The prognostic impact of tet oncogene family member 2 mutations in patients with acute myeloid leukemia: a systematic-review and meta-analysis

Titles and abstracts of initially selected studies were reviewed depending on inclusion and exclusion criteria. Once contradictions occurred, they would be solved by discussion. After articles meeting demands were chosen, their full-texts would be scanned to distinguish appropriate studies and quality assessment according to Newcastle-Ottawa Scale would be used in the final procedure to determine the studies that could be included in our meta-analysis.

Relevant information was extracted from ultimately determined studies and summarized. Basic extracted information included authors of studies, publication year, publication journal, region, sample size of cases, frequency of mutant TET2, median age of study cohort, detecting method, cohort type and therapy regimen, while clinical information contained sex ratio, laboratory results, French-American-British classification, cytogenetic risk classification, incidences of normal karyotype and complex karyotype and incidences of common genetic mutations (NPM1, FLT3-ITD, DNMT3A, IDH, CEBPA, ASXL1). Survival-related data were also extracted, including odds ratio (OR) for complete remission (CR) rates (or the cases achieving CR in patients with and without mutant TET2) and the hazard ratio (HR) for OS, event-free survival (EFS). Data from multivariate analyses were extracted with priority, but data from univariate analyses or calculated from Kaplan-Meier survival curves were also used in our meta-analysis when multivariate results were not available [17]. With respect to extraction of survival curves, we applied the software Engauge Digitizer 4.1 in this part. After inputting the pictures of survival curves and outputting the data of specific spots on curves, we used Excel to calculate HR and relevant data.

We used STATA 12.0 statistical software to analyze the data extracted. TET2 mutation was identified as relating to an unfavorable treatment outcome or survival when pooled ORs below or HRs over 1.000, which became statistically significant if 95% confidence interval (CI) did not cover 1.000 or a P value less than 0.05.

As for heterogeneity, Q test was used to measure it. I-square(I²) < 30%, 30–50%, 50–75, and > 75% meant low, moderate, substantial and considerable heterogeneity relatively [18]. Additionally, P < 0.10 represented a significant heterogeneity [19]. Random effect model was used for a significant heterogeneity observed, while fixed effect model was for studies without or with slight heterogeneity. Sensitivity test was used to determine the source of heterogeneity and evaluate the stability of results. Subgroup analyses were performed to evaluate the results in aspects of publication year, data type, region, study cohort, median age of patients as well as detecting method. Egger and Begg tests were used to assess publication bias, a P value less than 0.05 representing existence of publication bias.

At first, 686 studies were obtained from 5 databases, and we excluded 80 records because of duplication. Then we scanned the titles and abstracts of the remaining articles and removed 549 articles for no association with our study purpose. We read the full texts of the rest articles next and removed 41 studies due to the following reasons: 5 studies for focusing on pediatric AML, 7 studies for overlapped cohorts, 4 studies for reviews and meta-analysis, and 25 studies with incomplete data. Ultimately, there were 16 studies selected for our study. The selection procedure was illustrated in Fig. 1.

Characteristics of 16 studies with a total of 4378 cases were shown in Table 1: 4 of them from America, 6 from Europe, 5 from Asia, and one involving multiple regions [9, 10, 20‐33]. TET2 mutation had a frequency varying from 6.05 to 27.36% in AML patients and from 6.05 to 36.23% in patients with normal cytogenetics (CN-AML). Mutational frequency refers to the proportion of patients harboring TET2 mutations.

Eleven out of 16 studies revealed that TET2 mutations were distributed throughout the whole coding exon detected with no specific hot spots (Additional file 2: Table S1) [9, 22‐30, 32]. TET2 mutation was significantly associated with older age in 7 studies [9, 21, 24, 25, 27, 30, 33], higher white blood cell (WBC) count in 5 studies [21, 24, 25, 27, 30], higher hemoglobin level in 2 studies [30, 33] and with lower platelet count in 4 studies (Additional file 3: Table S2) [25, 27, 28, 30]. Besides, one study showed that TET2 mutations were rarer in M3 whereas another showed TET2 mutation was more frequent in M4 with respect to FAB classification [25, 28]. The others didn’t show any relationship between mutant TET2 and FAB subtypes (Additional file 3: Table S3). In addition, TET2 mutation was more frequent in AML patients with normal karyotype according to two studies [25, 30], and there was only one study showing the close association between mutant TET2 and intermediate-risk cytogenetics (Additional file 3: Table S3) [25]. But after a meta-analysis, we found a significantly close relationship between TET2 mutations and intermediate-risk cytogenetics with an OR of 1.765(95%CI: 1.212–2.569; P = 0.003; I² = 39.7%, P = 0.159), and TET2 mutations were more infrequent in favorable-risk cytogenetics(OR = 0.547, 95%CI: 0.325–0.920; P = 0.023; I² = 11.2%, P = 0.337). As for gene-gene association, a significantly higher frequency of NPM1 mutation was existed in patients with TET2 mutations compared with those with wild type TET2 according to 5 studies [21, 25, 28, 29, 31]. The same strong relationship was also noted between mutant ASXL1 and mutant TET2 in 3 studies [10, 20, 25]. On the contrary, IDH mutation was mutually excluded with TET2 mutation, which was observed in 7 studies (Additional file 3: Table S5) [9, 10, 24‐26, 28, 33], and the pooled OR was 0.123(95%CI: 0.071–0.212; P < 0.001). Then we conducted meta analyses and explored the exact close association of TET2 mutations with NPM1(OR = 1.735, P < 0.001), DNMT3A(OR = 2.361, P = 0.045) and ASXL1 mutations(OR = 2.743, P = 0.009).

With respect to response to therapy in AML patients, the combined ORs of TET2 mutation for CR rate were 0.802 (95%CI: 0.583–1.103; P = 0.176; heterogeneity: I² = 0.0%, P = 0.465; Fig. 2a). Fifteen studies were included to calculate pooled HRs for OS in AML patients, resulting in the result of 1.480 (95%CI: 1.241–1.766; P < 0.001; heterogeneity: I² = 34.3%, P = 0.094). With P < 0.1 in heterogeneity test, we conducted a sensitivity test and found recombined HRs for OS were 1.386 (95%CI: 1.217–1.577; P < 0.001; heterogeneity: I² = 19.7%, P = 0.239) when Lin’s study were excluded (Additional file 1: Figure S1a-1b). Considering EFS in AML patients, we combined 9 studies and pooled HRs were 1.594 (95%CI: 1.187–2.141; P = 0.002; heterogeneity: I² = 56.4%, P = 0.019; Fig. 2b). We conducted a sensitivity test owing to the significant heterogeneity and that result wasn’t altered by omitting any one of the studies. We therefore conducted subgroup analyses to evaluate the result (Table 2). Furthermore, we also explored the impact of mutant TET2 on OS in AML patients under 65 years of age, figuring out a pooled HRs of 1.310 (95%CI: 0.999–1.718; P = 0.051; heterogeneity: I² = 24.6%, P = 0.264; Fig. 2c), while the pooled HRs for EFS in the same cohort were 1.724 (95%CI: 1.007–2.954; P = 0.047; heterogeneity: I² = 56.1%, P = 0.077) and became 1.429 (95%CI: 1.041–1.962, P = 0.027; heterogeneity: I² = 37.4%, P = 0.202) after the study of Cher and colleagues was excluded (Additional file 1: Figure S1c-1d). Interestingly, we found that the summary HRs for OS in patients with IR-AML were 1.662 (95%CI: 1.312–2.105; P < 0.001; heterogeneity: I² = 46.1%, P = 0.046), and decreased to 1.474 (95%CI: 1.252–1.734; P < 0.001; heterogeneity: I² = 17.1%, P = 0.285) with the study of Patel and colleagues omitted (Additional file 1: Figure S1e-1f).

Among patients with CN-AML, the pooled ORs of TET2 mutation for CR rate were 0.803 (95%CI: 0.562–1.147; P = 0.228; heterogeneity: I² = 39.0%, P = 0.161; Fig. 3a). Despite the insignificance of ORs for CR rate, integrated HRs for OS were 1.425 (95%CI: 1.221–1.664; P < 0.001; heterogeneity: I² = 0.0%, P = 0.633; Fig. 3b), while the pooled HRs for EFS were 1.450 (95%CI: 1.199–1.754; P = 0.001; heterogeneity: I² = 45.1%, P = 0.105; Fig. 3c).

Among patients with CN-AML under 65 years of age, the pooled HRs for OS were 1.630 (95%CI: 0.817–3.250; P = 0.166; heterogeneity: I² = 54.0%, P = 0.114) and became 1.999 (95%CI: 1.317–3.033, P = 0.001; heterogeneity: I² = 0.0%, P = 0.352)(Additional file 1: Figure S2a-2b) with Damm’s study omitted, while the combined HRs for EFS were 1.741 (95%CI: 1.180–2.569; P = 0.005; heterogeneity: I² = 47.6%, P = 0.148; Fig. 3d).

We evaluated the prognosis of TET2 mutation based on ELN risk stratification. Among patients in ELN favorable-risk (CN-AML patients carrying NPM1mutwithout FLT3-ITD and/or CEBPAdouble mut), the combined ORs for CR rate were 0.460 (95%CI: 0.252–0.840; P = 0.011; heterogeneity: I² = 0.0%, P = 0.405; Fig. 4a). What’s more, the pooled HRs for EFS were 2.140 (95%CI: 1.476–3.101; P < 0.001; heterogeneity: I² = 8.8%, P = 0.334; Fig. 4b). The pooled HRs for OS were 2.034 (95%CI: 1.440–2.872; P < 0.001; heterogeneity: I² = 0.0%, P = 0.797; Fig. 4c). The results above showed an unfavorable prognostic impact of TET2 mutation in patients of this group. As for patients in ELN intermediate-Irisk group (all remaining patients with CN-AML), the pooled ORs for CR rate were 1.158 (95%CI: 0.408–3.280; P = 0.783; heterogeneity: I² = 70.5%, P = 0.034). Hence, we performed a sensitivity test as well although the number of studies included was only three. After excluding the study of Gaidzik and colleagues, the recombined ORs for CR rate decreased to 0.809 (95%CI: 0.465–1.409; P = 0.454; heterogeneity: I² = 45.7%, P = 0.175) (Additional file 1: Figure S3a-3b). Therefore, this study made up a part of the significant heterogeneity and has to be discussed further. In addition, the pooled HRs for EFS were 1.487 (95%CI: 1.117–1.978; P = 0.007; heterogeneity: I² = 6.3%, P = 0.344; Fig. 4d).

Among patients with AML, the study of Lin and colleagues made up a part of heterogeneity due to its outlier of HR for OS according to the sensitivity test, but it shouldn’t be excluded for the following reasons: At first, the sample size of cases from the entire cohort was adequate as well as the number of cases with TET2 mutation; Second, the maximum follow-up period of this study was around 120 months, which was long enough to diminish the bias; Next, the new result didn’t change too much with this study omitted, and TET2 mutation still conferred an unfavorable impact on OS in AML patients. Likewise, among patients with IR-AML, the sensitivity test showed that the study of Patel and colleagues was the biggest contribution of heterogeneity in pooled HRs for OS. This study showed a relatively stronger impact of TET2 on OS compared with other studies, but actually there was no need to exclude it due to the reasons above. More importantly, this study didn’t influence pooled HRs of TET2 mutation for the OS significantly. With regard to patients under 65 years of age, the pooled HRs for EFS decreased from 1.724 to 1.429 with a lower heterogeneity after the study of Cher and colleagues was excluded. The sample size of cases with TET2 mutation from Cher’s study was the smallest, which might be expected to result in the heterogeneity.

In patients with CN-AML under 65 years of age, significant heterogeneity in pooled HRs for OS was from Damm’s study. Actually, this study demonstrated a totally different result opposed to other two studies confirming the unfavorable impact of TET2 mutation in this cohort. The relatively lower percentage (6.75%) and fewer cases (11 cases) with TET2 mutation might be the reasons for the bias.

As for patients in ELN intermediate-Irisk group, the study of Gaidzik and colleagues could account for a part of the significant heterogeneity in pooled ORs for CR rate and the combined ORs decreased to 0.809 from 1.120 with this study excluded. The cohort in Gaidzik’s study had an age range from 18 to 60 years, nevertheless the patients from the other two studies had more broad age range (Ahn: from 15 to 84 years; Metzeler: from 18 to 83 years), including patients over 60 years old. Therefore the relatively younger patient cohort could explain the heterogeneity in OR for CR rate and might be the cause of relatively higher CR rate in Gaidzik’s study. The study of Gaidzik and colleagues had a large sample size of cases with TET2 mutation and a relatively long follow-up period. Hence, this study should be reserved and the impact on CR rate of TET2 mutation in patients with ELN intermediate-Irisk cytogenetics remained unclear.

As we stated previously, we performed subgroup analyses to further assess the prognosis of TET2 mutation on EFS in AML patients. We found that publication year, data types, regions of samples and study cohorts had no influence on OS for mutant TET2. In aspect of detection method, the studies using next-generation sequencing method had more significant results compared with those using direct sequencing method (Table 2). Direct sequencing couldn’t detect all the mutated spots and sometimes might miss the real TET2 mutation, while next-generation sequencing skill could detect all the genome, more sensitive and precise than direct sequencing, so it could reduce the incident of false-negative events and exactly represent the adverse impact of TET2 mutation.

We performed Egger tests and Begg tests to assess publication bias in this study (Additional file 3: Table S6). There existed no publication bias in other results in both Egger test and Begg test (Additional file 1: Figure S4A-S4D).

With respect to OS and EFS for mutant TET2, HRs and 95%CI without heterogeneity (I² = 0.00%) from the fixed effect model were the same as those from the random effect model. HRs and 95%CI with heterogeneity (I² > 0.00%) from the fixed effect model had slight changes compared with those from the random effect model but these changes did not influence the prognosis analyses. In addition, ORs and 95%CI for CR rate were slightly changed from the fixed effect model to the random effect model no matter whether there existed heterogeneity or not, which had no impact on prognosis analyses (Additional file 3: Table S7).

In this study, TET2 mutations exhibited markedly unfavorable impacts on prognosis of AML patients. Patients with TET2 mutations had reduced OS (HR: 1.386; P < 0.001) compared with those with wild type TET2. We also observed that mutant TET2 had unfavorable impacts on OS (HR: 1.310, P = 0.051) and EFS (HR: 1.429, P = 0.027) in patients under 65 years of age. Although 95%CI for OS covered 1.000, the P value was slightly over 0.050 and still conferred an adverse trend for OS. In addition, TET2 mutations could also significantly reduce the OS of patients with IR-AML (HR: 1.474, P < 0.001).

Likewise, the same adverse influences of TET2 mutations could be observed in patients with CN-AML (HR for OS: 1.425, P < 0.001; EFS: 1.450, P < 0.001). Among CN-AML patients, patients under 65 years of age harboring TET2 mutations had more reduced OS (HR: 1.999, P = 0.001) and EFS(HR: 1.741, P = 0.005) compared with those harboring wild type.

As patients with CN-AML could be subdivided into a favorable group (ELN favorable-risk group) and an unfavorable-risk group (ELN intermediate-Igroup), we then evaluated the prognostic influence of TET2 mutation on patients from different ELN subgroups. Among ELN favorable-risk group, TET2 mutations led to significantly decreased CR rate (OR: 0.460, P = 0.011) and reduced survival (HR for EFS: 2.140, P < 0.001; HR for OS: 2.034, P < 0.001), which showed the adverse impact of TET2 mutation in this group. As for patients in ELN intermediate-Igroup, because of the significant heterogeneity, we performed a sensitivity analysis and found that recombined ORs for CR rate was 0.809 after excluding Gaidzik’s study but still didn’t reach statistically significance. We had discussed that the diverse age range might be the cause for the higher CR rate of patients from Gaidzik’s study. According to several studies, age could be an independent prognostic factor in AML, and younger patients had a better treatment outcome compared with older patients. However, this study shouldn’t be excluded due to the large sample size of cases with mutant TET2 (23.3%) and the relatively long follow-up period in this study. Therefore, the result needs to be confirmed further. In addition, mutant TET2 had an unfavorable impact on EFS of patients in ELN intermediate-Igroup (HR: 1.487, P < 0.001). Because the hot spots of TET2 mutations were not explored in the whole exon, we were not able to evaluate the prognostic impact of the specific mutated spot of TET2.

The emergence of TET2 mutations represents an early event in the progression from normal hematopoiesis to AML. Nonetheless, some patients with mutations in AML-associated genes such as TET2 don’t develop AML. Many researchers proposed a conception of clonal hematopoiesis of indeterminate potential, or CHIP, which was linked to a very low rate of conversion to AML and was also age-related like TET2 mutations. However, it’s possible that all clones in CHIP would generate AML if given enough time and additional hits of genetic mutations. That might account for this relationship between TET2 mutations and elder age.

Our study explored the adverse prognostic effect of mutant TET2 on AML patients based on aspects of patients cohort, age, and risk stratification, which can bring great benefits to prognosis evaluation and therapy strategy guidance. In Itzykson’s study, azacitidine conferred a better response rate on patients with myelodysplastic syndromes and low blast count AML, which showed TET2 mutations’ predictive value of response to hypomethylated agents [34]. We also validated the superiority of next-generation sequencing skill compared with direct sequencing skill from the subgroup analyses of EFS in AML patients, which could predict the promising prospect of next-generation sequencing skill in gene detection due to its precision and a high sensitivity.

Our study confirmed the adverse effects of mutant TET2 on prognosis of AML patients in accordance with another meta-analysis which was operated by Liu Wenjian with 8 studies included [11]. Both studies evaluated OS and EFS of patients from the entire AML, CN-AML, ELN favorable-risk group and intermediate-Igroup. Our meta analysis included 16 studies containing 4577 cases altogether, more stable and more reliable than the previous analysis. Our research was built on a broad population involving many regions including America, Asia, Europe, which meant utilization.

Besides, we summarized the correlation of TET2 mutations and several major clinical features involving age, laboratory parameters, FAB subtypes, cytogenetics and other genetic alterations. We noted that mutant TET2 correlated with older age, higher WBC count, lower platelet count, intermediate-risk cytogenetics, mutations in NPM1, DNMT3A and ASXL1, and were mutually excluded with favorable-risk cytogenetics and IDH mutations.

More importantly, we explored more about prognostic influences of mutant TET2 on survival and treatment outcomes. TET2 mutation had negative impacts on CR rate (OR: 0.460, P = 0.011) and OS (HR: 2.034, P < 0.001) within patients in ELN favorable-risk group. Nevertheless, among patients with AML, CN-AML and those in ELN intermediate-Igroup, TET2 mutation had no impact on CR rate. We also investigated the correlation of TET2 mutation with OS and EFS of patients under 65 years of age, and that mutant TET2 was linked to unfavorable OS and EFS in this kind of patients cohort with AML (HR for OS: 1.310, P = 0.051; EFS:1.429, P = 0.027) and CN-AML (HR for OS: 1.999, P = 0.001; EFS: 1.741, P = 0.005). We also found the adverse impact of mutant TET2 on OS (HR: 1.474, P < 0.001) in patients with IR-AML. All the findings above were not presented in previous meta analysis.

Regardless of the fact that we made the effort to perfect our study, there still existed several limitations. To start with, we only brought studies in English and Chinese into our research, with studies in other kinds of languages omitted, which might influence the real consequences. Secondly, all the studies included were retrospective studies, which made it difficult control the selective criteria and guarantee homogeneity of chosen studies. Thirdly, although some of studies provided precious data of HRs and 95%CI in multivariate analyses, there were also some studies only providing data from univariate analyses, even some data needed extracting from Kaplan-Meier survival curves, all of which might reduce the stability of the results. Next, various studies included in our meta analysis didn’t provide the specific information about karyotype, WHO subtype, FAB subtype, genetic aberration and survival period of each patient. Therefore we couldn’t investigate the relationship between the impact of TET2 mutations on prognosis and the factors above. Finally, we were not capable of evaluate the effect of mutant TET2 on prognosis in patients over 65 years old due to the data deficiency, which made it difficult to compare the impact on survival of TET2 mutation between different age groups. In addition, some forest graphs in our analysis only contained 3 individual studies, which appeared to compromise the overall accuracy. Nonetheless, we rigidly operated the accurate procedures of meta analysis and that all included studies were evaluated of high quality. What’s more, after omitting the study of high heterogeneity, the width of confidence interval reduced, which meant that the accuracy actually increased.