Studies were considered eligible if they satisfied all of the following inclusion criteria: 1) type of studies: cohort study; 2) type of subject: patients with ALL, NHL, and other hematological malignancies receiving HDMTX chemotherapy, with no restrictions on ethnicity, gender, or age; 3) classification of exposure: patients were grouped by wild or mutant genotype of included genes within the HDMTX pathway (Table 1). There were 12 genetic polymorphisms in total, including but not limited to genes involved in the Pharmacogenomics Knowledge Base (PharmGKB, https://www.pharmgkb.org/guidelineAnnotations), which contains recommendations from the Clinical Pharmacogenetics Implementation Consortium (CPIC) and other national association of pharmacogenomics; and 4) types of outcomes measured: HDMTX-related toxicity and prognosis outcomes. The toxicity outcomes included the rate of hepatic toxicity, renal toxicity, oral mucositis, gastrointestinal (GI) toxicity, neurotoxicity, dermal toxicity, overall toxicity, and therapeutic interference due to toxicity. The toxicity outcomes were identified by the Common Terminology Criteria for Adverse Events established by American National Cancer Institute (NCI-CTC) or Toxicity Grading Scale for Determining the Severity of Adverse Events of Chemotherapeutic Drugs established by the World Health Organization (WHO) or other common criteria. Grade 3 to 4 (G3-4) indicates severe toxicity. The prognosis outcomes included overall survival (OS), progression-free survival (PFS), disease-free survival (DFS), event-free survival (EFS), relapse-free survival (RFS), and relapse/death. The exclusion criteria were as follows: duplicate publications; abstracts without available full texts; unqualified data; and studies not in accordance with the Hardy–Weinberg equilibrium (HWE) (Trikalinos et al., 2006) or not reporting the genotype distribution.

Electronic databases including PubMed, Embase, Cochrane Central Register of Controlled Trials (CENTRAL), and Clinical Trials.gov were searched for potentially relevant studies from inception to November 11, 2020. Specific search strategies were developed for each database. The combination of keywords (“Methotrexate”) AND ("Hematologic neoplasms” OR Hematologic malignancy” OR “Leukemia” OR “Lymphoma”) AND (“Gene” OR “Polymorphism” OR “Pharmacogenetics” OR “Polymorphism, single nucleotide”) were used to search the title and abstract of queried literature (Supplementary Material II). No restrictions were placed on study design or language. The reference lists of previous systematic reviews and included literature were searched manually.

Two authors (ZS and YH) independently assessed the eligibility of all studies based on the aforementioned inclusion and exclusion criteria after reviewing the study title, abstract, and full text in succession. Studies were included in only the systematic review (but not the meta-analysis) if their findings were relevant to the research question, but data were not available for quantitative analysis. Any disagreement among authors was discussed and reconciled by the corresponding author (RZ).

Two authors (ZS and YH) independently extracted data based on a predesigned standardized extraction form, including the first author and publication year, country, ethnicity, diagnosis, sample size and genotype distribution, gender (female/male), age (years), MTX dose, calculated p-value for HWE, outcomes, and individual results of the single study. Study authors were contacted for missing data.

Two authors (ZS and YH) independently assessed the quality of studies under the Newcastle–Ottawa Scale (NOS) (Stang, 2010), as recommended in the Cochrane Handbook. The NOS attributes a maximum of 9 points to studies based on methodological design and formal reporting, involving “selection of cohorts,” “comparability of cohorts,” and “assessment of outcome.” NOS scores ranging from 7 to 9 points indicate high quality, 5 to 6 indicate medium quality, and 0 to 4 indicate low quality. Disagreements regarding data extraction and quality assessment were resolved by consensus or, when necessary, by consulting the corresponding author (RZ).

A chi-square test was performed to verify genotype distributions using SPSS version 25.0. A p-value greater than 0.05 would indicate accordance with the HWE. Before conducting the meta-analysis, clinical heterogeneity was estimated by comparing the diagnosis, efficacy or toxicity criteria, and other clinical features among studies. If two or more studies reported the same outcome and obvious clinical heterogeneity was not observed, meta-analysis was performed to quantitatively integrate outcomes by using the Cochrane Collaboration review manager software 5.3 (RevMan 5.3). Otherwise, only a descriptive analysis was performed.

The meta-analysis was performed as follows: 1) odds ratios (ORs) and hazard ratios (HRs) were calculated to evaluate the genetic association of toxicity or prognosis outcomes, respectively. And if the corresponding 95% confidence intervals (95% CIs) of the OR value (HR value) did not overlap with the value of 1 and the p-value was less than 0.05, the association was considered statistically significant. 2) The pooled OR or HR was calculated under the dominant model (MM/Mm vs mm), recessive model (MM vs Mm/mm), and allelic model (M vs m), where M is the mutant allele such that the G allele at RFC1 80A>G, C allele at SLCO1B1 521T>C, T allele at ABCB1 3435C>T, and so on; m is the wild allele, such that the A allele at RFC1 80A>G, T allele at SLCO1B1 521T>C, C allele at ABCB1 3435C>T, and so on. 3) The fixed-effect model was used initially, and the random-effects model was adopted when unidentified significant heterogeneity was detected. 4) The heterogeneity across the studies was assessed using a chi-square-based Q-test and I² statistics. P_{heterogeneity} (P_het) values <0.05 and I² values >50% were considered to indicate significant heterogeneity (Sedgwick, 2015), and in these cases, the source of heterogeneity was investigated by examining the steps taken to check data and perform the subgroup analysis and sensitivity analysis. If the potential sources of heterogeneity remained unclear, the random-effects model was used, or the descriptive analysis was performed. 5) Subgroup analyses were performed based on patients’ age. Age subgroups were defined as pediatric, adult, and mixed-age. P_subgroup (P_sub) <0.05 indicated a statistically significant difference across subgroups.

Sensitivity analyses were conducted to assess the impact of individual studies on the pooled estimates and the stability of the pooled estimates. A pooled OR (HR) was recalculated after removing each single primary study one by one and replacing the statistical model of meta-analysis. Publication bias was assessed by inspecting the funnel plot visually, and it was considered to be valid when 10 or more studies were included (Sedgwick and Marston, 2015).

A total of 4395 candidate references were identified in electronic database searches, and no additional reference was identified using a manual search. Of the total 4395 candidates, 441 duplicate references were removed and then 3904 were excluded after careful review of the titles and abstracts. Only 50 references were recognized as relevant and then we assessed all full texts. The PRISMA 2020 flow diagram is shown in Figure 2. Of the total 50 references, five did not focus on HDMTX, six did not report the targeted outcomes, three did not accord with HWE, and two did not have qualified data. Finally, according to the aforementioned inclusion and exclusion criteria, 34 studies were included in our systematic review. Of the 34 studies (Laverdière et al., 2002; Kishi et al., 2003; Seidemann et al., 2006; Shimasaki et al., 2006; Imanishi et al., 2007; Ruiz-Argüelles et al., 2007; Ashton et al., 2009; Faganel Kotnik et al., 2010; D'Angelo et al., 2011; Faganel Kotnik et al., 2011; Liu et al., 2011; Chiusolo et al., 2012; Erčulj et al., 2012; Haase et al., 2012; Fukushima et al., 2013; Radtke et al., 2013; Yanagimachi et al., 2013; Avivi et al., 2014; Erculj et al., 2014; Suthandiram et al., 2014; den Hoed et al., 2015; Ma et al., 2015; Huang et al., 2016; Tsujimoto et al., 2016; Choi et al., 2017; Giletti et al., 2017; Liu et al., 2017; Yang et al., 2017; Yazıcıoğlu et al., 2017; Oosterom et al., 2018; Chae et al., 2020; Esmaili et al., 2020; Kotur et al., 2020; Chang et al., 2021) included, 31 studies were included in the meta-analysis and three studies were only included for the descriptive analysis since the meta-analysis was infeasible.

In total, 34 studies involving 4102 patients were included for investigating the associations between genetic polymorphisms and HDMTX outcomes. Of the studies included, 14 studies reported the polymorphisms of RFC1 (rs1051266), six studies SLCO1B1 (rs4149056), seven studies ABCB1 (rs1045642), one study FPGS (rs10106), one study FPGS (rs1544105), 26 studies MTHFR (rs1801133), 17 studies MTHFR (rs1801131), six studies TYMS (rs34743033), and one study ATIC (rs2372536). None of studies reported the polymorphisms of GGH (rs3758149) and DHFR (rs408626, rs442767) yet. Among included studies, 20 and 14 studies were conducted in the ethnicity of Caucasian and Asian, respectively. Patients’ diagnosis included ALL, acute myeloid leukemia (AML), diffuse large B-cell lymphoma (DLBCL), primary CNS lymphoma (PCNSL), and other NHL. A total of 25 studies included pediatric patients only, 8 studies included adult patients only, and one study did not impose restrictions on patient age. All the studies were in accord with the HWE. And 30 studies reported toxicities outcomes, while prognosis outcomes were involved in 14 studies. Regarding quality assessment, 2 studies earned a full NOS score of 9 points, 19 studies earned 8 points, and 11 studies earned 7 points, varying mainly in presentation of the outcomes at the start of study and outcome follow-up. The average NOS score of all studies was 7.6 points, indicating a relatively high quality of overall methodology. The main characteristics and NOS scores of the studies included are summarized in Table 2. And the detailed NOS scores of the studies included are given in Supplemental Material III (Supplementary Table S2).

The overall findings are summarized in Table 3. Regarding the cellular transport and metabolism, RFC1 (rs1051266) was associated with a reduced risk of hepatic toxicity and overall toxicity (in pediatric), while ABCB1 (rs1045642) was associated with an increased risk of hepatic toxicity. No association was observed in other toxicities outcomes and genetic polymorphisms. According to findings of prognostic outcomes from individual studies, RFC1 (rs1051266) was associated with worse 2y-OS and 2y-PFS (in adult), and SLCO1B1 (rs4149056) was associated with worse 5y-EFS (in pediatric). However, FPGS (rs1544105) was associated with better 2y-OS (in adult). No association was observed in other outcomes of relapse.

Regarding the target, the polymorphisms of MTHFR were the most extensively investigated genes. MTHFR (rs1801133) was associated with an increased risk of hepatic toxicity, renal toxicity, mucositis, and therapeutic interference (in pediatric). In contrast, MTHFR (rs1801133), MTHFR (rs1801131), and TYMS (rs34743033) were associated with a reduced risk of G3-4 hepatic toxicity, renal toxicity, and mucositis, respectively. Only one study investigated the polymorphisms of ATIC (rs2372536), and it reported the lack of association of the neurotoxicity. According to findings of prognostic outcomes from individual studies, no association was observed.

To assess the impact of individual studies on the overall pooled estimate and explore potential sources of heterogeneity, sensitivity analyses were conducted by removing each study one by one for each comparison. In a total of 66 meta-analyses in this study, substantial changes were indicated in a small proportion of comparisons. For the meta-analysis of RFC1 (rs1051266), the outcome of hepatotoxicity under the dominant model changed to OR = 0.39 with 95% CI = 0.19–0.80 after excluding Esmali 2020 (Esmaili et al., 2020). For the MTHFR (rs1801133), the statistically significant result of hepatotoxicity under the dominant model changed substantially after excluding Chang 2021 (Chang et al., 2021) or Suthandiram 2014 (Suthandiram et al., 2014) or Fukushima 2013 (Fukushima et al., 2013) or switching into the random-effects model (Figure 4). Similarly, the substantial changes of pooled OR (95% CI) were detected in the following comparisons of MTHFR (rs1801133): the result of hepatotoxicity under recessive and allelic models after excluding Chang 2021 (Chang et al., 2021) or Suthandiram 2014 (Suthandiram et al., 2014); the renal toxicity under recessive and allelic models after excluding Chang 2021 (Chang et al., 2021); the mucositis under dominant and allelic models after excluding Faganel 2011 (Faganel Kotnik et al., 2011); and the mucositis under the recessive model after excluding Suthandiram 2014 (Suthandiram et al., 2014). However, the pooled estimate of all the other comparisons did not change significantly when different data were used, indicating that the conclusions of this study had a certain degree of reliability.

Publication bias was evaluated by analyzing the funnel plots visually. Serious publication bias was not indicated in any of the outcomes. For instance, the funnel plot (Supplementary Figure S5) does not suggest any serious bias for hepatotoxicity under the dominant model of MTHFR (rs1801133). However, it is notable that publication bias for some outcomes could not be excluded entirely by visual inspection of the funnel plots.

Recently, investigations into the HDMTX response from the perspective of genetic variation have been relatively recent and wide in scope. In this study, we conducted a systematic review aiming to identify and summarize present evidence evaluating the associations between genetic polymorphisms with HDMTX toxicity and prognosis outcomes. With the aim of identifying relevant variants that could be implemented in clinical prediction, we reviewed and investigated genetic polymorphisms within the whole pathway of drug metabolism and targets in this study. As HDMTX is far more toxic than low-dose MTX (Schmiegelow, 2009), patients with malignant cancer receiving HDMTX are more easily associated with serious toxic responses than those with rheumatoid arthritis. Besides, the dosage and infusion regimens of HDMTX vary greatly in the treatment of hematological malignancies and osteosarcoma (Ramsey et al., 2018), so toxic responses can be markedly different in the two malignant cancers. Therefore, we pay more attention to patients with hematological malignancies in this present study.

The polymorphisms of RFC1 (rs1051266), ABCB1 (rs1045642), and MTHFR (rs1801133, rs1801131) were the research focus in current pharmacogenetic studies of HDMTX. The most investigated clinical outcomes were hepatic toxicity and mucositis, followed by renal toxicity, while prognostic outcomes were reported by a small proportion of studies (14/34). Generally speaking, our study confirmed that the MTHFR 677C>T (rs1801133) has a significant effect on the increased risk of HDMTX toxicities (including hepatotoxicity, mucositis, and renal toxicity), and the ABCB1 3435C>T (rs1045642) has a significant effect on the increased risk of hepatotoxicity, which corresponds to the findings in previous studies (Yang et al., 2012; Zhao et al., 2016; Zhu et al., 2018; Yao et al., 2019; Maagdenberg et al., 2021), whereas we found a tendency toward reduced risk of hepatotoxicity in carriers of RFC1 80GG and toward reduced risk of mucositis in those with TYMS 3R3R or 2R3R genotypes. Also, a tendency toward reduced risk of renal toxicity was observed in carriers with the MTHFR 1298 variant C allele, which was similar to the results of mucositis and GI toxicity (Zhao et al., 2016) and dermal toxicity (Yang et al., 2012). In other words, a protective effect of some genetic polymorphisms on developing individual toxicities is suggestive in our study. It is worth mentioning that Lopez-Lopez 2013 (Lopez-Lopez et al., 2013) reported no association between MTHFR 677C>T and MTX toxicity in pediatric ALL (recessive model), and Oosterom 2018 (Oosterom et al., 2018) reported no association between TYMS 2R>3R and MTX toxicity in pediatric ALL (dominant model), which were contrary to our findings. The inconsistent patient’s age and MTX dosage may explain the discrepancy in different findings. With regard to descriptive analysis, limited evidence from single studies showed significant associations in the following prognosis outcomes: RFC1 (rs1051266) and worse 2y-PFS and 2y-OS; SLCO1B1 (rs4149056) and worse 5y-EFS; and FPGS (rs1544105) and better 2y-OS. Besides, a lack of association was observed in other outcomes investigated in our review.

Since HDMTX-related tolerance and toxicities might be influenced by patients’ age in the real clinical practice (Zhu et al., 2018), subgroup analysis was conducted to explore the difference. Significant differences were suggestive in the following outcomes: hepatotoxicity (SLCO1B1 rs4149056; MTHFR rs1801133), renal toxicity (MTHFR rs1801133), and mucositis (MTHFR rs1801133). It is worth mentioning that in the meta-analysis of MTHFR (rs1801133), the aforementioned subgroup differences can explain the heterogeneity of those outcomes with positive findings (hepatotoxicity, renal toxicity, and mucositis) to a certain extent. And patients’ age is possibly identified as a potential contributor to the association with some certain toxicities.

As the associations between genetic variations and HDMTX toxicities have become a wide clinical concern, some systematic reviews or meta-analyses have been researched earlier. Initially, we conducted an umbrella review of systematic reviews about the pharmacogenetics of MTX toxicity in patients with osteosarcoma or hematological malignancies in 2019 (Song et al., 2019a). After performing an update search and review in July 2021, we found six similar meta-analyses (Yang et al., 2012; Lopez-Lopez et al., 2013; He et al., 2014; Zhao et al., 2016; Oosterom et al., 2018; Yao et al., 2019) had a discussion on this issue in patients with hematological malignancies. Besides, three meta-analyses (Hagleitner et al., 2014; Zhu et al., 2018; Maagdenberg et al., 2021) included cancer patients and did not distinguish between hematological malignancies and osteosarcoma, although the dosage and infusion regimens of HDMTX vary greatly in the two diseases (Table 4).

To the best of our knowledge, there have not been any previous meta-analysis to address the following points: 1) investigated 12 genetic polymorphisms within the whole HDMTX pathway and performed analysis on prognosis outcomes, while most previous studies did not discuss these issues (Yang et al., 2012; Lopez-Lopez et al., 2013; Hagleitner et al., 2014; He et al., 2014; Zhao et al., 2016; Oosterom et al., 2018; Zhu et al., 2018); 2) set strict restrictions on hematological malignancies (specified diseases) and HDMTX (specified dose ranges), since the side effect profile of MTX varies markedly as its dose changes in clinical practice; 3) set strict restrictions on the HWE of included pharmacogenetic research, since HWE is crucial for genetic research; 4) analyzed the associations under the dominant model, recessive model, and allelic model, since the genetic models of HDMTX toxicities remain incompletely understood and need validation, whereas most studies only assumed one genetic model; 5) removed restrictions on patient age and ethnicity, which enabled this review to include a greater number of studies and populations than previous studies. The main characteristics of our review and previous meta-analysis are summarized in Table 4. Consequently, our review has provided a comprehensive and up-to-date synthesis of present evidence, and a more reliable conclusion of the association could be reached.

Currently, the biological mechanisms linking genetic polymorphisms to HDMTX toxicities still remain incompletely understood. In theory, the delayed MTX clearance or prolonged and elevated exposure of MTX can potentially lead to an increased risk of MTX-induced toxicities. But what is interesting, in one study (Yanagimachi et al., 2013), although an association with higher MTX plasma levels was observed, a genetic association with renal toxicity could not be established, which reminds us that other clinical and genetic factors may play a role together.

The enzyme MTHFR plays a critical role in the folate metabolism by catalyzing the conversion of 5,10-methylene-tetrahydrofolate (THF) to 5-methyl-THF. This is the primary circulating form of folate, which is needed to reduce the toxic homocysteine to methionine. Through this process, folate is an important donor of methyl groups for all intracellular methylation processes (Taylor et al., 2021). The mutation 677C>T (rs1801133) causes a change of alanine to valine in the protein, and the mutation 1298A>C (rs1801131) causes the replacement of glutamate by valine, resulting in decreased enzyme activity of MTHFR(Robien and Ulrich, 2003). Therefore, the mutations lead to the reduction of folate and then might exert an influence in HDMTX-related toxicities. MTX is extruded by cells using different transporters including ABC family members, and P-glycolprotein (P-gp) is a representative cell membrane protein encoded by the gene ABCB1 (Castaldo et al., 2011). The ABCB1 3435C>T (rs1045642) mutations can affect the activity of P-gp and thus may play a role in HDMTX toxicities.

Several limitations should be considered for our review. First, the outcome measures we investigated were clinical outcomes of HDMTX-related toxicities or prognosis and not the plasma concentration or other pharmacokinetic outcomes of MTX, since the relationships of clinical outcomes and plasma concentration still remain to be verified. And notably, a recent systematic review has discussed on pharmacogenetic factors influencing HDMTX pharmacokinetics (Taylor et al., 2021). Second, data of prognosis outcomes was substantially lacking, which made the quantitative analysis unfortunately impossible for the most prognosis outcomes. Therefore, the genetic associations with prognostic outcomes still remain inconclusive in this study. Third, the sample size of some studies is too small (most are less than 200), and thus, its statistical power might be limited. Furthermore, in addition to the positive findings in the MTHFR rs1801133 and the TYMS rs34743033, statistically significant results of other genetic polymorphisms are only confirmed by meta-analysis of two to three studies. The fact that a meta-analysis of two studies reveals the association between MTHFR rs1801131 and a decreased risk of renal toxicity (allelic model) means that the trend may not reflect the actual situation, and the results should be cautiously explained. Last but not least, although overall heterogeneity was not observed, the baseline characteristics varied among studies included, including diverse treatment protocols, infusion hours of HDMTX, leucovorin rescue, and other therapy-related factors which might contribute to the associations observed in these studies.

The aforementioned limitations warrant future larger validation studies into genetic association of HDMTX-related clinical outcomes. In line with the research gaps, we would recommend that further studies pay more attention to prognosis outcomes and the genetic polymorphisms of FPGS (rs10106, rs1544105), GGH (rs3758149), DHFR (rs408626, rs442767), and ATIC (rs2372536). And the validation studies are encouraged to calculate the sample size, establish analysis strategies, and prospectively collect data of toxicity and prognosis. With the aim to construct clinically relevant prediction models, the future studies should not focus on the individual effect of single polymorphisms but take into account other polymorphisms within the whole pharmacokinetic and targets/folate pathway. Besides, the prospective cohort studies are encouraged to confirm the clinical benefits of genetic testing by comparing the differences between patients performing genetic testing and those who have not carried out testing.

In light of the findings in this study, associations between genetic variations and the increased/decreased risk of HDMTX toxicities are observed, whereas the association of prognosis outcomes still remains inconclusive due to lacking data. From a clinician or pharmacist’s point of view, we focus more on the role of genetic testing in predicting increased toxicity to tailor MTX therapy, while the data of the potentially protective effect may be less meaningful clinically. With regard to clinical implementation of pharmacogenomics (PGx), the certainty and reliability of supporting evidence are important factors to consider. For instance, the associations between MTHFR 677C>T (rs1801133) and an increased risk of hepatotoxicity and mucositis are demonstrated by meta-analysis of nine studies, while the associations between ABCB1 3435C>T (rs1045642) and hepatotoxicity are supported by meta-analysis of only three studies.

With the aim to provide practical recommendations, we also reviewed drug instructions of MTX, clinical guidelines or expert consensus of HDMTX, CPIC, and several other pharmacogenetics guidelines. The French National Network of Pharmacogenetics (RNPGx) states that methotrexate pharmacogenetic tests are potentially useful in cancer patients (Quaranta and Thomas, 2017). The evidence-based practice guideline of HDMTX medication of the Chinese Pharmacological Society (Wang et al., 2021) states that the genotyping of MTHFR 677C>T and 1298A>C polymorphisms can be considered for patients with hematological malignancies (weak recommendation, moderate quality evidence), and the genotyping of ABCB1 3435C>T may be considered under certain conditions (weak recommendation, low-quality evidence). Further combining the PGx implementation of HDMTX from Chinese perspective (Song et al., 2019b), we would recommend clinicians to consider genetic testing of MTHFR polymorphisms when necessary, and ABCB1 3435C>T can also be a potential candidate gene. Since patients with gene mutations (MTHFR 677C>T particularly) are at the risk of increased hepatotoxicity and/or mucositis, a relatively lower dose and closer monitoring of plasma MTX concentrations are advisable to these patients.

It is worth mentioning that although PGx research has been advanced rapidly in recent years, the clinical implementation of PGx has a long way to go (Guo et al., 2021). To reduce HDMTX-related toxicities and improve outcomes, in addition to the role of genetic polymorphisms, renal function evaluation prior to treatment, co-medications, hydration and urinary alkalization, therapeutic drug monitoring (TDM), and leucovorin rescue might be taken into full consideration. Renal toxicity is one of the most feared side effects of MTX, since the renal dysfunction significantly delays MTX clearance and may cause other toxicities (Schmiegelow, 2009). So far, there is no proven useful approach to predict the individual risk of acute renal failure from the perspective of genetic variation. However, the implementation of standardized hydration and urinary alkalinization and TDM during HDMTX therapy contributes a lot to prevent renal toxicity and maintain MTX elimination.