Synopsis:
The most common symptom of cancer and its treatments is fatigue. In patients with chronic fatigue syndrome and other neuropsychiatric conditions, as well as cancer patients, changes in the intestinal microbiome have been discovered. The connection between the intestinal microbiome and fatigue in patients with advanced cancers, on the other hand, has not been studied. Understanding the connection between the microbiome of the intestine and fatigue will allow for interventional and therapeutic opportunities to manipulate the microbiome to enhance fatigue and other patient-reported outcomes. The aim of this study was to see if there were any links between microbiome composition and fatigue in advanced cancer patients. We included 88 patients with advanced, metastatic, unresectable cancers who were in a washout phase from chemotherapy in this cross-sectional retrospective analysis at a tertiary cancer care center. The MD Anderson Symptom Inventory"Immunotherapy fatigue score was used to assess fatigue, and 16srRNA was used to examine the intestinal microbiome. Eubacterium hallii was found to be negatively correlated with fatigue intensity scores (r = 0.30, p = 0.005), while Cosenzaea was found to be positively associated with fatigue scores (r = 0.33, p = 0.0002). We discovered that the composition of microbial species varies between high-fatigued and low-fatigued cancer patients. More research is needed to see whether modifying the microbiome decreases the magnitude of cancer patients’ fatigue and enhances their quality of life.
The first paragraph is an introduction.
Fatigue is one of the most common cancer signs and treatments. Fatigue has been linked to a worsening of cancer outcomes1,2,3,4,5,6. The contact between microbial communities and the host (either directly via microbial metabolites or indirectly via the immune system) provides real-time information about the environment to the central nervous system7 and links the brain’s emotional and cognitive centers with peripheral intestinal functions8.
The intestinal microbiota of cancer patients differs from that of non-cancer patients, according to several studies. Furthermore, prior to radiation therapy, gut microbial dysbiosis can predict fatigue in cancer patients9, and chemoradiotherapy may alter the intestinal microbiome composition of colorectal cancer patients, contributing to fatigue10.
Cancer and its therapies induce fatigue, which is the most common and pervasive symptom. Despite commendable attempts to identify cancer-related fatigue, the exact cause of fatigue in cancer patients remains unclear, and treatment is only moderately effective1,4,11,12,13,14. Given recent studies indicating that fatigue decreases the quality of life and increases mortality in these patients2,5,14, understanding the mechanism(s) of cancer-related fatigue is critical for improving patient experience and outcomes.
As a result, we performed a retrospective study of fatigue in advanced cancer patients, which led to the discovery of associations between microbiota parameters and cancer patient fatigue severity. Identifying features of the intestinal microbiome linked to fatigue in cancer patients may help us better understand the mechanisms behind fatigue and contribute to the creation of therapeutic approaches that target the microbiome to avoid or minimize fatigue in these patients.
Methodologies:
Patients – Between February 28, 2017 and January 4, 2018, patients were recruited from MD Anderson’s Department of Investigational Cancer Therapeutics. Patients had to be at least 18 years old, be able to communicate in English, and have a pathological diagnosis of advanced, metastatic, unresectable cancer. Until enrolling, patients had to be off antibiotics for at least 30 days. Patients were omitted if clinical research workers felt they did not understand the study’s purpose or were unable to complete the symptom evaluation questionnaire. Patients’ clinical and demographic data, including cancer diagnosis, previous use of antibiotics, probiotics, drugs (including antidepressants), neuropsychiatric conditions, gastrointestinal illness, and prior anti-cancer care, were collected by examining their electronic medical records. Until enrolling, the patients had to complete a protocol-required washout cycle, which is usually 4 weeks or 5 half-lives of the previous therapy.
Measuring fatigue:
The MD Anderson Symptom Inventory (MDASI)-immunotherapy module (MDASI-Immunotherapy) was used to evaluate fatigue. The MDASI-Immunotherapy has been validated for the assessment of 20 symptoms, including seven immunotherapy-specific items and six interference items15. Before beginning immunotherapy, patients completed the module and graded their symptoms and interference on a scale of 0"“10 (0 = no symptom, 10 = worst symptom imaginable, total interference). We classified patients as having low fatigue (scores of 0"“4) or high fatigue (scores of 5"“10) based on their responses to the fatigue object. It has been proven that the MDASI-Immunotherapy for early phase trials (EPT) is valid and reliable15. Furthermore, it has been demonstrated that fatigue and its consequences are recorded along a single axis and can be represented as a single number16.
Collecting feces and analyzing the microbiome:
Stool samples were obtained using the OMNIgene-GUT kit (DNA Genotek, Inc.) and stored at 80 °C at the start of the study. The yield of bacterial DNA was maximized while background amplification was kept to a minimum using genomic bacterial DNA extraction methods. As previously described17, DNA extraction and bacterial 16S rRNA sequencing were carried out. The QIAamp DNA FFPE Tissue Kit was used to collect bacterial genomic DNA (Qiagen). Polymerase chain reaction was used to amplify the 16S rDNA V4 region, which was then sequenced on the MiSeq platform (Illumina). Each sample yielded between 4350 and 30,843 sequences (average 12,517). When calculating the Alpha Diversity, we rarefied the OTU counts across samples with a minimum of 4000 sequences because this metric is sensitive to variations in sequencing depth. To avoid losing essential details in the complete microbiome data, we used the full unrarefied data for all other downstream studies. We calculated OTU relative abundances by scaling OTU counts by their total counts in each sample to account for variations in read depth.
The microbiome analysis was carried out using the same pipeline as defined by Wang, Y et al.18. Nucleotide sequences were analyzed using VSEARCH19. Merged paired-end reads is de-replicated and sorted by length and height. To produce a preliminary list of OTUs, the sequences were quality-controlled, error-corrected, and chimera-filtered using the UNOISE algorithm. The Mothur method21 and the Silva database version 12822 were used to assign taxonomy to both OTUs and presumed chimeras in QIIME20. Furthermore, sequences that were rejected by the UNOISE algorithm but matched a database entry with a perfect score were restored to construct the final list of OTUs. We use UNOISE3 to correct PCR errors such as mutations and chimeras using a computational approach. However, UNOISE3 is not always reliable, particularly when it comes to chimera calling. To compensate for this, our pipeline contains a step in which rejected sequences are mapped to the curated 16S database Silva, and those rare rejected sequences that are found to have a 100 percent identical match in the dataset are no longer labeled as noise and are re-added to the other sequences.
Statistical analysis: We compared the demographic and clinical features of low- and high-fatigue patients using statistical measures. A chi-square test or a proportion test is used to assess categorical variables. To compare the distributions of continuous variables between the low- and high-fatigue classes, the Mann"“Whitney U test was used. Spearman correlation analyses were used to link continuous variables26. Many of the experiments were performed in pairs. Statistical significance was described as a P-value of less than 0.05.
Non-parametric tests were used to compare alpha and beta diversity between classes, and PCoA was used to visualize the results. Unlike principal component analysis, which uses Euclidean distances, PCoA allows users to choose from a variety of distance metrics, including microbiome-specific distances like the weighted Unifrac metric. The number of species found in the study, the Shannon index, and the Inverse Simpson index were used as alpha-diversity metrics. Differences in the richness and evenness of microbial composition between groups of interest are calculated using differential analysis of alpha diversity. The structural change in microbial compositions from one sample to the next is measured by beta diversity. To search for differences in beta diversities between groups of interest, we used the permutational analysis of variance (PERMANOVA).
Patients were graded as having low fatigue (MDASI-Immunotherapy fatigue scores of 0"“4) or high fatigue (MDASI-Immunotherapy fatigue scores of 5"“10). We need to robustly distinguish features that were substantially different between these two groups19,27 since microbial data is high-dimensional and heterogeneous. As a result, we devised a method for visualizing and comparing microbial data between the low- and high-fatigue classes. In each case, the “progressive permutation” approach permutes the microbiome’s grouping factor labels and conducts several Mann"“Whitney U tests. When the signal strengths of the top hits from the original data are compared to their output in the permuted data, a declining pattern emerges if the top hits are true positives found from the data. We were able to identify individual features that were significantly different between the two groups using this tool, in addition to evaluating the overall associations between microbiome features and fatigue severity.
The fragility index is a metric that assesses the consistency of a clinical trial’s findings 28,29. In our progressive permutation study, we used a similar concept to assess the robustness of each important taxon. A variable’s fragility index is the smallest number of permutation steps needed to shift the variable’s significance to non-significance. The higher the fragility index, the more stable the described taxa were in this analysis. As a result, we rated the taxa’s value based on their fragility indices. R version 3.6.1 was used to conduct all statistical analyses, including alpha and beta diversity analyses and Mann-Whitney U studies.
Consent to participate and ethics approval:
The University of Texas MD Anderson Cancer Center’s Institutional Review Board approved the procedure. The research followed the Declaration of Helsinki and the recommendations set out by the International Conference on Harmonization for Good Clinical Practice. Before being included in the study, all participants signed a written informed consent form.
The following are the outcomes:
Table 1 lists the characteristics of the 88 patients (45 women and 43 men) who participated in the study. The median age of the patients was 58.5 years. The majority of the patients (n = 73) were white. Colon cancer, prostate cancer, cervical cancer, and non-small cell lung cancer were the most common diagnoses.
The OTU table, clustered phylogenetic tree, and taxonomic assignment were then loaded into R 3.6.1 for further quantitative analysis. Individual OTU counts were normalized by the total number of OTUs in each sample, resulting in a scaled OTU abundance vector of one. Phyloseq24 and Vegan25 R packages were used to calculate Alpha diversity scores and UniFrac distance23 between samples. All other tests, including theory coordinate analysis (PCoA), research methods, and visualization performance, were done in R.
Scores on MDASI-immunotherapy fatigue
58 (66%) of the 88 patients were classified as having low fatigue (9 had no fatigue and 49 had moderate fatigue), while 30 (34%) were classified as having high fatigue (18 had moderate fatigue and 12 had severe fatigue). The MDASI-Immunotherapy fatigue score was 3 on average. The MDASI-Immunotherapy fatigue score was 3.6 on a scale of one to ten (SD = 2.3). Patients in the high fatigue category had a median MDASI-Immunotherapy fatigue score of 6, while those in the low fatigue group had a median MDASI-Immunotherapy fatigue score of 2.5.
Comparison of alpha and beta diversity:
There were no major variations in microbiome diversity between low- and high-fatigue patients, according to the alpha diversity review (observed organisms, p = 0.2; Shannon index, p = 0.57; Inverse Simpson index, p = 0.78; Fig. 1). There were no major variations in microbial composition between low- and high-fatigue patients, according to the beta diversity review. Since PERMANOVA provided non-significant results for both the weighted-UniFrac (r2 = 0.01, p = 0.45) and the Bray"“Curtis (r2 = 0.01, p = 0.40, Fig. 2b) metrics. We added the PCoA plot of individual distances to the centroid and the boxplot of inter-individual distances (Fig. 2d) to compare inter-individual differences between classes. The results of the Bray"“Curtis metric are shown in Figures c and d. The Tukey test yielded a non-significant outcome (p = 0.31). In the supplementary material, we have provided two types of plots for the weighted-UniFrac metric (Figure S3). In addition, the Tukey test yielded a non-significant result (p = 0.51).
Microbial taxonomic differences in relative abundance:
Microbial characteristics were grouped at the phylum, class, order, family, genus, and species taxonomic stages. Figure S2 shows the taxonomic compositions of low- and high-fatigue classes at the phylum-class-order-family-genus level (in the supplementary). 15 of these characteristics were found to be substantially different between the low- and high-fatigue groups (p 0.05; Figure S4) using Mann"“Whitney U measures. Based on the progressive permutation study, a subgroup of 12 features (labeled in Figure S4) were classified as robust findings among the 24 features. Figure S5 in the supplementary includes a trending curve of P-Values, fragility indices, and effect sizes.
We used the DESeq method30 to check the results, and the Benjamini"“Hochberg procedure to change the p-values31. In the volcano map, we showed the modified p-values (Fig. 3). DESeq identified eight features that match those identified by our Wilcoxon process. We also used LEfSe and used a barplot to display the effects (Figure S6 in the supplementary). LEfSe determines hits using both p-values and LDA scores; however, the p-values are not multiplicity adjusted. LEfSe chose 21 features in all, using a default threshold of 2 on the LDA scores32. LEfSe found eleven features that match those identified by our system. We have used the labdsv package in R to perform indicator species analysis. As shown in the supplementary material, we plotted the selected variables and their log10 p-values (Figure S6 in the supplementary). We discovered the same five features that our method described.
The most significant differences between the high- and low-fatigue classes were found in Eubacterium hallii and Cosenzaea. All of the above research techniques have detected both of them. The high-fatigue group had a significantly lower abundance of Eubacterium hallii (p = 0.005; Fig. 4a) but a significantly higher abundance of Cosenzaea (p = 0.0039) than the low-fatigue group. Figures S4A and S4B in the supplementary display the abundances of Eubacterium hallii and Cosenzaea in each patient sample, grouped by fatigue severity. Eubacterium hallii was found to be negatively correlated with fatigue severity score (r = 0.31, p = 0.0026), while Cosenzaea was found to be positively associated with fatigue severity score (r = 0.26, p = 0.014). The p-values in the correlation analysis are not changed to correct for multiple testing since the correlation analysis is primarily used for visualization and exploration rather than inference.
We used the Random Forest method to test the accuracy of the classification using the 12 features we picked. We divided the data into 70 percent training and 30 percent research groups at random. The training and testing process was replicated 50 times. Figure S9 in the supplementary contains a plot of the value of these selected features. The taxonomic function "Species. Clostridium dakarense" is shown to be the most significant. To infer the network structures, we used SpiecEasi (SParse InversE Covariance Estimation for Ecological Association Inference). SpiecEasi accurately estimates a sparse inverse covariance that can model the conditional independence between microbial features33, in contrast to CCREPE and SparCC, which are based on an estimation of unconditional correlation. Figure S8 in the supplementary material shows the network plot at the genus level (178 features). The low fatigue group network has 285 edges, while the extreme fatigue group network has 345 edges. This suggests that in the high-fatigue community, there are more co-occurring genera.
To demonstrate the associations between microbial abundances and clinical variables, we used Spearman correlation analysis. The correlation plot displayed the taxa that had been resolved to the genus and species stages. The findings showed that microbiome characteristics were not highly associated with clinical factors that could be linked to fatigue severity, such as hemoglobin, albumin, creatinine, Eastern Cooperative Oncology Community success status ranking, number of lines of treatment, metastatic status, and age. Scatterplots of the associations between Eubacterium hallii and fatigue severity score, as well as Cosenzaea and fatigue severity score, have been provided.
Discussion Points:
We discovered a potential correlation between microbiome composition and fatigue in patients with advanced cancers in this research. We found that Eubacterium hallii was negatively associated with fatigue scores, while Cosenzaea was positively associated with fatigue scores, using the MD Anderson Symptom Inventory to score fatigue and 16S rRNA sequencing to classify the intestinal microbiota of 88 patients with advanced, metastatic, unresectable cancer.
Eubacterium hallii, a key species in trophic interactions, may have a significant effect on metabolic equilibrium, influencing the gut microbiota, host homeostasis, and host health34. Eubacterium hallii has also been used to treat insulin resistance-related diseases such as dyslipidemia, type 1 diabetes, Cushing syndrome, and other endocrine diseases35. These results indicate that Eubacterium hallii plays a part in intestinal metabolism and immune homeostasis, which may have consequences for the gut-brain axis and fatigue.
Cosenzaea myxofaciens (previously known as Proteus myxofaciens) belongs to the family Enterobacteriaceae, which includes the genera Proteus, Providencia, and Morganella36. There is no information on the role of Cosenzaea in cancer. Proteus bacteria, which live in the atmosphere and in humans’ intestines, can cause urinary tract infections, wound infections, and meningitis in the right circumstances (in neonates and infants). The presence of Cosenzaea myxofaciens in high-fatigue cancer patients suggests that the bacteria can play a role in inducing inflammation.
Constraints:
This research had a small sample size, which may have prevented the discovery of substantial variations in microbiome diversity between high- and low-fatigue patients, as well as between cancer types. Furthermore, six patients were on antibiotics within 1"“3 months of the stool sample collection, which may have influenced the composition of their intestinal microbiome. Recent research has found no connection between antibiotic use in the previous 30 days and alpha diversity or the dysbiosis index37. Our cohort’s mean fatigue score was lower than that of cancer patients who received no treatment (4.52, SD = 3.3), patients who received chemotherapy (5.2, SD = 2.81), and patients who received bone marrow transplants (5.29, SD = 2.55)38. This could be clarified by the good performance status our patients were expected to have in order to meet the eligibility requirements for clinical trial participation, which could result in any selection bias. Also among patients with good performance status, we were able to find a substantial difference in fatigue scores (some with extreme fatigue and others with moderate fatigue). Finally, we did not have safe controls in this study; however, in future longitudinal studies, we expect to obtain healthy household samples.
Final Thoughts:
Our research found that cancer patients with varying degrees of fatigue have different gut microbiome compositions. Given the importance of the microbiome in mucosal immunity and the growing recognition of the connection between the gut-brain axis and fatigue and other symptoms, our findings highlight the need to investigate the microbiome in cancer patients with fatigue using longitudinal samples, and to see whether modulating the microbiome reduces fatigue severity and improves the quality of life.
File accessibility:
The datasets used and/or analyzed in this analysis are available from the corresponding author upon appropriate request, in compliance with current guidelines.
Author information
Author notes
These authors contributed equally: Joud Hajjar, Tito Mendoza, and Liangliang Zhang.
These authors jointly supervised this work: Robert Jenq and Aung Naing.
Affiliations
Department of Pediatrics, Baylor College of Medicine, Houston, TX, USA
Joud Hajjar
The William T. Shearer Center for Human Immunobiology, Texas Children"™s Hospital, Houston, TX, USA
Joud Hajjar
Department of Symptom Research, The University of Texas MD Anderson Cancer Center, Houston, TX, USA
Tito Mendoza
Department of Biostatistics, The University of Texas MD Anderson Cancer Center, Houston, TX, USA
Liangliang Zhang & Christine B. Peterson
Department of Investigational Cancer Therapeutics, Unit 0455, The University of Texas MD Anderson Cancer Center, 1515 Holcombe Blvd, Houston, TX, 77030, USA
Siqing Fu, Sarina A. Piha-Paul, David S. Hong, Filip Janku, Daniel D. Karp, Alexej Ballhausen, Jing Gong, Abdulrazzak Zarifa, Funda Meric-Bernstam & Aung Naing
Department of Genomic Medicine, The University of Texas MD Anderson Cancer Center, Houston, TX, USA
Robert Jenq
Contributions
J.H. contributed to concept and design, data analysis and interpretation, drafting the manuscript, critical revision of the manuscript, and final approval; T.M. contributed to the data acquisition, data analysis, and interpretation, drafting the manuscript, critical revision of the manuscript, and final approval; L.Z. contributed to data analysis and interpretation, drafting the manuscript, critical revision of the manuscript, and statistical analysis; S.F. contributed to data acquisition, critical review, and final approval of the manuscript; S.A.P.-P. contributed to data acquisition, critical review, and final approval of the manuscript; D.S.H. contributed to data analysis and interpretation, critical revision of the manuscript, technical, administrative, and material support, and final approval; F.J. contributed to data acquisition, critical review, and final approval of the manuscript; D.D.K. contributed to data acquisition, critical review and final approval of the manuscript; A.B. contributed to the data collection and final approval; J.G. contributed to the data and acquisition and technical, administrative, and material support; A.Z. contributed to the data acquisition and data analysis and interpretation; C.P. contributed to critical revision of the manuscript, statistical analysis, and final approval; F.M.-B. contributed to study design, data interpretation, critical review, and final approval of the manuscript; R.J. contributed to data analysis and interpretation, critical revision of the manuscript, and final approval; A.N. contributed to concept and design, data acquisition, data analysis, and interpretation, drafting the manuscript, critical revision of the manuscript, technical, administrative, and material support and final approval.
Corresponding author
Correspondence to Aung Naing.