Off TKI. Similar results were observed for the induction of the synthesis of GZB from NK cells, which was reduced 1.5-fold (p < 0.01) in patients who relapsed in comparison with patients Off TKI (Figure 2A, right graph). In CD3 + CD56 + CD16+ NKT cells, the synthesis of IFNγ and TNFα was reduced 11.6-(p < 0.01) and 7.4-fold (p < 0.05), respectively, in patients who relapsed in comparison with patients Off TKI ( Figure 2B, left and middle graphs). The synthesis of GZB from NKT cells after stimulation was similar to NK cells, as it was reduced 1.4-fold (p < 0.05) in patients who relapsed in comparison with patients Off TKI ( Figure 2B, right graph). Off TKI who relapsed (▼)). Each dot corresponds to one sample and lines represent mean ± SEM. Statistical significance was calculated using one-way ANOVA and Tukey's multiple comparisons test. * p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001.

Differential Expression of Stimulatory and Inhibitory NK Markers
The expression of NKG2D in CD56+ cells from patients On TKI was similar to that of healthy donors, but it was increased more than 2-fold in patients Off TKI regarding , patients On TKI ( ), patients Off TKI who did not relapse ( ), and patients Off TKI who relapsed ( )). Each dot corresponds to one sample and lines represent mean ± SEM. Statistical significance was calculated using one-way ANOVA and Tukey's multiple comparisons test. * p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001.

Differential Expression of Stimulatory and Inhibitory NK Markers
The expression of NKG2D in CD56+ cells from patients On TKI was similar to that of healthy donors, but it was increased more than 2-fold in patients Off TKI regarding healthy donors (p < 0.05) and patients On TKI (p < 0.01) ( Figure 3A). This expression returned to basal levels in patients who lost DMR during TFR. We did not observe significant differences between groups in the expression of NKG2C. However, NKG2C increased 3.2-fold on the surface of NK cells from patients On TKI, in comparison with healthy donors (Figure 3B), and it decreased 1.7-fold in patients Off TKI who did not relapse. No changes from basal levels were observed in patients who relapsed. The expression of NKG2A increased 2.2-fold in patients Off TKI in comparison with healthy donors (p < 0.05) and patients On TKI (p < 0.01) ( Figure 3C). This expression decreased 2.5-fold in patients who relapsed (p < 0.05). The expression of the inhibitory receptor KIR2DL5/CD158f was not significantly modified between groups ( Figure 3D), but interestingly it increased 1.6-fold in patients Off TKI and returned to basal levels in CD56+ cells from patients who relapsed.
healthy donors (p < 0.05) and patients On TKI (p < 0.01) ( Figure 3A). This expression returned to basal levels in patients who lost DMR during TFR. We did not observe significant differences between groups in the expression of NKG2C. However, NKG2C increased 3.2-fold on the surface of NK cells from patients On TKI, in comparison with healthy donors (Figure 3B), and it decreased 1.7-fold in patients Off TKI who did not relapse. No changes from basal levels were observed in patients who relapsed. The expression of NKG2A increased 2.2-fold in patients Off TKI in comparison with healthy donors (p < 0.05) and patients On TKI (p < 0.01) ( Figure 3C). This expression decreased 2.5-fold in patients who relapsed (p < 0.05). The expression of the inhibitory receptor KIR2DL5/CD158f was not significantly modified between groups ( Figure 3D), but interestingly it increased 1.6-fold in patients Off TKI and returned to basal levels in CD56+ cells from patients who relapsed. , patients Off TKI who did not relapse (▲), patients Off TKI who relapsed (▼)). Each dot corresponds to one sample and lines represent mean ± SEM. Statistical significance was calculated using one-way ANOVA and Tukey's multiple comparisons test. * p < 0.05; ** p < 0.01. , new CML diagnosis ( ), patients On TKI ( ), patients Off TKI who did not relapse ( ), patients Off TKI who relapsed ( )). Each dot corresponds to one sample and lines represent mean ± SEM. Statistical significance was calculated using one-way ANOVA and Tukey's multiple comparisons test. * p < 0.05; ** p < 0.01.

Treatment with TKIs Increased CD8± TCRγβ+ T Cell Populations
Levels of CD8+ T cells were reduced 1.8-fold in PBMCs from patients with new diagnosis of CML in comparison with healthy donors ( Figure 4A). Treatment with TKIs returned CD8+ T cell populations to basal levels (p < 0.05) and they were maintained in patients Off TKI (p < 0.05). CD8+ T cells slightly increased in patients who relapsed in comparison with patients Off TKI but this increase was not statistically significant. Unconventional CD8+ TCRγβ+ T cells were increased 3.6-fold in patients On TKI in comparison with healthy donors and patients with new diagnosis of CML, still untreated ( Figure 4B, left graph). This cell population increased 4.1-fold in patients Off TKI in comparison with healthy donors (p < 0.0001) but they were reduced to basal levels in patients who relapsed (p < 0.01). The population of CD8-TCRγβ+ cells did not change in patients On TKI, but it increased 1.9-fold in patients Off TKI, in comparison with healthy donors (p < 0.05) and patients On TKI (p < 0.05) ( Figure 4B, right graph). CD8-TCRγβ+ cells returned to basal levels in patients who relapsed (p < 0.05).
turned CD8+ T cell populations to basal levels (p < 0.05) and they were mainta patients Off TKI (p < 0.05). CD8+ T cells slightly increased in patients who rela comparison with patients Off TKI but this increase was not statistically significa conventional CD8+ TCRγβ+ T cells were increased 3.6-fold in patients On TKI in c ison with healthy donors and patients with new diagnosis of CML, still untreated 4B, left graph). This cell population increased 4.1-fold in patients Off TKI in com with healthy donors (p < 0.0001) but they were reduced to basal levels in patien relapsed (p < 0.01). The population of CD8-TCRγβ+ cells did not change in patie TKI, but it increased 1.9-fold in patients Off TKI, in comparison with healthy don 0.05) and patients On TKI (p < 0.05) ( Figure 4B, right graph). CD8-TCRγβ+ cells re to basal levels in patients who relapsed (p < 0.05). , patients Off TKI who did not relapse (▲), patients Off TKI who re (▼)). Each dot corresponds to one sample and lines represent mean ± SEM. Statistical signi was calculated using one-way ANOVA and Tukey's multiple comparisons test. * p < 0.05; * 0.01; **** p < 0.0001.

Application of Random Forest Algorithm
The Random Forest algorithm was applied to evaluate the accuracy of the analyzed parameters in which there were statistically significant differences in the expression patterns to predict the relapse of patients with CML who interrupted treatment. For this analysis, data of parameters that changed with statistical significance from 27 patients Off TKI and 15 patients who relapsed during TFR were included for this analysis. After the selection of training and testing sets as described above, an accuracy of 90.48 ± 6.73% was obtained for the six iterations of the outer loop of the nested K-fold cross validation for each or not (Off TKI) (n = 8). Each dot corresponds to one sample and lines represent mean ± SEM. Statistical significance was calculated using one-way ANOVA and Tukey's multiple comparisons test. * p < 0.05. Inhibitory genes KIR2DL2/CD158b and KIR2DL5/CD158f and activating genes KIR2DS2/CD158j and KIR2DS3 were more frequent (86%, 71% (p < 0.05), 86%, and 71% (p < 0.05), respectively) in patients who relapsed in comparison with patients Off TKI (43%, 14% (p < 0.05), 50%, and 13% (p < 0.05), respectively) ( Figure 5B, left graph). Most patients who relapsed (85.7%) showed KIR BX haplotype (p < 0.05), whereas patients who did not relapse (Off TKI) showed a balanced frequency of KIR AA (50%) and BX (50%) haplotypes ( Figure 5B, right graph).

Application of Random Forest Algorithm
The Random Forest algorithm was applied to evaluate the accuracy of the analyzed parameters in which there were statistically significant differences in the expression patterns to predict the relapse of patients with CML who interrupted treatment. For this analysis, data of parameters that changed with statistical significance from 27 patients Off TKI and 15 patients who relapsed during TFR were included for this analysis. After the selection of training and testing sets as described above, an accuracy of 90.48 ± 6.73% was obtained for the six iterations of the outer loop of the nested K-fold cross validation for each competing algorithm ( Figure 6A). Therefore, 26 out of 27 patients Off TKI (96.30%) and 12 out of 15 patients who relapsed during TFR (80.00%) were correctly assigned to the right group using these features ( Figure 6B). The wrong predictions corresponded to three false negatives (prediction that one patient did not relapse when he/she had lost DMR) and one false positive (prediction that one patient had relapsed when he/she had not lost DMR). Proportionally, the non-relapsed group produced more accurate results because more samples were available to perform the training process. Besides, the Gini VIM method determined that variations in immune cell populations such as NK, NKT and CD8+ TCRγβ+ T cells and their ability to release proinflammatory cytokines, as well as HLA-E heterozygosis, were more important than KIR haplotypes for the categorization of patients with higher likelihood to relapse ( Figure 6C).

Discussion
Treatment for life with TKIs was initially recommended for patients with CML in order to avoid relapse in case of discontinuation. However, it is possible to attempt controlled treatment interruption in patients with very low or undetectable levels of BCR/ABL1 for at least two to three consecutive years [45]. Several factors may influence the outcome during TFR but the control of the residual leukemic cells by the immune system is essential [16,19]. Ninety-three patients at the chronic stage of CML, most of them (72%) with low Sokal risk, were recruited to analyze immunological parameters in PBMCs that may be useful as predictive biomarkers of relapse during controlled treatment interruption. Similar percentages of males and females were observed in all groups, although the relationship between gender and CML relapse was not conclusive. Patients who relapsed during TFR were 11 years younger than patients who did not relapse, suggesting that younger age could be a negative factor for maintaining DMR during TFR. However, due to controversial previous reports [46][47][48], current recommendations for the clinical management of CML patients do not consider age an essential factor [49].
TKIs show a direct activity against BCR-ABL+ cells but they are also immunomodulatory drugs that enhance the immune response. Treatment with dasatinib induces the expansion of LGLs, but this event has been associated with a concomitant viral infection such as CMV [16,18,[50][51][52][53][54]. This might indicate that further stimulation of the immune system with a chronic viral infection could trigger a sustained response against residual leukemic cells during TFR. In our study, most patients were seropositive for CMV in both groups of relapsed and sustained TFR ( Figure S1). Therefore, although the presence of CMV may enhance the antileukemic response, it was not a decisive factor to maintain this response and avoid relapse. On the other hand, the level of NK cells expressing CD56 was greatly increased in CML patients on treatment with TKIs, which is in agreement with previous reports that describe NK cells as the main component of LGLs [17,55]. Moreover, patients on TFR without relapsing maintained these high levels of CD56+ cells, whereas patients who relapsed showed very low levels of these cells. Therefore, the maintenance of a sustained population of NK cells seemed essential for a continuous antileukemic control in the absence of treatment. The level of NK and NKT activation was also different in both groups of patients as the expression of most essential activating NK markers was reduced in cells from patients who relapsed, such as CD16 and some CD94/NKG2 receptors. Although without statistical significance, the expression of NKG2D was reduced in patients who relapsed. The activation of NKG2D induces the reactivity of NK cells against tumor cells [56], but the expression of this marker is negatively modulated in patients with CML and other types of cancer, which facilitates the escape of tumor cells [57]. However, treatment with TKIs modulated NKG2D expression, which would favor the lytic activity of NK cells [25]. We also observed a statistically significant reduction in the expression of NKG2A in patients who relapsed, which conversely has been previously related to better prognosis [24]. The high expression of NKG2C has also been associated to reduced CML relapse [25], but it was not significantly different between patients who relapsed or not during treatment discontinuation. Accordingly, in our study, the expression of several CD94/NKG2 receptors was reduced overall in patients who relapsed, which was in accordance with the decreased functionality of NK cells in blood from these patients. This was demonstrated by the reduced synthesis of proinflammatory cytokines such as IFN or TNFα and cytotoxic proteases such as GZB after the stimulation of both NK and NKT cells from patients who relapsed.
Other cytotoxic cells such as CD8+ T lymphocytes (CTLs) may also participate in the control of leukemic cells during CML, although it was described that they cannot eliminate the cancerous clone completely due to progressive exhaustion [58]. However, the presence of specific CTLs, even at low frequency and with impaired functionality, may be a positive factor towards the control of CML during TFR since complete depletion of CTLs in animal models leads to rapid progression of the disease and death [59]. We did not find significant differences in the expression of the exhaustion marker PD-1 between PBMCs of different groups of patients ( Figure S2). However, we observed a significant increase in unconventional CD8± T cells expressing TCRγβ+ in PBMCs from patients on TFR who did not relapse. These patients had not only been on treatment with dasatinib, as described before [50], but also with other TKIs such as imatinib or nilotinib. Conversely, the presence of these cells was greatly reduced in patients who relapsed, independently of the treatment they received. CD8± TCRγβ+ cells have been appointed as a cell-based immunotherapy against cancer [60] due to their potent lytic activity that is not restricted by MHC presentation [29,30]. Therefore, high levels of CD8± TCRγβ+ cells may also contribute to maintaining DMR during TFR.
Low polymorphic HLA-E molecules are ligands of CD94/NKG2, thereby regulating lysis mediated by NK, NKT and TCRγβ+ cells during the cytotoxic response [31,61,62]. In our study, most patients who relapsed (76.9%) were homozygous for HLA-E*0103. This frequency was 1.7-fold higher than that estimated for the general population in Europe (43.3%) [63]. HLA-E*0103 has been associated with higher susceptibility to cancer diseases such as acute leukemia [64] and ovarian cancer [65], as well as with an increased risk of mortality in patients with chronic lymphocytic leukemia [66]. However, no association between cancer and HLA-E*0101 has been reported so far [67]. Both alleles differ only in one amino acid (Arg107Gly) that leads to higher stability of HLA-E*0103 on the cell surface, which increases half-life and prolongs the interaction with immune effector cells [68]. Both alleles present a limited set of peptides derived from class I leader sequences, although HLA-E*0101 may also present non-canonical peptides in the absence of HLA-class I molecules [69]. Therefore, the beneficial role of HLA-E*0101 for avoiding relapse of CML could be related to its decreased expression on the cell surface, which could result in lower inhibitory activity of NKG2A+ cells [70], and to a less restrictive peptide repertoire presentation than that found with HLA-E*0103 [71]. Furthermore, HLA-E*0103 allele has been associated with high levels of soluble HLA-E (sHLA-E), which provides protection against NK-mediated lysis [72] and usually correlates with disorders such as cancer and autoimmune diseases [67]. To our knowledge, this is the first report that describes a possible association between higher susceptibility to CML relapse and HLA-E*0103 homozygosis, which highlights the usefulness of this allele as biomarker.
The activity of NK and TCRγβ+ T cells is also modulated by the KIR repertoire expressed on the cell surface. The role of specific KIR genes as risk factors for leukemia is controversial but the absence of KIR ligands has been associated with lower risk of relapse in patients with CML who received allogeneic HSC transplantation from unrelated donors [26]. In our study, KIR2DS3 was more frequent in patients who relapsed after treatment interruption. This transmembrane glycoprotein mediates NK cytotoxicity and, although its expression is reduced in patients who develop acute leukemia [27], it is an important risk factor for CML relapse [26]. KIR2DL5B was also present in most patients who relapsed during TFR. KIR2DL5 is an NK receptor with two isoforms A and B that inhibits cytotoxicity, preventing the lysis of target cells [73]. The presence of KIR2DL5B is a predictor of inferior molecular response in CML, as well as reduced transformation-free survival and event-free survival [74]. KIR haplotype was determined for the patients included in our study based on KIR genes content. B haplotypes contain one or more Bspecific genes such as KIR2DS1, KIR2DS2, KIR2DS3, KIR2DS5, KIR2DL2, and KIR2DL5 [75]. Most patients who relapsed during TFR showed KIR haplotype BX, which may contain inhibitory KIR2DL5B and activating KIR2DS3 genes. According to our results, other studies described previously that BX haplotype is significantly associated with CML relapse [76]. However, although there is still controversy about using KIR haplotyping as a predictive parameter of CML remission outcome [17,77], the importance of this feature as a predictive biomarker was lower than that of the high levels of functional cytotoxic cells and the presence of HLA-E*0103.
Consequently, in this study we defined several parameters related to the antileukemic immune response that showed significantly different expression in PBMCs from patients who relapsed during TFR in comparison with patients on sustained TFR. Some of these parameters were in accordance with previous reports [78] and other were newly described. In order to determine the applicability as predictive biomarkers of CML relapse of the parameters evaluated in this study, we used a Random Forest algorithm. Although there was some class imbalance between the number of relapsed and non-relapsed patients, we overcame this issue by using a nested K-fold cross validation procedure for each competing algorithm. According to the model, the selected parameters could predict the classification of patients with CML in groups of relapse or non-relapse during TFR with 90.48% accuracy.

Conclusions
Several immune factors should be considered as predictors of CML relapse after treatment interruption such as low levels of CD56+ cells with low expression of CD16 and CD94/NKG2 receptors, impaired synthesis of proinflammatory cytokines or cytotoxic proteases from NK and NKT cells, low levels of CD8± TCRγβ+ T cells, as well as KIR haplotype BX and homozygosis for HLA-E*0103, which is a potential predictive biomarker never explored before. According to our results, the development of functional cell populations with cytotoxic potential in patients on successful TFR was not related to treatment with a particular TKI or to a previous infection with CMV but it seemed more dependent on the ability of the immune system of each patient to answer continuously against the residual leukemic cells. Applied to clinical practice, these biomarkers will help to decide which patients with CML may safely initiate a controlled treatment interruption with a high likelihood of non-relapse. The applicability of these biomarkers will be tested in a longitudinal study with a larger cohort of patients.

Supplementary Materials:
The following are available online at https://www.mdpi.com/2077-038 3/10/1/42/s1, Figure S1: Titers of IgGs against CMV in plasma from the different groups of patients with CML, Figure S2: Changes in the expression of PD1 in mononuclear cells from peripheral blood of patients with CML. Institutional Review Board Statement: All individuals gave informed written consent to participate. Confidentiality and anonymity were protected by current Spanish and European Data Protection Acts. This study was performed in accordance with the Helsinki Declaration and it was previously approved by the Ethics Committee (Instituto de Salud Carlos III, IRB IORG0006384, reference: CEI PI 46_2018).