Background
Myasthenia gravis (MG) is a typical autoimmune disease that targets the neuromuscular junction by autoantibodies and is clinically manifested as skeletal muscle fatigue [
1]. The prevalence of MG is 150–250 cases per million people [
2]. There are two types, ocular MG and generalized MG, depending on whether the myasthenic weakness is limited to the ocular muscles. Approximately 80% of generalized MG cases and only 50% of patients with ocular MG have detectable antibodies against the acetylcholine receptor (AChR) [
3,
4]. Pediatric MG accounts for about 10–15% of all MG cases in Europe and North America [
5], but is relatively common in China, where up to 50% of patients have disease onset under 15 years [
6]. There are many different characteristics between pediatric and adult MG, including symptoms, clinical severity, and antibody titer [
7]. Approximately 80% of pediatric MG is ocular MG in China [
6]. The diagnosis of pediatric MG can be challenging, because most children do not respond well to MG’s conventional diagnostic methods, such as fatigue test, repetitive nerve stimulation, and neostigmine test.
The gastrointestinal tract is a complex ecosystem containing many resident microorganisms [
8], which is a symbiotic relationship with the host, and affects human nutrition, metabolism, and immune function [
9‐
11]. The gut microbiome’s impact on human health is becoming increasingly apparent; an imbalance in its composition may contribute to various diseases, including obesity, inflammatory bowel disease, and immune diseases [
12]. Sequencing methods of microbiome initially focused on amplicon sequencing of the 16S ribosomal RNA (rRNA) gene. However, nowadays, it is recommended to perform a genome-wide association study (MWAS) based on the microbiome’s whole-genome shotgun sequencing. This method can detect the microbiome’s genome composition at the species level of bacteria, archaea, fungi, and even viruses and analyze its functional biological characteristics [
13]. Recent studies have reported that the gut microbiota in MG patients is altered compared to healthy people [
14,
15], and differential gut microbiota and fecal metabolites are related to the clinical subtypes of MG [
16]. However, the studies are based on 16S research focusing on adult patients. Therefore, we present our work for pediatric MG patients with whole-genome shotgun sequencing data. Key metabolites produced by gut microbiota are short-chain fatty acids (SCFAs) [
17], in which content in fecal is significantly lower in the adult MG group [
14]. We will explore the relationship between pediatric MG’s gut microbiome and the SCFAs.
In this study, we identified potential microbial marker species that could be used for MG identification by whole-genome shotgun sequencing analysis of fecal samples of pediatric MG patients and healthy controls (HC). Based on functional annotation, we found that the gut microbiome affects the production of short-chain fatty acids (SCFAs). We confirmed this result by determining SCFA levels in the serum of the participants. Our study elucidates the underlying roles and mechanisms of gut microbiota in pediatric MG pathogenesis. The novel microbiota-targeted markers will significantly help for the diagnosis of pediatric MG.
Methods
Inclusion criteria
We recruited a discovery cohort of 53 pediatric MG patients and 46 HC that were matched for age and gender; this study was conducted at the Center of Treatment of Myasthenia Gravis Hebei Province, People’s Hospital of Shijiazhuang. Clinical, grading of all patients was class I (any ocular muscle weakness; may have weakness of eye closure; all other muscle strength is normal.), based on the Myasthenia Gravis Foundation of America (MGFA) clinical classification [
18]. The diagnosis of pediatric MG was based on the clinical presentation of the patients and was confirmed with at least one diagnostic test, including positive antibodies (AChRAb), electromyography, or fatigue test and response to a therapeutic trial (Additional file
4: Table s-1). We further recruited 19 pediatric MG patients as the validation cohort based on the same criteria. Participants were excluded if they had one of these conditions: (1) age < 2 years and 10 months, or no age information; (2) antibiotic usage (except β-lactam) within 3 months of the study; (3) received any MG-related treatments (e.g., pyridostigmine, glucocorticoid, human immunoglobulin, or Chinese herbal medicine); (4) using of any drugs or unknown status; (5) with a family history or onset after birth; and (6) complicated with other diseases. The study was reviewed and approved by the ethics committee of the People’s Hospital of Shijiazhuang (NO. 66). Informed consent was obtained from all recruited patients’ parents and healthy individuals’ parents.
DNA extraction and sequencing
The DNA extraction and library construction are followed the instruction of TruSeq DNA Nano Reference Guide (1000000040135) (
https://sapac.support.illumina.com/downloads/truseq-dna-nano-reference-guide-1000000040135.html) and Hiseq 2500 System Guide (15035786) (
https://support.illumina.com/downloads/hiseq_2500_user_guide_15035786.html) from Illumina. Sterile plastic cupDNA concentration was assessed using a fluorometer reader (Qubit Fluorometer, Invitrogen). DNA sample integrity and purity were confirmed with agarose gel electrophoresis (concentration of agarose gel: 1%, voltage: 150 V, electrophoresis time: 40 min). One microgram genomic DNA from each sample was randomly fragmented by Covaris (LE220), and sequences of an average size of 200–400 bp were selected by magnetic beads (Ampure XP). The fragments were end repaired and 3′-adenylated, and adaptors were ligated to their ends. This process was performed to amplify fragments with the adaptors from the previous step. PCR products were purified using magnetic beads. The double-stranded PCR products were heat denatured and circularized by the splint oligo sequence. The single-strand circular DNA (ssCir DNA) was formatted as the final library. The library was amplified with the phi29 DNA polymerase to make DNA nanoballs (DNBs), which were loaded into the patterned nanoarray. Paired-end 100/150 bp reads were generated via the combinatorial Probe-Anchor Synthesis (cPAS). After library construction, the samples were sequenced on the Illumina HiSeq 2500 platform.
Quality control
We performed quality control of the sequencing data according to the following steps before analyses: (1) filtered low-quality reads and (2) removed contamination of the human genome sequence. We used fastp (Version: 0.21.1) [
19] with its default parameters to screen out low-quality reads and sequence adapter sequences, aligned the reads to the human genome (hg38) with bowtie2 (Version: 2.3.5) [
20], and used samtools (Version: 1.9) to screen out paired reads that did not align to the human genome. Clean data were used in subsequent analyses.
Taxonomy annotation and functional annotation
We used metaphlan2 (Version: 2.1.0) to map high-quality reads to the mpa_v20 marker gene database according to the metagenomics data analysis pipeline of The Huttenhower Lab; this enabled us to obtain microbial taxonomy abundance profiles at different taxonomic levels for each sample. We used the utils of metaphlan2 merge_metaphlan_tables.py to combine the results of all samples and utilized an in-house script to generate a combined profiling table for different taxonomic levels. Humann2 (Version: 2.0) was used to map high-quality reads to uniref90 and chocophlan (humann2 recommended databases) for profiling of gene family abundance and pathway abundance. We subsequently used the humann2 utils: humann2_join_table(s), humann2_renorm_table, and humann2_split_stratified_table to merge the abundance of all samples, normalize the abundance, and to stratify profiles with taxonomy annotations, respectively.
Statistical analysis
After obtaining the species and functional abundance profiles, we used the Wilcox.test two.sided function in R (Version: 3.5.3) to assess differences in bacterial abundance between the two groups. P values in the results were corrected according to the Benjamini–Hochberg (BH) method to obtain q values (false discovery rate, FDR), which were used for identifying significantly different species and pathways between the pediatric MG patients and HC.
For the calculation of the Shannon index, we used in-house scripts to calculate the α-diversity of each sample using the taxonomy abundance data. Using the same input data, the vegdist method of the Vegan package (Version: 2.5-5) in R (with the parameter “method = dist_method”) was used to calculate β-diversity.
To explore relationships among samples, we performed principal component analysis (PCA) and principal coordinate analysis (PCoA) with the taxonomy abundance data using the Ade4 package (Version: 1.7-15) of R. We use the corr.test of the R package with “method = spearman, use = pairwise, adjust = BH” to calculate the Spearman’s rank correlation between the clinical phenotypes and significantly different microbial species and functions.
Random forest model
Significantly different microorganisms between MG patients and HC at all taxonomy levels were assigned as candidate microbial markers. Using the relative abundance of the candidate markers at each taxonomy level, we used a random forest model to build classifiers to distinguish between HC and MG patients with RandomForest package (Version: 4.6-14) in R. The samples we used in this step were labeled with the two-value disease situation. We applied the package in classification with default parameters, like the number of trees grow in the process is 500. Subsequently, a fivefold cross-validation method was used to evaluate the performance of the predictive model. We calculated the minimum error in the average cross-validation error curve and the average curve and used the standard deviation of this point as the cut-off point of the filter prediction model. The set of candidate markers, which contained the minimum number of candidate markers in all groups and had an error below the critical value, was selected as a marker set to build the diagnostic classifier. The probability of MG was then calculated based on this optimal set; the receiver operating characteristics (ROC) of the discovery and validation cohort were plotted using the pROC package (Version: 1.16.2).
Quantification and analysis of SCFAs in serum samples
Approximately 2 ml of fresh peripheral blood of MG patients and HC individuals were collected in the early morning before meals. Samples were collected before patients were treated. These samples were centrifuged at 3000 rpm for 10 min after being left at room temperature for 30 min; the serum was collected and stored at − 80 °C until further analysis. The SCFAs quantification protocol was referenced from Sandin et al., Vemuri et al., and Sivaprakasam et al. [
21‐
23].
The standard SCFAs solution was prepared as following: 1 mg/ml SCFAs was diluted with ultrapool water into 1 ml solution with concentrations of 1 μg/ml, 2 μg/ml, 5 μg/ml, 10 μg/ml, 50 μg/ml, 100 μg/ml, and 200 μg/ml, respectively. Take ≥ 50 μL fluid, add 400 μL saturated sodium chloride solution and 50 μL saturated hydrochloric acid sodium chloride solution with 3 mmol, vibrate it to dissolve completely, and use low-temperature ultrasound for 20 min. Then, the standard curve was drawn with the concentration of standard SCFAs (μg/ml) as the horizontal coordinate and the peak area as the vertical coordinate.
To prepare the testing solution, 50 g samples were mixed with 400 μL saturated sodium chloride solution and 50 μL with 3 mmol saturated hydrochloric acid sodium chloride solution. The fully dissolved mixed solution was carried on ice bath ultrasound for 20 min. After the ultrasound, 400 μL ice ether was added and the mixed solution was shaken for 10 min to fully extract. The solution was then centrifuged for 10 min at 12000 rpm at 4 °C. Collect the supernatant and add 50 mg anhydrous sodium sulfate to oscillate for 3 min and centrifuge for 5 min at 4500 rpm at 4 °C. The supernatant was used in the followed GC-MS analyses.
The analytical instrument used in testings was Thermo Trace1300-Thermo TSQ9000 GC/MS. SIM mode was used for data acquisition, and Tracefinder (Thermo Fisher Scientific, Waltham, MA, USA) was used for data processing.
The column used in gaseous phase was Agilent HP-FFAP (25 m × 0.50 mm × 0.32 μm), the temperature was processed as follows: 100 °C for 1 min and 5 °C/min until 160 °C for 2 min. In mass spectrometry, the ion source temperature was 280 °C. The mass spectrometer was set to scan mode from m/z 35-200 and monitoring in SIM Ion m/z 43,45,60 for acetic acid, m/z 45,57,74 for propanoic acid, m/z 43,73,88 for butyric acid, m/z 41,60,73,88 for isobutyric acid, m/z 43,60,87 for N-Valeric acid, m/z 41,60,73,87 for isovaleric, and m/z 60,73,87 for caproic acid. The Pearson correlation between SCFAs and the abundance of species and the corresponding p value were calculated using the Hmisc package of R with default parameters.
Discussion
In this study, we conducted gut microbiome analysis of pediatric MG cases, using whole shotgun genome metagenomics sequencing. We collected samples from a Chinese cohort comprising pediatric MG patients and age-matched HC. We determined the microbial markers that were enriched or depleted in the MG group at different taxonomy levels. Based on this analysis, we identified and tested an MG classifier that can distinguish MG patients and HC based on the abundance of microbial markers. Subsequently, functional analyses implicated several microbiome pathways in the MG cohort. Thus, our study shows how the gut microbiome influences MG, and illustrates excellent microbial markers for diagnosing MG.
Our data indicated that MG patients had no significant differences in alpha diversity indices compared to the HC. This finding is notably inconsistent with a previous study on MG, which is based on 16S research focusing on adult patients [
15]. Two plausible explanations could help explain these apparent differences. One is the possibility of systematic error between the 16S rRNA gene sequencing data and metagenomic sequencing data; the second is that our cohort comprised pediatric MG cases, whereas the other examined cohorts in the previous study comprised adult MG cases.
We then conducted a Wilcoxon differential test to detect taxa with significantly different abundance between the MG and HC cohorts. Among the 14 significantly different genera, 10 were enriched in MG patients, and four were absent in the MG group. Based on the identified marker species F. mortiferum, P. stercorea, P. copri, M. funiformis, and M. hypermegale, we constructed an MG classifier. For cross-validation, the classifier obtained 0.94 AUC. For the independent validation cohort, the classifier achieved 0.84 AUC. This validation demonstrated that gut microbial features could be used to distinguish MG patients from HC.
Recent advances in sequencing and analysis of metagenomic data have facilitated discovering new viruses and improved our ability to catalog viral communities in an unbiased manner [
31,
32]. Notably, in addition to bacterial abundance, our data revealed differential viral abundance between the MG and HC cohorts. Specifically, we detected human adenovirus in the gut microbiome of 10 patients; no adenovirus was detected in the HC. The corresponding species-level viruses are human adenovirus D and human adenovirus F. Viral infection has been proposed as a possible important factor in the initial stage of MG. Cavalcante et al. (2010) proposed the link between viral infection (such as Epstein-Barr virus) and thymus pathology [
33]. Other viruses, such as cytomegalovirus, human foamy virus, and Nile virus, may also be related to MG [
34,
35]. Further research on the intestinal virus could be helpful for revealing the pathogenesis of pediatric MG.
The Metacyc [
36] annotation data showed no significantly different unstratified pathways in the comparison between MG and HC cohorts. We divided the MG cases into AChRAb-positive and AChRAb-negative groups according to AChRAb levels, which resulted in the detection of 15 and 16 significantly different unstratified pathways, respectively, with four and three pathways related to the production of SCFAs, respectively. A common aspect of these significantly different unstratified pathways was altered pyruvate and acetate metabolism. In the AChRAb-positive MG patients, acetyl-CoA was reduced, accompanied by an increased amount of coenzyme A. In the AChRAb-negative MG patients, the amount of pyruvate was reduced.
SCFAs are known to mediate various interactions between the gut microbiome and host metabolism [
37] .Both prebiotics and probiotics rely on SCFAs to act on the host [
38]. Three molecules that comprise more than 95% of the SCFAs in the human intestine are acetate, propionate, and butyrate [
39]. In Type 2 diabetes-related studies, the levels of butyrate and propionate in feces were related to the patient’s insulin response and risk of disease [
40]. It is also known that gut microbial dysbiosis of inflammatory bowel disease (IBD) patients and a significantly increased number of mucositis cells are related to reducing butyrate-producing microorganisms in the gut [
41]. Butyrate can directly affect immune cells in the intestinal mucosa, increase the number and activity of Tregs, inhibit the neutrophils, macrophages, and dendritic cells, and reduce effector T cell activity [
42]. To verify the impact of SCFAs, we tested the SCFA levels in the blood and found that butyric acid and isobutyric acid levels were significantly reduced in pediatric MG patients compared to those in the age-matched HCs.
There were some limitations that should be acknowledged. First, the clinical diagnostic potential of the microbial markers identified by us should be confirmed using multicenter independent samples; more MG patients enrolled from different regions might make our results more solid and reasonable. Second, a previous study noted that the levels of SCFAs are not directly equivalent in the blood and the intestinal environment [
43]. Intestinal SCFAs testing would provide more evidence for the roles of altered intestinal flora in the pathogenesis of MG. Third, further cell and animal experiments are needed to determine with certainty whether the relationship between the gut microbiome and MG is causal or incidental.
We detected several gut microorganisms whose abundance was significantly related to the level of serum SCFAs, and we should notice the correlation coefficients were at a low level, like around 0.2. In other words, the correlation does exist but not determinants. The difference between serum SCFAs and fecal SCFAs should be a cause to consider. The relationship between SCFAs in serum and fecal is still under research. Serum SCFA is not as intuitive as fecal SCFA, and SCFA metabolism from the gut harbored microorganism. Wolever et al. reported that acetate and propionate had demonstrated a 40% reduction in serum [
44]. However, studies in recent years have also shown that serum SCFAs are positively correlated with the level of fecal SCFAs [
45] and can reflect the trend of fecal SCFAs to a certain extent [
46]. Our results did not strongly link SCFA produced bacteria with serum SCFAs, but still confirmed the reduction of SCFAs in the gut microbiota of MG patients.
Conclusion
In summary, we identified a group of microbial markers (Fusobacterium mortiferum, Prevotella stercorea, Prevotella copri, Megamonas funiformis, and Megamonas hypermegale) associated with pediatric MG. Our study demonstrated the potential of pediatric MG diagnosis by our identified microbial markers. Among the microorganisms enriched in MG patients, we found adenovirus for the first time, which is worthy of further investigation. At the same time, the relationship between the SCFAs and MG needs in-depth understanding. Our findings provide a new approach for studying of the pathogenesis of pediatric MG and developing new diagnostic and therapeutic tools for MG.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.