Abstract Background Children with cancer receiving chemotherapy commonly report a cluster of psychoneurological symptoms (PNS), including pain, fatigue, anxiety, depression, and cognitive dysfunction. The role of the gut microbiome and its functional metabolites in PNS is rarely studied among children with cancer. This study investigated the associations between the gut microbiome–metabolome pathways and PNS in children with cancer across chemotherapy as compared to healthy children. Methods A case–control study was conducted. Cancer cases were recruited from Children’s Healthcare of Atlanta and healthy controls were recruited via flyers. Participants reported PNS using the Pediatric Patient-Reported Outcomes Measurement Information System. Data for cases were collected pre-cycle two chemotherapy (T[0]) and post-chemotherapy (T[1]), whereas data for healthy controls were collected once. Gut microbiome and its metabolites were measured using fecal specimens. Gut microbiome profiling was performed using 16S rRNA V4 sequencing, and metabolome was performed using an untargeted liquid chromatography–mass spectrometry approach. A multi-omics network integration program analyzed microbiome–metabolome pathways of PNS. Results Cases (n = 21) and controls (n = 14) had mean ages of 13.2 and 13.1 years. For cases at T[0], PNS were significantly associated with microbial genera (e.g., Ruminococcus, Megasphaera, and Prevotella), which were linked with carnitine shuttle (p = 0.0003), fatty acid metabolism (p = 0.001) and activation (p = 0.001), and tryptophan metabolism (p = 0.008). Megasphaera, clustered with aspartate and asparagine metabolism (p = 0.034), carnitine shuttle (p = 0.002), and tryptophan (p = 0.019), was associated with PNS for cases at T[1]. Gut bacteria with potential probiotic functions, along with fatty acid metabolism, tryptophan, and carnitine shuttle, were more clustered in cancer cases than the control network and this linkage with PNS needs further studies. Conclusions Using multi-omics approaches, this study indicated specific microbiome–metabolome pathways linked with PNS in children with cancer across chemotherapy. Due to limitations such as antibiotic use in cancer cases, these findings need to be further confirmed in a larger cohort. Supplementary Information The online version contains supplementary material available at 10.1186/s12967-024-05066-1. Keywords: Children, Gut microbiome, Metabolome, Solid tumor, Chemotherapy, Multi-omics, Gut–brain axis Background Children with cancer receiving intensive chemotherapy frequently report cooccurring psychoneurological symptoms (PNS), including pain, fatigue, anxiety, depression, and cognitive dysfunction [[45]1]. Collectively, these symptoms are defined as the PNS cluster, which can develop up to 6 months after treatment and even continue into survivorship [[46]2]. Unfortunately, poor management and treatment of PNS can significantly reduce a child’s quality of life (QOL) and future psychosocial functioning [[47]3, [48]4]. A symptom experience framework presented by Hockenberry and Hooke identified multiple antecedents that influence children’s experience of PNS across cancer treatment, including personal (e.g., sex and developmental stage), environmental (e.g., child’s hospitalization), and disease-related (e.g., type of cancer, length of treatment, treatment frequency, and chemotherapy drugs) factors [[49]5]. Subsequent literature proposed that the PNS cluster may share common biological mechanisms [[50]6], such as proinflammatory cytokines (e.g., IL-6 and TNF-α), Hypothalamic–Pituitary–Adrenal (HPA) axis, and monoamine neurotransmission system [[51]7–[52]9]. Nevertheless, the biological mechanisms of the PNS cluster are still largely unknown in cancer populations, particularly in pediatric oncology [[53]10]. Recently, investigations of the microbiome–gut–brain (MGB) axis [[54]11, [55]12] suggest that the gut microbiome (i.e., a collection of microorganisms and their genomes in the gastrointestinal tract) can signal the brain via functional metabolites and activation of other pathways (e.g., neurotransmitters), ultimately resulting in PNS for patients with cancer receiving chemotherapy [[56]13, [57]14]. Chemotherapy has the potential to negatively interfere the MGB axis through a diverse set of pathways, including dysregulating the diversity and composition of bacteria in lumen, altering the gut microbiome-derived metabolites, and activating neuroimmune signaling [[58]11, [59]12, [60]15]. As a commonly used treatment modality in children with cancer, chemotherapy can potentially lead to PNS via the MGB axis. Although limited, promising work has demonstrated enriched abundance of Bacteroides among adult patients with low PNS and enriched abundance of Blautia for those with high PNS [[61]16]. Additionally, adult patients with head and neck cancer with high PNS had higher abundance of gut microbial Bacteroidota, Ruminiclostridium, and Tyzzerella compared to those with low PNS, while patients with low PNS had higher abundance of Lactococcus and Phascolarctobacterium compared to those with high PNS [[62]13]. However, the role of the gut microbiome in PNS for children with cancer (CWC) has yet to be elucidated [[63]17]. Microbiome-derived metabolites represent the functional role of the gut microbiome, as they are the drivers of gut–brain communication and carry out signals of a disturbed gut microbiome [[64]18]. Communications between the gut and the brain occur following a network of pathways involving key microbial metabolites, such as short-chain fatty acids (SCFAs) [[65]19] and tryptophan for kynurenine pathway metabolism [[66]18]. SCFAs are part of a group of key microbial metabolome pathways associated with psychological functioning [[67]20]. Alterations in the SCFA metabolism can result in disturbances to the central nervous system [[68]21], although the effects of SCFAs on PNS have primarily been studied in animal models [[69]20]. Additionally, tryptophan, an essential amino acid, is another key metabolite in the MGB axis, with dual emphasis on the regulation of serotonin and melatonin synthesis, and the control of kynurenine pathway [[70]18, [71]22]. Tryptophan must be obtained from dietary or microbial sources [[72]18] and can be synthesized from chorismate by bacterial phyla Pseudomonadota, Actinomycetota, and Bacillota [[73]23]. In humans, untargeted metabolomics analysis showed that increased pain was associated with decreased tryptophan, and increased fatigue was associated with decreased arachidonic acid [[74]24] in women with breast cancer receiving chemotherapy. Targeted metabolomics analysis further indicated moderate-to-strong correlations between changes in pain and tryptophan, as well as between changes in depressive symptoms and serotonin levels [[75]24, [76]25]. Decreased tryptophan, increased kynurenine, and subsequent altered tryptophan/kynurenine ratio were associated with a higher level of PNS among cancer survivors [[77]25]. Among children with cancer receiving chemotherapy, fatty acids pathways were associated with pain, and both tryptophan and carnitine shuttle pathways were associated with the PNS cluster [[78]14]. A growing body of preclinical studies support the impact of the gut microbiome and microbial metabolites on the gut–brain communications via neuronal, immunological, and endocrinological pathways [[79]26]. However, research on this mechanistic pathway in the context of chemotherapy-related PNS is still very limited. Furthermore, current work primarily adopts single-omics approaches (e.g., microbiome analysis or metabolomics analysis independently) in human health and disease. On the other hand, multi-omics approaches provide an opportunity to examine multiple layers of molecules (e.g., microbiome and metabolites) [[80]27] to interpret health outcomes. Thus, there is paucity of research regarding how the interrelationship between the gut microbiome and their metabolites can influence PNS among patients with cancer receiving chemotherapy. Considering the severe PNS burden among children with chemotherapy and the unknown biological mechanisms of PNS, uncovering the multi-omics biological pathways within the MGB axis will pave a way for precision medicine (e.g., diet and probiotic interventions) to manage and treatment-related psychoneurological toxicities among CWC. The purpose of this study was to investigate the associations between the gut microbiome–metabolome pathways and PNS among CWC receiving chemotherapy (pre-cycle two chemotherapy [T[0]] and post-chemotherapy within 4 weeks [T[1]]) compared to a group of healthy children (HC). An integrative multi-omics approach (i.e., metabolomics coupled to amplicon microbiome data) was adopted to examine the interrelationship of PNS-associated microbial taxa and their functional metabolites in CWC across chemotherapy. This study adopted a multi-omics network integration program xMWAS [[81]28] to analyze associations of microbiome–metabolome pathways with PNS. Material and methods Design and setting This study adopted a case–control study design. After informed consent (with child assent) was obtained from parents, children aged 7–18 years with solid tumors were recruited from the AFLAC Cancer and Blood Disorder Center at Children’s Healthcare of Atlanta in Atlanta, Georgia. Age-, sex-, race-, and body mass index (BMI)-matched HC were recruited via flyers, online e-news blast, and ResearchMatch™ [[82]29] in the Greater Atlanta Area. Approval was obtained from the Institutional Review Board at Emory University (IRB No. 00102775). Participants This study included two groups of children: CWC (n = 21) and HC (n = 14). Eligible CWC were: (1) 7–18 years old, (2) diagnosed with solid tumors (e.g., sarcomas, excluding brain tumors), (3) those who received at least one cycle of chemotherapy, (4) receiving treatment at the Aflac Cancer and Blood Disorder Center at Children’s Healthcare of Atlanta, and (5) agreed to participate in the study. Age-, sex-, race-, and BMI-matched HC were included if they were: (1) 7–18 years old, (2) not on antibiotics within the past 4 weeks, and (3) not involved in interventions (e.g., dietary program) that may influence the gut microbiome and metabolome. CWC with stem cell transplant, or relapses, or brain tumors, or whole abdominal radiotherapy within the past 4 weeks were excluded. For both CWC and HC, those with cognitive impairment (determined by treating physicians and neuropsychologists with objective cognition testing) or chronic diseases (e.g., inflammatory bowel diseases) that affect the gut microbiome and metabolome were excluded. Measures Gut microbiome Fecal specimens were collected to analyze the gut microbiome. Following the Human Microbiome Project protocol [[83]30], children were instructed to collect fecal samples using an at-home collection kit that has been tested in our project with > 80% compliance [[84]31]. Parents received instructions to assist their child to collect fecal samples. The provided spoon was used to transfer an aliquot of fecal sample into the collection tube (Fisher Scientific LLC., Pittsburgh, PA). Subsequently, the tubes were capped, placed into a biohazard bag, and then packed into a padded, labeled freezer bag with an ice pack. Samples were immediately placed into a freezer until shipped via FedEx. The FedEx shipment took approximately 24 h (range 16–24 h). Once received at the Emory Nursing Biobehavioral Laboratory, fecal samples were stored at − 80 °C until DNA extraction and assaying. Metabolites Metabolomic profiling of fecal metabolites in the gut was conducted following a well-validated protocol at Emory Lipidomics & Metabolomics Core [[85]32], which identifies localized metabolic processes in the large intestine, colon, and rectum [[86]33, [87]34]. An average of 100 mg (range from 95 to 105 mg) fecal specimen was aliquoted for each sample for untargeted metabolomics analysis. The untargeted metabolomics approach was utilized to acquire data for species, annotating metabolites, and reviewing both known and unknown metabolic changes. An advantage to untargeted data is its hypothesis-generating nature, which provides a foundation for further analysis using targeted approaches [[88]35]. PNS Children reported their PNS (e.g., pain, fatigue, anxiety, depressive symptoms, and cognitive dysfunction) using the Pediatric Patient Reported Outcomes Measurement Information System (PROMIS) [[89]36, [90]37]. All the PNS reported by PROMIS scales aligned with clinical anchors (i.e., low blood counts) in children with cancer [[91]38] and anchor-based methods using expert or patient judgment suggested a minimally important difference of 3 points on the PROMIS T-score scale for children [[92]39]. The various PROMIS scales utilized in this study were scored using a T-score with a reference mean of 50 (standard deviation [SD] = 10) by the Health Measures Scoring Center. Previous reliability testing of the PROMIS short form (PROMIS-SF) system in adolescents reported Cronbach’s α coefficients ranging between 0.88 and 0.96 for initial surveys and exceeding 0.91 for subsequent visits. Furthermore, Cronbach’s α coefficients for pooled PROMIS-SF data across all visits ranged from 0.91 to 0.97 [[93]40]. Pain The one-item PROMIS Pain Intensity Scale was used to evaluate the child’s pain within the previous 7 days, with scores ranging from “No pain at all” (0) to “Worst pain” (10). This scale has demonstrated great construct validity and good feasibility for use in children aged 7–18 years with cancer [[94]38]. The 8-item PROMIS Pain Interference Scale-SF was used to assess the influence of pain on the child’s social, cognitive, emotional, physical, and recreational activities over the past 7 days. A higher total score of pain interference indicates more pain impact on the child’s life. Fatigue The 10-item PROMIS Fatigue Scale-SF was used to assess the child’s fatigue within the previous 7 days, with scores ranging from “Not at all” (0) to “Always” (4). This scale has demonstrated good construct validity for use in children ages 7–18 years treated for cancer [[95]38]. Anxiety and depressive symptoms The 8-item PROMIS Anxiety Scale-SF was used to assess the child’s fear, anxiety, and somatic symptoms within the previous 7 days. The 8-item PROMIS Depressive Symptoms Scale-SF assessed the child’s depressive symptoms within the previous 7 days. Scores for each item on both scales ranged from “Never” (0) to “Almost always” (4). Both scales have demonstrated good construct validity for use in children aged 7–18 years treated for cancer [[96]38]. Cognitive function The 7-item PROMIS Cognitive Function Scale-SF assessed perceived difficulties in cognitive abilities. Scores were on a 5-point scale, with a higher total score indicating higher cognitive dysfunction. This scale has demonstrated excellent internal consistency and item-scale correlations in children [[97]41]. Demographic and clinical variables Children’s demographics (e.g., age, sex, race/ethnicity, and BMI percentile) and health history (e.g., use of antibiotics and disease history) were reported by their parents during the clinical visit. Cancer and treatment-related variables (e.g., type of cancer, cancer stage, and cycle of chemotherapy) were either reported by parents or extracted from the electronic medical records. Collection procedure All the data for CWC were collected pre-cycle 2 chemotherapy (T[0]) and post-chemotherapy (T[1], with an average 2 weeks post-chemotherapy [range 1–4 weeks]). Children confront various stressors from tumor diagnosis, treatment plans, and painful procedures during the first cycle of chemotherapy. To reduce psychological burden for the family, pre-cycle two chemotherapy period was selected for consent and data collection, with a mean of 3.7 weeks (range 2–8) from the first cycle chemotherapy in our participants. An average of 6 months between T[0] and T[1] were reported in this study. Only one timepoint of data was collected for HC. Children with solid tumors receiving chemotherapy were recruited during their routine outpatient clinic visits. Clinical collaborators from Children’s Healthcare of Atlanta identified eligible patients while a member from our research team consented parents (or children) and assented age-eligible patients. All PROMIS questionnaires were distributed for children to complete, and parents were provided pictorial instruction on at-home fecal specimen collection. The electronic medical records of CWC were used to collect demographic, clinical, and health-related variables. For HC, all procedures were identical, excluding the use of electronic medical records. DNA extraction Based on the Human Microbiome Project protocol, microbial DNA was extracted from fecal samples using the PowerSoil isolation kit (Mo Bio Laboratories, Carlsbad, CA, USA) at the Environmental Microbial Genomics Laboratory, Georgia Institute of Technology. 16S rRNA amplicon libraries were prepared for the 16S rRNA V4 gene region [[98]42]. These 16S rRNA amplicons were generated using KAPA HiFi HotStart ReadyMix (KAPA Biosystems, KK2600) and primers specific to 16S V4 region of bacteria and indices were attached using the Nextera XT Index kit (Illumina, FC-131-1001). Clean-up was performed on the indexed libraries using AMPure XP beads. The 16S libraries were pooled in equal amounts based on fluorescence quantification. Each run included a control template to test for polymerase chain reaction (PCR) accuracy and possible contamination. Final library pools were quantitated via qPCR (Kapa Biosystems, catalog KK4824). The pooled library was sequenced on the Illumina miSeq system using miSeq v3 600 cycle chemistry (Illumina, catalog MS-102-3003) at a loading density of 8 pM with 20% PhiX at PE300 reads. The microbial sequencing produced paired-end sequences. High-resolution untargeted metabolomics (HRM) An HRM protocol established at the Emory Lipidomics & Metabolomics Core was adopted for liquid chromatography–mass spectrometry (LC–MS) analysis. Metabolic features were extracted from fecal samples using a 1:1 mixture of Acetonitrile: Methanol. 200 µL 1:1 Acetonitrile: Methanol was added to 50 μL samples, which was vortexed for 3 s, incubated on ice for 30 min, and then centrifuged at 20,000×g for 10 min to pellet precipitated protein. The supernatant was then transferred to an amber autosampler vial for LC–MS analysis. For quality control, a pooled quality control sample was created by combining 5 µL of each sample extract into a separate vial. This sample was run in triplicate at the beginning, the end, and intermittently over the course of analysis. Next, an untargeted HRM approach was performed using an ID-X™ Tribrid™ mass spectrometer coupled to a Vanquish Ultra-High-Performance Liquid Chromatography (UHPLC, Thermo Fisher Scientific Inc., San Jose, CA). Metabolic features from the fecal extracts were resolved on a SeQuant ZIC-HILIC™ 3.5 μm, 100A 150 × 2.1 mm column. For chromatography, water was used as Solvent A and Acetonitrile as Solvent B, both of which contained 0.1% Formic Acid. 1 μL extract was injected into the LC–MS system for analysis. A full scan MS1 spectrum for each sample was obtained at resolution of 120,000 and mass-to-charge ratio (m/z) range 67–1000. The mass spectrometer was operated in both positive and negative ionization modes. Uniquely detected ions consisted of accurate mass m/z, retention time and ion abundance, referred to as m/z features. Data were processed using Thermo Compound Discoverer software, which scans our metabolic data against internal and external databases. Raw data was uploaded into the software with m/z values and retention times aligned. Signal intensities are normalized by the pooled quality control sample and corrected to compensate for any variation of signal for batch correction. Statistical analysis T-scores of the Pediatric PROMIS scales were calculated for PNS (excluding pain intensity). For their respective PROMIS questionnaires, T-scores ≥ 50 indicated significant fatigue, anxiety, and depressive symptoms, while a T-score ≤ 45 indicated significant cognitive dysfunction [[99]43]. Independent sample t-test was used to compare the PNS between CWC and HC; paired sample t-test was applied to compare the PNS between T[0] and T[1] for CWC. QIIME 2 default parameters were used to analyze the composition of the gut microbiome [[100]44]. 16S rRNA sequence quality was filtered with dada2 to infer ASVs. Using the Silva132 database with a 97% identity threshold, a Naive Bayes classifier was trained to assign our ASVs to taxonomy at the phylum and genus levels to integrate into analysis. Silva database was selected due to its checked quality and regular updates of aligned 16S subunit rRNA sequences for bacteria. Alpha diversity (within-sample diversity, i.e., Shannon, observed OTUs, Pielou_e, and Faith_PD) and beta diversity (between-sample diversity, i.e., Jaccard and unweighted UniFrac distance) parameters reported associations of the gut microbiome with PNS. Meanwhile, filtering of metabolic data was performed to remove m/z features with median coefficient of variation (CV) within technical replicates ≥ 75%. Only samples with Pearson correlation within technical replicates ≥ 0.7 were selected for downstream analysis. Metabolite intensities were log[2]-transformed, and quantile normalized. Metabolites associated with PNS were annotated by matching m/z and retention time to currently confirmed metabolites via standardized laboratory references or