Changes in gut microbiota predict neutropenia after induction treatment in childhood acute lymphoblastic leukemia

During the past decades, intensification of chemotherapy treatment has significantly improved survival rates in pediatric acute lymphoblastic leukemia (ALL).¹ However, this progress has come with a cost of prolonged immunosuppression, with increased risk of infections and delay in subsequent antineoplastic treatment, thus jeopardizing the control of leukemia.^2,3 

Neutrophils play a pivotal role in the innate immune system as a first-line defense and are crucial for combating bacterial infections through early recruitment to sites of infection and tissue damage.⁴ Understanding the factors driving neutrophil recovery after chemotherapy is essential for devising improved strategies to limit infectious complications and treatment delays in ALL treatment.

Emerging evidence suggests a key role of the gut microbiota in regulating neutrophil production and migration to extravascular tissues.⁵ Experimental mice studies have substantiated these findings, demonstrating that certain gut bacteria facilitate neutrophil recovery after chemotherapy-induced neutropenia.^5,6 In clinical settings, extensive antibiotic use has consistently been linked to an increased risk of prolonged neutropenia in various patient groups.^7,8 Although the underlying mechanisms behind prolonged neutropenia are poorly understood, they may involve loss of the gut microbiota-induced stimulation of granulopoiesis within the bone marrow as well as increased extravascular migration of neutrophils due to overgrowth of proinflammatory bacteria.^5,9-11 

Although neutrophil production is controlled by colony-stimulating factors, among those granulocyte colony-stimulating factor (G-CSF),¹² the recruitment of circulating neutrophils is regulated by chemokines, a superfamily of immune-modulatory small proteins, the primaries and most well-described being CXCL1 and CXCL8.^13,14 These are released from antigen-presenting cells and damaged tissues, including the gut, independently of the peripheral blood neutrophil count.¹⁵ 

This study aimed to investigate the dynamics of the relationship between gut microbiota composition and neutropenia in children undergoing ALL induction treatment. Moreover, we explored associations between the microbiota and signaling proteins controlling neutrophil production and migration.

In this prospective, observational multicenter study, 51 children with newly diagnosed ALL treated according to the Nordic Society of Pediatric Hematology and Oncology ALL 2008 protocol were included at University Hospital of Copenhagen Rigshospitalet, Odense University Hospital, and Aarhus University Hospital, Denmark, between 2015 and 2018. Patients received a 29-day induction chemotherapy treatment, as previously described.¹⁶ All time points reported refer to the treatment day of the 29-day induction period. The cohort of this study has been previously published as part of larger studies.^17-19 

All patients received trimethoprim-sulfamethoxazole twice weekly as prophylaxis for Pneumocystis jirovecii infection. The first-line antibiotics for neutropenic fever were meropenem (20 mg/kg × 3 daily) + gentamicin (5 mg/kg daily) in Copenhagen and piperacillin/tazobactam (100 mg/kg × 4 daily) in Odense and Aarhus. Positive blood cultures and antibiotic administrations were registered from the patients’ medical records.

Of the 51 families, 36 were screened for sibling eligibility, of which 12 refused participation and 5 did not meet the inclusion criteria (age 1-25 years, without chronic diseases, and sharing the same parents and housing as the patient).¹⁸ Therefore, 19 siblings were included in the microbiota assessment.

The peripheral blood absolute neutrophil count (ANC) was determined daily part of the clinical routine by flow cytometry (Sysmex XN, Norderstedt, Germany). Prolonged neutropenia was defined as ANC <0.5 × 10⁹ cells per L 1 week after the end-of-induction treatment (day 36), the time point for initiation of consolidation therapy in patients with ANC >0.5 × 10⁹ cells per L.

EDTA-anticoagulated blood was collected from each patient on treatment days 8, 15, 22, and 29 (±2 days). Ten unrelated healthy young blood donors (aged 20-31 years, 50% male) served as controls for neutrophil signaling marker measurements, as healthy siblings did not provide blood samples. Concentrations of CXCL1, CXCL8, and G-CSF were measured using the Bio-Plex Pro Human Chemokine Assay (Bio-Rad, Hercules, CA) on the Luminex platform (Luminex Corporation, Austin, TX), according to the manufacturer's instructions, as previously described.¹⁷ 

Fecal samples were collected on treatment days 1 (range, −2 to 5), 8 (5-11), 15 (13-18), 22 (20-25), and 29 (27-34), for a total of 185 samples (median, 4; range, 1-5, per patient). The number of samples available at each time point is shown in supplemental Table 1. The fecal samples were stored at −20°C immediately after collection and transferred to −80°C within 2 to 5 days. Siblings had fecal samples collected at 2 to 62 days from the patients’ start of chemotherapy. None of the siblings had received antibiotics in the month before sampling.

DNA extraction and 16S ribosomal RNA (rRNA) gene amplicon sequencing were performed as previously described.¹⁸ In short, DNA was extracted from fecal samples and 11 blank controls using the Qiagen QIAamp Fast DNA Stool mini kit, modified to include a bead-beating step (Qiagen PowerBead Tubes, Garnet 0.70 mm). High-throughput amplicon sequencing of the V3–V4 region of the 16S rRNA gene was conducted using Illumina MiSeq technology (Illumina Inc., San Diego, CA). Samples were resequenced if the read count after chimera removal (described below) was <9000.

Primer sequences were removed with cutadapt (version 1.16)²⁰ at a tolerated maximum error rate of 15%. Reads were further processed using the R package DADA2 (version 1.8) for inference of high-resolution amplicon sequence variants (ASVs).²¹ Forward and reverse reads were truncated at 280 and 200 base pairs, respectively, to allow overlap for merging. Chimeras were identified sample-wise and removed from the whole data set (removeBimeraDenovo function, method “consensus”). Taxonomy was assigned with SILVA reference database version 132 formatted for DADA2.²² No technical batch effects by sequencing run, extraction round, extraction kit, or extraction date were detected by principal coordinates analysis (PCoA) with different distance metrics (Bray-Curtis, Jaccard, Raup-Crick, Chao and Horn-Morisita) using the R package phyloseq. A total of 266 putative contaminant ASVs were removed using the R package decontam, following the “combined” method with a probability threshold of 0.07²³ and manually assessing borderline sequences. Subsequently, only ASVs with at least 5 reads in at least 2 samples were selected for further analysis using the kOverA function of the R package gene filter. After this step, the median sequencing depth was 20 649 reads (range, 5038-326 811).

The study was approved by the Scientific Ethics Committee of the Capital Region of Denmark (H-16016942) and the Danish Data Protection Agency (RH-2016-214) and was conducted in accordance with the Declaration of Helsinki. Oral and written informed consent was obtained from the parents or legal guardians.

Statistical analyses were performed using R (3.4.0, R Foundation for Statistical Computing, Austria).²⁴ Sequencing data were integrated using the R package phyloseq. The within-sample bacterial diversity (α-diversity) was calculated as the number of observed ASVs (richness) and the Shannon index (accounting for both richness and evenness of the ASVs present). Between-sample diversity (β-diversity) was computed using the Bray-Curtis dissimilarity at the ASV level. As the primary end point, this study focused on neutropenia persisting at day 36 (hereafter referred to as prolonged neutropenia), since day 36 is the scheduled start of consolidation therapy if ANC >0.5 × 10⁹ cells per L was achieved.

A mixed model with an unstructured variance-covariance matrix was applied to compare chemokine levels, α-diversity, and relative abundance of the microbial community at the genus level over time and between groups. The interaction between time points and variables was tested using the likelihood-ratio test. Chemokine levels, observed richness, and relative abundance were log-transformed using the lowest nonzero value as a pseudocount. For the analyses of α- and β-diversity at individual time points, all weekly time points were analyzed and only reported if P value was <.05.

Correlation analyses were performed using Spearman rank-order correlation analysis. Logistic regression models were used to assess the risk for prolonged neutropenia. All potential risk factors listed in Table 1 were tested in univariate analyses and included in a multivariate model if they showed statistically significant associations with outcome variables and/or microbiota, as indicated in the results.

β-diversity was visualized with PCoA plots and tested for inference using permutational multivariate analysis of variance (adonis2 from the package vegan with 999 permutations).

A supervised sparse partial least squares (sPLS) regression model was used to estimate the predictive power of multiple bacteria at the genus level in relation to prolonged neutropenia. Relative abundances were used as input features after filtering (day 1 samples, genera with a prevalence >25%). We tuned the sPLS models with a range of included genera (1, 2, 4, 8, 16, 25, 40, 50, and 65) to be kept for 1 to 5 components. We selected the optimum number of input variables (bacterial genera) using 5-time repeated 10-fold cross-validation of the area under the curve (AUC) statistic to avoid overfitting. The final model was chosen based on the highest median AUC value and included the total number of included genera (n = 65) and 1 component. The predicted values of the left-out folds were combined into a PLS score. Further validation of the set of selected genera through a data-splitting approach was not feasible due to the relatively small data set. Finally, the relative abundance of all genera with a prevalence >25% in day 1 samples was compared between groups using the univariate Wilcoxon rank-sum test. Multiple test correction was performed for all analyses regarding the abundances of bacterial genera using the Benjamini-Hochberg false discovery rate (FDR) method. Both P values and FDR-corrected P values are reported.

The clinical characteristics of 51 patients are listed in Table 1. The healthy siblings were older than the patients (median: 7.1 [range, 1.4-17.6] years vs 3.7 [1.1-17.3] years; P = .006) and represented less males (50%) than the patient cohort (75%).

Median peripheral blood ANC declined during induction treatment reaching nadir at day 15 (0.08 × 10⁹ cells per L, range, 0.0-0.6) (Figure 1A) and then gradually recovered toward day 29 (median, 0.7 × 10⁹ cells per L; range, 0.0-10.0). At diagnosis, before the start of chemotherapy, 24 (47%) patients exhibited signs of leukemia-induced bone marrow suppression with neutropenia (ANC <0.5 × 10⁹ cells per L). On day 15, after the initiation of chemotherapy, 44 (91%) of the patients had neutropenia. Prolonged neutropenia, defined as neutropenia persisting on treatment day 36 (the day of the scheduled start of consolidation therapy), was observed in 18 (37%) patients. The median duration of neutropenia before consolidation therapy was 24 days (range, 0-59).

Due to suspected bacterial infection, 47 (92%) patients received antibiotic treatment during the induction period, which was initiated before day 1 in 44 (86%) patients, with a median duration of 17 days (range, 0-29 days). Bacteremia was confirmed in 13 (27%) patients on median day 13 (range, 5-20). Details regarding antibiotics and microbial findings are provided in supplemental Tables 2 and 3.

Prolonged neutropenia was not associated with sex, age, ALL phenotype, induction regimen, ANC, lactate dehydrogenase, or blasts (%) in peripheral blood at treatment initiation (day 1), duration of antibiotics, or documented bacteremia during the induction period (all P > .05).

Plasma levels of CXCL1, CXCL8, and G-CSF were significantly increased in the patients than in the healthy controls, reaching a peak on day 15 (Figure 1B-D). Older patients had lower levels of CXCL1 (16% decrease per year; 95% confidence interval [CI], −10 to −21; P = .01) and CXCL8 (9% decrease per year; 95% CI, 5-14; P < .001) overall during the induction treatment but not G-CSF (P = .08). Patients with bacteremia had higher levels of CXCL8 than those without bacteremia (86% higher; 95% CI, 18-296; P = .008). Levels of CXCL1, CXCL8, and G-CSF were not associated with sex, ALL phenotype, induction regimen, or duration of antibiotics (all P > .05).

Patients with prolonged neutropenia had overall significantly higher levels of CXCL1 (177% higher; 95% CI, 48-417; P = .002), CXCL8 (122%; 95% CI, 46-239; P < .001) and G-CSF (56% higher; 95% CI, 2-138; P = .04) during the induction period than patients without prolonged neutropenia (Figure 1G-I). This remained significant after adjusting for age (CXCL1, 116% higher; 95% CI, 20-288; P = .01; CXCL8, 100%; 95% CI, 38-191; P < .001). Similarly, ANC correlated inversely with same-day levels of both chemokines and G-CSF at all measured time points (Spearman's coefficient [rho] = −0.48 to −0.71; all P < .01).

The overall α-diversity was significantly reduced in the patients compared with the healthy siblings on all measured days for the Shannon index (P = .03 to <.001) and days 8 to 29 for observed richness (all P < .001), reaching a nadir on day 8 for both observed richness (supplemental Figure 1) and Shannon index (Figure 1E). Patients receiving antibiotics on day 1 had overall lower α-diversity during induction treatment than patients not receiving antibiotics on day 1 (Shannon, −0.5; 95% CI, −0.06 to −0.9; P = .03), and older patients had marginally higher α-diversity measures (Shannon, 0.05 per year; 95% CI, 0.01-0.08; P = .01). Diversity metrics were not associated with sex, ALL phenotype, induction regimen, treatment center, type of antibiotics (supplemental Figure 2), or bacteremia (all P > .05).

Low α-diversity on day 1 was associated with an increased risk of prolonged neutropenia (odds ratio = 0.06 per doubling of Shannon index; 95% CI, 0.002-0.8; P = .05) (Figure 1J), and this remained significant after adjusting for patient age (supplemental Table 4 including the results of additional covariates). Additionally, α-diversity only on treatment day 15 correlated with ANC on days 15 (Shannon, rho = 0.53; P = .002) and 22 (rho = 0.52; P = .007) and was negatively correlated with neutropenia duration (Shannon, rho = −0.52; P = .002). The α-diversity metrics at treatment day 15 correlated negatively with chemokine levels on days 15 and 22 (Figure 2).

No significant associations between microbiota diversity metrics and G-CSF levels were found.

Next, we examined the dissimilarities in microbial composition between fecal samples, known as β-diversity, throughout the induction course using PCoA (Figure 3A). On day 1, the healthy siblings tended to form a rather homogenous group distinct from the patients, suggesting that they shared a similar gut microbial composition compared with the patients (P = .02 by Adonis 2) (Figure 3B).

Given the similarity in β-diversity across treatment days among patients (Figure 3A), we opted to focus our analysis of β-diversity in relation to outcomes on the final treatment day (day 29).

β-diversity on day 29 was significantly associated with prolonged neutropenia (Figure 3C) and correlated with day 29 levels of CXCL1 (P = .05), CXCL8 (P = .004), and G-CSF (P = .02). Likewise, β-diversity at day 29 was associated with same-day chemokine levels dichotomized at the median (Figure 3D-E), but not with G-CSF levels (P = .07; Figure 3F).

Taxonomic analyses of the specific bacterial microbes were performed at the genus level. Here, we included genera represented in >25% of the patient fecal samples.

First, we investigated changes in relative abundance over time throughout the induction treatment. Among the 65 included bacterial genera, Enterococcus (P = .01) and Romboutsia (P = .001) were enriched during the induction period compared with day 1, whereas only the commensal bacteria Faecalibacterium (P = .01) and Erysipetotrichacea UCG-003 (P = .001) were depleted when looking at all patients as 1 group.

However, patients with prolonged neutropenia had significantly lower relative abundance of Ruminococcacea UCG-002, Ruminococcacea UCG-003, Ruminococcacea UCG-005, Ruminoclostridium_9, Agathobacter, Fusicatenibacter, Lachnospira, Roseburia, Lachnospiraceae UCG-004, Collinsella, Romboutsia, and Sutterella throughout the induction treatment, and experienced significantly greater overgrowth of Enterococcus during the induction period compared with the patients who recovered from neutropenia (Figure 4).

Looking at the time of ALL diagnosis, the patients had reduced abundances of 16 genera, primarily from the Ruminococcaece and Lachnospiraceae families, compared with the healthy siblings, whereas Enterococcus and Eggerthella were more abundant in the patients (Figure 5A).

To predict the risk of prolonged neutropenia from the microbiota on day 1 (n = 12 [38%]), we implemented a multivariate approach using an sPLS regression model. Using this model, we identified a pattern of various Ruminococcaece and Lachnospiraceae genera together with Veillonella, Sutterella, Collinsella, and Erysipelotrichaceae UCG-003 to be the bacterial genera with the lowest negative loadings, whereas Enterococcus had the highest loading (Figure 5B). This reflects a combined pattern of low vs high abundances of these specific genera on day 1, predicting the risk of prolonged neutropenia (median [range] cross-validated AUC, 0.68 [0.63-0.69]; Figure 5C).

Figure 5D shows the relative abundances of the genera on day 1 with significant differences (Wilcoxon rank-sum univariate test, P < .05) in relative abundance between groups (prolonged neutropenia). After FDR correction, none of the univariate analyses were statistically significant.

To assess the correlations between the abundance of genera (prevalence >25%, n = 65) over time and neutrophil signaling markers, we analyzed the individual’s maximum levels of CXCL1, CXCL8, and G-CSF throughout the induction period (CXCL1_max, CXCL8_max, and G-CSF_max). After FDR correction, CXCL8_max correlated inversely with 10 genera, of which 8 belonged to Ruminococcaece or Lachnospiraceae families, whereas 5 genera were among those found to be associated with recovery from neutropenia (Figure 6A). Likewise, CXCL8_max correlated positively with Enterococcus.

CXCL1_max and G-CSF_max were not associated with any of the 65 investigated bacterial genera.

To visualize the associations between chemokine levels and the abundance of bacterial genera, we dichotomized patients by the median CXCL8_max level. The genera that were significantly associated, after FDR correction, with CXCL8 _max, both when analyzed as a continuous variable and when dichotomized, are shown in Figure 6B.

All genera presented in Figure 6B remained significantly correlated with CXCL8_max after adjusting for patient age, except Enterococcus (FDR-corrected P = .06).

This study delves into the intricate interplay between the gut microbiota and neutrophil recovery during the induction treatment of childhood ALL. Our findings demonstrate an association between gut microbial composition and the duration of neutropenia, uncovering potential implications for the management and improvement of outcomes in patients undergoing chemotherapy.

One of the key observations in our study is the association between loss of gut microbiota diversity and prolonged neutrophil recovery after chemotherapy-induced neutropenia. Notably, alterations in diversity were evident at the onset of chemotherapy, suggesting that the richness of the gut microbiota at diagnosis provides valuable insights into the potential immunosuppressive burden of intestinal dysbiosis. This points to the potential role of microbial α-diversity as a predictive indicator for the immune reconstitution process, although its clinical utility as a biomarker appears less likely due to the large variation in the data.

We identified specific commensal bacteria, including Veillonella, members of the Lachnospiracea family (Coprococcus 1, Roseburia, and Dorea), and the Ruminococcaceae family (UGG-002, UGG-003, and Subdogranulum), as potential contributors to neutrophil recovery dynamics. The absence of these commensals or an increased relative abundance of Enterococcus at the initiation of chemotherapy treatment emerged predictive of the duration of chemotherapy-induced neutropenia. When analyzing the changes in microbiota composition throughout the induction period, similar patterns were found with an increased relative abundance of Enterococcus and a loss of various Lachnospiraceae and Rumonicoccaeae genera during the induction treatment, which was associated with a prolonged period of neutropenia.

Although the study design did not allow for exploring events preceding hospital admission for leukemia, we speculate that antibiotic treatment before leukemia diagnosis and disease-induced diet changes could potentially explain the substantial variation in gut microbiota composition observed among patients already at the time of leukemia diagnosis. These speculations are substantiated when considering the significant differences in gut microbiota composition observed in healthy siblings from the same home environment.

Most of the taxa we found associated with reduced risk of prolonged neutropenia, including Ruminococcacea spp., Coprococcus, Dorea, and Roseburia, are well-known to diminish inflammation by inhibiting the NF-κB pathway,^25-27 which is known to be activated during chemotherapy-induced gut barrier injury,²⁸ and by producing short-chain fatty acids (SCFAs), especially butyrate.^29-31 Butyrate plays a multifaceted role, both systemically with effects on the host immune system as well as locally, by stabilizing the gut mucosal barrier as an energy source for colonocytes.³² Furthermore, during chemotherapy treatment these SCFA-acid-producing microbes may add to the integrity of the intestinal barrier through increased mucus production, improved tight junction integrity, and reduced inflammation.^33-36 Veillonella is, particularly, known to promote general gut health through SCFA production by lactate fermentation.³⁷ 

Although the detailed biological functions of many Ruminococcaceae genera remain incompletely understood, recent studies suggest their immunomodulatory properties.^38-40 Specifically, an experimental mouse study associates Ruminococcaceae UCG-014 with granulopoiesis after chemotherapy-induced neutropenia, through a mechanism involving T-cell production of interleukin-17a, which subsequently stimulates G-CSF secretion.⁶ 

In line with our results, Schluter et al established a connection between the high abundance of Ruminococcaceae genera and rapid neutrophil engraftment after hematopoietic stem cell transplantation (HSCT),⁴¹ whereas Ingham et al found high abundances of mainly Ruminococcaceae and Veillonella to be associated with improved adaptive immune cell reconstitution - both T cell subset and B cells - after pediatric HSCT.⁴² Notably, prior studies in adult patients with HSCT have consistently revealed that the loss of the taxonomic families of Ruminococcaceae and Lachnospiraceae is associated with both increased treatment-related mortality and low overall survival.^43-47 Moreover, members of Ruminococcaceae were found associated with increased efficacy of chimeric antigen receptor T-cell therapy.⁴⁸ The role of Sutterella, which we also found associated with reduced risk of prolonged neutropenia, is poorly understood, although its capacity to adhere to intestinal epithelial cells also implies a potential immunomodulatory function for this genus.⁴⁹ 

Enterococcus, though generally considered harmless in healthy individuals, has emerged as a potential proinflammatory contributor in the context of gut microbiota dysbiosis.⁵⁰ Through disturbed interaction with the host, which might be facilitated through reduced mucosal barrier integrity, these bacteria can produce substances and cell wall components acting as pathogen-associated molecular patterns to activate innate immune responses in the host,⁵¹ including the release of chemokines.¹¹ 

The well-documented expansion of Enterococcus after chemotherapy has been associated with adverse effects, such as bacteremia, systemic inflammation, and enterocyte damage.^18,52,53 The present findings extend these insights and contribute to the claim that Enterococcus overgrowth stimulates proinflammatory responses, including the secretion of chemokines, thereby attracting neutrophils to extravascular tissue.

Consequently, the overgrowth of Enterococcus and depletion of barrier-stabilizing commensals, coupled with depressed bone marrow function, may reduce peripheral blood ANC. The resulting proinflammatory responses, which remove neutrophils from circulation, might extend the duration of neutropenia despite hematopoietic recovery in the bone marrow. In the present study, this hypothesis was substantiated by the finding of a positive correlation between a higher abundance of Enterococcus and elevated chemokine levels.

In immunocompetent individuals, enhanced extravascular neutrophil migration can be easily compensated by upregulated granulopoiesis mediated by colony-stimulating factors. However, in the present cohort of patients with ALL, this compensatory mechanism is comprised due to chemotherapy-induced bone marrow impairment, which may explain our observation of neutropenia despite elevated G-CSF levels.

Although antibiotics lead to improved overall survival in patients with cancer,^54,55 their use is accompanied by substantial alterations in the gut microbiota.^56,57 Long-term use of antibiotics followed by neutropenia has been observed in patients with various infections.^7,8 This dual impact of antibiotics, acting as both a shield and a potential risk factor, has sparked considerable debate and emphasizes the need for more targeted use of antibiotics, especially during periods of increased vulnerability, such as chemotherapy-induced neutropenia, as highlighted in this study.

The complex interplay between the gut microbiota and neutrophils is suggested to be bidirectional.⁵⁸ Nevertheless, in this study, the predictive value of the microbiota composition at treatment onset emphasized that particularly the loss of specific commensals may contribute to prolonged neutropenia, which, additionally, may explain previous observations of antibiotic-induced neutropenia.⁷ Additionally, the study design with time-serial sampling strengthens our ability to discern the chronological order in the relationship between gut microbiota changes and neutropenia. Yet, due to the observational nature of this study, conclusions regarding causality cannot be drawn.

A notable limitation of our study is the sample size. Even though not maintaining statistical significance after adjusting for multiple tests, this exploratory observational study yields valuable insights into the potential role of specific gut bacteria in influencing prolonged neutropenia. Notably, our findings align with those of previous studies and are supported by hypotheses concerning the impact of SCFA-producing bacteria on the host immune system.

In conclusion, this study highlights the intricate relationship between gut microbiota composition and neutrophil recovery during induction treatment of childhood ALL. Our results indicate that perturbation of the gut microbiota may influence blood neutrophil counts by upregulating chemokine production, leading to increased neutrophil extravasation, rather than by suppressing growth-promoting signaling in the bone marrow. Although the study does not establish causality, it lays the groundwork for future investigations exploring the mechanisms behind these associations and their clinical implications for developing and optimizing gut microbiota-sparring strategies to improve treatment outcomes in pediatric patients with leukemia.

Sequence preprocessing was performed using the Danish e-Infrastructure Consortium National Life Science Supercomputer at the Technical University of Denmark. M.E.S. is a postdoctoral fellow of the BRIDGE Translational Excellence Program at the University of Copenhagen, funded by the Novo Nordisk Foundation. The authors express their deepest gratitude to the children and their families for their contribution to the study.

This work was supported by grants from the BRIDGE Translational Excellence Program at the Faculty of Health and Medical Sciences, University of Copenhagen, funded by the Novo Nordisk Foundation (NNF20SA0064340), the Danish Childhood Cancer Foundation, and the Dagmar Marshall Foundation.

Contribution: M.E.S. conducted data acquisition, analysis, interpretation, and drafted the manuscript; K.M. supervised the study from conception to data acquisition, analysis, and interpretation; U.B. and J.S. developed the bioinformatics pipeline and contributed to statistical analyses; S.D.P., S.W., M.R., H.H., B.A.-N., C.E., and S.P. contributed to study implementation and data acquisition; and all authors had access to the study data, made significant contributions to the data analysis and interpretation, and provided intellectual input throughout the study.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Maria Ebbesen Sørum, Department of Pediatrics and Adolescent Medicine, University Hospital Rigshospitalet, Blegdamsvej 9, 2100 Copenhagen, Denmark; email: [email protected].