bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

1 Type 1 Diabetes: an Association Between Autoimmunity the Dynamics of Gut Amyloid- 2 producing E. coli and Their Phages

3

4 George Tetz1,2*, Stuart M. Brown3, Yuhan Hao4,5, Victor Tetz 5 6 1Human Microbiology Institute, New York, NY 10013, USA. 7 2Tetz Laboratories, New York, NY 10027, USA. 8 3 New York University School of Medicine, Department of Cell Biology, New York, NY, 10016, 9 USA 10 4 Center for Genomics and Systems Biology, New York University, New York, New York 11 10012, USA 12 5 New York Genome Center, New York, New York 10013, USA. 13 14 15 *Correspondence and requests for materials should be addressed to G.T. 16 (email: [email protected]) 17

1 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

18 Summary 19 The etiopathogenesis of type 1 diabetes (T1D), a common autoimmune disorder, is not 20 completely understood. Recent studies suggested the gut microbiome plays a role in T1D. We 21 have used public longitudinal microbiome data from T1D patients to analyze amyloid- 22 producing bacterial composition and found a significant association between initially high 23 amyloid-producing Escherichia coli abundance, subsequent E. coli depletion prior to 24 seroconversion, and T1D development. In children who presented seroconversion or developed 25 T1D, we observed an increase in the E. coli phage/E. coli ratio prior to E. coli depletion, 26 suggesting that the decrease in E. coli was due to prophage activation. Evaluation of the role of 27 phages in amyloid release from E. coli biofilms in vitro suggested an indirect role of the 28 bacterial phages in the modulation of host immunity. 29 This study for the first time suggests that amyloid-producing E. coli, their phages, and bacteria- 30 derived amyloid might be involved in pro-diabetic pathway activation in children at risk for 31 T1D. 32 33 Keywords: E.coli, amyloid, autoimmunity, microbiome, bacteriophages, prophages 34 35 Introduction

36 Type 1 diabetes (T1D) is an autoimmune disorder driven by T cell-mediated destruction of the 37 insulin-secreting β-cells of the pancreatic islets that often manifests during childhood (Lehuen et 38 al., 2010). There are a number of factors associated with the development of T1D (Atkinson and 39 Eisenbarth, 2001; Soltesz et a., 2007; Rewers, M., and Ludvigsson, 2016). The strongest 40 susceptibility alleles for T1D encode certain human leukocyte antigens (HLAs), which play a 41 central role in autoreactive T-cell activation in T1D pathogenesis (Barrett et al., 2009; Ikegami et 42 al., 2011). Only a portion (3–10%) of children carrying HLA risk alleles will develop T1D, 43 pointing out the role of environmental factors in disease initiation; however, the role of 44 environmental factors in T1D etiology remains elusive (Achenbach et al., 2005; Noble and Valdes, 45 2007; Rewers, M., and Ludvigsson, 2016).

46 T1D is generally associated with a long pre-diabetic seroconversion period, during which 47 autoantibodies to antigens of pancreatic β-cells or insulin are produced (Vehik et al., 2011). There 48 are a few known factors that drive autoimmunity during infancy, such as so-called “natural 49 apoptosis” or spontaneous cell death within the β-cell population, deposition of islet amyloid 50 polypeptide (IAPP) aggregates, microbiota alterations, or viral infection that specifically targets 51 pancreatic β-cells and leads to islet cell death, which contribute to the formation of β-cell antigens,

2 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

52 activation of dendritic cells (DC), and antigen presentation (Drescher et al., 2004; Vaarala et al., 53 2008; Eizirik et al., 2009; Lehuen et al., 2010; Todd, 2010, Paulsson et al., 2014; Knip and 54 Siljander, 2016). However, the exact roles of these mechanisms in disease onset remain poorly 55 understood (Lehuen et al., 2010).

56 Pancreatotropic viruses, such as enterovirus, coxsackie B, mumps, rubella, and 57 cytomegalovirus, have been evaluated as triggers of T1D since the discovery of antiviral antibodies 58 in diabetic patients and because of the role of viral infection in elevated levels of type I interferon 59 (IFN), which is known to play a role in T1D progression (Flodström et al., 2002; Op de Beeck and 60 Eizirik 2016; Hyöty 2016). However, in many patients who develop T1D, there is no clear link 61 with viral infections (Filippi and von Herrath, 2008). Moreover, the mimicry between β-cell 62 autoimmune antibodies and viral epitopes of coxakieviruses, which are the most strongly 63 associated with T1D, has recently been shown not to underlie T1D pathogenesis (Schloot et al., 64 2001; Harrison 2019). Natural apoptosis has been described only in certain patients and highlights 65 the gaps in our knowledge about disease initiation (Lehuen et al., 2010;). Therefore, despite 66 decades of research, the etiology of autoimmune response induction remains elusive (Harrison, 67 2019). 68 Previous studies have established a correlation between seroconversion and chronic 69 inflammation resulting from impaired gut barrier function and increased intestinal permeability, 70 which is suggested as a hallmark of T1D (Vaarala et al., 2008). Given the overarching influence 71 of gut bacteria on human health, including intestinal permeability and its regulatory role in 72 autoimmunity, the role of the microbiota in T1D development has been recently explored (Knip 73 and Siljander, 2016; Kostic et al., 2016; Paun et al., 2017). The microbiota of the human intestinal 74 tract is comprised of bacteria, fungi, and eukaryotic and bacterial viruses (bacteriophages). 75 Bacteria in the human gut live within surface-associated microbial communities termed biofilms, 76 which are characterized by the presence of self-produced extracellular matrix and a surface film 77 that protects the microorganisms from the outer environment (Costerton et al., 1999; Tetz et al., 78 2009; Flemming and Wingender, 2010; O’Toole et al., 1999). The extracellular matrix consists of 79 different biomolecules, including extracellular nucleic acids, polysaccharides, and proteins. 80 Several microorganisms within the human microbiome, predominantly members 81 of Enterobacteriaceae, also produce amyloid proteins that can form so-called “curli fibers” which 82 provide biofilms with unique mechanical properties (Chapman, 2002; Barnhart and Chapman 83 2006). Bacterial curli proteins of Enterobacteriaceae form highly ordered β-sheet rich amyloid 84 composed of a major subunit, CsgA, and a minor subunit, CsgB (Schwartz and Boles 2013). 85 Bacterial amyloids share certain similarity with human amyloids; however, whereas amyloid in 86 bacteria is suggested to be physiological, in mammals, the formation of misfolded, insoluble,

3 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

87 highly ordered prion-like cross-β structures is commonly associated with a variety of diseases 88 (Fowler et al., 2007). In humans, pathological deposition of insoluble amyloid aggregates has been 89 shown to be associated with the development of Creutzfeldt-Jakob disease, Alzheimer’s disease, 90 Parkinson’s disease, and T1D, where an increased islet amyloid polypeptide (IAPP) concentration 91 may constitute a risk factor for β-cell destruction (Paulsson et al., 2014; Collinge 2016). Moreover, 92 similarities between bacterial and eukaryotic amyloid β-sheets have been implicated in inter- 93 kingdom communication through the toll-like receptor signaling pathway, leading to the 94 deposition of misfolded protein in the brain and altered amyloidogenesis in Caenorhabditis 95 elegans and rats following colonization with amyloid-producing E. coli (Chen et al., 2016). 96 Notably, in this work, the authors showed that bacterial curli fibers triggered T-cell immunity 97 through the TLR-2-MyD88-NF-kB signaling pathway, which is also known to be associated with 98 T1D progression; however, no microbiome studies have been conducted to reveal the association 99 of amyloid-producing bacteria with diabetes (Tukel et al., 2010). Moreover, only a limited number 100 of studies on the human microbiome have revealed alterations that may be associated with T1D. 101 Kostic et al. and Zhao et al. found that an increased abundance of Blautia spp., Ruminococcus spp., 102 and OTUs belonging to the genus Bacteroides positively correlated with T1D progression, 103 suggesting that the primary pathological implication in autoimmunity might be through an altered 104 metabolite profile associated with these bacteria, including triglyceride and branched-chain amino- 105 acid production Davis-Richardson et al. 2014; Kostic et al., 2016). Notably, a previous microbiota 106 study revealed microbiome alterations that emerged only after the appearance of autoantibodies, 107 suggesting that such alterations are involved in clinical progression to T1D rather than in disease 108 initiation (Knip, and Siljander 2016). 109 There are multiple causes of gut bacterial alterations in T1D, and bacteriophages are 110 considered to be one of the most influential, but are poorly studied in this regard (Zhao et al., 111 2017). Data from a phagobiota analysis of children with T1D are even more limited, and showed 112 that alterations in Bacteroides phages are discriminative for disease status, but not clearly 113 associated with disease progression (Zhao et al., 2017). Bacteriophages as possible agents that may 114 negatively affect mammalian health have attracted scientific attention only recently (Tetz and Tetz 115 2018). We have previously shown that bacteriophages can alter the mammalian microbiota, 116 leading to increased intestinal permeability and triggering chronic inflammation, which potentially 117 participate in protein misfolding and might contribute to the onset of Parkinson’s disease Tetz et 118 al., 2017; Tetz et al., 2018 ). 119 In this study, we conducted a detailed comparative metagenomic analysis of amyloid- 120 producing gut bacteria and the associated phagobiome in a prospective, longitudinal cohort of 121 HLA-matched infants up to 3 years of age. We retrieved a dataset of short sequence reads

4 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

122 generated by Kostic et al. (2015) from the NCBI Sequence Read Archive (SRA) 123 (https://www.ncbi.nlm.nih.gov/bioproject/382085) and analyzed amyloid-producing bacteria and 124 bacteriophage diversity using MetaPhlAn and a custom method (Segata et al., 2012; Kostic et al., 125 2016). 126 This study for the first time revealed alterations in the abundance of intestinal amyloid- 127 producing bacteria during early life due to prophage activation that that correlated with the 128 development of T1D-associated autoimmunity. These data were supported by findings in vitro that 129 revealed significant amyloid release from E. coli biofilms upon induction by prophages, suggesting 130 a potential involvement of E. coli-derived curli fibers in pro-diabetic activation pathways in 131 children at risk for T1D. 132 133 Results 134 Autoimmunity Development is Associated with an Alteration in Amyloid-producing 135 Bacterial Abundance 136 To explore the potential link between the abundance of amyloid-producing bacteria and T1D- 137 associated seroconversion, we used longitudinal shotgun metagenomics sequencing data of the 138 fecal microbiome from 10 children who exhibited autoantibodies (six seroconverters and four 139 children who developed T1D) and eight non-seroconverted HLA-matched control individuals. All 140 patient characteristics were previously reported by Kostic et al. (Table S1) (Kostic et al., 2016). 141 HiSeq-2500 sequencing was used, producing an average of ~2.5 Gb per sample. Amyloid- 142 producing bacteria were represented by E. coli, Staphylococcus aureus, and Salmonella spp. 143 (Barnhart and Chapman 2006; Schwartz and Boles 2013). Among these amyloid-producing 144 bacteria, E. coli was the major group, while the other curli-producing bacteria were identified only 145 in a single sample and at a few collection times and thus, were disregarded in subsequent analysis 146 (Table S2). We first compared the association between E. coli abundance, disease phenotype, and 147 collection time using a two-tailed Mann–Whitney U test (Figure 1; Table S3). We clustered the 148 samples into time-point bins of 300 days, and studied the dynamics of E. coli abundance over 1300 149 days within and among patient groups. E. coli abundances demonstrated a different dynamic 150 relationship across time for case and control groups. E. coli tended to disappear in T1D and 151 seroconverters over the time, whereas in controls, E. coli abundance tended to increase and did not 152 change significantly over time. In both the seroconverter and T1D groups, we found a statistically 153 significant decrease in E. coli abundance when comparing the first 0-300 day period and the 154 following 300-600, 600-900, 900-1300 time bins (p < 0.05). Notably, the initial abundance of E. 155 coli was significantly higher in cases than in controls at 0–300 days. The case groups responded

5 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

156 differently over time, and importantly, data from the control group revealed that there was no 157 variation in E. coli associated with age. 158 To explore the relation between E. coli abundance and autoimmunity further, we next 159 analyzed whether alterations in the abundance of E. coli could be associated with the development 160 of autoantibodies in more detail, and whether or not E. coli abundance could distinguish the T1D 161 disease state. To this end, we studied the appearance and disappearance of E. coli in each patient 162 in dynamic association with the appearance of autoantibodies (Figure 2). We found a positive 163 correlation between the total disappearance or presumed disappearance (defined as a >50-fold 164 decrease) in E. coli and autoantibody appearance in the T1D and seroconverter groups, but not in 165 controls (Walterspiel et al., 1992). All T1D patients showed disappearance or an episode of 166 disappearance of E. coli before the detection of autoantibodies and disappearance of E. coli prior 167 to the diagnosis of diabetes (Table S4). In seroconverters, E. coli disappeared in all patients, except 168 E003989 and T013815. For patient E10629, data were difficult to interpret because of the long 169 period between sample collections (556 days between the last sample taken before seroconversion 170 and the first sample after seroconversion). In any case, E. coli was absent in the first sample 171 following seroconversion; thus, it is reasonable to assume that it had disappeared before 172 autoantibody appearance. Notably, in certain control patients, E. coli disappeared at some time 173 points; however, this was followed by a restoration of the E. coli population in all cases, which 174 was not regularly seen in case subjects. 175 Next, we examined changes in the absolute abundance of E. coli in each patient 176 individually among different periods, and we presented the data as a heatmap (Figure 3). We 177 detected alterations in E. coli abundance within children from the control group, followed by 178 restoration of the E. coli population to the levels equal to or above the pre-seroconversion levels, 179 whereas restoration of the E. coli population to preceding levels was noted in only a few children 180 from the case groups. This more detailed analysis has also revealed some previously overlooked 181 data, for patient T013815. Although in this subject, E. coli did not disappear prior to the 182 seroconversion, the appearance of autoantibodies occurred in the period when E. coli started to 183 significantly decrease from the highly elevated abundance compared to other patients. Another 184 observation that attracted our attention was that the initial abundance of E. coli in T1D and 185 seroconverter groups was significantly higher than that in the control group, which confirmed the 186 data presented in Figure 1. The result further supports that the initial higher abundance of E. coli 187 and the decrease of amyloid-producing bacteria abundance in case group were signatures 188 associated with autoimmunity and disease progression. 189 In addition, we studied the potential correlation between the disappearance of E. coli and 190 certain HLA alleles or the appearance of particular autoantibodies, IAA, GADA, IA2A, ZNT8A

6 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

191 or ICA, in case groups. No correlations were observed (data not shown). We also detected no 192 association between E. coli disappearance and breastfeeding duration, Bifidobacterium abundance, 193 or antibiotic usage, which are known to influence Enterobacteriaceae (Tables S5, S6) (Candela et 194 al., 2008). 195 196 E. coli Bacteriophages as Signatures Associated with Seroconversion 197 As we identified E. coli as T1D-discriminative bacteria, we next studied the relationships between 198 this microorganism and bacteriophages of E. coli, as phages are known as main regulators of 199 bacterial populations. First, we detected 63 E. coli phage species in case and control groups. We 200 evaluated possible differences in E. coli phages, using α-diversity, in the pre-seroconversion period 201 (Table S7). For the samples from the control group, we set the medium time to seroconversion in 202 case groups (540 days) as an artificial benchmark. Phage richness was statistically different 203 between control and seroconversion individuals as indicated by ACE and Chao1 indexes (ACE: p 204 = 0.0181; Chao1: p = 0.0137), but was not statistically significantly different between control and 205 T1D, most likely because of the small T1D patient cohort (ACE: p = 0.6086; Chao1: p = 0.6359). 206 E. coli phage diversity tended to be lower in the control subjects than in seroconverters (Shannon, 207 p = 0.1526; Simpson, p = 0.5437), and was significantly lower in controls vs. T1D cases (Shannon, 208 p = 0.0055; Simpson, p = 0.0248) (Mann–Whitney test) ( Table S7). 209 We observed that nearly all E. coli phages were lysogenic, whereas strictly virulent lytic 210 phages, such as Enterobacteria phage IME10 or Enterobacteria phage 9g, were found only in a 211 few samples (Table S2). The predominance of lysogenic E. coli phages clarifies why α-diversity 212 indices revealed trends of decreased evenness and diversity of phages in case groups compared to 213 controls in pre-seroconversion samples, which most likely reflected lower E. coli abundance in 214 controls (Table S2). 215 Next, we analyzed the E. coli phage/E. coli bacterial cell ratio, which represents the “lytic 216 potential” (Figure 4; Table S8; Table S9) (Waller et al., 2014). To this end, we first normalized 217 the E. coli phage abundance to that of E. coli. In theory, the phage/bacteria ratio reflects whether 218 or not a prophage is stably integrated within the host bacterial genome (Tetz et al., 2018). A 219 low ratio indicates that prophage is most likely absent in the genomes of part of the bacterial 220 host population, whereas a high ratio suggests active, productive phage-induced bacterial lysis. 221 The phage/bacteria ratio increased in subjects from all groups prior to the decrease in E. coli 222 abundance, indicating that productive phage infection was the cause of E. coli depletion. 223 Notably, that observation reflected a statistically significant elevation in the phage/bacteria ratio 224 before the appearance of autoantibodies in the T1D and seroconverter groups (Table S10).

7 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

225 To evaluate the implication of the increase in the phage/bacteria ratio as a driving force 226 behind E. coli depletion further, we analyzed which phages had a correlation with the depletion of 227 E.coli abundance across most subjects. The most frequently found lysogenic phages with an 228 inverse relationship with E.coli abundance are the representatives of the 229 Peduovirinae subfamily and of the unclassified Lambdavirus subfamily within the Caudovirales 230 order. The increase in number of the lysogenic E. coli phages along with the decrease in E. coli 231 abundance indicated that there was a productive bacteriophage infection that led to bacterial 232 host death and the release of phage progeny (Waller et al., 2014). These data revealed 233 previously overlooked particularities of the phagobiota in T1D, suggesting a primary role of 234 induction of certain E. coli prophages in E. coli depletion, and an association with autoimmunity 235 and T1D development. 236 237 Role of Prophage Induction in the Release of Amyloid from E. coli Biofilms 238 We evaluated how the die-off of E. coli populations due to prophage induction could lead to 239 amyloid release. We used 48-h old E. coli biofilms with confirmed curli expression (curli 240 formation on 48-h colonies was visible with the naked eye on petri dishes (data not shown). 241 Prophages were induced with mitomycin C. Lytic bacteriophage development was confirmed as 242 an increase in plaque forming units (PFU) and a reduction in colony-forming units (CFU). The 243 presence of phage in E. coli biofilm was confirmed by the production of plaques on a non-induced 244 E. coli control culture between 4 h and 10 h. The effect of prophage induction in E. coli biofilm 245 on CFU and PFU is summarized in Table 1. Phages started to appear in the biofilm as of 4 h after 246 prophage induction, with maximum PFU at 8 h, which coincided with a decrease in CFU. 247 Next, we evaluated the effect of prophage induction on the amount of amyloid fibers 248 released into the biofilm supernatant. To quantify the amount of aggregated amyloid, we evaluated 249 its binding to the amyloid-diagnostic dye Congo red (CR) (Reichhardt et al., 2015). We observed 250 a significant increase in CR absorbance upon binding to supernatant 8 h following prophage 251 induction as compared with control samples (Figure 5). Like for other curli fibers, heating of the 252 E. coli supernatant induced a spectral shift of the CR solution, with a maximum difference in 253 absorbance between CR alone and CR bound to amyloid fibers at ~541 nm (Chapman et 254 al.,2002).These data suggested that the phage-mediated lysis of E. coli led to the release of amyloid 255 aggregates into the biofilm. 256 257 258 Discussion

8 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

259 This study revealed an association between the dynamics of intestinal amyloid-producing E. coli 260 prior to seroconversion in HLA-matched children and the development of T1D-associated 261 autoimmunity. Several lines of evidence support roles of E. coli, E. coli phages, and E. coli-derived 262 amyloid in T1D. 263 First, in seroconverter and T1D patients, we observed a statistically significant elevation in E. coli 264 abundance from day 0 to 300, followed by E. coli depletion . Notably, we observed the total 265 disappearance or a pronounced depletion of E. coli prior to the appearance of autoantibodies in all 266 children who progressed to T1D. A similar trend was noticed among most patients in the 267 seroconverter group. Based on comparison with the control group, we confirmed that the observed 268 changes in amyloid-producing bacteria were associated with disease state, not with age. Despite 269 the relatively small patient cohort studied, the longitudinal analysis and the large total number of 270 samples (120) analyzed provided sufficient statistical power to identify significant changes (Zhao 271 et al., 2017). There are multiple factors that could lead to E. coli depletion within the gut 272 microbiome, including interplay with other gut microorganisms, which constantly change during 273 infancy. E. coli is a member of Enterobacteriaceae, which are known to positively correlate with 274 Bifidobacteriaceae, which are affected by breastfeeding (Candela et al., 2008; Kostic et al., 2016). 275 With that in mind, we evaluated a possible association between these families; however, we found 276 no association between E. coli, duration of the breastfeeding, and Bifidobacteriaceae abundance. 277 Another environmental factor that can affect the intestinal E. coli population is antibiotic usage. 278 Although some children in this study had been exposed to antibiotics, no association between the 279 timing of antibiotic usage and E. coli depletion was noticed (Mills et al., 2013). However, the most 280 profound regulators of bacterial populations are bacteriophages. Notably, we found a positive 281 correlation between the marked decline in E. coli abundance and an increase in the E. coli phage/E. 282 coli bacterial cell ratio. The strongest association was observed for lysogenic E. coli phages from 283 the Peduovirinae subfamily in the family Myoviridae and the number of unclassified 284 Lambdavirus, in the family Siphoviridae in the order Caudovirales suggesting these phages might 285 play a role as a diabetogenic indirectly leading to the release of E. coli amyloid-DNA complexes. 286 Most likely, prophage induction that led to E. coli depletion was triggered by an external stimulus 287 (Penadés et al., 2015). Prophage induction signals vary among bacteriophages, but generally are 288 activated in response to environmental and internal factors, such as bacterial SOS response 289 molecules triggered by antimicrobial agents, oxidative stress, and DNA damage (De Paepe et al., 290 2014; Tetz and Tetz, 2018). However, we found no relation between the timing of antibiotic usage 291 and E. coli disappearance relevant to prophage induction. 292 Significant die-off of the host population by prophage activation results in microbial biofilm 293 disruption, leading to the release of pathogen-associated molecular patterns (PAMPs) (Marrack et

9 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

294 al., 200). Accordingly, the E. coli die-off observed in this study might have led to the release of a 295 variety of PAMPs, such as LPS, DNA, and amyloid; however, for the majority of PAMPs, their 296 role in T1D remains elusive. Moreover, LPS derived from E. coli possess anti-diabetic activity, 297 inhibit innate immune signaling, and enhance endotoxin tolerance in non-obese diabetic mice 298 (Vatanen et al., 2016). The predominant component expressed in E. coli biofilms that is unique to 299 amyloid-producing Enterobacteria is curli amyloid fiber, which forms highly immunogenic 300 complexes with DNA (Tursi et al., 2017). Amyloid-DNA composites are known to intensively 301 stimulate TLR2 and TLR9, inducing immunogenic reactions, and they have been associated with 302 the triggering of other autoimmune disorders, including systemic lupus erythematosus (Gallo et 303 al., 2015; Tursi et al., 2017). Therefore, it is reasonable to speculate that die-off of the intestinal 304 E. coli population, particularly considering its initial high abundance among children from the case 305 groups, might lead to the prominent release of curli fibers and DNA-amyloid complexes that in 306 turn act as pro-diabetic protein and trigger the autoimmune cascade in susceptible hosts. In favor 307 of the suggested diabetogenic role of E. coli amyloid, it has been previously implicated in 308 autoimmunity associated with increased type I IFN-induced activation and protein misfolding 309 (Di Domizio et al., 2012; Gallucci et al., 2015). Type I IFN is a signature cytokine in T1D patients 310 and is known to be one of the injury agents during viral infection, which actually led to the 311 hypothesis of the viral nature of pancreatic damage in T1D (Schloot et al., 2001). Type I IFN is 312 implicated in early T1D stages by triggering β-cell apoptosis and the formation of β-cell 313 autoantigens (Op de Beeck and Eizirik 2016). Notably, E. coli curli-DNA composites have the 314 potency to stimulate the type I IFN response via two ways: through TLR stimulation of β-cells and 315 of DCs (Gallo et al., 2015). Indeed one of the pathways underlying the potential diabetogenic 316 action of E. coli amyloid-DNA complexes might be its direct triggering of pancreatic β-cell death 317 by activating TLR2 and TLR9, which in turn leads to the production of type I IFN and the 318 subsequent well-defined cascade of T1D autoimmune alterations (Honda and Taniguchi 2006; Op 319 de Beeck and Eizirik 2016). Alternatively, E. coli-derived amyloid might interact with antigen- 320 presenting cells. Recent studies on other autoimmune diseases, such systemic lupus erythematosus, 321 have shown that microbial curli fibers can lead to the activation of TLR2 and TLR9 on DCs, thus 322 stimulating immune responses, including increased type I IFN production (Honda and Taniguchi 323 2006; Gallo et al., 2015; Spaulding et al., 2015; Tursi et al., 2017). Moreover, microbial amyloid 324 has been shown to engage TLRs and trigger systemic inflammation through interaction with 325 Peyer’s patches and lymph follicles (Teng et al., 2016; Friedland, R.P. and Chapman, M.R., 2017). 326 We can speculate that finally, bacterial amyloid might lead to β-cell destruction and β-antigen 327 release by stimulating IAPP aggregation and deposition in the pancreas (Westermark et al., 2017). 328 IAPP plays multiple roles in T1D progression: on the one hand, it leads to β-cell destruction, and

10 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

329 on the other hand, it reportedly is an autoantigen for T1D triggering (Delong et al., 2011; Denroche 330 and Verchere, 2018). How bacterial amyloid can contribute to IAPP accumulation remains 331 unclear; however, recent studies have revealed a trans-kingdom interplay between E. coli curli 332 fibers and alpha-synuclein aggregation in the gut and brain of Parkinson’s disease models, and in 333 triggering amyloidosis in systemic lupus erythematosus following Salmonella typhimurium curli- 334 DNA treatment (Gallo et al., 2015; Chen et al., 2016). 335 Next, we evaluated the diabetogenic potential of E.coli prophages by the fact that their induction 336 (that modulated an elevated E.coli phage/E.coli ratio from microbiome analysis) could 337 contribute to the release of bacterial amyloid (Nanda et al., 2015). We have determined that in 338 vitro activation of E.coli prophages with mytomicin C results in the pronounced amyloid release 339 from preformed microbial biofilms, which according to data from metagenomics analysis, 340 suggests that the same process occurs in the gut of those children who further develop 341 autoimmunity and T1D (Reyrolle et al., 1982). 342 For E. coli-derived gut amyloid to activate either one of the above signaling pathways 343 triggering autoimmune cascade and T1D, it has to reach antigen-presenting cells or pancreatic cells 344 (Hartmann et al., 2012). This can happen in the case of increased intestinal permeability, leading 345 to the translocation of microbial byproducts to the lamina propria, or through interaction with 346 Peyer’s patches (Teng et al., 2016; Friedland and Chapman, 2017). Although we did not analyze 347 intestinal permeability, altered gut barrier function is well described in T1D, with certain studies 348 pointing it out as an important pathogenic factor for T1D (Carratu et al., 1999; Vaarala et al., 349 2008). Therefore, the model we propose might shed another light on the role of increased intestinal 350 permeability–the mechanistic impact of which on T1D pathogenesis was unclear to date– 351 suggesting that leaky gut allows E. coli amyloid-DNA complexes to pass to the lamina propria and 352 trigger autoimmunity and T1D progression (Carratu et al., 1999). 353 We cannot conclusively explain the disappearance of E. coli in some control patients (who 354 also had HLA risk for T1D); the lower abundance of E. coli prior to the decrease in these subjects 355 compared to the T1D and seroconverters groups attracted our attention (Figure 4). It is possible 356 that the lower amount of E. coli resulted in insufficient amyloid release to trigger autoimmunity. 357 Another reasons for the differential effects of E. coli depletion across patients could be alteration 358 of permeable state for intestinal barrier and the presence of E. coli strains with different levels of 359 curli expression or amyloid synthesis under the different growth conditions; however, these 360 possibilities need further study. 361 Finally, it is reasonable to speculate that the released E. coli phages themselves might have 362 played a particular role in triggering T1D autoimmunity. The role of bacteriophages in the 363 interplay with the immune system is poorly described; however, recent studies have suggested that

11 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

364 bacterial viruses interact with eukaryotic cells, including immune cells (Górski et al., 2017 Nguyen 365 et al., 2017). 366 Our analysis revealed previously overlooked correlations between an initially high level of 367 amyloid-producing E. coli in the intestine, followed by their depletion, most likely due to 368 prophage induction, and the initiation of autoimmunity and T1D progression. The diabetogenic 369 role of E. coli prophages was supported by the fact that the activation of E. coli prophages with 370 mitomycin C resulted in pronounced amyloid release from preformed microbial biofilms in 371 vitro. Together with the data from metagenomics analysis, these findings suggested the same 372 process might occur in the gut of children who developed autoimmunity and T1D. Combing 373 with existing data on the immunogenic role of enterobacterial amyloid, our findings suggest that 374 curli released by E. coli might trigger autoimmunity in susceptible children, highlighting the 375 need to pay specific attention to the relationships between amyloid-producing bacteria and their 376 bacteriophages in genetically susceptible hosts. Determining the exact role of E. coli-derived 377 amyloid in the progression of T1D may lead to novel diagnostics and interventional approaches; 378 however, obviously, further, detailed studies are required. 379 380 Author Contributions 381 G.T. designed the analysis. S.B., Y.H. and G.T. conducted a metagenomics analysis. G.T. and V.T. 382 supervised data analysis and wrote the manuscript. 383 384 Declaration of Interests 385 The authors declare no competing interests. 386 387 388 389 390 391 392 393 394 395 396 397 398

12 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

399 References 400

401 1. Achenbach, P., Bonifacio, E., and Ziegler, A. G. (2005). Predicting type 1 402 diabetes. Current diabetes reports, 5, 98-103. 403 2. Atkinson, M. A., and Eisenbarth, G. S. (2001). Type 1 diabetes: new perspectives on 404 disease pathogenesis and treatment. The Lancet 358, 221-229. 405 Bailey, B., Lithgow T., Lazarou, M., Luque, A., Rohwer, F., Blumberg , R., and Luque, A. 406 (2017). Bacteriophage transcytosis provides a mechanism to cross epithelial cell 407 layers. MBio, 8(6), e01874-17. 408 3. Barnhart, M. M., and Chapman, M. R. (2006). Curli biogenesis and function. Annu. Rev. 409 Microbiol 60, 131-147. 410 4. Barrett, J. C., Clayton, D. G., Concannon, P., Akolkar, B., Cooper, J. D., Erlich, H. A.,et 411 al. (2009). Genome-wide association study and meta-analysis find that over 40 loci affect 412 risk of type 1 diabetes. Nature genetics, 41, 703-707. 413 5. Candela, M., Perna, F., Carnevali, P., Vitali, B., Ciati, R., Gionchetti, P., Rizzello, F., 414 Campieri, M. and Brigidi, P. (2008). Interaction of probiotic Lactobacillus and 415 Bifidobacterium strains with human intestinal epithelial cells: Adhesion properties, 416 competition against enteropathogens and modulation of IL-8 production. International 417 Journal of Food Microbiology 125,286-292. 418 6. Caporaso, J. G., et al. (2010). QIIME allows analysis of high-throughput community 419 sequencing data. Nature methods 7, 335. 420 7. Carratù, R., Secondulfo, M., de Magistris, L., Iafusco, D., Urio, A., Carbone, M. G., Grazia, 421 M., Gabriele, M., Marilena, C., and Prisco, F. (1999). Altered intestinal permeability to 422 mannitol in diabetes mellitus type I. Journal of pediatric gastroenterology and 423 nutrition, 28, 264-269. 424 8. Chapman, M. (2002). Role of Escherichia coli Curli Operons in Directing Amyloid Fiber 425 Formation. Science 295,851-855. 426 9. Chapman, M. R., Robinson, L. S., Pinkner, J. S., Roth, R., Heuser, J., Hammar, M., 427 Normark, S., and Hultgren, S. J. (2002). Role of Escherichia coli curli operons in directing 428 amyloid fiber formation. Science 295, 851-855. 429 10. Chen, S., Stribinskis, V., Rane, M., Demuth, D., Gozal, E., Roberts, A., Jagadapillai, R., 430 Liu, R., Choe, K., Shivakumar, B., Son, F., Jin, S., Kerber, R., Adame, A., Masliah, E. and 431 Friedland, R. (2016). Exposure to the Functional Bacterial Amyloid Protein Curli Enhances 432 Alpha-Synuclein Aggregation in Aged Fischer 344 Rats and Caenorhabditis elegans. 433 Scientific Reports 6, 34477.

13 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

434 11. Collinge, J. (2016). Mammalian prions and their wider relevance in neurodegenerative 435 diseases. Nature, 539(7628), 217. 436 12. Costerton, J. W., Stewart, P. S., and Greenberg, E. P. (1999). Bacterial biofilms: a common 437 cause of persistent infections. Science 284, 1318-1322. 438 13. Davis-Richardson, A., Ardissone, A., Dias, R., Simell, V., Leonard, M., Kemppainen, K., 439 Drew, J., Schatz, D., Atkinson, M., Kolaczkowski, B., Ilonen, J., Knip, M., Toppari, J., 440 Nurminen, N., Hyaty, H., Veijola, R., Simell, T., Mykkanen, J., Simell, O. and Triplett, E. 441 (2014). Bacteroides dorei dominates gut microbiome prior to autoimmunity in Finnish 442 children at high risk for type 1 diabetes. Frontiers in Microbiology, 5. 443 14. De Paepe, M., Leclerc, M., Tinsley, C.R. and Petit, M.A., 2014. Bacteriophages: an 444 underestimated role in human and animal health?. Frontiers in cellular and infection 445 microbiology, 4, p.39. 446 15. Delong, T., Baker, R.L., Reisdorph, N., Reisdorph, R., Powell, R.L., Armstrong, M., 447 Barbour, G., Bradley, B. and Haskins, K. (2011). Islet amyloid polypeptide is a target 448 antigen for diabetogenic CD4+ T cells. Diabetes, DB_110288. 449 16. Denroche, H.C. and Verchere, C.B. (2018). IAPP and type 1 diabetes: implications for 450 immunity, metabolism and islet transplants. Journal of molecular endocrinology 60, R57- 451 R75. 452 17. Di Domizio, J., Dorta-Estremera, S., Gagea, M., Ganguly, D., Meller, S., Li, P., Zhao, B., 453 Tan, F.K., Bi, L., Gilliet, M. and Cao, W., 2012. Nucleic acid-containing amyloid fibrils 454 potently induce type I interferon and stimulate systemic autoimmunity. Proceedings of the 455 National Academy of Sciences, 201206923. 456 18. Drescher, K., Kono, K., Bopegamage, S., Carson, S. and Tracy, S. (2004). Coxsackievirus 457 B3 infection and type 1 diabetes development in NOD mice: insulitis determines 458 susceptibility of pancreatic islets to virus infection. Virology 329 381-394. 459 19. Eizirik, D., Colli, M. and Ortis, F. (2009). The role of inflammation in insulitis and β-cell 460 loss in type 1 diabetes. Nature Reviews Endocrinology 5,219-226. 461 20. Filippi, C. and von Herrath, M. (2008). Viral Trigger for Type 1 Diabetes: Pros and Cons. 462 Diabetes, 57(11), pp.2863-2871. 463 21. Flemming, H. C., and Wingender, J. (2010). The biofilm matrix. Nature Reviews 464 Microbiology 8, 623. 465 22. Flodström, M., Maday, A., Balakrishna, D., Cleary, M., Yoshimura, A. and Sarvetnick, N. 466 (2002). Target cell defense prevents the development of diabetes after viral infection. 467 Nature Immunology 3, 373-382.

14 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

468 23. Fowler, D. M., Koulov, A. V., Balch, W. E., and Kelly, J. W. (2007). Functional amyloid– 469 from bacteria to humans. Trends in biochemical sciences 32, 217-224. 470 24. Friedland, R.P. and Chapman, M.R. (2017). The role of microbial amyloid in 471 neurodegeneration. PLoS pathogens 13, e1006654. 472 25. Gallo, P.M., Rapsinski, G.J., Wilson, R.P., Oppong, G.O., Sriram, U., Goulian, M., 473 Buttaro, B., Caricchio, R., Gallucci, S. and Tükel, Ç. (2015). Amyloid-DNA composites 474 of bacterial biofilms stimulate autoimmunity. Immunity 42,1171-1184. 475 26. Gallucci, S., Gallo, P., Rapsinski, G., Oppong, G., Sriram, U., Wilson, R., Buttaro, B. and 476 Tukel, C., 2015. Bacterial amyloids promote type I interferon production and accelerate 477 autoimmunity (BA6P. 124). 478 27. Georgopoulos, M., Vass, C., Vatanparast, Z., Wolfsberger, A., and Georgopoulos, A. 479 (2002). Activity of dissolved mitomycin C after different methods of long-term 480 storage. Journal of glaucoma 11, 17-20. 481 28. Górski, A., Dąbrowska, K., Międzybrodzki, R., Weber-Dąbrowska, B., Łusiak- 482 Szelachowska, M., Jończyk-Matysiak, E., and Borysowski, J. (2017). Phages and 483 immunomodulation. Future microbiology 12, 905-914. 484 29. Hao, Y., Yang, L., Galvao Neto, A., Amin, M. R., Kelly, D., Brown, S. M., and Pei, Z. 485 (2018). HPViewer: sensitive and specific genotyping of human papillomavirus in 486 metagenomic DNA. Bioinformatics 34, 1986-1995. 487 30. Harrison, L. C. (2019). Type 1 Diabetes. In Clinical Immunology (Fifth Edition) pp, 957- 488 966. 489 31. Hartmann, P., Haimerl, M., Mazagova, M., Brenner, D.A. and Schnabl, B. (2012). Toll- 490 Like Receptor 2–Mediated Intestinal Injury and Enteric Tumor Necrosis Factor Receptor I 491 Contribute to Liver Fibrosis in Mice. Gastroenterology 143,1330-1340. 492 32. Honda, K. and Taniguchi T. (2006). IRFs: master regulators of signalling by Toll-like 493 receptors and cytosolic patternrecognition receptors. Nat Rev Immunol 2006, 6:644–58. 494 33. Hyöty, H. (2016). Viruses in type 1 diabetes. Pediatric Diabetes 17, 56-64. 495 34. Ikegami, H., Noso, S., Babaya, N. and Kawabata, Y. (2011). Genetics and pathogenesis of 496 type 1 diabetes: prospects for prevention and intervention. Journal of Diabetes 497 Investigation 2,415-420. 498 35. Jaccard, P. (1912). The distribution of flora in the alpine zone. New Phytologist 11, 37–50. 499 36. Jensen, K. F. (1993). The Escherichia coli K-12" wild types" W3110 and MG1655 have 500 an rph frameshift mutation that leads to pyrimidine starvation due to low pyrE expression 501 levels. Journal of bacteriology 175, 3401-3407.

15 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

502 37. Knip, M.,and Siljander, H. (2016). The role of the intestinal microbiota in type 1 diabetes 503 mellitus. Nature Reviews Endocrinology 12, 154. 504 38. Kostic, A., Gevers, D., Siljander, H., Vatanen, T., Hyötyläinen, T., Hämäläinen, A., Peet, 505 A., Tillmann, V., Pöhö, P., Mattila, I., Lähdesmäki, H., Franzosa, E., Vaarala, O., de 506 Goffau, M., Harmsen, H., Ilonen, J., Virtanen, S., Clish, C., Orešič, M., Huttenhower, C., 507 Knip, M. and Xavier, R. (2016). The Dynamics of the Human Infant Gut Microbiome in 508 Development and in Progression toward Type 1 Diabetes. Cell Host & Microbe 20, 121. 509 39. Langmead, B. and Salzberg, S. (2012) Fast gapped-read alignment with Bowtie 2. Nature 510 Methods 9, 357-359. 511 40. Lehuen, A., Diana, J., Zaccone, P. and Cooke, A. (2010). Immune cell crosstalk in type 1 512 diabetes. Nat. Rev. Immunol. 10, 501-513. 513 41. Maniatis, T., Fritsch, E. F., and Sambrook, J. (1982). Molecular cloning: a laboratory 514 manual (Vol. 545). Cold Spring Harbor, NY: Cold spring harbor laboratory. 515 42. Marrack, P., Kappler, J. and Kotzin, B.L. (2001). Autoimmune disease: why and where it 516 occurs. Nature medicine 7, 899. 517 43. McDonald, J. E., Smith, D. L., Fogg, P. C., McCarthy, A. J., and Allison, H. E. (2010). 518 High-throughput method for rapid induction of prophages from lysogens and its 519 application in the study of Shiga toxin-encoding Escherichia coli strains. Applied and 520 environmental microbiology 76, 2360-2365. 521 44. McMurdie, P. J. and Holmes, S. (2013) phyloseq: An R Package For Reproducible 522 Interactive 519 Analysis And Graphics Of Microbiome Census Data. PloS one 8, e61217. 523 45. Mills, S., Shanahan, F., Stanton, C., Hill, C., Coffey, A. and Ross, R.P., 2013. Movers and 524 shakers: influence of bacteriophages in shaping the mammalian gut microbiota. Gut 525 microbes 4, 4-16. 526 46. Nanda, A. M., Thormann, K., and Frunzke, J. (2015). Impact of spontaneous prophage 527 induction on the fitness of bacterial populations and host-microbe interactions. Journal of 528 bacteriology 197, 410-419. 529 47. Nguyen, S., Baker, K., Padman, B. S., Patwa, R., Dunstan, R. A., Weston, T. A., Schlosser, 530 K., 531 48. Noble, J. and Valdes, A. (2011). Genetics of the HLA Region in the Prediction of Type 1 532 Diabetes. Current Diabetes Reports 11,.533-542. 533 49. Nurk, S., Meleshko, D., Korobeynikov, A. and Pevzner, P. A. (2017). metaSPAdes: a new 534 versatile metagenomic assembler. Genome research, gr-213959. 535 50. Op de Beeck, A. and Eizirik, D. (2016). Viral infections in type 1 diabetes mellitus — why 536 the β cells?. Nature Reviews Endocrinology, 12(5), pp.263-273.

16 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

537 51. Paulsson, J., Ludvigsson, J., Carlsson, A., Casas, R., Forsander, G., Ivarsson, S., Kockum, 538 I., Lernmark, Å., Marcus, C., Lindblad, B. and Westermark, G. (2014). High Plasma Levels 539 of Islet Amyloid Polypeptide in Young with New-Onset of Type 1 Diabetes Mellitus. PLoS 540 ONE 9, e93053. 541 52. Paulsson, J., Ludvigsson, J., Carlsson, A., Casas, R., Forsander, G., Ivarsson, S., Kockum, 542 I., Lernmark, Å., Marcus, C., Lindblad, B. and Westermark, G. (2014). High Plasma Levels 543 of Islet Amyloid Polypeptide in Young with New-Onset of Type 1 Diabetes Mellitus. PLoS 544 ONE 9, p.e93053. 545 53. Paun, A., Yau, C., and Danska, J. S. (2017). The influence of the microbiome on type 1 546 diabetes. The Journal of Immunology 198, 590-595. 547 54. Penadés, J.R., Chen, J., Quiles-Puchalt, N., Carpena, N. and Novick, R.P., 2015. 548 Bacteriophage-mediated spread of bacterial virulence genes. Current opinion in 549 microbiology 23, 171-178. 550 55. Reichhardt, C., Jacobson, A., Maher, M., Uang, J., McCrate, O., Eckart, M. and Cegelski, 551 L. (2015). Congo Red Interactions with Curli-Producing E. coli and Native Curli Amyloid 552 Fibers. PLOS ONE 10, e0140388. 553 56. Rewers, M., and Ludvigsson, J. (2016). Environmental risk factors for type 1 diabetes. The 554 Lancet 387, 2340-2348. 555 57. Reyrolle, J., Chopin, M. C., Letellier, F., and Novel, G. (1982). Lysogenic strains of lactic 556 acid streptococci and lytic spectra of their temperate bacteriophages. Applied and 557 environmental microbiology 43, 349-356. 558 58. Reyrolle, J., Chopin, M.C., Letellier, F. and Novel, G. (1982). Lysogenic strains of lactic 559 acid streptococci and lytic spectra of their temperate bacteriophages. Applied and 560 environmental microbiology 43, 349-356. 561 59. Schloot, N., Willemen, S., Duinkerken, G., Drijfhout, J., de Vries, R. and Roep, B. (2001). 562 Molecular mimicry in type 1 diabetes mellitus revisited: T-cell clones to GAD65 peptides 563 with sequence homology to Coxsackie or proinsulin peptides do not crossreact with 564 homologous counterpart. Human Immunology 62,299-309. 565 60. Schwartz, K., and Boles, B. R. (2013). Microbial amyloids–functions and interactions 566 within the host. Current opinion in microbiology 16, 93-99. 567 61. Segata, N., Waldron, L., Ballarini, A., Narasimhan, V., Jousson, O., & Huttenhower, C. 568 (2012). Metagenomic microbial community profiling using unique clade-specific marker 569 genes. Nature methods 9, 811.

17 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

570 62. Soltesz, G., Patterson, C. C., Dahlquist, G., and EURODIAB Study Group. (2007). 571 Worldwide childhood type 1 diabetes incidence–what can we learn from 572 epidemiology?. Pediatric diabetes, 8, 6-14 573 63. Spaulding, C.N., Dodson, K.W., Chapman, M.R. and Hultgren, S.J. (2015). Fueling the 574 fire with fibers: bacterial amyloids promote inflammatory disorders. Cell host & 575 microbe 18, 1-2. 576 64. Teng, F., Klinger, C. N., Felix, K. M., Bradley, C. P., Wu, E., Tran, N. L., Umesaki, Y., 577 and Wu, H. J. J. (2016). Gut microbiota drive autoimmune arthritis by promoting 578 differentiation and migration of Peyer’s patch T follicular helper cells. Immunity 44, 875- 579 888. 580 65. Teng, F., Klinger, C.N., Felix, K.M., Bradley, C.P., Wu, E., Tran, N.L., Umesaki, Y. and 581 Wu, H.J.J. (2016). Gut microbiota drive autoimmune arthritis by promoting differentiation 582 and migration of Peyer’s patch T follicular helper cells. Immunity 44, 875-888. 583 66. Tetz, G. and Tetz, V., 2018. Bacteriophages as New Human Viral 584 Pathogens. Microorganisms 6, 54. 585 67. Tetz, G. V., Artemenko, N. K., and Tetz, V. V. (2009). Effect of DNase and antibiotics on 586 biofilm characteristics. Antimicrobial agents and chemotherapy 53, 1204-1209. 587 68. Tetz, G. V., Hao, Y, Zhou, H., Stuart B, and Tetz, V. (2018). Parkinson’s disease and 588 bacteriophages as its overlooked contributors. Scientific reports 8, 10812. 589 69. Tetz, G. V., Ruggles, K. V., Zhou, H., Heguy, A., Tsirigos, A., and Tetz, V. (2017). 590 Bacteriophages as potential new mammalian pathogens. Scientific reports 7, 7043. 591 70. Tetz, G., and Tetz, V. (2018). Bacteriophages as New Human Viral 592 Pathogens. Microorganisms 6, 54. 593 71. Todd, J. (2010). Etiology of Type 1 Diabetes. Immunity, 4,457-467. 594 72. Tükel, Ç., Nishimori, J. H., Wilson, R. P., Winter, M. G., Keestra, A. M., Van Putten, J. 595 P., and Bäumler, A. J. (2010). Toll‐like receptors 1 and 2 cooperatively mediate immune 596 responses to curli, a common amyloid from enterobacterial biofilms. Cellular 597 microbiology 12, 1495-1505. 598 73. Tursi, S.A., Lee, E.Y., Medeiros, N.J., Lee, M.H., Nicastro, L.K., Buttaro, B., Gallucci, S., 599 Wilson, R.P., Wong, G.C. and Tükel, Ç., (2017). Bacterial amyloid curli acts as a carrier 600 for DNA to elicit an autoimmune response via TLR2 and TLR9. PLoS pathogens 13, 601 e1006315. 602 74. Vaarala, O., Atkinson, M. A., ans Neu, J. (2008). The “perfect storm” for type 1 diabetes: 603 the complex interplay between intestinal microbiota, gut permeability, and mucosal 604 immunity. Diabetes 57, 2555-2562.

18 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

605 75. Vatanen, T., Kostic, A.D., d’Hennezel, E., Siljander, H., Franzosa, E.A., Yassour, M., 606 Kolde, R., Vlamakis, H., Arthur, T.D., Hämäläinen, A.M. and Peet, A. (2016). Variation 607 in microbiome LPS immunogenicity contributes to autoimmunity in humans. Cell 165, 608 842-853. 609 76. Vehik, K., Haller, M., Beam, C., Schatz, D., Wherrett, D., Sosenko, J. and Krischer, J. 610 (2011). Islet Autoantibody Seroconversion in the DPT-1 Study: Justification for repeat 611 screening throughout childhood. Diabetes Care, 34 pp.358-362. 612 77. Waller, A. S., Yamada, T., Kristensen, D. M., Kultima, J. R., Sunagawa, S., Koonin, E. V., 613 and Bork, P. (2014). Classification and quantification of bacteriophage taxa in human gut 614 metagenomes. The ISME journal 8, 1391. 615 78. Walterspiel, J. N., Morrow, A. L., Cleary, T. G., and Ashkenazi, S. (1992). Effect of 616 subinhibitory concentrations of antibiotics on extracellular Shiga-like toxin I. Infection 20, 617 25-29. 618 79. Westermark, G.T., Krogvold, L., Dahl-Jørgensen, K. and Ludvigsson, J. (2017). Islet 619 amyloid in recent-onset type 1 diabetes—the DiViD study. Upsala journal of medical 620 sciences 122, 201-203. 621 80. Yuan, Y., Gao, M., Wu, D., Liu, P., & Wu, Y. (2012). Genome characteristics of a novel 622 phage from Bacillus thuringiensis showing high similarity with phage from Bacillus 623 cereus. PLoS One, 7(5), e37557. 624 81. Zhao, G., Vatanen, T., Droit, L., Park, A., Kostic, A., Poon, T., Vlamakis, H., Siljander, 625 H., Härkönen, T., Hämäläinen, A., Peet, A., Tillmann, V., Ilonen, J., Wang, D., Knip, M., 626 Xavier, R. and Virgin, H. (2017). Intestinal virome changes precede autoimmunity in type 627 I diabetes-susceptible children. Proceedings of the National Academy of Sciences p, 628 201706359. 629 630 631 632 633 634 635 636 637 638 639

19 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

640 Methods 641 Study Population 642 We used data from a prospective, longitudinal cohort study by Kostic et al. of 16 HLA-matched 643 infants followed from birth until 3 years of age (Kostic et al., 2016). The children, from Finland 644 and Estonia, were recruited to the study between September 2008 and August 2010 645 (http://www.diabimmune.org/). The study generated shotgun metagenomics sequencing data of 646 the fecal microbiome from 10 children who exhibited autoantibodies (six who developed serum 647 autoantibodies with no progression to T1D, and four who were seroconverted and developed T1D) 648 and eight non-seroconverted control individuals (Kostic et al., 2016). Inclusion criteria were: 649 presence of HLA DR-DQ alleles associated with T1D development. Data on diabetes-associated 650 autoantibodies are presented in the original study. Demographic parameters of the study 651 participants and data collection details are presented in Table S1. 652 653 Microbiota Sequencing and Processing 654 High-throughput shotgun sequencing was performed on the Illumina HiSeq 2500 platform, 655 generating ~2.5 Gb of sequence per sample with 101-bp paired-end reads. Human contamination 656 was removed with the BMTagger (ftp://ftp.ncbi.nlm.nih.gov/pub/agarwala/bmtagger/). Bacterial 657 and phage contents were quantified separately using the SRA shotgun metagenomic sequencing 658 data. Bacterial content was quantified by taxa directly from SRA reads using Metaphalan (v. 2.0), 659 which maps sequence reads to a database of predefined clade-specific marker genes (Segata et al., 660 2012). All bacterial taxa with relative abundances <0.01 in all samples were excluded from 661 statistical analysis. Phage content was assessed using a custom method. First, reads from each SRA 662 file were assembled de novo into contigs with metaSPAdes (v. 3.11.1) (Nurk et al., 2017). Then, 663 contigs >200 bp were aligned to the EBI collection of phage genomes 664 (https://www.ebi.ac.uk/genomes/phage.html) by BLAST, with a threshold e-value <1e-5 and 665 alignment length >50% of contig length. All of the original reads were then re-mapped with 666 Bowtie2 (v. 2.3.4.1) to the contigs with good phage BLAST matches in order to increase sensitivity 667 and to more accurately count the abundance of reads from each type of phage (Langmead and 668 Salzberg, 2012). Phage read counts per contig were combined per phage genome (taxa) and 669 normalized to relative abundance. A detection threshold of two reads per sample (>90% identity 670 to the phage genome) was used, based on a previous report (Hao et al., 2018). 671 672 Statistical Analysis of Microbial Community Composition and Differential Abundance 673 The QIIME pipeline (v1.9.1) was used for quality filtering of bacterial and bacteriophage DNA 674 sequences, chimera removal (with USEARCH software), taxonomic assignment, and

20 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

675 calculation of α-diversity, as previously described (Caporaso et al. 2010; Tetz et al., 2017). 676 Downstream data analysis and calculation of diversity metrics were conducted in R v3.5.1, using 677 ggplot2 and phyloseq libraries; DESeq2 was used to calculate logarithm of fold change. Bacterial 678 and bacteriophage communities at the genus, family, and species levels were characterized based 679 on α- and β-diversities. α-Diversity indices (ACE, Chao 1 richness estimator, Shannon and 680 Simpson indexes) were calculated using the phyloseq R library (McMurdie and Holmes, 2013). 681 Differences in α-diversities between datasets were examined by the Mann-Whitney test; p < 0.05 682 was considered statistically significant. 683 Differences among groups where two variables exist (phenotype and time point) were 684 analyzed by two-way ANOVA and Tukey’s multiple comparison test. When one variable was 685 compared, an unpaired two-tailed t-test was used. Data were visualized using multidimensional 686 scaling (MDS). 687 Correlations between E. coli disappearance, Bifidobacterium abundance, breastfeeding 688 and antibiotic usage were assessed pairwise using the Jaccard similarity index (Jaccard, 1912). 689 Differences were considered statistically significant at p < 0.05.

690 Bacteria and Bacteriophages

691 E. coli strain MG1655 was used as a host for prophage induction experiments (Jensen 1993). 692 Bacteria were subcultured from freezer stocks onto Mueller–Hinton agar plates (Oxoid) and 693 incubated at 37 °C overnight. All subsequent liquid subcultures were derived from colonies 694 isolated from these plates and were grown in Mueller–Hinton broth (Oxoid). 695 Bacteriophage λ from our collection was employed. Bacteriophage suspensions were

696 routinely stored in TM buffer (10 mM Tris-HCl, 10 mM MgSO4, pH 7.2) at 4 °C. E. coli 697 lysogenic strains were obtained by infection of E. coli MG1655 with the phage and titration of 698 cells on Mueller–Hinton agar plates as previously described (Maniatis et al. 1989).

699 Prophage Induction

700 Mitomycin C (Sigma-Aldrich) was prepared in 0.1 M phosphate buffer (pH 7.2), filtered through 701 a 0.22-μm-pore filter (Millipore Corp., Bedford, MA), and stored at 4 °C (Georgopoulos et al., 702 2002). A modified protocol according to Reyrolle et al. designed for prophage induction on 96- 703 well plates was used (Reyrolle et al., 1982; McDonald et al., 2010). Twenty microliters of an 704 overnight E. coli MG1655 culture, with an absorbance at 600 nm of 0.2, was dispensed in each 705 well of a 96-well flat-bottom polystyrene tissue culture microtiter plate (Sarstedt, Numbrecht, 706 Germany), after which 180 μl of MHB (Oxoid) was added. The plates were incubated at 37 °C

21 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

707 for 48 h. Induction of prophages in the E. coli biofilms was provoked by the addition of 708 mitomycin C (Sigma) to a final concentration 1 μg/ml to each well. The plates were further 709 incubated at 37 °C and samples were taken at 4, 6, 8, and 10 h following prophage induction.

710 Determination of CFU and PFU

711 To estimate CFU, biofilms were thoroughly scraped (Tetz et al., 2009). Well contents were 712 aspirated, placed in 1 ml of isotonic phosphate buffer (0.15 M, pH 7.2), and the total CFU number 713 was determined by serial dilution and plating on Mueller–Hinton agar. 714 The number of phage virions produced after induction was estimated by phage titration 715 (using strain MG1655 as a host) and phage plaque assay. Following induction, at the indicated 716 times, 200-μl samples of biofilms were collected. Then, 30 μl of chloroform was added to each 717 sample, the mixture was vortexed and centrifuged at 3000 ×g for 5 min in a microcentrifuge 718 (Eppendorf 5415D). A supernatant fraction of the bacterial lysate was further used. The phage 719 titer (number of phages per ml) was determined using the double agar overlay assay method as 720 described previously (Yuan et al., 2012). To this end, 2.5 μl of each titration point of phage 721 lysate was spotted on Mueller−Hinton 722 agar (Oxoid). Then, a mixture of 1 ml of E. coli MG1655 culture and 2 ml of 0.7% nutrient agar

723 (heated to 45 °C) supplemented with MgSO4 and CaCl2 (to a final concentration of 10 mM 724 each) was poured over the plate. Plates were incubated at 37 °C overnight. Each experiment 725 was repeated in triplicate.

726 Supernatant CR Depletion Assay

727 The amount of bacterial amyloid following prophage induction was measured in the supernatant 728 of E. coli MG1655 biofilms. Biofilms were obtained as described above. Biofilm supernatant was 729 taken at 4, 6, 8, and 10 h following prophage induction. To isolate aggregated amyloid fibers from 730 the supernatant, the supernatant was centrifuged at 10,000 × g, filtered through a 0.22 µm filter to 731 separate bacterial cells, and treated with proteinase K to eliminate contaminating proteins. CR 732 (Sigma-Aldrich) was added to a final concentration of 10 μg/ml from a filtered stock solution of 1 733 mg/ml (Reichhardt et al., 2015). After 5 min of equilibration, absorbance spectra were recorded 734 from 400 to 600 nm (SmartSpec 3000, Bio-Rad). Each trace represents the average of 5 735 accumulated spectra. For all samples, spectra of corresponding nutrient MHB solutions with CR 736 were used as blanks. 737 738 Data availability

22 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

739 The sequencing datasets generated and/or analyzed in the current study are available from the 740 corresponding author on reasonable request.

23 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

741 Figure Legends

742 Figure 1. E. coli Abundance Over Time 743 Boxplot showing absolute E. coli abundances in the seroconverter, T1D, and control groups in 744 the indicated periods. E. coli abundance across different time points was compared using two- 745 tailed Mann–Whitney U test. Statistically significant variation between selected conditions 746 (*p < 0.05, **p < 0.01 and*** p < 0.005). 747 748 Figure 2. Association Between the Dynamics of the Disappearance of E. coli and 749 Seroconversion

750 Each row represents an individual, with each symbol indicating a single stool sample. The X-axis 751 indicates age at sample collection. E. coli abundance is >0.01. Green indicates presence of E. coli, 752 and red indicates disappearance or presumed disappearance (over 50-fold reduction when 753 compared with initial abundance) of E. coli. Diamonds represent the time point at which 754 autoantibodies were detected. 755 756 Figure 3. Absolute Abundances of E. coli Across Samples. Fecal bacterial communities were 757 analyzed by high-throughput Illumina Hiseq2500 sequencing. Threshold for E. coli abundance is 758 >0.01. Each row represents an individual, with E. coli abundance in each stool sample (indicated 759 with different symbols) being color-coded using a color gradient, ranging from red (low 760 abundance) to purple (maximum abundance). Samples with no E. coli are marked with white 761 circles. The X-axis indicates age at sample collection. Diamond sign represents the time of 762 seroconversion detected. 763 764 Figure 4. Association Between E. coli phage/E. coli Ratio Across Samples and 765 Seroconversion. 766 Each row represents an individual, with each symbol representing the phage/bacterial ratio, 767 calculated as E. coli phage abundance normalized to that of the E. coli abundance, of a sample. 768 Abundance is represented as a color gradient, ranging from purple (lowest ratio) to yellow 769 (maximum ratio). Samples in which E. coli phages but no bacterial cells were present are marked 770 with white circles. Samples with total E. coli disappearance or presumed disappearance with over 771 50-fold decrease in E. coli are indicated by red circles. The X-axis indicates age at sample 772 collection. The diamond symbol represents the time point at which seroconversion was detected.

773

24 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

774 Figure 5. Prophage Induction Promotes Amyloid Release from E. coli in vitro 775 CR absorbance upon binding to amyloid from intact E. coli supernatant (dashed line) and from E. 776 coli supernatant after prophage induction (solid line). Both curves are approximately equal to zero 777 at the isosbestic point (403 nm), and the maximum difference in absorbance was observed at 541 778 nm. Both spectra were normalized to the absorbance spectrum of free CR. 779

25 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

780 Table 1. Effect of prophage induction in MG1655 wild-type lysogenic strain on PFU and 781 CFU 782 PFU/ml CFU/ml 4 h 6 h 8 h 10 h 4 h 6 h 8 h 10 h E. coli 0 0 0 0 9.3 8.9 9.1 9.5 control log10 log10 log10 log10 +/- 0.6 +/- 0.5 +/- 0.6 +/- 0.6 log10 log10 log10 log10 E. coli + 2.3 log10 9.3 10.4 11.2 8.8 8.2 7.6 7.4 mytomicin +/- 0.2 log10 log10 log10 log10 log10 log10 log10 C log10 +/- 0.5 +/- 0.6 +/- 0.5 +/- 0.4 +/- 0.7 +/- 0.7 +/- 0.7 log10 log10 log10 log10 log10 log10 log10 783 784

26 bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

* * * * *** ** Phenotype

Sample date bioRxiv preprint not certifiedbypeerreview)istheauthor/funder,whohasgrantedbioRxivalicensetodisplaypreprintinperpetuity.Itmade

T1D Seroconverter Control Seroconversionassociated is with E.coli disappearance doi: https://doi.org/10.1101/433110 available undera ; this versionpostedOctober3,2018. CC-BY-ND 4.0Internationallicense The copyrightholderforthispreprint(whichwas . (presumed disappearance) (presumed disappearance) = Serconversion bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

E. coli abundance Control Control Seroconverter Seroconverter

= serconversion = zero E.coli disappearance / presumed disappearance T1D T1D bioRxiv preprint not certifiedbypeerreview)istheauthor/funder,whohasgrantedbioRxivalicensetodisplaypreprintinperpetuity.Itmade

T1D Seroconverter Control doi: https://doi.org/10.1101/433110 available undera ; this versionpostedOctober3,2018. CC-BY-ND 4.0Internationallicense The copyrightholderforthispreprint(whichwas . = zero= E.coli = Serconversion bioRxiv preprint doi: https://doi.org/10.1101/433110; this version posted October 3, 2018. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-ND 4.0 International license.

0.4

0.3

Treatment Supernatant after prophage induction 0.2 Intact supernatant Absorbance

0.1

0.0

400 450 500 550 600 Wavelength