Gut microbiome and anti-islet c - Diabetes

2 downloads 44 Views 4MB Size Report
Mar 7, 2014 - Atkinson4, Desmond Schatz4, Ezio Bonifacio5,6, Eric W. Triplett3**, Anette-G. ... 5Center for Regenerative Therapies, Dresden, and Paul ...
Diabetes

Compromised gut microbiota networks in children with anti-islet cell autoimmunity Running title: Gut microbiome and anti-islet cell autoimmunity David Endesfelder1,2*, Wolfgang zu Castell1*, Alexandria Ardissone3, Austin G. DavisRichardson3, Peter Achenbach2, Michael Hagen1, Maren Pflueger2, Kelsey A. Gano3, Jennie R. Fagen3, Jennifer C. Drew3, Christopher T. Brown3, Bryan Kolaczkowski3, Mark Atkinson4, Desmond Schatz4, Ezio Bonifacio5,6, Eric W. Triplett3**, Anette-G. Ziegler2**

1

2

Scientific Computing Research Unit, Helmholtz Zentrum München, Germany

Institute of Diabetes Research, Helmholtz Zentrum München, and Forschergruppe Diabetes, Klinikum rechts der Isar, Technische Universität München, Germany 3

Department of Microbiology and Cell Science, Institute of Food and Agricultural Sciences, University of Florida, Gainesville, FL, U.S.A. 4

5

Department of Pediatrics, University of Florida, Gainesville, FL, U.S.A.

Center for Regenerative Therapies, Dresden, and Paul Langerhans Institute Dresden, Technische Universität Dresden, Germany 6

Institute for Diabetes and Obesity, Helmholtz Zentrum München, Germany *

**

these authors contributed equally

these authors co-directed the project

Corresponding author contact information: Prof. Dr. med. Anette-G. Ziegler

Diabetes Publish Ahead of Print, published online March 7, 2014

Page 2 of 44

Page 3 of 44

Diabetes

Institute of Diabetes Research, Helmholtz Zentrum München, Ingolstaedter Landstr. 1, 85764 Neuherberg, Germany Phone: 0049-89-3187 2896, Fax: 0049-89-3187 3144 E-mail: [email protected]

Number of Figures: 4 Number of Tables: 0 Word count: 4000

Diabetes

Abstract The gut microbiome is suggested to play a role in the pathogenesis of autoimmune disorders such as type 1 diabetes. Evidence of anti-islet cell autoimmunity in type 1 diabetes appears in the first years of life, however little is known regarding establishment of the gut microbiome in early infancy. Here, we sought to determine whether differences were present in early composition of the gut microbiome in children who developed anti-islet cell autoimmunity. We investigated the microbiome of 298 stool samples prospectively taken up to age 3 years from 22 case children who developed anti-islet cell autoantibodies, and 22 matched control children who remained islet autoantibody negative in follow-up. The microbiome changed markedly during the first year of life, and was further affected by breast-feeding, food introduction, and birth delivery mode. No differences between anti-islet cell autoantibody positive and negative children were found in bacterial diversity, microbial composition, or single genus abundances. However, substantial alterations in microbial interaction networks were observed at age 0.5 and 2 years in the children who developed anti-islet cell autoantibodies. The findings underscore a role of the microbiome in the pathogenesis of antiislet cell autoimmunity and type 1 diabetes.

Page 4 of 44

Page 5 of 44

Diabetes

Introduction Type 1 diabetes is the result of a complex interplay of genetic susceptibility (1) and environmental determinants leading to anti-islet cell autoimmunity against pancreatic islet beta cells and autoimmune beta cell destruction (2). Anti-islet cell autoimmunity precedes the clinical onset of type 1 diabetes and often develops within the first years of life (3). This suggests that early shaping of the immune system in children is critical for the initiation of autoimmunity (3). There is increasing evidence that the immune response is shaped by factors that include how the host establishes a stable ecosystem with a large cohort of accompanying bacteria (4-7). With this, the role of microbiota in type 1 diabetes pathogenesis has become an important subject of investigation (8-12). The largest community of bacteria is established in the gastrointestinal tract (13, 14) where beneficial host-bacteria interactions have been demonstrated for food degradation or pathogen defense (14-16). Relatively few studies of the human gut microbiome have been performed in children less than 5 years old. These studies suggest that the phylogenetic composition of the bacterial communities evolves towards an adult-like configuration within the three-year period after birth (14, 17-19). Hence, it is conceivable that the evolution of the microbiome in infancy could influence the risk of anti-islet cell autoimmunity in susceptible children. Indeed, studies from Finland have provided evidence for this hypothesis (10, 20). The aim of our study was to investigate gut bacterial community structures during the early period from birth to the age of 3 years from the perspective of complex interaction networks. We estimated interaction on the basis of co-variation of bacterial abundances to compare children who developed anti-islet cell autoantibodies with children who did not develop such autoantibodies. We took advantage of the prospective BABYDIET study (21) where infants at increased risk of type 1 diabetes were monitored from birth for the development of anti-islet cell autoantibodies and type 1 diabetes. The gut microbiome composition was estimated based on measurements of

Diabetes

16S rRNA gene sequences from fecal samples that were obtained at 3 months intervals up to the age of 3 years. Analyses were focused on bacterial diversity, community composition, individual bacterial species and microbial interaction networks. Results show that complex bacterial interaction networks, rather than single genera, appear to be relevant to early preclinical type 1 diabetes.

Page 6 of 44

Page 7 of 44

Diabetes

Research Design and Methods BABYDIET study material Analysis of microbiota was performed on 298 stool samples from 44 children participating in the BABYDIET study (21). The BABYDIET study randomized 150 infants with a first degree relative with type 1 diabetes and with the type 1 diabetes risk HLA genotypes DR3/4DQ8 or DR4/4-DQ8 or DR3/3 to gluten exposure at 6 months or at 12 months of age. The intervention had no effect on anti-islet cell autoimmunity outcome. Blood and stool samples were collected at 3 month intervals from age 3 to 36 months and subsequently at 6 month intervals. Anti-islet cell autoantibodies (i.e., autoantibodies to insulin, GAD, insulinomaassociated antigen-2, and zinc transporter 8) were measured at each study visit. Written informed consent was obtained from the parents. The study was approved by the ethics committee of the Ludwig-Maximilian-University, Munich, Germany (Ethikkommission der Medizinischen Fakultät der Ludwig-Maximilians Universität No. 329/00). Stool samples chosen for the study included 147 samples from the 22 BABYDIET cohort children who developed persistent anti-islet cell autoantibodies at a median age of 1.54 years (IQR: 0.90 years and maximum 2.45 years), and 151 samples from 22 children who remained anti-islet cell autoantibody negative, and were matched for date of birth. Of the 22 children with persistent islet autoantibodies, 15 had developed persistent multiple islet autoantibodies, and 10 developed diabetes after a median follow-up of 5.3 years. For the 44 children, stool samples were taken from age 0.24 to 3.2 years with an average of 6.8 probes per child (Supplementary Table 1). Sample processing and deep sequencing Stool samples were collected at home and shipped by express courier overnight to the clinical study center where they were processed and immediately frozen at -80°C. DNA was extracted

Diabetes

from the stool samples as described previously (10). Bacterial 16S rRNA genes present within fecal samples were amplified using the primers 515F and 806R (22) modified with a sample specific barcode sequence and Illumina adapter sequences. PCR was performed at an initial denaturation temperature of 94oC for 3 min, followed by 20 cycles of 94oC for 45 sec, 50oC for 30 sec, and 65oC for 90 sec. A final elongation step at 65oC was run for 10 min.

PCR products were purified using the Qiagen™ PCR

purification kit following the manufacturer’s protocol. Illumina high throughput sequencing of 16S rRNA genes was conducted as described (23). Illumina sequencing was performed with 101 cycles each. Sequences were trimmed based on quality scores using a modified version of Trim2 (24) and the first 11 bases of each paired read were removed to eliminate degenerate bases derived from primer sequences. The prokaryotic database (25) used for 16S rRNA gene analysis was formatted using TaxCollector (26). Sequences were compared to the TaxCollector-modified RDP database using CLC Assembly Cell version 3.11 utilizing the paired reads and global alignment parameters. Two parameters were used in this step, a 98% length fraction and similarity values dependent on the desired taxonomic level, i.e., 80% at Domain/Phylum, 90% to Class/Order/Family, 95% to Genus levels (27). Pairs that matched different references at the species level were classified at the lowest common taxonomic level. Unresolved pairs were discarded. Henceforth, successfully paired reads are referred to as reads. Confounding variables Data on breastfeeding (yes, no), the duration of breastfeeding (weeks) and the introduction of solid food (gluten free and gluten-containing cereals, vegetables, fruits), were analysed from daily food records as previously described (21). Data on Caesarean section (yes, no) were obtained from obstetric records.

Page 8 of 44

Page 9 of 44

Diabetes

Statistical analysis Shannon evenness and Chao richness indices were estimated at genus level as described (28, 29). To correct bacterial diversity for the influence of confounding factors, stepwise multiple regression was performed with diversity as dependent variable. Age, breast-feeding at sampling time, introduction of solid food, first gluten exposure and delivery by Caesarean section were used as confounding variables. Akaike’s Information Criterion (AIC) (30) was used in the stepwise regression procedure to select confounding factors associated to diversity. To avoid bias due to violations of normality, rank regression (31) was used to estimate p-values of the regression coefficients corresponding to confounding factors associated with diversity. The R package fields (32) was used for cubic spline regression of age versus Shannon evenness. Chao richness was corrected for the influence of Caesarean section by using the residuals of a regression model with richness as dependent and Caesarean section as independent variable. Diversity analyses were performed on the entire age range and after grouping reads into three age classes of 0.5±0.25, 1.0±0.25 and 2.0±0.5 years. At most one single probe closest to 0.5, 1 and 2 years was used, respectively, for each child. For further analyses, phyla and genera with less than 0.01% abundance within the total number of reads were neglected. This reduced the number of genera from 452 to 75 and the number of phyla from 21 to 8. For the analysis of bacterial community compositions, BrayCurtis distances (33) were estimated on Hellinger transformed data (34). Differences in community compositions were tested with the non-parametric Multivariate Analysis of Variance (npMANOVA) (35) method. To visualize the results, Principal Coordinate Analysis (PCoA) was performed. Relative abundances of individual phyla and genera were compared by Wilcoxon-Mann-Whitney tests. To account for heterogeneity in variances, BrunnerMunzel tests (36) were used for bacteria where Bartlett’s test (37) showed evidence for unequal variances (P