Clinical characterization of hyperbilirubinemia

Within the 1547 patients of the AIEOP ALL-BFM 2000 study population with available toxicity information, 540 (34.9%) had normal and 1007 (65.1%) had increased bilirubin levels during induction/consolidation (protocols IA/IB) (Table 1). This included 825 (53.3%) patients with moderate hyperbilirubinemia (grades 1-2; 707 during induction, 575 during consolidation, 355 in both phases) and 182 (11.8%) with high hyperbilirubinemia (grades 3-4; 158 reports during induction, 50 during consolidation, 26 in both phases). Comparing patients without to patients with hyperbilirubinemia (grade 0 vs. grades 1-4), we noticed a larger proportion of older patients (P < 0.001) and more T cell ALL patients (P = 0.022) among those affected. The group of B cell ALL patients exhibiting hyperbilirubinemia contained fewer hyperdiploid patients (P = 0.002). However, no differences with regard to other genetic subgroups were observed (ETV6-RUNX1, BCR-ABL1 and KMT2A-AFF1) (Table 1). Considering the entire course of therapy, 245 patients (16.0%) had high hyperbilirubinemia in at least one treatment element. We determined a median time to protocol day 78 (after completion of induction and consolidation) of 89 ± 11 days (range 64-185 days), analyzing 1453 of 1547 individuals of our study population with available information. Compared to patients with moderate or no hyperbilirubinemia, patients with hyperbilirubinemia grades 3-4 experienced more therapy delays, requiring 91 days (range 64-174 days) to complete induction/consolidation vs. 88 days (range 70-154 days) for grades 1-2 and 89 days (range 70-185 days) for grade 0 (P = 0.002). No alterations of therapy in response to hyperbilirubinemia were noted.

Table 1 Characteristics of 1547 patients according to serum bilirubin levels during induction/consolidation therapy for acute lymphoblastic leukemia

In outcome analyses, patients with high hyperbilirubinemia (grades 3-4) during induction/consolidation fared significantly worse compared to patients with moderate or no hyperbilirubinemia: 5-year EFS 76.7 ± 3% vs. 87.7 ± 1% (P < 0.0001), and vs. 85.2 ± 2% (P = 0.0031), respectively (Fig. 1A). The corresponding 5-year cumulative incidences of relapse (CIR) were 15.6 ± 3% for high, 9.0 ± 1% for moderate hyperbilirubinemia, and 11.1 ± 1% for patients without hyperbilirubinemia (Fig. 1B).

Fig. 1

A Probability of 5 year event free survival (EFS) [%] and (B) corresponding cumulative incidence of relapse (CIR) according to the maximum total bilirubin toxicity during induction/consolidation – protocols IA/IB –in the in the AIEOP ALL-BFM 2000 study population with available bilirubin information (n = 1547 patients). Bilirubin toxicity grading was according to the Common Toxicity Criteria (CTC) of the National Cancer Institute, version 2; standard error (SE) and the number of included individuals are indicated for each category

In our study cohort, 1443 (93%) patients were observed with elevated hepatic transaminase activity levels (grades 1-4) during induction/consolidation (Table 1), 765 of which demonstrated grades 3-4. Transaminase levels were positively correlated with hyperbilirubinemia and, in particular, patients with high hyperbilirubinemia were at risk for concurrent high transaminase levels (grades 3-4) compared to the remaining patients (70% vs. 40%, odds ratio (OR) = 21.3, 95% confidence interval (CI) = 5.1-88.2, P = 2.53 × 10− 5) (Suppl. Table 2).

Interestingly, 5-year EFS and CIR did not differ between patients with moderate (grades 1-2), high (grades 3-4) or absent transaminasemia during induction/consolidation (Suppl. Fig. 2A and B). The 5-year EFS of patients with concurrent high hyperbilirubinemia and high transaminasemia was 76.2 ± 4% and 78.1 ± 6% in patients with high hyperbilirubinemia accompanied by moderate or no transaminasemia (grades 0-2, P = 0.63; Suppl. Fig. 2C). The corresponding CIR were 17.5 ± 3% and 11.0 ± 4% (P = 0.13; Suppl. Fig. 2D).

Multivariate analyses including established prognostic factors in AIEOP-BFM trials identified high hyperbilirubinemia as an independent predictor of outcome, while severe transaminasemia (CTC grade 4) did not demonstrate an impact here (Table 2).

Table 2 Estimated hazard ratiosa from the multivariable Cox proportional model on event-free survival and hazard of relapse in patients of the study cohortCharacteristics of the GWAS discovery cohort

The finally pruned GWAS discovery cohort included 650 of 1547 patients with available genome-wide genotyping information (Suppl. Table 3). Both the distribution of hyperbilirubinemia and the clinical characteristics were comparable to those of the entire study population (Table 1 and Suppl. Tables 3 and 4).

Of the 650 patients in this discovery cohort, 215 (33%) patients had normal and 435 (67%) had increased bilirubin levels during induction/consolidation: 367 (56.5%) patients demonstrated moderate hyperbilirubinemia (grades 1-2; 313 during induction, 248 during consolidation, 159 in both phases) and 68 (10.5%) had high hyperbilirubinemia (grades 3-4; 59 reports during induction, 21 during consolidation, 12 in both phases). Pre-treatment hyperbilirubinemia at diagnosis was rare, but more frequent among patients developing chemotherapy-related hyperbilirubinemia during induction/consolidation compared to those not (7.5% (19/252) vs. 1.7% (2/116), P = 0.025). Considering the entire course of therapy, 91 patients (14.0%) had high hyperbilirubinemia in at least one treatment element.

Genome-wide association study

When comparing the 435 patients with hyperbilirubinemia (grades 1-4) during induction/consolidation to the 215 patients with normal bilirubin levels, the five loci most associated with this phenotype were the UGT1A gene cluster, MARK2P5, SULF2, MIR924HG and USH2A (Suppl. Table 5). The strongest associations were observed for variants residing in the UGT1A (UDP glucuronosyltransferase family 1 member A) locus at 2q37. The only variant reaching near genome-wide significance (OR = 2.1, 95% CI = 1.6-2.7, P = 7.3 × 10− 8), rs6744284 was also the index SNV of a 189 kb region of high linkage disequilibrium (LD). Information on the complex UGT1A cluster with its overlapping genes and results of imputation are presented in supplementary material (Suppl. Fig. 3, Suppl. Tables 6 and 7).

To examine whether inclusion of age and immunophenotype would influence allelic association, we compared results from crude and adjusted logistic regression analyses. We did not detect any differences in conferred risk for the variant rs6744284 T allele with reference to the wild-type C allele (unadjusted allelic OR = 2.1, 95% CI = 1.6-2.7, P = 1.8 × 10− 7; adjusted OR = 2.1, 95% CI = 1.6-2.8, P = 1.2 × 10− 7) (Suppl. Table 5).

The genotypic association of rs6744284 with frequency and risk of hyperbilirubinemia during induction/consolidation increased stepwise (Table 3). Compared to wild-type patients (CC; 58% with hyperbilirubinemia grades 1-4), heterozygotes (TC; 71% with hyperbilirubinemia grades 1-4) demonstrated a 1.7-fold higher risk of hyperbilirubinemia, while homozygosity for the T allele (TT; 90% with hyperbilirubinemia grades 1-4) conferred an OR of 6.5 (95% CI = 2.9-14.6, P = 7.0 × 10− 6) (Fig. 2A and Table 3). Inclusion of age and immunophenotype as covariates or as stratifying variables did not significantly alter these results (Suppl. Tables 7, 8 and 9). Notably, TT-homozygotes were at particular risk of developing high hyperbilirubinemia (grades 3-4, OR with reference to CC genotype 16.4, 95% CI = 6.1-43.8, P = 2 × 10− 8).

Table 3 Association between rs6744284 genotype and risk of hyperbilirubinemia during different treatment elementsFig. 2

Frequency of rs6744284 genotype by bilirubin toxicity grading in induction/consolidation (protocols IA/IB) treatment according to the Common Toxicity Criteria of the National Cancer Institute, version 2 (CTC) (A), and by UGT1A1*28/*37 genotype (B). The number of patients (n) for each rs6744284 genotype is given above the columns. A analysis based on 650 patients of the discovery cohort; (B) based on a subset of 544 patients subsequently genotyped for UGT1A1*28/*37 depending on availability of DNA

Independent replication analysis

We performed independent replication analyses in a cohort of 224 ETV6-RUNX1-rearranged pediatric ALL patients (Suppl. Table 10), results of which supported our initial GWAS findings. The allelic OR for hyperbilirubinemia (grades 1-4) during induction/consolidation conferred by the variant rs6744284 T allele versus the wild-type C allele was 2.3 (95% CI = 1.5-3.7, P = 2.4 × 10− 4). Genotypic OR in comparison to wild-type patients (CC) were 2.4 (95% CI = 1.3-4.3, P = 3.8 × 10− 3) for heterozygotes (TC) and 6.1 (95% CI = 1.7-21.6, P = 5.6 × 10− 3) for homozygous variant patients (TT) (Table 3). Similar to our findings in the GWAS discovery cohort, patients possessing the rs6744284 TT genotype were at particular risk of high hyperbilirubinemia (OR with reference to CC genotype 13.6, 95% CI = 2.6-71.8, P = 0.002).

Association testing in the combined discovery and replication cohorts resulted in genome-wide significance. Compared to the rs6744284 wild-type allele, presence of the T allele was associated with an OR of 2.1 (CI = 1.7-2.7) for hyperbilirubinemia (grades 1-4) during induction/consolidation at a significance level of P = 5.7 × 10− 11.

UGT1A rs6744284 genotype in subsequent treatment elements

Similar to initial findings, we observed that the rs6744284 TT genotype was also strongly associated with hyperbilirubinemia during extracompartment therapy (Protocol M, OR = 4.1, 95% CI = 2.2-7.9; P < 0.001), re-intensification (Protocols II and III, OR = 9.1, 95% CI = 4.5-18.6, P < 0.001) and high-risk (HR) block elements (OR = 15.3, 95% CI = 1.8-126.6; P = 0.012) (Table 3, Suppl. Fig. 4). Thus, the effect of rs6744284 on risk of hyperbilirubinemia was not limited to early chemotherapy, but was generalizable to all intensive treatment phases for pediatric ALL.

UGT1A rs6744284 and Gilbert’s syndrome-associated variants

The UGT1A enzyme family is crucial for bilirubin glucuronidation and related impairing genetic alterations form the mechanistic basis of the Gilbert’s syndrome (GS) [18,19,20]. Therefore, we genotyped the GS-related functional genetic variations UGT1A1*28 and *37 [21, 22] in 544 (84%) patients of our discovery cohort with available remission DNA (Suppl. Table 11). Comparable to rs6744284, we observed a strong association with hyperbilirubinemia: the allelic OR for *28/*37 vs. wild-type (*1) was 1.9 (95% CI = 1.4-2.5, P = 5.0 × 10− 6). Genotype-based analyses demonstrated a stepwise increase of frequency and risk of hyperbilirubinemia for the variant alleles. Out of 544 patients 62 (11%) were homozygous for either UGT1A1*28/*28 or *37/*37 – this cannot be differentiated by our assay – and had the highest rate and risk of hyperbilirubinemia (89% compared to 58% for *1/*1; OR in comparison to *1/*1 5.8; 95% CI = 2.5-13.3; P = 3.3 × 10− 5) (Suppl. Table 12). Homozygous variant patients were at particular risk of developing high hyperbilirubinemia (grades 3-4, OR = 12.4, 95% CI = 4.4-34.8, P = 1.9 × 10− 6). The strong interrelationship of rs6744284 with UGT1A1*28/*37 is depicted in Fig. 2B. Extended haplotype analyses including eight additional GS-related variants further documented a strong association with rs6744284 (see Suppl. Table 13 and related additional information). Of note, none of the GS-related variants showed a stronger association with hyperbilirubinemia than rs6744284.

Hyperbilirubinemia, transaminase levels, rs6744284 genotype, treatment delay and outcome in the GWAS discovery cohort (n = 650)

Similar to the patients of the entire study population (n = 1547), patients in our discovery cohort with high hyperbilirubinemia during induction/consolidation tended to take 2 days longer to complete consolidation (P = 0.072) (for details see Suppl. Information, page 25). Of interest, we did not observe significant differences between rs6744284 genotypes: 88 days for TT vs. 89 days for CT and 90 days for CC (P = 0.122).

Consistent with the results obtained for the entire study cohort, outcome analyses of the discovery cohort showed that high hyperbilirubinemia during induction/consolidation was associated with a poor 5-year EFS of 71.8 ± 5%, compared to 87.4 ± 2% and 81.7 ± 3% in patients with moderate and without hyperbilirubinemia, respectively (Fig. 3A). The corresponding 5-year CIR were 19.3 ± 5% for high hyperbilirubinemia, 10.7 ± 2% for moderate, and 13.2 ± 2% for no hyperbilirubinemia (Fig. 4A). Although rs6744284 was strongly associated with high hyperbilirubinemia and the proportion of patients with TT genotype among those with high hyperbilirubinemia was 28% (19/68), there were no differences between rs6744284 genotypes related to EFS or CIR (Figs. 3B and 4B) in the discovery cohort. However, within high hyperbilirubinemic patients those carrying the TT genotype had a better EFS (84.2 ± 8% vs 66.9 ± 7%, P = 0.110) and a lower CIR (5.3 ± 5% vs 24.9 ± 6%, P = 0.039) at 5 years compared to the remaining genotypes (TC, CC) (Figs. 3D and 4D).

Fig. 3

Event-free survival (EFS) at 5 years in ALL patients from the discovery cohort according to (A) maximum total bilirubin toxicity grade during induction/consolidation (protocols IA/IB); (B) rs6744284 genotype (CC, TC, TT); and (C) homozygosity for the UGT1A1*28 or *37 allele. D illustrates the EFS by rs6744284 genotype (CC/TC and TT) restricted to patients with high bilirubinemia (n = 68; grades 3 and 4). Bilirubin toxicity grading was according to the Common Toxicity Criteria (CTC) of the National Cancer Institute, version 2; standard error (SE) and the number of included individuals are indicated for each category

Fig. 4

Cumulative incidence of relapse (CIR) at 5 years in ALL patients from the discovery cohort according to (A) maximum total bilirubin toxicity grade during induction/consolidation (protocols IA/IB); (B) rs6744284 genotype (CC, TC, TT); and (C) homozygosity for the UGT1A1*28 or *37 allele. D shows the effect of the rs6744284 genotype (CC/TC and TT) on CIR in the group of patients with high bilirubin levels (n = 68; grades 3 and 4). Bilirubin toxicity grading was according to the Common Toxicity Criteria (CTC) of the National Cancer Institute, version 2; standard error (SE) and the number of included individuals are indicated for each category

In the GWAS discovery cohort 604 (93%) patients were observed with elevated hepatic transaminase levels (grades 1-4) during induction/consolidation (Suppl. Table 3), 315 of which demonstrated high grades 3-4. Similar to the results obtained for the 1443 patients in the entire study cohort with available information, transaminase levels were positively associated with hyperbilirubinemia. Especially patients with high hyperbilirubinemia were at increased risk for concurrent high transaminase levels compared to the remaining patients (75% vs. 41%, OR = 16.8, 95% CI = 2.2-127.1, P = 0.006) (Suppl. Table 14). Of importance, transaminasemia was not associated with rs6744284 genotype (P = 0.74). Five-year EFS and CIR did not differ between patients with no (grade 0), moderate (grades 1-2) or high (grades 3-4) transaminasemia during induction/consolidation (Suppl. Fig. 5A and B).

Patients with high hyperbilirubinemia and concurrent high transaminasemia tended to have a higher 5-year EFS of 74.3 ± 6% compared to 64.7 ± 12% in patients with high hyperbilirubinemia accompanied by moderate or no transaminasemia (grades 0-2, P = 0.480; Suppl. Fig. 5C). Corresponding CIR were 21.8 ± 6% and 11.8 ± 8% (P = 0.260, Suppl. Fig. 5D).

Multivariate analyses including established prognostic factors in AIEOP-BFM trials identified high hyperbilirubinemia as an independent predictor of outcome, while rs6744284 TT genotype demonstrated only a tentative protective effect in these analyses (Table 4). However, in multivariate analysis restricted to patients with prognostically unfavorable high hyperbilirubinemia in induction/consolidation, the rs6744284 TT genotype was associated with a statistically significant 14-fold lower relapse-risk compared to rs6744284 wild-type or heterozygous variant patients (CC or TC) (Suppl. Table 15).

Table 4 Estimated hazard ratiosa from the multivariable Cox proportional model on event-free survival and hazard of relapse in patients of the discovery cohort