Intestinal Metaproteomics Reveals Host-Microbiota ... · Dysbiosis of the gut microbiota has been implicated in disease pathogenesis of many immune-mediated diseases, including type

Intestinal MetaproteomicsReveals Host-MicrobiotaInteractions in Subjects at Riskfor Type 1 DiabetesDiabetes Care 2018;41:2178–2186 | https://doi.org/10.2337/dc18-0777

OBJECTIVE

Dysbiosis of the gut microbiota has been linked to disease pathogenesis in type 1diabetes, yet the functional consequences to the host of this dysbiosis are unknown.We investigated the functional interactions between the microbiota and the hostassociated with type 1 diabetes disease risk.

RESEARCH DESIGN AND METHODS

We performed a cross-sectional analysis of stool samples from subjects with recent-onset type 1 diabetes (n = 33), islet autoantibody–positive subjects (n = 17), low-riskautoantibody-negative subjects (n = 29), and healthy subjects (n = 22). Metaproteomicanalysis was used to identify gut- and pancreas-derived host and microbial proteins,and these data were integrated with sequencing-based microbiota profiling.

RESULTS

Both human (host-derived) proteins andmicrobial-derived proteins could be usedto differentiate new-onset and islet autoantibody–positive subjects from low-risksubjects. Significant alterations were identified in the prevalence of host proteinsassociated with exocrine pancreas output, inflammation, and mucosal function.Integrative analysis showed that microbial taxa associated with host proteinsinvolved in maintaining function of the mucous barrier, microvilli adhesion, andexocrine pancreas were depleted in patients with new-onset type 1 diabetes.

CONCLUSIONS

These data support that patients with type 1 diabetes have increased intestinalinflammation and decreased barrier function. They also confirmed that pancre-atic exocrine dysfunction occurs in new-onset type 1 diabetes and show for thefirst time that this dysfunction is present in high-risk individuals before diseaseonset. The data identify a unique type 1 diabetes–associated signature in stool thatmay be useful as a means to monitor disease progression or response to therapiesaimed at restoring a healthy microbiota.

Type 1 diabetes is caused by T-cell–mediated destruction of insulin-producing cellsand is influenced by both genetic susceptibility and the environment. Dysbiosis ofthe gut microbiota has been implicated in disease pathogenesis of many immune-mediated diseases, including type 1 diabetes (1). The composition of the gutmicrobiota is shaped by both genetics and environmental agents, such as diet andhygiene (2,3). Environmental factors such as infant diet and exposure to microbial

1The University of Queensland Diamantina In-stitute, The University of Queensland, Transla-tional Research Institute, Brisbane, Queensland,Australia2Translational Research Institute, Brisbane,Queensland, Australia3Barbara Davis Center for Childhood Diabetes,University of Colorado Denver, Aurora, CO

Corresponding author: Emma E. Hamilton-Williams, [email protected].

Received 10 April 2018 and accepted 16 July2018.

This article contains Supplementary Data onlineat http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc18-0777/-/DC1.

P.G.G. and J.A.M. contributed equally to thiswork.

J.A.M. is currently affiliated with AgResearch,Grasslands Research Centre, Palmerston North,New Zealand.

K.-A.L.C. is currently affiliated with MelbourneIntegrative Genomics, School of Mathematicsand Statistics, University ofMelbourne, Parkville,Victoria, Australia.

M.M.H. is currently affiliated with QIMRBerghofer Medical Research Institute, Brisbane,Queensland, Australia.

D.Z. is currently affiliated with Innate Biotech-nologies, LLC, Denver, CO.

© 2018 by the American Diabetes Association.Readers may use this article as long as the workis properly cited, the use is educational and notfor profit, and the work is not altered. More infor-mation is available at http://www.diabetesjournals.org/content/license.

Patrick G. Gavin,1 Jane A. Mullaney,1

Dorothy Loo,2 Kim-Anh Le Cao,1

Peter A. Gottlieb,3 Michelle M. Hill,1

Danny Zipris,3 and

Emma E. Hamilton-Williams1

2178 Diabetes Care Volume 41, October 2018

PATH

OPHYSIOLO

GY/COMPLICATIONS

agents have been linked to type 1 di-abetes risk (4). How changes in the gutmicrobiota alter the intestinal environ-ment and immune function in type 1diabetes is unknown.Disparate findings among studies

have not led to a clear consensus of whatconstitutes a type 1 diabetes–associatedmicrobiota (5). Dysbiosis of the gut mi-crobiota is hypothesized to exacerbategut inflammation before disease onset(1). Evidence suggests that individualswith islet autoimmunity have increasedintestinal permeability (6), and studiesof duodenal biopsy specimens from in-dividuals with type 1 diabetes have foundevidence of mild inflammation (7,8).An understanding of the functional prop-erties of the microbiota and the inter-actions between the gut microbes andthe host is needed to provide insight intothe role of the intestinal microbiome intype 1 diabetes development. Metapro-teomics of stool proteins is a relativelynew approach that moves beyond com-positional analysis of the microbiota toprovide a noninvasive means to assayintestinal function and examine host-microbiota relationships (9).In this study, we performed a meta-

proteomic analysis of stool samples from101 individuals, including patients withnew-onset type 1 diabetes, high- or low-risk first-degree relatives (FDRs), andhealthy control subjects. We have un-covered a network of human and micro-bial stool proteins linked to epithelialbarrier function as well as exocrine pan-creas that is disturbed both in patientswith new-onset type 1 diabetes and inhigh-risk individuals with ongoing isletautoimmunity and linked these changesto specific microbial taxa.

RESEARCH DESIGN AND METHODS

Subject CharacteristicsAll subjects resided in the Denver, Colo-rado, area. Stool samples were collectedeither at home or at the Barbara DavisCenter for Diabetes as previously de-scribed (10). Samples collected at homewere stored at 220°C before deliveryto the center on ice. Individuals withgastrointestinal disease or known infec-tions or who had taken antibiotics in the4 weeks prior were excluded. Subjectcharacteristics are summarized in Table 1.Subjects included patients with newlydiagnosed type 1 diabetes (,6 months,new-onset patients) recruited during

their visits to the Barbara Davis Centerfor Childhood Diabetes. Individuals withone to four autoantibodies and seroneg-ative FDRs of patients with type 1 di-abetes were recruited from the Type 1Diabetes TrialNet Pathway to Preventionstudy. Individuals without any family his-tory of autoimmunity were recruited fromUniversity of Colorado employees andchildren of employees. Three seronega-tive subjects and one seropositive subjectwere vegetarian, and one seronegativesubject, one seropositive subject, andtwo new-onset patients consumed agluten-free diet. All subjects gave in-formed consent according to the studyprotocol approved by the institutionalreview board, University of Colorado Den-ver. The study was approved as an an-cillary study by Type 1Diabetes TrialNet.

Protein Extraction and DigestionSamples were resuspended in potassiumphosphate buffer containing proteaseinhibitors (phenylmethylsofonyl fluoride,leupeptin, aprotinin) before centrifuga-tion to remove debris. Total protein fromthe supernatant was deglycosylated withPNGase F (Promega, Madison, WI). Pro-teins were solubilized, reduced, and alkyl-ated followed by digestion with modifiedpig trypsin (Promega). Detergents wereprecipitated and salts removed using C18tips (Glygen, Columbia, MD). Tryptic pep-tides were resuspended in 0.1% formicacid for mass spectrometry analysis.

Mass SpectrometryTryptic peptides were used for liquidchromatography–tandem mass spec-trometry analysis on a Q Exactive massspectrometer (Thermo Fisher Scientific,Waltham, MA). Peptides were sepa-rated on an EASY-Spray analytical column(Thermo Fisher Scientific). A 90-min gra-dient from 3 to 25% acetonitrile wasperformed over 60 min followed by 25–40% acetonitrile over 12 min and 95%acetonitrile for 15 min containing 0.1%formic acid at a flow rate of 250 nL/min.Survey mass spectrometric scans fromcharge/mass ratio 350–1,400 were ac-quired in the Orbitrap analyzer (ThermoFisher Scientific) with resolution r =70,000, automatic gain control of 33 106,and maximum injection time of 100 ms.The top 20 most intense ions were se-lected for tandem mass spectrometryanalysis with a normalized collisionenergy of 29%. Dynamic exclusion wasenabled for 30 s, with a mass exclusion

widthof10partspermillion. Secondstageof mass spectrometry was acquired witha mass/charge ratio 200–2,000 at a reso-lution r = 17,500, automatic gain control of5 3 105, and maximum injection time of55 ms.

Database Searching and ProcessingPeptide spectrum matching, proteininference, grouping, and quantitationwere performed using the MetaPro-IQstrategy (11), with exceptions as follow.The X! Tandem algorithm was imple-mented using rTANDEM, and the finaldatabase search was performed withSpectrum Mill (Agilent, Santa Clara, CA).For the final search, carbamidomethyla-tion of cysteine was included as afixed modification. Oxidation of methi-onine and deamidation of asparaginewere considered as variable modifica-tions. A maximum of two missed cleav-ages was allowed. A decoy database,prepared by reversing the search data-base with a false discovery rate of 0.01,was used. Proteins that shared one ormore peptides were grouped. Proteinswith only a single detected peptide werediscarded. One sample with ,500 pro-teins identified and 1 outlier were ex-cluded on the basis of a very high ratioof human/microbial proteins present (8SDs above the mean) (SupplementaryFig. 1). Raw spectra, peak lists, the cus-tom sequence database, and quantifi-cation results have been deposited inthe PRIDE (Proteomics Identifications)partner repository with the identifierPXD008870.

Microbiome AnalysisBacterial abundance in stool was deter-mined by 16S rRNA gene sequencing ofthe V4 variable region and was previouslydescribed (10).

Statistical AnalysesProteins detected in.40% of any singlesubject group were included in univar-iate and multivariate analyses. All pro-teins were considered for functionalpathway assignments. Human and non-human proteins were scaled to thesum of the total human or total non-human intensity, respectively. Proteinintensities were normalized by constantlog-ratio transformation, with an offsetcorresponding to theminimum detectedintensity across the data set. One-wayANOVA was performed after adjust-ment for age and sex. Correction for

care.diabetesjournals.org Gavin and Associates 2179

multiple hypotheses used the Benjaminiand Hochberg procedure. Proteins withadjusted P , 0.05 were further investi-gated with post hoc Tukey tests.Supervised multivariate models were

developed using supervised sparse par-tial least squares discriminant analysis(sPLS-DA) with the mixOmics R pack-age (12). Differences between groupswere evaluated by permutation ANOVA(PERMANOVA) using the vegan packagein R and 100,000 permutations. Gener-alized linear models were developedusing the 10 proteins with the lowestunivariate P values (combined controland seronegative subjects vs. new-onsetpatients) and evaluated using leave-one-out cross validation. Integration of proteo-mic and 16S rRNA sequencing data wereperformed using DIABLO in mixOmicswith a value of 0.5 in the design matrix(12). Relevance association networkswere plotted in Cytoscape. Pearson cor-relations and hierarchical clustering wereperformed using R. Kegg ortholog (KO)assignments of microbial proteins wereretrieved from the Integrated ReferenceCatalog of the Human Gut Microbiome(http://meta.genomics.cn/meta/home).The intensity of proteins that shared aKO were summed then normalized, fil-tered, and evaluated in the same way asindividual proteins. KOs were assignedto gut metabolic modules (GMMs) andevaluated using GOmixer (13). The lastcommon ancestor of each peptide wasdetermined using Unipept (14).

RESULTS

Metaproteomic Analysis of StoolSamplesWe used metaproteomics of fecal pro-teins to test the hypothesis that intesti-nal inflammation is associated with type 1diabetes progression and to determinewhether the functional characteristicsof the microbiota are altered in individ-uals with islet autoimmunity or clinicaldisease. To this end, we analyzed solublefecal proteins, which are enriched forhuman proteins but also contain solu-ble microbial proteins. The four subjectgroups included new-onset patients, se-ropositive individuals, seronegative FDRs,and unrelated healthy control subjects(Table 1). A total of 470 human proteinand 10,908 nonhuman protein clusterswere identified, with a mean of 1,5666305 protein clusters identified persample. Although a smaller proportionof human protein identifications weremade, human origin proteins repre-sented 52 6 22% of the total intensity.No significant difference was foundin the total number of human andmicrobial proteins identified or the pro-portion of human-to-bacterial proteinintensity present among the groups(Supplementary Fig. 1). No individualproteins correlated significantly withage or sex after adjustment for multipletesting. We then investigated whetherthe abundance of specific proteins dif-fered between individuals with and with-out ongoing islet autoimmunity.

Multivariate Analysis of Stool ProteinsDifferentiates Individuals With IsletAutoimmunity From Low-Risk andHealthy IndividualssPLS-DA was used to identify proteinsthat may discriminate the four groups(15). This analysis showed that healthycontrol subjects and seronegativeindividuals overlapped, whereas thenew-onset patients and seropositive in-dividuals separated from these groups(Supplementary Fig. 2A). A statisticalcomparison found that although over-all the four subject groups differed (P =0.014), the healthy control versus sero-negative groups did not (P = 0.08). Onthe basis of this result, we pooled thehealthy control and seronegative groupsto increase the power to detect differ-ences in individuals with ongoing isletautoimmunity from low-risk individuals.sPLS-DA of these three groups clusteredthe majority of the new-onset patientsaway from the combined control groupon component 1, whereas the majorityof the seropositive group separatedfrom the control group on both compo-nents 1 and 2 (Fig. 1A) (PERMANOVAP = 0.0048). Both microbial and humanproteins contributed to segregating thesethree groups (Fig. 1B). Two-way sPLS-DAshowed that all new-onset patients andalmost all seropositive individuals couldbe separated from the control subjects(P = 0.036 and P = 0.021) (SupplementaryFig. 2B and C). Hence, the stool meta-proteome was able to distinguish individuals

Table 1—Cohort characteristics

Cohort

Healthy control (n = 22) Seronegative (n = 29) Seropositive (n = 17) New-onset (n = 33)

SexFemale, n 8 14 9 15Male, n 14 15 8 18

Age (years), median (range) 11.5 (4.0–45.0) 12.0 (3.0–45.0) 9.0 (4.0–36.0) 11.0 (2.0–20.0)

HLA, n (%)3 alone 4 (18.2) 11 (37.9) 4 (23.5) 10 (30.3)3 plus 4 3 (13.6) 9 (31.0) 4 (23.5) 8 (24.2)4 alone 4 (18.2) 7 (24.1) 9 (52.9) 12 (36.4)x/x 11 (50.0) 2 (6.9) 0 (0.0) 3 (9.1)

Autoantibodies, n (%)0 22 (100.0) 29 (100.0) 0 (0.0) 1 (3.0)1 0 (0.0) 0 (0.0) 3 (17.6) 12 (36.4)2–4 0 (0.0) 0 (0.0) 14 (82.4) 20 (60.6)

HbA1c (%), median (range) d d 4.9 (4.2–5.3) 7.9 (5.5–15.0)

Age at onset (years), median (range) d d d 10.0 (2.0–20.0)

Disease duration (weeks), median (range) d d d 5.1 (0.3–17.3)

Impaired glucose metabolism, n (%) 4 (23.5)

2180 Intestinal Metaproteomics in Type 1 Diabetes Diabetes Care Volume 41, October 2018

Figure 1—The fecal meta-proteome distinguishes individuals with islet autoimmunity and new-onset (NO) patients from control subjects. A: sPLS-DAof all human and microbial proteins comparing NO, seropositive (SP), and combined healthy control and seronegative (CO/SN) groups. B: Component1 contributing variables. Red text, human proteins; black text, microbial proteins; 158742018-stool1_revised_scaffold9355_1_gene110855abbreviated as gene11085. C: Log-normalized intensity of significant proteins comparing NO, SP, and CO/SN subject groups by one-way ANOVA(adjusted P, 0.05). Data are mean6 SEM. D and E: sPLS-DA of human and microbial proteins, respectively, comparing NO, SP, and CO/SN subjectgroups. A PERMANOVA comparing the three groups is shown. *P , 0.05; **P , 0.01; ***P , 0.001.

care.diabetesjournals.org Gavin and Associates 2181

with islet autoimmunity from low-riskhealthy individuals in this cohort.

Identification of Individual HumanProteins in Stool That Differ AmongSubject GroupsWe next assessed whether individualhuman proteins distinguish subjects withdisease or islet autoantibody positivityfrom control subjects. Comparison ofthe new-onset, seropositive, and thecombined control groups by one-wayANOVA (Supplementary Table 1) re-sulted in 24 proteins with a P , 0.05,with seven significant after multipletesting adjustment (galectin-3, fibrillin-1,CUZD1, neutral ceramidase, CELA3A, CLCA1,and IGHA1) (Fig. 1C). Galectin-3 andfibrillin-1 were increased in new-onset pa-tients,whereasCUZD1,neutralceramidase,CELA3A, CLCA1, and IGHA1 were de-creased in new-onset patients and/orseropositive subjects compared with con-trol subjects. sPLS-DA analysis showedthat the three subject groups could bedifferentiated by using human proteinsalone (P = 0.007) (Fig. 1D). The differen-tially abundant proteins included threeproduced by the exocrine pancreas(CELA3A, CUZD1, and neutral ceramidase).The CELA3A protein was grouped withthe highly homologous CELA3B, and thetwo are both known as pancreatic elas-tase 1. Other differentially abundant pro-teins were involved with IgA antibodyproduction (IGHA1), inflammation andneutrophil activation (galectin-3), extra-cellular matrix (fibrillin-1), tumor growthfactor-b (TGF-b) bioavailability (fibrillin-1),and goblet cell mucous production (CLCA1).These observations confirm the findingsof others that exocrine pancreas outputis decreased in individuals with type 1diabetes (16). The current findings sug-gest for the first time in our knowledgethat reduced output of exocrine enzymesbegins in islet autoantibody–positiveindividuals. Increased abundance ofgalectin-3 and fibrillin-1 in new-onset pa-tients raises the hypothesis that proteinsassociated with inflammation are releasedinto the stool in type 1 diabetes.

Microbial Protein Identification in theStool of Individuals With IsletAutoimmunityWe next identified individual microbialproteins associated with type 1 diabe-tes. Although 34 proteins had a P, 0.01and 143 proteins had a P , 0.05, nonewere significant after multiple testing

adjustment (Supplementary Table 2).The five microbial proteins with thelowest unadjusted P values (P , 0.001)are shown in Supplementary Fig. 3. TheKO assignments for these five proteinswere phosphotransferase system, sugar-specific enzyme IIA component [EC:2.7.1.-],elongation factor thermo unstable,basic membrane protein A and relatedproteins, unassigned and ferredoxin hy-drogenase [EC:1.12.7.2]. The taxonomicassignments of these five proteins wereFaecalibacterium prausnitzii, Clostridiales,and unassigned. Although individualmicrobial proteins did not significantlydifferentiate subject groups, multivari-ate sPLS-DA using microbial proteins onlywas able to separate new-onset, sero-positive, and control subjects (P = 0.007)(Fig. 1E). These results indicate thatcollectively, bacterial-derived proteinsdiscriminate subjects with islet autoim-munity from control subjects.

Functional Changes in the StoolMetaproteome May Be Associated WithIslet AutoimmunityTo probe the biological pathways as-sociated with the microbial proteinsdriving the differences observed, wecollapsed the microbial proteins byKO. However, one-third of microbialproteins identified did not have an as-signed KO. Of the 545 KOs represented,none were significantly different be-tween groups after adjusting for multi-ple testing (Supplementary Table 3).sPLS-DA showed that KO drivers wereclose to differentiating subject groups(P = 0.06) (Supplementary Fig. 4). Asan alternative approach, the packageGOmixer was used to infer GMMsassociated with the identified microbialproteins. A comparison of new-onset pa-tients and combined control subjectsshowed that mucin degradation was in-creased in the new-onset patients (P =0.02). A comparison of seropositive indi-viduals with combined control subjectsshowed that five GMMs had a P , 0.05,including butyrate production throughtransferase (P = 0.023), which was re-duced in seropositive individuals. All iden-tified GMMs are shown in SupplementaryTable 4. Although none of these differ-ences were significant after adjustingfor multiple testing, they suggest thatmucin degradation and butyrate pro-duction may be altered in individualswith islet autoimmunity.

Integrative Network Analysis RevealsCorrelations Between DifferentiallyExpressed Proteins and Bacterial TaxaAssociated With Disease RiskWe next performed an integrative anal-ysis to explore correlations between thehuman and microbial proteins and bac-terial taxa that best discriminated eachsubject group with other highly corre-lated features in the data by using thenewly developed DIABLO method (12).The abundance of bacterial taxa waspreviously determined by 16S rRNAgene sequencing analysis (10). We firstcompared combined control subjectswith new-onset patients, and the selectedfeatures and similarity scores were usedto generate a similarity matrix visualizedin a network diagram (Fig. 2A). This anal-ysis resulted in three clusters, the firstdriven by Prevotella, which was nega-tively correlated with Bacteroides anda cluster of human proteins, includingadhesion molecules CDHR5 and CDH1and brush border enzymes MGAM andNAALADL1. The second cluster was drivenby Alistipes, which was positively corre-lated with mucous layer proteins MUC2and FCGBP and adhesion moleculeCEACAM5 and a cluster of exocrinepancreas–produced proteins (CUZD1,CELA3A, GP2, and neutral ceramidase)but negatively correlated with fibrillin-1and a cluster of heavy- and light-chainantibodyvariableregions.BecauseAlistipeswas previously found to be significantlymore abundant in seronegative indi-viduals compared with new-onset pa-tients (10), these data suggest thatAlistipesmay promote mucous produc-tion and a healthy epithelial barrier,whereas the presence of certain anti-body specificities, which were abundantin new-onset patients, may oppose thisinteraction.

We then repeated the DIABLO analysisby comparing combined control subjectswith seropositive individuals, resulting intwo clusters (Fig. 2B). One cluster againcontained Prevotella, which negativelycorrelated with Bacteroides, gobletcell–produced CLCA1, and two antibodycomponents. The second cluster linkeda group of F. prausnitzii–derived proteinswith MUC2 and Clostridium. This clusterpositively correlated with the adhesionmolecule CDHR5 and brush border en-zymes MGAM and NAALADL1. MUC2also was linked to Ruminococcus, whichwas positively correlated with adhesion

2182 Intestinal Metaproteomics in Type 1 Diabetes Diabetes Care Volume 41, October 2018

molecules CDHR2 and CEACAM5 and exo-crine pancreas proteins CELA3A, CUZD1,GP2, and DPEP1. These data suggest thatproteins that were significantly increasedin control subjects (e.g., CELA3A, CUZD1)are positively correlated with proteinsand taxa that have been previously linkedwith gut health, such as MUC2 andF. prausnitzii (17).We next calculated Pearson correla-

tion coefficients for each pair of proteinsor taxa selected in the two DIABLO net-works by using all subject group dataand grouped these using hierarchal clus-tering (Supplementary Fig. 5). This clus-tering analysis clearly identified threemajor groupings linked to 1) Prevotella,2) Alistipes and F. prausnitzii, and 3)Bacteroides. This showed that Alistipesand F. prausnitzii proteins both posi-tively correlated with mucin-layer com-ponents, adhesion molecules, and exocrinepancreas–derived proteins, suggestingthat these taxa may occupy similar func-tional niches. We conclude from thesedata that proteins and taxa that are

increased in healthy control subjectsand seronegative individuals are asso-ciated with functions and taxa linked toanti-inflammatory effects and improvedbarrier function.

Evaluation of the DiagnosticPerformance of Stool Proteins asBiomarkersFinally, we assessed how the top 10differentially abundant proteins per-formed as a potential biomarker sig-nature. A general linear model wasdeveloped using combined control sub-jects versus new-onset patients. Receiveroperating characteristic analysis withleave-one-out cross validation resultedin an area under the curve of 0.85 (Fig.3A). We used this model to predict theprobability of seropositive individualsidentifying as developing new-onset di-abetes (Fig. 3B) and found that sero-positive individuals were significantlymore likely to be identified as new-onsetthan seronegative or healthy control in-dividuals (P , 0.001). Our model did

not significantly correlate with sex (P =0.22) or HbA1c in new-onset patients (P =0.968), indicating that it was unlikelyto have been induced by dysglycemia.Of note, a significant negative correla-tion with age was found (SupplementaryFig. 6A). However, if the older subjectswere excluded, the correlation with agewas lost (Supplementary Fig. 6B), butthe model still significantly distinguishedthe subject groups (Supplementary Fig.6C). These results demonstrate the po-tential diagnostic value of stool proteinsignatures in type 1 diabetes risk strat-ification and further suggest involvementof intestinal/stool proteins in diseaseprogression.

CONCLUSIONS

We have used metaproteomics to func-tionally characterize human and micro-bial fecal proteins linked to type 1diabetes in the largest study of itskind to our knowledge. We have madethe following observations: 1) Humanand microbial proteins present in stool

Figure 2—Integrative multivariate analyses identify multiomics signatures associated with features that discriminate new-onset (NO) patients orseropositive (SP) individuals from seronegative (SN) and healthy control (CO) individuals. DIABLO method was used to integrate three data sets:microbial taxonomic abundance data, microbial protein abundance, and human protein abundance. This model was used to identify features highlycorrelated with variables that discriminate NO patients from combined CO/SN subjects (A) and SP individuals from CO/SN subjects (B). Similaritymatrix among the identified features were obtained from DIABLO and represented using relevance networks. Relative bacterial abundance of eachmicrobial taxon is shown in box-and-whisker plots. Microbial proteins have been manually grouped according to taxonomic identity.

care.diabetesjournals.org Gavin and Associates 2183

appear able to discriminate individualswith ongoing islet autoimmunity fromlow-risk and healthy individuals; 2) bothpatients with new-onset diabetes andislet autoantibody–positive individualshave reduced markers of exocrine pan-creas output compared with control sub-jects; 3) proteins linked to inflammationare increased in abundance in the stool ofpatients with new-onset type 1 diabetes;and 4) a functional relationship existsbetween proteins and bacterial taxa,which are reduced in new-onset patientsand seropositive individuals with muco-sal barrier function and exocrine pan-creas output.Of the significantly altered human

proteins, several had known immuno-modulatory functions. Fibrillin-1 formsa key constituent of extracellular matrixmicrofibrils where it stores latent TGF-b,reducing TGF-b activity and signaling(18). Because regulatory T-cell inductionin the intestine is TGF-b dependent (19),increased fibrillin-1 may reduce regula-tory T-cell frequency. In addition, fibrillin-1plays a role in macrophage chemotaxis(20). Galectin-3 is believed to be a proin-flammatory mediator in a wide rangeof inflammation-associated disorders(21) and is chemotactic for inflamma-tory cells through reactive oxygen spe-cies production (22). Overproduction ofthese proteins supports our hypothesisthat increased intestinal inflammation

is associated with type 1 diabetes path-ogenesis. However, the source of theseinflammatory proteins is unknown andmay be intestinal or pancreatic.

The abundance of the fecal inflam-mation marker calprotectin (comprisingS100A8 and S100A9) was not altered.Calprotectin is produced by activatedneutrophils (e.g., as in active inflamma-tory bowel disease). Mild intestinal in-flammation has been reported in studiesof duodenal and jejunal biopsy speci-mens from patients with type 1 diabetes(7,23). This included overexpression ofHLA class II, ICAM-1 in the epithelium,and interleukin-1a and -4 in the laminapropria (7) and increased infiltration ofmacrophages and monocytes (8). In-creased neutrophilic infiltration hasnot been described in the type 1 diabetesgut, which may explain the unchangedcalprotectin. Studies of biopsy samplesfrom the ileum or colon in type 1 diabe-tes have not been reported; therefore,the involvement of inflammation in thesetissues is unknown.

Although IGHA1 (IgA1 heavy-chainconstant region) was reduced in sero-positive individuals, several Ig variableregions contributed to the varianceamong the groups. Some studies havereported fecal IgA levels to be lower inpatients with type 1 diabetes (24),whereas others found no difference(25,26). The reduced soluble IgA levels

possibly indicate increased coating ofbacteria by IgA, as reported by others(27). Alternatively, reduced IgA levelscould be related to the reduction inbacterial diversity reported to precedethe onset of clinical disease in high-riskindividuals (28).

In addition to immunomodulatory pro-teins, exocrine pancreas proteins weredecreased in new-onset patients andseropositive individuals. Both ultrasoundimaging and fecal pancreatic elastase-1measurement have found a deficiency ofexocrine pancreas function in patientswith type 1 diabetes (16,29). Consistentwith our findings, fecal elastase wasreported as decreased in new-onsetpatients and continued to drop withdisease duration (29). Fecal elastasepositively correlated with C-peptideand negatively correlated with HbA1c(29). Our demonstration of reducedfecal elastase in islet antibody–positiveindividuals suggests that the processof exocrine dysfunction begins beforediagnosis. The causes of exocrine pan-creas dysfunction are unknown, withmultiple hypotheses proposed (16).Studies have suggested a possible rolefor immune involvement in exocrinepancreas abnormality (30). Of note, ourintegrative analysis suggests that a re-lationship exists between reduced exo-crine pancreas function in type 1 diabetesand the gut microbiota.

Figure 3—Diagnostic biomarker performance of a multivariate prediction model. A: Receiver operating characteristic curve of cross-validated linearmodels. Each sample was classified as combined control and seronegative (CO/SN) or new-onset (NO) using a generalized linear model developedwithout that sample. B: Probability of being classified as NO. Probabilities were determined using a linear model developed on all CO/SN subjectsand NO patients. The 10 proteins selected for this model were MH0003_GL0005194, T2D.17A_GL0015017, V1.FI37_GL0010556, galectin-3,MH0133_GL0027226, fibrillin-1, CELA3A, neutral ceramidase, MH0012_GL0028086, and V1.UC30.0_GL0016764. NS, not significant. ***P , 0.001;****P , 0.0001. SP, seropositive.

2184 Intestinal Metaproteomics in Type 1 Diabetes Diabetes Care Volume 41, October 2018

By using multivariate integrative analy-sis, we found that an increased abundanceof exocrine pancreas proteins correlateswith proteins associated with a healthyepithelial barrier (e.g., MUC2, FCGBP,CDHR5, CDH1) (31,32) and with Alistipes,which was significantly increased in se-ronegative individuals (10). This health-associated cluster contained butyrateproducers F. prausnitzii and ClostridiumaswellasAlistipes,Ruminococcus,Barnesiella,andDorea. Colonicmicrobial-produced bu-tyrate is known to exert several bene-ficial effects (33). A recent meta-analysisof 3,048 data sets found that Barnesiellaand Alistipes were universally associatedwith health (34).Inantagonism to thehealth-associated

cluster were Bacteroides- and Prevotella-associated clusters. Several studies havereported an increased abundance ofBac-teroides species either in islet autoantibody–positive individuals or around diseaseonset (26,35,36). Prevotella is knownto predominate in individuals who con-sume a plant-rich diet, whereas Bac-teroides is associated with high-fat,high-protein western diets (3). Despitethis association of Prevotella with a plant-based diet, many studies have foundan increased abundance of Prevotellaspecies in inflammatory diseases (37).These seemingly contradictory find-ings may be due to the large diversitywithin the Prevotella genus. The cur-rent analyses do not support a protectiverole for Prevotella in preventing progres-sion to islet autoimmunity or type 1diabetes.Seropositive individuals can be classi-

fied into disease stages by number of isletautoantibodies and the presence of dys-glycemia (38). Stage 1 is two or moreautoantibodies, stage 2 is two or moreautoantibodies and dysglycemia, andstage 3 is hyperglycemia. Of the 17 sero-positive individuals in this study, 3 hadone autoantibody; however, 1 of thesehad dysglycemia and could not be stagedby this method, and another 3 were stage2. Because of the low numbers, we wereunable to determine whether our modelfitted with disease stage. Longitudinalstudies will be needed to determine de-finitively whether proteins in stool pre-dict disease progression.Although we have detected a striking

disease-associated signature in our sam-ples, the differences in protein abun-dance need to be validated in a larger,

independent cohort. Furthermore, wedid not measure C-peptide, so we werenot able to make correlations betweenb-cell function and stool proteins. Shot-gun proteomics is limited by a lack ofsensitivity for detecting low-abundanceproteins and by low reproducibility com-pared with other methods. Therefore,the changes we have observed shouldbe validated with other methods. Areduction in CUZD1 and CELA3A in pa-tients with type 1 diabetes has beenreported in other studies (39,40), whichsuggests that the current findings arereproducible. Another limitation wasthat we analyzed only soluble stoolproteins. Although this analysis resulted inan enrichment of human proteins, itbiased the microbial protein detectiontoward those that are secreted or re-leased from lysis-prone bacteria, suchas gram-negative species. In addition,many identified bacterial proteins wereuncharacterized and could not be in-cluded in our functional analyses. Thesubjects in our study were of a relativelywide age range. Because disease path-ogenesis mechanisms may vary betweenchildrenandadults, itwill benecessary tovalidate our findings separately in thesepopulations. We also did not collect de-tailed information on dietary intake,which may be an additional confounderto our data.

Despite these limitations, we were ableto describe novel associations amonggut-, pancreas-, and microbiota-derivedproteins related to type 1 diabetes.These studies identify a negative corre-lation between the presence of specificstool antibodies and taxa associatedwith gut health and demonstrate anassociation between the abundance ofexocrine pancreatic proteins and anti-inflammatory microbial taxa and mucosalbarrier function. This study establishesfor the first time in our knowledge theutility of using stool-based metaproteo-mics to link changes in the gut micro-biota with the health of the intestinalmucosa and exocrine pancreas in type 1diabetes.

Acknowledgments. The authors thank theTranslational Research Institute ProteomicsCore facility, the individuals who participatedin this study, Cini James (The University ofQueensland, Brisbane, Australia) and CaseyWright (The University of Queensland, Brisbane,Australia) for technical assistance, Nicholas

Matigian (The University of Queensland, Brisbane,Australia) for statistical advice, and RanjenyThomas (The University of Queensland, Brisbane,Australia) and Mark Harris (Lady Cilento Chil-dren’s Hospital, Brisbane, Australia) for helpfuldiscussions.Funding. J.A.M. was funded by a JDRF post-doctoral fellowship (PDF-2014-222-A-N). Thestudy was funded by JDRF grant 2-SRA-2015-306-Q-R (to D.Z.). E.E.H.-W. is funded by a JDRFcareer development fellowship (2-2013-34).Type 1 Diabetes TrialNet is funded by Na-tional Institute of Diabetes and Digestive andKidney Diseases grant U01-DK-85509.Duality of Interest. No potential conflicts ofinterest relevant to this article were reported.Author Contributions. P.G.G. performed thedata analysis, statistical analysis, and visualizationand revised the manuscript. J.A.M. designedthe approach, developed the methodology,performed experiments, performed the dataanalysis, supervised students, and revisedthe manuscript. D.L. performed experimentsand provided technical advice. K.-A.L.C. providedstatistical oversight. P.A.G. and D.Z. providedresources and revised the manuscript. M.M.H.provided technical advice, supervised students,and revised the manuscript. E.E.H.-W. concep-tualized and led the project, designed experi-ments, supervised students, and wrote themanuscript. E.E.H.-W. is the guarantor for thiswork and, as such, had full access to all the datain the study and takes responsibility for theintegrity of the data and the accuracy of thedata analysis.Data Availability. The mass spectrometry pro-teomics data have been deposited to the Pro-teomeXchange Consortium through the PRIDEpartner repository with the data set identifierPXD008870.

References1. Atkinson MA, Chervonsky A. Does the gutmicrobiota have a role in type 1 diabetes? Earlyevidence from humans and animal models ofthe disease. Diabetologia 2012;55:2868–28772. Mullaney JA, Stephens JE, Costello ME, et al.Type 1 diabetes susceptibility alleles are asso-ciated with distinct alterations in the gut micro-biota. Microbiome 2018;6:353. Wu GD, Chen J, Hoffmann C, et al. Linkinglong-term dietary patterns with gut microbialenterotypes. Science 2011;334:105–1084. Rewers M, Ludvigsson J. Environmental riskfactors for type 1 diabetes. Lancet 2016;387:2340–23485. Ardissone A, Kemppainen KM, Triplett EW.Type 1 diabetes and intestinal microbiota: howgeographic differences between human cohortscan influence interpretation of associations. Di-abetes Case Rep 2017;2:1286. Bosi E,Molteni L, RadaelliMG, et al. Increasedintestinal permeability precedes clinical onset oftype 1 diabetes. Diabetologia 2006;49:2824–28277. Westerholm-Ormio M, Vaarala O, Pihkala P,Ilonen J, Savilahti E. Immunologic activity in thesmall intestinal mucosa of pediatric patients withtype 1 diabetes. Diabetes 2003;52:2287–22958. Pellegrini S, Sordi V, Bolla AM, et al. Duodenalmucosa of patients with type 1 diabetes showsdistinctive inflammatory profile and microbiota.J Clin Endocrinol Metab 2017;102:1468–1477

care.diabetesjournals.org Gavin and Associates 2185

9. Lichtman JS, Sonnenburg JL, Elias JE. Moni-toring host responses to the gut microbiota.ISME J 2015;9:1908–191510. Alkanani AK, Hara N, Gottlieb PA, et al. Al-terations in intestinal microbiota correlatewith susceptibility to type 1 diabetes. Diabetes2015;64:3510–352011. Zhang X, Ning Z, Mayne J, et al. MetaPro-IQ:a universal metaproteomic approach to studyinghuman and mouse gut microbiota. Microbiome2016;4:3112. Rohart F, Gautier B, Singh A, Le Cao KA.mixOmics: an R package for ‘omics feature selec-tion and multiple data integration. PLOS ComputBiol 2017;13:e100575213. Darzi Y, Falony G, Vieira-Silva S, Raes J.Towards biome-specific analysis of meta-omicsdata. ISME J 2016;10:1025–102814. Mesuere B, Devreese B, Debyser G, AertsM,Vandamme P, Dawyndt P. Unipept: tryptic peptide-based biodiversity analysis of metaproteomesamples. J Proteome Res 2012;11:5773–578015. Le Cao KA, Boitard S, Besse P. Sparse PLSdiscriminant analysis: biologically relevant fea-ture selection and graphical displays for multi-class problems. BMC Bioinformatics 2011;12:25316. Hardt PD, Ewald N. Exocrine pancreatic in-sufficiency in diabetes mellitus: a complicationof diabetic neuropathy or a different type ofdiabetes? Exp Diabetes Res 2011;2011:76195017. Sokol H, Pigneur B, Watterlot L, et al. Fae-calibacterium prausnitzii is an anti-inflammatorycommensal bacterium identified by gut micro-biota analysis of Crohn disease patients. ProcNatl Acad Sci U S A 2008;105:16731–1673618. Zeyer KA, Reinhardt DP. Fibrillin-containingmicrofibrils are key signal relay stations for cellfunction. J Cell Commun Signal 2015;9:309–32519. Coombes JL, Siddiqui KR, Arancibia-CarcamoCV, et al. A functionally specialized populationof mucosal CD103+ DCs induces Foxp3+ regulatoryT cells via a TGF-beta and retinoic acid-dependentmechanism. J Exp Med 2007;204:1757–176420. GuoG,BoomsP,HalushkaM, et al. Inductionof macrophage chemotaxis by aortic extracts of

the mgR Marfan mouse model and a GxxPG-containing fibrillin-1 fragment. Circulation 2006;114:1855–186221. deOliveiraFL,GattoM,BassiN,etal.Galectin-3in autoimmunity and autoimmune diseases. ExpBiol Med (Maywood) 2015;240:1019–102822. Madrigal-Matute J, Lindholt JS, Fernandez-Garcia CE, et al. Galectin-3, a biomarker linkingoxidative stress and inflammation with the clin-ical outcomes of patients with atherothrombo-sis. J Am Heart Assoc 2014;3:e00078523. Badami E, Sorini C, CocciaM, et al. Defectivedifferentiation of regulatory FoxP3+ T cells bysmall-intestinal dendritic cells in patients withtype 1 diabetes. Diabetes 2011;60:2120–212424. Lassenius MI, Fogarty CL, Blaut M, et al.;FinnDiane Study Group. Intestinal alkaline phos-phatase at the crossroad of intestinal healthand disease - a putative role in type 1 diabetes.J Intern Med 2017;281:586–60025. de Groot PF, Belzer C, Aydin O, et al. Distinctfecal and oral microbiota composition in humantype 1 diabetes, an observational study. PLoSOne 2017;12:e018847526. de Goffau MC, Luopajarvi K, Knip M, et al.Fecal microbiota composition differs betweenchildren with b-cell autoimmunity and thosewithout. Diabetes 2013;62:1238–124427. Palm NW, de Zoete MR, Cullen TW, et al.Immunoglobulin A coating identifies colitogenicbacteria in inflammatory bowel disease. Cell2014;158:1000–101028. Kostic AD, Gevers D, Siljander H, et al.;DIABIMMUNE Study Group. The dynamics ofthe human infant gut microbiome in develop-ment and in progression toward type 1 diabetes.Cell Host Microbe 2015;17:260–27329. Cavalot F, Bonomo K, Perna P, et al. Pan-creatic elastase-1 in stools, a marker of exocrinepancreas function, correlates with both residualbeta-cell secretion and metabolic control intype 1 diabetic subjects. Diabetes Care 2004;27:2052–205430. Rodriguez-Calvo T, Ekwall O, Amirian N,Zapardiel-Gonzalo J, von Herrath MG. Increased

immune cell infiltration of the exocrine pancreas:a possible contribution to the pathogenesis oftype 1 diabetes. Diabetes 2014;63:3880–389031. Crawley SW, Shifrin DA Jr., Grega-Larson NE,et al. Intestinal brush border assembly driven byprotocadherin-based intermicrovillar adhesion.Cell 2014;157:433–44632. Pelaseyed T, Bergstrom JH, Gustafsson JK,et al. The mucus and mucins of the goblet cellsand enterocytes provide the first defense line ofthe gastrointestinal tract and interact with theimmune system. Immunol Rev 2014;260:8–2033. Furusawa Y, Obata Y, Fukuda S, et al. Com-mensal microbe-derived butyrate induces thedifferentiation of colonic regulatory T cells. Na-ture 2013;504:446–45034. Mancabelli L, Milani C, Lugli GA, et al. Iden-tification of universal gut microbial biomarkersof common human intestinal diseases by meta-analysis. FEMS Microbiol Ecol 2017;93:fix15335. Davis-Richardson AG, Ardissone AN, Dias R,et al. Bacteroides dorei dominates gut micro-biome prior to autoimmunity in Finnish childrenat high risk for type 1 diabetes. Front Microbiol2014;5:67836. Murri M, Leiva I, Gomez-Zumaquero JM,et al. Gut microbiota in children with type 1diabetes differs from that in healthy children:a case-control study. BMC Med 2013;11:4637. Larsen JM. The immune response to Pre-votella bacteria in chronic inflammatory disease.Immunology 2017;151:363–37438. Insel RA, Dunne JL, Atkinson MA, et al.Staging presymptomatic type 1 diabetes: a sci-entific statement of JDRF, the Endocrine Society,and the American Diabetes Association. DiabetesCare 2015;38:1964–197439. Pinto E, Anselmo M, Calha M, et al. Theintestinal proteome of diabetic and control chil-dren is enriched with different microbial and hostproteins. Microbiology 2017;163:161–17440. Heintz-Buschart A, May P, Laczny CC, et al.Integrated multi-omics of the human gut micro-biome in a case study of familial type 1 diabetes.Nat Microbiol 2016;2:16180

2186 Intestinal Metaproteomics in Type 1 Diabetes Diabetes Care Volume 41, October 2018