Community Research and Development Information Service - CORDIS

Final Report Summary - INDIGO (Investigation of Novel biomarkers and Definition of the role of the microbiome In Graves’ Orbitopathy (INDIGO).)

INDIGO, a partnership of 3 universities; Cardiff (UK) Essen (Germany) and the University of Milan (Italy) and 2 SMEs; Cultech (UK) and PTP (Italy), investigated the role of the gut microbiome in two related autoimmune conditions; Graves’ disease (GD) and Graves’ orbitopathy (GO). GD hyper-thyroidism is caused by thyroid stimulating antibodies (TSAB) to the thyrotropin receptor (TSHR). Some GD patients develop GO, in which orbital tissue remodelling produces inflammation and protruding eyes leading to double-vision and even blindness. Current GO treatments are inadequate and patients have reduced quality of life. GD and GO are chronic and cost Europeans €1.4-2.8 billion/annum in health costs & lost earnings etc. The project applied state-of-the-art ‘omics platforms and bioinformatics to address 4 questions 1. Are gut micro-organisms (the microbiota) associated with GD and progression to GO? 2. Can probiotic modulation of GD microbiota improve outcomes, especially GO progression? 3. Can gut microbiota manipulation improve in vivo models of GD/GO? 4. Could bio-markers predict GD patients most at risk of GO progression?
Findings: 1. In humans, analysis of the microbiota composition by 16S rRNA sequencing at recruitment in GD (n=65), GO (n=56) and healthy controls (n=42) revealed no significant differences in α and β diversity. At the phylum level, Bacteroidetes were significantly more abundant in controls (38.5%) than in GD (24.2%) and GO (27.3%) patients, Firmicutes were more abundant in GD (59%) and GO patients (60.5%) than controls (53.2%) and the firmicutes:bacteroidetes (F:B) ratio was significantly higher in GD than controls but similar in GD and GO. In 2 GD patients who developed GO there was a decrease in the genus Bacteroides (BH adjusted p<0.0001), confirmed using traditional microbiology. Enteroccoccus gallinarum counts, a pathobiont reportedly involved in triggering autoimmunity, though low were significantly higher in GD and GO than controls. 2. Twenty-nine GD patients were randomized to receive LAB4 (n=14, 10 hyperthyroid and 4 euthyroid) or placebo (n=15, 14 hyperthyroid, 1 euthyroid) in addition to standard GD therapy. Patients on probiotic were significantly less likely to have hyperthyroid relapse and had lower levels of circulating IgG and IgA hinting at a systemic effect. Microbiota composition was more stable in probiotic receiving patients who also displayed a significant reduction in counts of the Firmicutes phylum compared to placebo (P=0.033). 3. A TSHR-induced animal model of GD/GO was reproduced in Essen but displayed differences in disease phenotype when compared with the original location. Analysis of the gut microbiota in the 2 centres revealed significant differences in α and β diversity and we identified disease-associated taxonomies, which support a role for the gut microbiota in shaping the autoimmune response. Subsequent experiments employed antibiotic, probiotic or human GO faecal material transfer (hFMT) to modify the gut microbiota in TSHR immunized mice. Antibiotic more significantly modified the gut microbiota than probiotic or hFMT, probably due to the earlier and continuous nature of its administration. hFMT induced a human GO-like gut microbiota. Probiotic had limited effects but increased Treg and decreased T effector cell numbers. However the absence of pathological TSAB and orbital pathology, combined with reduced Treg in vancomycin treated mice confirm a role for the gut microbiota in promoting autoimmune GD and GO. 4. We analysed miRNA and proteins in blood from 14 GD, 19 GO and 13 healthy controls using high-throughput proteomics and miRNA sequencing to identify potential biomarkers. We detected 3025 miRNAs and 1886 proteins and Multi-Dimensional Scaling revealed good separation of the 3 groups. Biomarkers were identified by combined DE and lasso-penalized predictive models; accuracy of predictions was 0:86 and found 5 miRNA and 20 proteins including Zonulin and Fibronectin. Functional analysis identified relevant pathways, e.g. bacterial invasion of epithelium and mRNA surveillance. Proteomic and miRNA analyses, combined with robust bioinformatics, identified circulating biomarkers applicable to diagnose GD and predict GO disease status.


Graves' orbitopathy (GO) is a clinically burdensome autoimmune condition; in autoimmunity the body’s defence system starts to attack itself, this is described as loss of tolerance. GO is a complication of Graves’ disease (GD) which is caused by thyroid stimulating autoantibodies (TSAB), producing an over-active thyroid (hyperthyroidism) and swelling in the neck (goitre). All GD patients have some eye involvement; half develop eye changes that reduce quality of life and 5-10% have severe disease classified as GO. Extensive orbital tissue remodelling produces protruding eyes (proptosis) which underpins all GO signs and symptoms, including compression of the optic nerve which may lead to blindness. The project aimed to reduce the number of GD patients who develop severe GO, by addressing 4 linked and complementary objectives:
• Are gut micro-organisms (the microbiota) associated with GD and progression of GD to GO
• Can probiotic manipulation of GD microbiota improve outcomes, especially GO progression
• Can gut microbiota manipulation improve in vivo models of GD/GO
• Identify bio-markers to predict patients at risk of progression from GD to GO

Mothers transmit their genes and gut microbiota to their children. The microbial populations in the gastro-intestinal tract may affect metabolism and health but progress in exploring mechanisms is hindered by the inability to culture them. Technological advances make it possible to identify them by sequencing one of their genes (16S rRNA), which is highly variable from one species of micro-organism to another, rapidly and cheaply. The sequencing generates data called a microbiome. INDIGO tested the hypothesis that in GD patients micro-organisms inducing tolerance are under-represented or those promoting inflammation are over-represented.
• INDIGO applied 16S rRNA gene sequencing to characterise the microbiomes of healthy individuals vs GD and GO patients, in order to assess whether this information can be used to develop diagnostics for individuals at risk of GD and then of developing GO.

Probiotics are live microorganisms which may have a beneficial effect on health by modifying the host microbiota. Probiotics may improve immune function; an example is the probiotic mixture LAB4 (Cultech Ltd, UK), which has been reported to be safe for human use and improves allergy in infants.
• INDIGO conducted a double-blind, placebo controlled, randomised clinical trial of the effects of LAB4 on the gut microbiota composition in GD patients and the persistence of any changes observed.

Animal models are essential for identifying and testing novel treatment targets. Reports of GO induction by immunizing with the autoantigen targeted by TSAB autoantibodies, the thyrotropin receptor (TSHR), were not reproducible in other academic centres. The laboratories in which the studies were performed were not ‘specific pathogen free’, leading us to hypothesize that the microbial environment may play a role, as occurs in other models of autoimmunity.
• INDIGO investigated whether modification of the gut microbiota in a GD mouse model could lead to an improved and more reproducible GO model, useful both for GO and as a prototype for in vivo microbiome studies of other autoimmune disorders.

The TSHR is implicated in GO since most have TSHR autoantibodies, however the mechanisms breaking tolerance are not known but genetic and environmental factors contribute. In other autoimmune diseases components of common foods or microbial antigens trigger autoimmunity. Interactions between environmental antigens and pattern recognition receptors used by some immune cells can stimulate immune responses. Depending on the pattern recognition receptor involved, soluble factors called cytokines are released which direct development of other immune cells and can influence the balance between Regulatory T cells (Treg protect) and Th17 +cells which are associated with autoimmunity.
• INDIGO compared antibody responses in GD patients and controls to identify whether microbial or food derived antigens are involved in triggering disease or associated with GO progression.

Antibodies to the TSHR are currently the only biomarker able to predict GO onset, which occurs mainly in GD patients who respond poorly to treatment.
• INDIGO used high through-put analysis to identify proteins and small nucleic acids (miRNA, similar to DNA but much smaller and probably has a regulatory rather than a coding role) in serum to search for additional biomarkers that can be used to predict eye disease progression and hence inform the early selection of the most appropriate treatments.

The gut microbiota affects the immune system, as described above. At any one time more than half of our lymphocytes are in the gut associated lymphoid tissue or GALT. Microbial products interact with pattern recognition receptors called Toll-like receptors and influence the cytokines released, which can influence T cell development. Probiotics have been shown to enhance production of anti-inflammatory cytokines using in vitro models.
• INDIGO developed in vitro culture systems using gut epithelium and immune cells, to assess the effects of micro-organisms from patients and probiotics on cytokine release by GALT.

INDIGO was a partnership of 3 universities; Cardiff (CU, UK) Essen (UK Essen, Germany) and the University of Milan (UMIL, Italy) and 2 small-medium enterprises; Cultech (Italy) and PTP (UK).

The project comprised three phases: 1) the collection or creation of biological materials and data using innovative approaches; 2) the biotechnological analysis of the materials using state of the art technologies; and 3) the interpretation of the data produced. Management and dissemination of the results, which may be implemented in the context of patient care or in further research, spanned all three phases. The project was constructed in 9 Work Packages (WPs): 4 WPs addressed Phase 1 (WP 2, 3 and 4 with WP5 linking these WPs), 3 WPs addressed Phase 2 (WP 6. 7 and 8), while Phase 3 was addressed by a single WP (WP 9). Management was carried out in WP1 which also addressed extended training needs within the consortium and dissemination of the project activities and the results obtained to the wider community of stakeholders, potential users and the public.

WP1 Project Management, Training and Dissemination (Leader PTP SME)

Prof Ludgate CU, coordinated the project assisted by PTP for administrative management. The management structure comprised The Project Board (PB), The Project Executive Committee (PEC) and The Ethical Review Committee (ERC). Dissemination was achieved by creation of a web-site summarising the project in layman’s terms to ensure visibility of INDIGO to 3rd parties. Results have been/will be presented at scientific conferences, published in scientific journals and in the popular press (layman’s terms). INDIGO was publicised at Science Fairs and in meetings of patient support groups. Annual face-to-face meetings were timed/located to coincide with EUGOGO executive meetings. A 2 day closing conference presented project outputs in the context of latest research and clinical practice.

WP2 In vitro models (Leader CULTECH SME)

WP2 optimised and extended in vitro models based on cell lines for human gut epithelium and monocytes. The models compared how the gut microbiota from GD patients and healthy controls (WP3) affects cytokines produced within the systems (WP6) and whether LAB4 (WP5), can modulate the quality and quantity of this response.

WP3 Patient Studies (Leader UMIL University)

Samples of serum, nasal swabs, tears and faeces from controls, GD & GO patients were collected to identify biomarkers (WP6) and the role of the microbiome in human (WPs 2, 5 & 8) and murine (WP4) disease. Lifestyle data sought links between dietary or nutritional factors with immune responses (WP7) and a trial evaluated whether probiotics can modify the GD microbiome.
The recruited trial manager liaised with INDIGO partners and EUGOGO centres to help obtain ethical approvals. Biological samples from controls, GD and GO patients were collected according to standardised procedures formulated by the manager. A patient database was established and samples were transferred to UMIL with identities anonymised (database links biological samples with clinical information). CULTECH devised a questionnaire (dietary habits, use of antibiotics and probiotics, smoking patterns) for completion by all recruited patients and controls. A trial assessed the impact of a probiotic (LAB4) on the microbiome of GD patients undergoing standard treatment for the disease.

WP4 In vivo mouse model (Leader UKESSEN University)

This WP aimed to extend existing GD/GO models, which use TSHR plasmids to prime recipient mice. The GD model was established in Essen by immunizing mice with the TSHR A-subunit plasmid using a novel in vivo electroporation protocol. The effects of modifying the gut microbiota composition were then investigated. This experiment involved treatment with a broad spectrum antibiotic, treatment with LAB4, the same probiotic mixture as employed in the human clinical trial, or by colonizing with bacteria from GO patients (‘contrabiotic’ WPs 3 & 5) or no additional intervention (control).
Faeces were collected at various time points for analysis in WP8. Following sacrifice, biological samples were collected to measure TSAB, assess microbiota composition and phenotype immune cells etc. WP4 recruited an experienced researcher in UKESSEN for 12 months.

WP5 Probiotics and Contrabiotics (Leader Cultech SME)

WP5 involved production of LAB4, for use in the WP3 trial and WP4 in vivo model, followed by cultures of gut micro-organisms from GD/GO patients and controls (WP3) for analysis in the various gut model culture systems (WP2) and in vivo models (‘contrabiotics’ WP4). WP5 also analysed human and mouse gut contents, using traditional microbiology methods, to complement results obtained using 16S rRNA next-generation gene sequencing.
WP5 was achieved by the 6 month secondment of a UMIL researcher to CULTECH and the 12 month secondment of a UKESSEN researcher to CULTECH; both will also work on WP2.

WP6 Proteomic and Genomic analyses (Leader PTP SME)

WP6 undertook proteomic and genomic analyses of samples from patients (WP3) and in vitro models (WP2) to identify early biomarkers of disease onset and prognosis. Serum from controls, GD and GO patients (WP3) were analysed using high through-put proteomics techniques to generate profiles, which were analysed within and between groups to search for protein patterns diagnostic of GD and GO. miRNA was analysed in parallel serum samples to identify and quantify the miRNA species present and seek correlations with disease states. Supernatants collected from in vitro models (WP2) were also analysed using proteomic profiling. Protein and miRNA profiles from patients will be correlated with microbiome analyses carried out in WP8 to investigate the association between the microbiome-gut interaction and GO autoimmune responses. Proteomic analysis (focussing on cytokines) was attempted on supernatants from in vitro cell cultures exposed to different micro-organisms to reveal variation in cellular responses. Had time allowed, assays would have been developed for proteins (eg ELISA, HPLC or electrophoresis) and miRNA (eg qPCR or arrays) identified as putative biomakers to confirm or refute their association with disease status. As a substitute we applied predictive models, (Lasso penalized multinomial logistic regression).
WP6 was achieved by the 12 month secondment of a CU researcher to PTP.

WP7 Immune response (Leader PTP SME)

WP7 planned to analyse GD & GO patient antibody responses (WP3) to environmentally derived antigens to identify risk factors associated with the gut-microbiome or exogenous antigens expressed on common micro-organisms or in foods. An ELISA was developed to measure antibodies to common gut commensals but was not pursued. We focussed on the link between thyroid autoimmunity and coeliac disease and measured IgG and IgA responses to transglutaminase and gliadin. An ELISA was also used to measure responses to 40 food antigens.
WP7 was achieved by a 6 month secondment of the UMIL researcher to PTP.

WP8 Microbiome (Leader CU University)

WP8 explored whether there are variations in the gut microbiome of GD and GO sufferers (WP3) and whether this can be linked to the disease in animal models (WP4). The microbiomes of patients versus controls (360 samples planned, 330 samples analysed) and from mouse models (100 samples planned, >200 samples analysed) were investigated by analysing the highly variable regions of bacterial 16S rRNA genes using next generation sequencing (NGS) and bioinformatics. DNA was extracted from faecal samples from mice (WP4, controls, standard GD model and mice after contra-biotic) and humans (WP3, including the GD patients treated with probiotic). Samples were processed and variations in the 16S rRNA gene used to define the species present. Target 16S RNA gene regions were amplified using universal PCR primers in the highly variable V1-V2 regions, amplicons were sequenced to generate ~3000 bp reads/sample and provide a quantitative measurement of the microorganisms. A library of 16S rRNA gene sequences of micro-organisms was created for mice using published data and for humans using information from the human microbiome project. The V1-V2 region NGS data was analysed with a bioinformatic pipeline to determine species present and identify statistically significant variations between the microbiomes for the mouse models and among human controls, GD and GO patients. The same approach was used to investigate the impact of pro-biotics on the microbiome by defining changes that occur in micro-organism numbers and strains during probiotic administration and again following withdrawal.
WP8 was achieved by the secondment of a PTP junior researcher to CU for 12 months.

WP 9 Consolidation (Leader CU)

WP9 consolidated WP outputs, analysed correlations and defined results to be carried forward in future trials or research. The 7 RTD WPs in the project applied different and complementary approaches to obtain samples and produce data. WP9 examined which of the various outputs can be combined to confirm or strengthen individual findings.
WP9 was managed by the coordinator Prof Ludgate (CU) and recruited an experienced statistician for the final 12 months of the project.

The Science & Technology Results of INDIGO are reported following the 4 main project objectives.

• Are Gut Micro-organisms (the Microbiota) Associated with GD and Progression of GD to GO?


Patients Studied & Recruitment.
Patients were recruited mainly from the 3 INDIGO academic centres but additional participants were enroled by colleagues who are members of EUGOGO (European Group on Graves’ Orbitopathy) in Brussels, Newcastle, Belgrade and Pisa plus Moorfield’s Hospital London UK.
Patient samples (stool, sera, plasma, nasal swabs and tears) were collected between October 2014 and May 2017. Appropriate ethical approval was obtained from all centres and informed written consent was obtained from all participants. Ethical approvals were monitored throughout the project to ensure compliance and that they were up-to-date. All subjects were treated in accordance with the tenets of the Declaration of Helsinki.
The INDIGO study enrolment criteria were: i) newly diagnosed or relapsing GD patients under anti thyroid treatment within six weeks from the onset of hyperthyroidism, ii) GD patients at euthyroidism, iii) newly diagnosed GD patients with overt GO. Hyperthyroid GD patients were defined based on their suppressed TSH and high T4 levels and positive TRAB titers. Euthyroid GD patients were defined by the T4 thyroid hormone levels being in the normal range.
Thyroid function tests were measured using routine automated assays at each recruiting centre and TSHR autoantibodies were measured using Siemens Immunlite, which detects thyroid stimulating immunoglobulins (TSI, synonym for TSAB), and Cobas Roche TRAK which detects antibodies inhibiting binding to the TSHR.

Samples for Study of the Gut Microbiota (‘microbiome’)
A total of 146 patients (mean age 47) enrolled in centres from four European countries (UK, Italy, Belgium, Germany) provided at least one faecal sample during the course of the study. Samples were requested at baseline (BL) i.e. the time of enrolment, when anti-thyroid drug therapy had restored euthyroid status (EU) and at the end of the follow up (EFU). To facilitate the collection of the faecal sample at home, patients were provided with a packaged kit comprising instructions for sampling, a sterile collection tube and a transport tube to be returned to the clinic in anaerobic storage bags, where they were kept frozen at -20°C, including during transportation to Cardiff, where they were processed.
Patients were further subdivided into Graves’ disease patients with no sign of eye disease (GD) and Graves’ orbitopathy (GO) with either mild, moderate-severe or sight-threatening signs of eye disease, based on the assessment of the EUGOGO guidelines (Bartalena et al., 2008). Healthy donors from each recruitment centre, matched for patients’ age and gender, were all euthyroid, free of thyroid disease with no signs of any eye disease and negative for TRAB.

DNA extraction and 16S rRNA-gene sequencing
A total of ~300 faecal samples were kept frozen at -20°C for a maximum of two months prior to processing. Up to 180-220 mg of slowly thawed faeces were individually placed in 2mL FastPrep tubes prefilled with 0.1mm silica spheres (FastPrep lysing matrix B, MP Biomedicals, UK) and dissolved in 1 mL InhibitEX buffer (Qiagen Ltd, West Sussex, UK). A bead-beating step (Beadbug microcentrifuge homogenizer, Benchmark Scientific, USA) was conducted for 3 x 60 sec at 5m/s with 5 min rest in-between. DNA extraction followed manufacturer’s instruction provided in the QiAmp Fast DNA Stool Mini kit (Qiagen Ltd, UK). Total genomic DNA was eluted in 200ul TAE buffer in sterile tubes and stored at -20°C until used. The 16S rRNA gene sequencing was performed at Research and Testing RTL Genomics (Lubbock, Texas, USA), using primers detecting the V1-V2 regions of the 16S rRNA gene plus bifidobacteria regions (28F-combo) to generate 10,000 paired-ends reads per sample on a Illumina MiSeq (Illumina, San Diego, USA).

Bioinformatics processing
Demultiplexed paired-end reads from 16S rRNA-gene sequencing were first checked for quality using FastQC (Andrews, 2010) for an initial assessment. Forward and reverse paired-end reads were joined into single reads using the C++ program SeqPrep (John, 2011). After joining, reads were filtered for quality based on: i) maximum three consecutive low-quality base calls (Phred < 19) allowed; ii) fraction of consecutive high-quality base calls (Phred > 19) in a read over total read length >= 0.75; iii) no “N”-labeled bases (missing/uncalled) allowed. Reads that did not match all the above criteria were filtered out. All remaining reads were combined in a single FASTA file for the identification and quantification of OTUs (operational taxonomic units, equates ± to a single bacterial species with 97% accuracy). Reads were aligned against the SILVA closed reference sequence collection release 123, with 97% cluster identity (Quast et al., 2013; Yilmaz et al., 2014), for taxonomic identification along the main taxa ranks down to the genus level (domain, phylum, class, order, family, genus). By counting the abundance of each OTU, the OTU table was created and then grouped at each phylogenetic level. OTUs with total counts lower than 10 in fewer than 2 samples were filtered out. All of the above steps, except the FastQC reads quality check, were performed with the QIIME open-source bioinformatics pipeline for microbiome analysis.

Alpha and Beta Diversity Indices
The gut microbial diversity was assessed within- (alpha diversity) and across- (beta diversity) samples. All indices (alpha and beta diversity) were estimated from the complete OTU table (at the OTU level), filtered for OTUs with more than 10 total counts distributed in at least two samples. Besides the number of observed OTUs directly counted from the OTU table, within-sample microbial richness, diversity and evenness were estimated using the following indexes: Chao1 and ACE (Abundance-based Coverage Estimator) for richness, Shannon, Simpson and Fisher’s alpha for diversity, Simpson E and Pielou’s J (Shannon’s evenness) for evenness. The across-sample gut microbiota diversity was quantified by calculating Bray-Curtis dissimilarities. Prior to the calculation of the Bray-Curtis dissimilarities, OTU counts were normalized for uneven sequencing depth by cumulative sum scaling (CSS). Among groups (GD, GO, controls) and pairwise Bray-Curtis dissimilarities were evaluated non-parametrically using the permutational analysis of variance approach (999 permutations).

Reads from 16S rRNA gene sequencing were processed with the QIIME pipeline. The ACE index and sample-base rarefaction were estimated using our own in-house scripts for Python ( and R ( The Tax4Fun R package, which employs the ‘nearest neighbor’ identification with a minimum sequence similarity to link the representative 16S rRNA gene sequences to functional annotations of prokaryotic genomes, was used to predict the metagenome from metataxonomy and the functional profile of the microbiome. Plots were generated using the ggplot2 R package. Additional data handling was performed with the R environment for statistical computing.

Benjamini-Hochberg (BH) for false discovery rate (FDR)
The False Discovery Rate is simply the proportion of positive tests that are false positives (false positives) / (all positive tests); the Benjamini-Hochberg procedure attempts at estimating this proportion based on the distribution of obtained p-values. While a p-value of 0.05 means that 5% of all tests are on average false positives, a FDR of 0.05 implies that 5% of positive results may on average be false. Thus BH is more stringent than simple p-values, but less stringent than Bonferroni correction.

Predictive Model; Random Forest Analysis
Random Forest (RF) is a statistical learning method based on the construction of a forest of “decision trees” for classification or regression. RF was used to predict to which group (GD, GO healthy control) a sample belonged based on their microbiota composition (classification), and to identify genera driving the classification (variable importance). Relative abundance counts with non-zero values in at least 20% samples were retained and scaled to centre. To control the accuracy of the calculation, a repeated cross-validation (repeatedcv) method with number=10 and repeats=3 was used. The tuning hyperparameter mtry, calculated as the square root of the number of columns of the dataset, was tested from 10 to 50 and 5,000 or 10,000 number of trees (ntree) using the R package Caret. RF was next run using the identified parameters providing the best fit during the cross-validation step using the R package RandomForest. The mean decrease Gini was used for the variable importance selection.

Baseline Characteristics of Gut Microbiota in GD, GO, controls
We have so far analysed the baseline data in 65 GD, 56 GO and 42 healthy controls. The within-sample alpha and between-sample beta diversity were similar in patients at recruitment and controls. When considering phylum composition, there were differences noted between the 3 groups Bacteroidetes were significantly more abundant in controls (38.5%) than in GD (24.2%) and GO (27.3%) patients, while Firmicutes were more abundant in GD (59%) and GO patients (60.5%) than controls (53.2%). Consequently the Firmicutes:Bacteroidetes ratio was significantly higher in GD/GO than controls, but similar between GD and GO. Enteroccoccus gallinarum counts, a pathobiont reported to be involved in triggering autoimmunity, though low overall were significantly higher in GD and GO than controls.

Change in gut microbiota composition when GD progresses to GO
We had the opportunity to study the gut microbiota composition in 2 women whose GD evolved to GO during the study. Subject 1004 was a new GD patient at BL who developed active GO, with a clinical activity score (CAS) of 2, by the EU time point two months later. In contrast subject 1013 was seen at BL, for a relapse of her GD and 3 months later at EU, had active GO with a CAS of 3. In both cases we observed a dramatic shift in their microbiota, with the common feature being a decrease in the genus Bacteroides (BH adjusted p<0.0001).

Our BL data illustrate substantial perturbation of the gut microbiota composition in GD/GO. Future analyses will explore associations between taxonomic profiles and TSAB, thyroid function and GO disease severity and whether they are affected by treatment (Manuscript in preparation).

• Can Probiotic Manipulation of GD Microbiota Improve Outcomes, Especially GO Progression?

We conducted a double-blind randomized controlled trial, at the University Hospital of Milan (Italy) approved by the local Ethics Committee, to assess whether administration of a well-characterized probiotic mixture (LAB4) would modify the microbiota composition in GD/GO patients, improve their immunologic status and prevent GD relapse and/or GO progression.

Inclusion criteria:
Group A: Untreated Graves’ hyperthyroidism (or within 6 weeks of initiating anti-thyroid drug [ATD] treatment)
- definition hyperthyroidism: TSH decreased, FT4 and /or FT3 increased
- definition Graves’: diffusely enlarged thyroid gland either by palpation or echograpy, and/or homogeneous thyroid uptake at scintigraphy, or positive TSHRAb
- recurrence of Graves’ hyperthyroidism defined as relapse after 6 weeks from ATD withdrawal
- minimal or no eye signs, defined as lid retraction / lid lag but no other signs.
Planned treatment with ATD either titration regimen or block-and-replace regimen for 18 months.

Group B. Untreated Graves’ hyperthyroidism (or within 6 weeks of initiating ATD treatment) with overt signs of GO as defined by EUGOGO (ref).
1. Mild GO: patients whose features of GO have only a minor impact on daily life insufficient to justify immunosuppressive or surgical treatment. They usually have only one or more of the following: minor lid retraction (<2 mm), mild soft tissue involvement, exophthalmos <3 mm above normal for race and gender, transient or no diplopia, and corneal exposure responsive to lubricants)
2. Moderate-to-severe GO: Patients without sight-threatening GO whose eye disease has sufficient impact on daily life to justify the risks of immunosuppression (if active) or surgical intervention (if inactive). Patients with moderate-to-severe GO usually have any one or more of the following: lid retraction R2 mm, moderate or severe soft tissue involvement, exophthalmos >3 mm above normal for race and gender, inconstant, or constant diplopia.
3. Sight -threatening GO: Patients with dysthyroid optic neuropathy (DON) and/or corneal breakdown.

Control group: local unrelated people with no known autoimmunity, a BMI below 30, no recent antibiotic use (see above) and matched (where possible) for age, gender and diet (main categories such as vegetarians, vegans)

Main exclusion criteria:
Previous or planned treatment with 131I or thyroidectomy; sight threatening GO requiring decompression; drugs interfering with the natural course of GO; steroids, immunosuppressants, thiazolidinediones, antibiotics/antifungals/antivirals (topical or systemic for at least 4 weeks prior to recruitment to the study); acute diarrhea (gastroenteritis for at least 4 weeks prior to recruitment to the study); Drugs interfering with thyroid function: amiodarone, lithium, iodine supplements; Drug or alcohol abuse; no informed consent; Age less than 18; Pregnancy.

Primary Endpoints:-
➢ Compare Firmicutes:Bacteroidetes (F:B) Ratio in probiotic treated and placebo control groups at various time points
Secondary Endpoints:-
➢ Changes in TSHR autoantibody titre, IgG & IgA levels in the two groups pre/post treatment
➢ Thyroid function tests (TFTs) pre/post treatment
➢ Clinical features of GD pre/post treatment
➢ GO severity and clinical activity score pre/post treatment

The probiotic consortia LAB4 (produced at Cultech, UK) comprises 25 billion colony-forming-units/capsule of two Lactobacillus acidophilus strains (CUL21 NCIMB 30156, CUL60 NCIMB 30157) plus Bidifobacterium bifidum(CUL20 NCIMB 30153) and Bidifobacterium animalis subsp lactis (CUL34 NCIMB 30172). Twenty-nine GD patients, recruited when untreated or within 4 weeks of initiating ATD, were randomized to receive either LAB4 (n=14, 10 hyperthyroid and 4 euthyroid) or placebo (n=15, 14 hyperthyroid, 1 euthyroid) plus ATD for 6 months. Blood for biochemical analyses [TSHR autoantibodies measured using Siemens Immulite for TSI & Roche Cobas for TRAK; TSH, FT4, FT3 measured using electrochemiluminescent immunoassay and IgG/IgA levels measured using Immunoturbidimetric assay, both Roche Diagnostics] and faecal samples for 16S rRNA gene sequencing (samples processed and analysed as above) were collected at baseline, when euthyroid (EU) and at the end of treatments (more than 6 months, EFU).


There was no significant difference in age or female:male ratio in the LAB4 (11F, 3M, 46.4 ±11.9 years) and placebo (13F, 2M, 44.8±12.4 years) groups at recruitment baseline (BL). In the placebo/probiotic groups there were 5 GD/10 GO and 5 GD/9 GO patients respectively. At the EU time point, hyperthyroidism started to resolve in both groups but serum FT3 levels in the placebo group were higher at EFU than the probiotic group (P<0.04), who remained euthyroid both at EU and through EFU.
Autoantibody titres to the thyrotropin receptor (TSHR) were slightly higher in the probiotic treated patients at BL but were significantly reduced in all patients by standard block/replace therapy with no difference noted between treatment groups.
Circulating IgG and IgA were similar in both groups at BL and EFU but probiotic treatment significantly reduced IgG (p<0.02) and IgA (p<0.03) levels at EU suggesting a transient systemic effect.
Severity or progression of GO were not different whether patients received placebo or probiotics.

The microbiota composition was more stable in probiotic receiving patients. The most abundant phyla in the gut microbiota were the firmicutes, bacteroidetes and actinobacteria. There was no difference in alpha diversity between the two groups or in the firmicutes :bacteroidites (F:B) ratio. However the actinobacteria:bacteroidetes (A:B) ratio, calculated to assess the contribution of LAB4 to the gut microbiota, was higher in the probiotic treated, but just missed significance. We also observed a significant reduction in counts of the Firmicutes phylum in the probiotics group compared to placebo (P=0.033) at EU.

Although severity or progression of GO were not different in the two treatment groups our results indicate that probiotic reduces the risk of hyperthyroid relapse, compared with placebo, after 6 months of ATD therapy. Additional analyses will investigate possible associations between microbiota composition and smoking, thyrotropin receptor autoantibodies and thyroid function. (Manuscript in preparation)

• Can Gut Microbiota Manipulation Improve in vivo Models of GD/GO?

Prior to investigating the effect of modifying the gut microbiota on a TSHR induced GD/GO animal model, we had to establish the model in one of the INDIGO partner’s institutions. Female BALBc mice were immunized with an expression plasmid for the TSHR A subunit (βgal empty plasmid control) using in vivo electroporation. All animals, whether TSHR or βgal controls, received a total of four plasmid injections at three week-intervals and were sacrificed nine weeks after the last immunization (18 weeks) to permit the development of the chronic phase of the disease in the TSHR group. Appropriate biological samples were obtained and observations made to provide:- i) evaluation of clinical symptoms, ii) the determination of the thyroid hormone thyroxine blood levels (fT4) and TRAB (both stimulating TSAb and blocking TSBAb) antibodies, iii) the measurement of the expansion of fat cells (adipogenesis) and muscular atrophy in the orbit. A proportion of immunized mice developed a GO-like disease and comparison with the same model induced in London indicates some differences e.g. hyperthyroidism induced in London but not in Essen. Please see attached PDF for links to all papers published to date from the project.
We then compared the gut microbiota of immunized TSHR BALB/c mice established in Essen (Essen) and London (london), to observe whether the gut microbiota may have an impact on the GO preclinical mouse model in different laboratories. Mice in london were maintained conventionally in open cages in one room and co-housed at a maximum of 3 animals per cage. In Essen, the mice were co-housed according to their immunizations, 2-4 animals per individually ventilated cage in one room. All mice were provided by different outlets of the same supplier (Harlan Ltd or Harlan laboratories BV). In both centres, mice received autoclaved water and had been fed ad libitum similar commercial chow from different suppliers (Rat and Mouse no.1 Maintenance from Special Diet Services, LBS Biotech UK for London and Rat/Mouse Maintenance V1534-300 from Ssniff Spezialadiaten GmbH, Germany, for Essen). Also the cage bedding was from different suppliers. After sacrifice murine intestines were snap frozen and the contents scraped out for DNA extraction and 16S rRNA gene sequencing as described above.
From 16S rRNA gene sequencing (V1-V2 regions), a total of 5,333,798 reads were obtained which reduced to 4,047,186 reads after quality filtering. Following alignment, we obtained an average of 20,534 reads per sample, ranging from 3,502 to 134,901. Subsampling per library size resulted in a 96% average coverage per OTU definition at 3,052 reads per sample. The averaged coverage and subsampling was sufficient to describe gut bacterial communities according to sequence-based rarefaction curves (data not shown). We identified a total of 4,281 OTUs: 1,037 OTUs had more than 10 counts across samples, and were retained.
We compared the richness and diversity (alpha-diversity) and differences in the community composition (beta-diversity) of the London and Essen mice. Comparison of the alpha diversity indices showed a significant reduction in the richness (P=0.01), but not in the diversity of the Essen microbial community (P>0.05, Figure 1A). The gut microbiota composition from the two centres showed a good separation according to the Spearman distance and Ward hierarchical clustering and a PERMANOVA test on the weighted Unifrac distances revealed a spatial difference between bacterial communities (P=0.005 with 999 permutations, Figure 1B). At the phylum level, Bacteroidetes and Firmicutes were the most abundant among the 7 phyla identified, with no differences between the two centres (P=0.99). Lactobacillaceae, Ruminococcaceae and Porphyromonadaceae families were more abundant in Essen than in London TSHR mice (P<0.01, Figure 1C) whilst E.coli and coliforms were least abundant in the Essen TSHR mice. The 16S rRNA sequencing data were largely confirmed using traditional microbiological culture which also indicated significantly higher yeast counts in London TSHR-immunized mice.
(Please see attached PDF for all final report figures.)

We also compared the gut microbiota between immunization groups within Essen, comprising the TSHR or βgal control immunized mice and naïve untreated mice. We observed a shift of the TSHR immunized mice bacterial communities described by the beta-diversity weighted Unifrac.
Finally we investigated associations between disease features and taxonomy in the two centres. We observed a positive correlation between levels of TSAb and Deferribacteres phylum, which include one-genus Mucispirillum, in London (Rho=0.92, P=0.028). In contrast we obtained a strong negative correlation of the Firmicutes genus Intestinimonas and the levels of TSBAb in the London (Rho=-0.89, P<0.05), but not in the Essen counterpart. No significant correlation was observed between OTUs from the genus Intestinimonas and levels of TSAb or levels of free thyroxine hormone (fT4).
In Essen Bacteroidetes and Firmicutes were also negatively correlated (Rho=-0.99, P<0.0001). We found a significant positive correlation (Rho=0.6, P=0.009) between the OTUs from the Firmicutes and the orbital adipogenesis value and a negative correlation of this value with the phylum Bacteroidetes (Rho= -0.57, P=0.014), specifically in TSHR immunized mice. The correlation pattern we found (Firmicutes positively correlated, Bacteroidetes negatively correlated) was also recapitulated at the genus level (figure 2). Among the genera of the Firmicutes, three, within the Clostridia family (Butyricicoccus, Parvimonas and Fusibacter) and the genus Lactobacillus were correlated positively with adipogenesis; while three Bacteroidetes genera (Anaerophaga, Paraprevotella and Tannerella) correlated negatively with the orbital adipogenesis values. A strong positive correlation (Rho=0.82, P=0.007) was observed between orbital adipogenesis and the total anaerobes counts obtained from the traditional microbial cultures of TSHR immunized mice, but not in the controls.
(see Figure nr. 2 in the document attached)

We concluded that the data support a role for the gut microbiota in modulating the induced response. Furthermore, the identification of disease-associated taxonomies suggests that the gut microbiota may contribute to the heterogeneity of induced response.

We then conducted experiments to modify the gut microbiota composition. Female BALBc mice were treated with Antibiotic (vancomycin), Probiotic (LAB4) or Contrabiotics (GO patient human faecal material transfer (hFMT)) to modify the gut microbiota, or received no additional treatment, prior to immunization with TSHR A-subunit or βgal.


Production of freeze-dried faecal material for transplant (hFMT = contrabiotic) and LAB4 probiotic
Six patients with sight-threatening Graves’ Orbitopathy (GO) were enrolled at the University Hospital of Duisburg-Essen and their faecal samples were processed in WP5 at Cultech Ltd. (Port Talbot, UK) for the production of a freeze-dried faecal material to be transferred in mice (hFMT). Faeces were pooled together and prepared for a sequential culture method in maximum recovery diluent broth (MRD) and incubated overnight at 37°C under aerobic or anaerobic conditions. The mixture was further inoculated into 500mL pre-reduced MRD, followed by an overnight incubation at 37°C under aerobic or anaerobic conditions. After centrifugation the pellet was weighed and poured into petri dishes, where they were supplemented with 10% w/v skimmed milk powder as a cryoprotectant agent, and placed at -80°C until completely frozen. The freeze-dried process was performed in a freeze-dryer machine from overnight to several days. The resulting powder was aliquoted into small vials containing 0.125g and shipped to the University Hospital of Duisburg-Essen (Germany) for administration to mice via gavage.
The probiotic LAB4 was also produced at Cultech.
GO animal model and treatments
Female BALB/c mice used in this study were bred at the University Hospital of Duisburg-Essen (Germany) facility, in order to administer the treatments from an early-stage of life. The antibiotic vancomycin was provided in the drinking water (autoclaved) at a dose of 0.2 g/l initially to dams and then to the pups from their first day of life for the entire course of the experiment.
The probiotic LAB4 is a consortium of lactic acid-producing bacteria comprising two strains of Lactobacillus acidophilus CUL60 (NCIMB 30157) and CUL21 (NCIMB 30156), Bifidobacterium lactis CUL34 (NCIMB 30172) and Bifidobacterium bifidum CUL20 (NCIMB 30153) and was administered at a total of 1x1010 CFU in 50 μl autoclaved water per gavage. The hFMT powder was dissolved in autoclaved water and provided at a final concentration of 1x1010 CFU in 50 μl autoclaved water per gavage. A group of mice receiving autoclaved water was included as a control. Administration of both interventions and control was performed through gavage on pups a total of four times from the first day after birth, at weaning, before and in the mid of the immunisation procedure, (TSHR-A subunit or b-gal control, as described above), as summarised in figure 3. (see the attachment)

A total of 98 mice provided faecal pellets throughout the experiment and intestinal contents (small, colon or whole gut) after sacrifice. DNA was extracted from these and underwent 16S rRNA gene sequencing, as described above. Relevant biological samples were also collected for measurement of TSHR autoantibodies, T4 levels, to assess orbital pathology and to phenotype T cell subsets in lymph nodes (flow cytometry).

Serum markers of thyroid function and autoimmunity
Antibodies to the hTSHR, measured by inhibition of TSH binding (TRAB), were detected in all hTSHR immunized mice but not b-gal controls. Similar results were obtained when hTSHR antibodies were measured using a bioassay to detect thyroid stimulating antibodies (TSAB) with the exception of the vancomycin treated group where no significant difference was observed between TSHR and b-gal immunized mice. Furthermore TSAB levels in these animals were weakly positive in < 50% of the TSHR immunized and were significantly lower than the equivalent control and probiotic treated mice. Total thyroxine (T4) levels were significantly higher (ANOVA) in hTSHR compared with b-gal immunized mice only in probiotic treated animals. Results are summarised in table 1. (see the attachment)

Orbital Pathology and Lymph Node Phenotype
In contrast to the previous experiment, no significant differences were observed in total orbital adipose tissue between hTSHR and b-gal immunized mice in any of the treatment groups; highest values were observed in 2 TSHR immunized/probiotic treated mice. The areas of total fat tissue and brown fat tissue were determined using imageJ and the percentage of brown fat area was calculated. We observed a significantly higher percentage of brown adipose tissue (BAT) induced in hTSHR immunized mice compared with b-gal only in the control and probiotic treated groups and not in those receiving antibiotic or hFMT, whilst significant muscle atrophy was detected only in control TSHR immunized mice and not the other 3 groups. Flow cytometry analysis of lymph nodes revealed that CD25+ Treg numbers were significantly lower/higher in antibiotic and probiotic treated animals respectively. Results are summarised in table 1 (please see the attachment).

Summary of the sequencing outcomes
Sequencing of the V1-V2 plus bifidobacteria regions of the 16S rRNA gene produced a total of 13,782,107 after the “join” function in QIIME 1.9, with an average of 2,297,017.83 (± 1,820,298.366). Filtering of reads with a Phred > 19, allowing about 1 error in 100 bases, retained a total of 12,884,785 sequences with an average of 2,147,464,17 (± 1,726,134.85), which resulted in a 6.5% sequences removed. While the control, hFMT and the LAB4 treatments show very similar numbers of reads, the vancomycin treatment group show twice the number of reads. A smaller number of reads were obtained from the six GO patients (plus some replications) providing the samples for the hFMT production.

GO model gut microbiota recapitulates the treatments received and showed disease-associated differences in taxonomy
After sacrifice, the small (ileum, jejunum and duodenum) and the large intestinal (colon) contents were collected from the vancomycin, hFMT groups and their respective controls, while entire intestines were obtained from the LAB4-treated mice and a group of controls. As previously reported by others, the small intestine showed a significantly reduced richness and diversity of the bacterial communities compared to those of the entire gut and the colon (P<0.001), while the entire gut and the large intestine alpha diversity indices were more similar (P>0.05).
The long-term vancomycin treatment reduced richness, diversity and equitability indices in all microbiota sources, with a less severe effect on the small intestines’ Shannon diversity (P=0.111) and equitability (P=0.252) indices, compared to those of the control groups. In particular, diversity and equitability indices in the small intestines were more similar in vancomycin treated mice compared to their controls (P>0.05), than those of the vancomyin treated mice compared to the hFMT group (Post-hoc P=0.0007 and P=0.003, respectively).
Significant differences between the two immunisations, b-gal and TSHR, were observed in the small intestines in the Shannon (P=0.046) and equitability (P=0.013) indices in the control group and in the colon-entire Shannon index of the vancomyin-treated mice (P=0.002).
At the taxonomic levels, the vancomyicin treatment led to a unique microbiota, with a reduction of most Actinobacteria, Firmicutes and Bacteroidetes genera and an increase in Proteobacteria (e.g. Salmonella, Shigella/E.coli, Enterobacter) and more than 70 taxonomies differentially abundant compared to controls, LAB4 and contrabiotics. Differences were also observed between the LAB4 and the contrabiotics compared to the controls, in particular, the genus Bacteroides, previously associated with the disease status (e.g. decreased in GD patients) was significantly decreased in the TSHR-contrabiotics group compared to controls.
In order to confirm the success of the engraftment (i.e. transfer of taxonomies from donors to recipients) provided by the administration of a contrabiotics to mice from the early stage of life, we employed the SourceTracker Bayesian model to calculate the percentage of similarity of the gut microbiota in the contrabiotics-receiving to that of the human donors and compared this against the control mice. It appeared that the contrabiotics-receiving mice had a higher percentage of similarity to human samples compared to controls mice at the baseline (after 3 gavages from birth), while the similarity inverted in the mid-timepoint in controls and at the endpoint both controls and contrabiotics mice showed an equal similarity with the human samples in the colon samples, although more variable in the contrabiotics mice. Such an analysis indicated that the engraftment was more successful at the early stage of life at least for some mice, possibly influencing the immune system of the mice as well.

Antibiotic treatment more significantly modified the gut microbiota composition than probiotic or hFMT probably due to the earlier (pre-birth) and continuous nature of this treatment compared with the LAB4 or hFMT (from birth and at time intervals). We were able to induce a human GO-like gut microbiota using the contrabiotic but this was not sustained. Furthermore our human microbiome data indicate more significant differences between GD and patients with mild-GO. Probiotic had limited effects but increased Treg and decreased T effector cell numbers. However the absence of pathological TSAB and orbital pathology, combined with reduced Treg in vancomycin treated mice confirm a role for elements of the gut microbiota in promoting autoimmune GD and GO.

• Identify Bio-Markers to Predict Patients at Risk of Progression from GD to GO

Currently, the diagnosis of GD relies on examination of the thyroid, measurement of serum thyroid hormones concentrations and TRAB assay, while predicting the progression from GD to GO is imperfect and relies mostly on clinical signs. Biomarkers (a characteristic that is objectively measured and evaluated as an indicator of normal biologic processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention) could complement diagnosis and monitoring of this disease by: i) increasing accuracy, ii) reducing costs, iii) reducing time to diagnosis. Additionally, the loss of immune-tolerance mechanisms in GD/GO is poorly understood. Thus we analysed circulating proteins and microRNA (miRNA); the two most widely used classes of blood biomarkers for clinical research.

Sample collection
All samples were collected within the same timeframe, with ethical approval and from the same patient groups as described above. We analysed samples from a total of 46 individuals, GD 14, GO 19 and 13 healthy controls. Blood samples were collected from all subjects enrolled, either in EDTA-coated test tubes for plasma or gel-filled test tubes for serum separation (BD Vacutainer), and processed by centrifugation at 1200 g for within one hour from collection, then stored at until used.


microRNA sequencing
miRNAs were extracted from plasma samples using NucleoSpin miRNA Plasma kits and quantified using a NanoDrop 1000 spectrophotometer. We prepared libraries of extracted small RNA using the Truseq SmallRNA kit, amplified them by 15 cycles of PCR, pooled and then resolved them on a Pippin Gel cassette. DNA fragments from 140-160 bp (the length of miRNA inserts plus the 3’ and 5’ adaptors) were recovered in manufacturer’s elution buffer and then purified by Qiagen MinElute PCR Purification kit. The indexed libraries were quantified on an ABI9700 qPCR instrument using the KAPA Library Quantification Kit in triplicates, subsequently the pooled library was used for sequencing using Illumina HiSeq2000 with a 50 Single-Read sequencing module in a total number of 4 lanes.

Nano LC/QTOF proteomic mass-spectrometry
Serum samples were processed through the ProteoExtract® Albumin/IgG removal kit before proteomic analyses. Protein content was determined in the eluted extracts using the Bio-Rad Protein Assay kit with bovine γ-globulin as standard. Then, proteins were loaded in 4-20% TGX Stain-Free TM precast mini gel (Bio-Rad) and separated by SDS-PAGE electrophoresis. The lanes from each sample were manually cut into four slices of the same size, and proteins were reduced with DTT, alkylated with iodoacetamide and overnight digested using trypsin. Tryptic peptides were analysed using nanoscale liquid chromatography coupled to a hybrid quadrupole-time-of-flight (Q-TOF) mass spectrometer. With this aim, a nano LC Agilent 1260 Chip Cube source and an Agilent 6550 IFunnel Q-TOF mass spectrometer (Agilent Technologies, Santa Clara, CA, USA) were used. MS/MS spectra of peptides were used for protein inference via database searching in Spectrum Mill MS Proteomics Workbench. Carbamidomethylation of cysteine was set as a fixed modification while trypsin selected as enzyme for protein digestion, accepting two missed cleavages per peptide. Inference was set with the following parameters: Scored Peak Intensity (SPI) ≥50%, precursor mass tolerance of ±10 ppm and product ions mass tolerance of ±20 ppm. The search was conducted against the proteome of Homo sapiens (UniProt, downloaded July 2016). The database was concatenated with the reverse one and 1% false discovery rate was selected for validation purposes. Finally, a label-free quantitation was carried out using summed peptide abundance.

Bioinformatics processing
Fastq files resulted from miRNA sequencing were first checked for sequence quality in terms of GC content, duplication level, length and Phred score. Reads were trimmed on base-call quality (Phred score >15 over a 4bps sliding-window; this gives a probability lower than 0.03 that a base is called incorrectly; see Ewing et al. for details on Phred) and length >15 bps, as miRNAs usually have length in the range 20-22 bps. Adapters used for reverse transcription and amplification were also trimmed off the reads. Trimmed reads were used along with the human reference genome (assembly GRCh38) and a database of known mature miRNA and precursors to detect known and novel miRNAs in each sample. miRNA detection and quantification was carried out using the miRDeep2 algorithm ; both novel and known miRNAs were quantified in terms of counts per sample. All miRNA bioinformatics processing was streamlined using an informatic pipeline developed in-house at the Bioinformatics Unit of PTP Science Park (


Analysis of sequenced miRNAs and proteins
Querying the miRDeep2 database with trimmed reads, and alignment against the human reference genome, yielded a total of 3 025 miRNAs (1 881 known, 1 144 novel). Filtering for CPM≥2 in more than two samples left 1 332 miRNA available for subsequent analyses (777 known, 555 novel). From proteomic mass-spectrometry, 1886 proteins were detected in blood samples. After filtering for proteins expressed in more than two samples, 831 proteins were retained for later analyses.

Cluster analysis
Distances between samples (GD, GO and controls) were estimated from miRNA counts and protein relative abundances. Three Euclidean distance matrices were compared: based on miRNA, proteins and miRNA+proteins combined. Clearly separate clusters were not obtained from miRNA counts alone, while protein abundances and, especially, proteins and miRNA data combined did form separate clusters. In particular, controls were clearly distinguished from Graves’ disease cases, while a less clear boundary between GD and GO patients was obtained. The first two dimensions from multidimensional scaling accounted for 39.2% (29.6%, 9.6%), 76.6% (57.6%, 19.0%) and 77.3% (57.5%, 19.8%) of total variation, when using, respectively, only miRNA, only proteins, miRNA and proteins together (Figure 4 in attachment).

Metabolic pathway analysis
From the KEGG (Kyoto Encyclopedia of Genes and Genomes) database of human genes and annotated metabolic pathways, the metabolic pathways associated with the genes corresponding to the identified candidate biomarkers have been retrieved and ordered by the significance of association. From the combined set of genes based on miRNA targets and proteins, there were 16 pathways with p-value ≤0.05. These included: regulation of actin cytoskeleton, PI3K-Akt signaling pathway, Arrhythmogenic right ventricular cardiomyopathy (ARVC), Hypertrophic cardiomyopathy (HCM), Dilated cardiomyopathy (DCM), Cell adhesion molecules (CAMs), Oxytocin signaling pathway, Focal adhesion, Hippo signaling pathway, PPAR signaling pathway, Bacterial invasion of epithelial cells, Complement and coagulation cascades, ECM-receptor interaction, Longevity regulating pathway, mRNA surveillance pathway, Circadian entrainment.

Some of the protein biomarkers identified (pleasesee Figure 5 in attachment) are relevant to GD/GO pathology; e.g. gut permeability (zonulin) and tissue remodelling (fibronectin). Proteomic and miRNA analyses, combined with robust bioinformatics, identified circulating biomarkers applicable to diagnose GD, predict GO disease status and optimize patient management. The work has been published please see relevant links in the attachment.

In a second investigation related to the topic, we investigated the immune response of recruited patients to food antigens. Subjects had completed a questionnaire about dietary habits, but unfortunately it was not helpful in the identification of the most relevant food antigens to test. Since many GD patients describe bowel discomfort and there is an association between coeliac disease and GD, we decided to measure anti-gliadin (deamidated gliadin, DGP, which gives positive responses in coeliac disease and gluten sensitivity) and anti-transglutaminase (tTG, which is also detected in patients with coeliac disease and those with inflammatory bowel disease, juvenily type 1 diabetes and some forms of arthritis). To complement these specific assays, we also used a combined anti-food ELISA kit to assess the immune response of sera against 40 different food antigens clustered in 21 food groups (corn, oat, rice, rye, wheat, cow’s milk, egg white, egg yolk, white fish mixture, shellfish mixture, soya, legume bean mixture, mustard mixture, gluten, apple+pear, berries mixture, citrus mixture, nut mixture, yeast, chicken+turkey and pork+beef).

We tested 105 GD patients (89F/16M, mean age 44±14 years) using commercially available ELISA kits to detect IgA and IgG responses to DGP and tTG. GO was present in 48 of the patients and results were compared with the epidemiological prevalence of coeliac disease. Serum IgG and IgA concentrations were measured in all tested patients using Immunoturbidimetric assay (Roche Diagnostics).
For the combined food antigen ELISA, we tested 71 GD patients (33 also had GO) and 24 healthy controls, all women (to reduce variables) with mean age 46±13 years.

We observed that 6 out of 105 sera (5.7%) were tTG positive; 16 and 7 out of 108 (15 and 6.5%) were DGP-IgA and DGP-IgG positive respectively; a significantly higher prevalence in GD compared to the worldwide prevalence of celiac disease (1%) (chi-squared test; p-value < 0.001). Since total IgG and IgA levels were all found to be in the normal range the anti-tTG and anti-DGP positivity levels do not need to be adjusted.
Twenty-three out of 71 (32.3%) GD sera showed sensitivity against a food antigens, compared to 25% (6 out of 24) positive results among healthy controls (chi-squared test; p-value=0.4).

GD patients have a higher proportion of antibody reactivity to antigens relevant to coeliac disease and also demonstrate more immune responses to common food antigens than healthy controls, particularly milk and egg white. (Manuscript in preparation)

Reported by

United Kingdom


Life Sciences
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top