The adult human gastrointestinal microbiome is a stable and complex community comprised of trillions of microorganisms (Cho and Blaser, 2012). Humans are born with an essentially sterile gastrointestinal tract, with microbial colonization starting during birth, and stabilizing into an adult-like composition around three years of age (Milani et al., 2017; Kennedy et al., 2023). The establishment of the microbial community in infants has become widely recognized as a fundamental developmental process (Bokulich et al., 2016; Tamburini et al., 2016; Chu et al., 2017; Milani et al., 2017).

The human gut microbiome plays several essential roles for health and physiology of the host, including influencing the development of the immune system, protection from pathogens, and the production of vitamins and short-chain fatty acids. It has been suggested by (Levy et al., 2017), that the gut microbiome may be influenced by host factors that could be used to indicate gut dysbiosis. In the case of Ulcerative colitis (UC) and Crohn’s disease (CD), shifts in the bacterial community with depletion of Faecalibacterium prausnitzii have been found (Zhang et al., 2016) and similar shifts have been observed in colon cancer (Lopez-Siles et al., 2016).

The link between cancer and the microbiome may be causal, as some microbes are considered oncogenic (Chang and Parsonnet, 2010). However, it is still unclear whether the general composition of the gut microbiota can influence the development of cancer (Zitvogel et al., 2017), e.g. if certain microbial community profiles increase individual susceptibility to the disease. It is also possible that the composition of the gut microbiota is altered by changes in the host physiology resulting from or by cancer, with important implications for host health and treatment prospects.

Childhood cancer is still the predominant cause of death among children over a year in Western countries despite significantly improved treatment and increased knowledge of the molecular mechanisms of various cancers (Siegel et al., 2021). Pediatric cancers are entirely different in comparison to adult cancers, in terms of frequencies, clinical behavior, and the potential factors involved in the onset of the disease. Current knowledge about the role of the gut microbiome in childhood cancer is scarce (Bai et al., 2018). It has, however, been shown that the gastrointestinal microbial community differs between pediatric versus adolescent Acute Lymphomatic Leukemia (ALL) patients in comparison to a healthy control group at the time of diagnosis (Rajagopala et al., 2016), indicating that the gastrointestinal microbiome may play a role in the establishment and progression of the disease. Further, it has been speculated that in pediatric acute leukemia (AL), the perturbation of the gut microbiome may extend the actual treatment with antibiotics, chemo- or immune therapeutics and thus, contribute to long term side-effects in AL survivors (Masetti et al., 2021).

To our knowledge, this pilot study is one of the first studies investigating the gut microbiota of children diagnosed with cancer (D’Amico et al., 2022; Masetti et al., 2022 Cancers). We collected fecal samples from patients diagnosed with cancer prior to the beginning of the treatment regime. In addition, as normal controls, we obtained fecal samples from children attending two different kindergartens in Oslo. We performed 16S rRNA gene amplicon sequencing of the DNA extracted from fecal samples in order to look for differences in the gut microbiota composition between children diagnosed with cancer and a group of healthy children.

Of the 33 patients included in the final analyses, 12 had been diagnosed with tumors of the central nervous system (CNS), 13 with solid tumors (ST), and eight with leukemia ( Table 1 ). The cohort of patients included children within an age range from newborn to 17 years. In particular in the ST group was the median age two years (inter quartile range 0-11) relative to 4.5 (3.5-6.5) for leukemia and 9.5 (3-15) for CNS tumors. A PERMANOVA model did not find any significant effect of sex (p=0.95). However, non-metric multidimensional scaling (NMDS) analysis demonstrated the large effect age had on the microbiota in this cohort ( Figure S2 ). Yatsunenko et al. (2012) and others (Stewart et al., 2018) have reported that the gut microbiota develops an adult composition by three years of age. Thus, we divided the patients into those less than three years old and those three years or older. A PERMANOVA model including both, diagnosis and age group, did not demonstrate any significant effects on diagnosis (p=0.27), while age group was highly significant (p<0.001).

Diversity was significantly reduced ( Figures S3A, B ) in the patients younger than three years, both relative to those three years and older and healthy controls three years and older (henceforth referred to as controls), in terms of species richness and Shannon entropy (p<0.002 for all comparisons, Wilcoxon rank sum test). This result was in line with the expectations (e.g. Yatsunenko et al., 2012). In order to carry out a more realistic comparison with our kindergarten controls, we excluded children younger than three years (11 in total) from our patient cohort in the following analyses. Within the patients of age three years and over, there was no significant effect of age on the microbiota (p=0.57, PERMANOVA). Furthermore, all patients will henceforth be treated as a single group in the comparisons with the control group.

Children diagnosed with cancer had a significantly different microbiota composition (R² = 0.09, p<0.001, PERMANOVA) compared to the microbiotas of the cohort of children sampled from the kindergartens ( Figure 1 ). This result was confirmed by a machine learning approach using a random forest classification model. This resulted in a classification accuracy of 100.0%, with no misclassified samples out of 43 (p<0.001, permutation test). It should be noted that inclusion of children two years of age in both the patient and the healthy cohort did not alter the discrimination of the microbiotas between the groups (R² = 0.09, p<0.001, PERMANOVA).

We observed a marginally higher Shannon entropy in the patient group relative to controls (p<0.09, Wilcoxon rank sum test, Figure S3B ), while ASV richness did not differ significantly between the groups ( Figure S3A ). At the phylum level, we observed an increased mean relative abundance of Actinobacteria in the patient group (p=0.042, Wilcoxon rank sum test) ( Figure S4 ), but no other significant differences.

ASV level analysis using DESeq2 identified 13 ASVs with significantly (Benjamini-Hochberg corrected p<0.05) differential occurrence between the patient and the control group ( Table 2 ). six of these ASVs were depleted in children diagnosed with cancer. Strikingly, three of these eight were classified as Faecalibacterium prausnitzii (ASV2, ASV3, and ASV6). The combined relative abundance of these three ASVs were significantly higher in the control cohort (p<<0.001) ( Figure 2A ). All three groups were highly prevalent in the data ( Figures 2B–D ) and additional testing confirmed the DESeq2 results (Wilcoxon rank sum test ASV2=p<0.0001, ASV3 p<0.012, and ASV6 p<0.001). In order to investigate the robustness of our results if the young children were included, we checked that no significant relationship was observed between the relative abundance of these three ASVs and the age parameter in the patient group ( Figures S5A–C ). Furthermore, all three ASVs were found in both the patients and the controls. Since systematic bias could be introduced by differences in sample collection, we compared the combined relative abundances of ASVs assigned to F. prausnitzii in fecal samples that were frozen (7 samples, 3 individuals) with paired samples collected in ethanol (7 samples, 3 individuals). No statistically significant difference was observed (Wilcoxen rank sum test p=0.32). The other three ASVs depleted in patients were Haemophilus parainfluenzae, Sutterella wadsworthiensis, as well as Ruminococcus. Among the ASVs that were significantly overrepresented in the patient group, were Prevotella copri, Steptococcuss, and Dorea longicatena.

ASV2, ASV3, and ASV6 were also variables of high importance for successful discrimination between the patients and the controls in the random forest model ( Figures S6A, B ). The random forest model also identified the following short chain fatty acid producing taxa as being among the top 30 variables of highest importance for successful classification: Roseburia spp., Ruminococcus spp., Anaerostipes hadrus, Coprococcus comes, Intestinimonas, Lachnospiracea, Roseburia inulinivorans, and Blautia faeces.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ncbi.nlm.nih.gov/genbank/, PRJNA814879.

The studies involving humans were approved by Regional Committee for Medical and Health Research Ethics (REK nr: 2018/2491 and 2019/66945). The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation in this study was provided by the participants’ legal guardians/next of kin.

LB and MM-K initiated the overall study. LB and MM-K conceived and designed the childhood cancer, EM, PT, NN, and PF the control study. EM, PT, NN, PF, VS, and JH tested and established various methods for the study. NR and JH collected, prepared and managed the samples from children diagnosed with cancer, EM, NN, and PF from children of the control study from Norwegian kindergartens. EM, PT, and VS performed DNA extraction and sequencing. EM and PT established the analysis pipeline and carried out the data analysis. EM, PT, MM-K, and LB interpreted the data. EM and PT wrote the draft of the manuscript, JH, MM-K, NN, and PJ critical reviewed the manuscript, and EM and LB finalized it. All authors read and approved the final manuscript. All authors contributed to the article.