Page 1 of 67 Diabetes

1 Original Article 2 3 Insulin-Like Growth Factor Dysregulation Both Preceding and Following Type 1

4 Diabetes Diagnosis

5 Running Title: Insulin-like Growth Factors in T1D Development 6 7 Melanie R. Shapiro1; Clive H. Wasserfall1; Sean M. McGrail1, Amanda L. Posgai1; Rhonda

8 Bacher2; Andrew Muir3; Michael J. Haller4; Desmond A. Schatz4; Johnna D. Wesley5;

9 Matthias von Herrath5; William A. Hagopian6; Cate Speake7; Mark A. Atkinson1,4; Todd

10 M. Brusko1 11 12 1Department of Pathology, Immunology, and Laboratory Medicine, University of Florida

13 Diabetes Institute, Gainesville, Florida, USA.

14 2Department of Biostatistics, University of Florida, Gainesville, Florida, USA.

15 3Department of Pediatrics, Emory University, Atlanta, Georgia, USA.

16 4Department of Pediatrics, University of Florida Diabetes Institute, Gainesville, Florida,

17 USA.

18 5Novo Nordisk Research Center, Seattle, Inc., Seattle, Washington, USA.

19 6Pacific Northwest Research Institute, Seattle, Washington, USA.

20 7Benaroya Research Institute at Virginia Mason, Seattle, Washington, USA.

21 Correspondence to: Todd M. Brusko, PhD, Department of Pathology, University of

22 Florida, College of Medicine, Box 100275, 1600 SW Archer Road, Gainesville, FL 32610;

23 (352) 273-9255; Fax (352) 273-9339; Email: [email protected]

24 Word count (4,019/4,000); Abstract (198/200); References (51/50); Figures (4); Tables

25 (4); Supplementary Figures (9); Supplementary Tables (7)

Diabetes Publish Ahead of Print, published online December 11, 2019 Diabetes Page 2 of 67

26 Insulin-like growth factors (IGFs), specifically IGF1 and IGF2, promote glucose

27 metabolism with their availability regulated by IGF binding proteins (IGFBPs). We

28 hypothesized that IGF1 and IGF2 levels, or their bioavailability, are reduced during

29 type 1 diabetes development. Total serum IGF1, IGF2, and IGFBP1–7 levels were

30 measured in an age-matched, cross-sectional cohort at varying stages of

31 progression to type 1 diabetes. IGF1 and IGF2 levels were significantly lower in

32 autoantibody (AAb)+ compared to AAb- relatives of type 1 diabetes subjects. Most

33 high-affinity IGFBPs were unchanged in individuals with pre-type 1 diabetes,

34 suggesting that total IGF levels may reflect bioactivity. We also measured serum

35 IGFs from a cohort of fasted type 1 diabetes subjects. IGF1 levels significantly

36 decreased with disease duration, in parallel with declining β-cell function.

37 Additionally, plasma IGF levels were assessed in an AAb+ cohort drawn monthly

38 for a year. IGF1 and IGF2 showed longitudinal stability in single AAb+ subjects, but

39 IGF1 levels decreased over time in subjects with multiple AAb and those who

40 progressed to type 1 diabetes, particularly post-diagnosis. In sum, IGFs are

41 dysregulated both before and after the clinical diagnosis of type 1 diabetes and

42 may serve as novel biomarkers to improve disease prediction. Page 3 of 67 Diabetes

43 Public health care screening for presymptomatic type 1 diabetes does not yet exist, but

44 efforts are underway to evaluate feasibility of population-based genetic testing and

45 screening for disease-related autoantibodies (AAb) (1-3). A recent consensus article

46 presented a two-step definition of pre-type 1 diabetes: the first being seroconversion to

47 two or more AAb against β-cell antigens (Stage 1) with Stage 2 occuring when multiple

48 AAb+ subjects demonstrate dysglycemia following an oral glucose tolerance test (OGTT)

49 (4). The duration of progression to overt hyperglycemia is highly variable, ranging from

50 weeks to decades beyond seroconversion (5), and OGTTs are time-consuming and

51 require multiple venous blood sample collections. This exemplifies the need for minimally

52 invasive biomarkers reflective of metabolic dysregulation to complement AAb screening

53 in predicting type 1 diabetes prior to clinical manifestation (Stage 3).

54 Type 1 diabetes is thought to result from deficient immunoregulation together with

55 impaired β-cell viability and/or function (6). The insulin-like growth factor (IGF) axis,

56 particularly IGF1 and IGF2, are candidates for correcting these deficiencies (7). IGFs are

57 hormones produced primarily by the liver that induce cellular proliferation via the widely-

58 expressed IGF1 receptor (IGF1R) (8). IGF1 and IGF2 are highly homologous to insulin in

59 structure (9, 10) and can mediate similar metabolic effects, but are unable to compensate

60 for the loss of insulin production in type 1 diabetes. IGF production is temporally regulated

61 such that IGF2 is thought to primarily be important for embryonic and fetal development,

62 whereas IGF1 is vital for postnatal growth (8). Possibly impacting type 1 diabetes

63 pathogenesis, IGFs have been shown to promote the regulation of T cell-mediated

64 inflammation (11-13). IGFs are also known to support endocrine (14, 15) and exocrine

65 pancreatic growth (16, 17), which, if lacking, might contribute to the profound deficiency Diabetes Page 4 of 67

66 of total pancreas mass in subjects with type 1 diabetes (18, 19). However, it remains

67 unclear whether insufficient levels or bioavailability of IGF-family ligands may directly

68 promote the pathogenesis of type 1 diabetes and/or serve as reliable biomarkers of

69 disease staging. Additionally, the regulation of IGF levels by insulin (20) begs the question

70 of whether defects in IGF levels may be primary or secondary to the loss of insulin

71 production known to occur before (21) and after clinical diagnosis (22).

72 A recent report described suppressed levels of IGF1 in plasma as children

73 seroconvert to AAb positivity (23), yet current literature lacks information on the circulating

74 levels of IGF2 during the progression to type 1 diabetes. Additionally, circulating IGFs are

75 generally bound to IGF-binding proteins (IGFBP) that modulate IGF activity by blocking

76 binding to their cognate receptors (24). A longitudinal study suggested that IGFBP3,

77 which is the most prevalent IGFBP in circulation (25), increases prior to seroconversion

78 to at least one type 1 diabetes-related AAb (23), potentially modulating IGF bioavailability.

79 Three out of seven IGFBPs, which are expressed at differing levels and have varying

80 affinities for IGF1 and IGF2 (25), are increased in serum of subjects with established type

81 1 diabetes (26). However, the majority of these proteins have not been assessed in pre-

82 type 1 diabetes. Herein, we test the hypothesis that IGF1 and IGF2 levels or

83 bioavailability, as inversely related to IGFBPs, are altered during progression to type 1

84 diabetes. Page 5 of 67 Diabetes

85 RESEARCH DESIGN AND METHODS

86 Subject Enrollment

87 All procedures were approved by Institutional Review Boards at each institution and

88 conducted in accordance with the Declaration of Helsinki. Informed consent was obtained

89 from participants (or their legal guardian in the case of minors) prior to enrollment.

90

91 Cross-Sectional Subjects

92 Subjects were recruited from clinics at the University of Florida (UF; Gainesville, FL);

93 Nemours Children’s Hospital (Orlando, FL); and Emory University (Atlanta, GA).

94 Peripheral blood samples were collected from non-fasted subjects by venipuncture in

95 serum separator vacutainer tubes (BD Biosciences) and rested overnight prior to

96 processing to minimize variance due to transport times. Sera were stored in the UF

97 Diabetes Institute (UFDI) biorepository at -20C prior to batch processing. Subjects were

98 selected from the biorepository by age, sex, ethnicity, and BMI-matching to subjects

99 positive for at least one AAb against GAD65 (GADA), IA-2 (IA-2A), or ZnT8 (ZnT8A).

100 Type 1 diabetes subjects were recruited by the Benaroya Research Institute (BRI;

101 Seattle, WA) under the BRIDge IRB protocol. Peripheral blood samples were collected

102 from fasted or non-fasted subjects in serum separator vacutainer tubes (BD Biosciences),

103 processed within 24 hours, and stored at -80C until batch processing. Detailed subject

104 demographic information is presented in Table 1.

105

106 Longitudinal Subjects Diabetes Page 6 of 67

107 Subjects at-risk for type 1 diabetes were recruited and enrolled from 2013–2014 in the

108 Type 1 Diabetes Longitudinal Biomarker Trial (T1DBIT; NCT01846312, manuscript in

109 preparation) by the Pacific Northwest Research Institute and Novo Nordisk Research

110 Center Seattle, Inc. Subjects were followed for up to 18-months post-enrollment.

111 Individuals included in this study were positive for at least one islet AAb at the time of

112 enrollment (GADA, IA-2A, or insulin (IAA)), age ≥4 and <40 years, non-pregnant or –

113 breast-feeding, had no chronic disorders, and used no medication that might impact

114 immune status or progression to type 1 diabetes. Subject demographic information is

115 included in Table 2. Samples were shipped to the clinical site at ambient temperature,

116 processed within 24 hours of draw, and stored at -80°C.

117

118 AAb Measurement

119 For UFDI samples, GADA, IA-2A, and ZnT8A were measured from serum via an Islet

120 Autoantibody Standardization Program (IASP) validated enzyme-linked immunosorbent

121 assay (ELISA) as previously described (27). For T1DBIT samples, IAA, GADA, IA-2A, IA-

122 2A, and ZnT8A were measured from serum via radioimmunoassay (RIA) with positivity

123 defined as the 97.5th percentile of control sera, as previously described (28). After

124 enrollment into T1DBIT, seroconversion was defined as at least two consecutive

125 timepoints with AAb positivity.

126

127 Glucose Metabolism

128 For BRI subjects, glucose, C-peptide, and HbA1c from mixed meal tolerance test (MMTT)

129 visit were tested at the University of Washington’s Northwest Lipid Metabolism and Page 7 of 67 Diabetes

130 Diabetes Research Laboratories (NWRL; Seattle, WA). Serum C-peptide was measured

131 by two-site immunoenzymometric assay on a Tosoh II 600 autoanalyzer and area under

132 the curve (AUC) C-peptide calculated as previously described (29). HbA1c was measured

133 either at or within three months of the time of draw.

134 For T1DBIT subjects, plasma C-peptide was measured by two-site

135 immunoenzymometric assay performed on a Tosoh II 600 autoanalyzer. HbA1c was

136 measured using a commercial HPLC-based assay at each draw by a Clinical Laboratory

137 Improvement Amendments (CLIA)-certified clinical lab. Subjects were considered pre-

138 diabetic upon HbA1c >5.7% and diagnosed with type 1 diabetes at HbA1c >6.4% (30).

139

140 IGF1 and IGF2 Quantification

141 Total IGF1 and IGF2 were measured under blinded conditions in duplicate from human

142 sera (UFDI, BRI) or plasma (T1DBIT) via commercially available ELISAs (ALPCO),

143 according to the manufacturer’s instructions. This assay provides more complete

144 recovery of IGFs compared to traditional acid-ethanol extraction methods (31). Briefly,

145 samples were pre-treated with an acidic buffer to dissociate IGFs from IGFBPs, followed

146 by sample neutralization immediately prior to plating such that the high-affinity interaction

147 between IGF and capture antibody occurs prior to IGF—IGFBP re-association. Data were

148 collected with SpectraMax M5 plate reader (Molecular Devices) and analyzed with

149 SoftMax Pro version 4.8. Data were excluded if the coefficient of variation (CV) between

150 duplicates was >25%.

151 The mean intra-assay CV for duplicates from an initial subset of samples showed

152 IGF1 (5.3%, n=36) and IGF2 (3.5%, n=32) measurements were reproducible and precise. Diabetes Page 8 of 67

153 Additionally, IGF1 and IGF2 concentrations were measured across seven plates using

154 manufacturer-provided control sera to assess inter-assay reproducibility in our hands.

155 Inter-assay CVs from low or high concentration control sera were 11.6% or 2.4%,

156 respectively, for IGF1 and 13.9% or 9.4%, respectively, for IGF2.

157

158 IGFBP Quantification

159 IGFBP1-7 levels were measured under blinded conditions using Luminex assay on a

160 Milliplex platform (Millipore EMD), according to the manufacturer’s instructions. Data were

161 analyzed with Milliplex Analyst version 5.1.0.0. For reported analytes, values outside of

162 the standard curve were assigned the value of the limit of detection (LOD) in the five-

163 parameter logistic regression, as determined by Milliplex Analyst software. Although a

164 variety of methods may be used to impute values outside of the LOD, selection of “fill in”

165 values provides a relatively unbiased estimate of data variance when <30% of data are

166 outside such limits (32), as seen in our data set.

167

168 Statistical Analysis

169 BMI percentiles were calculated according to subject age and sex with available height

170 and weight data using a publicly-available SAS program based on growth charts from the

171 Center for Disease Control and Prevention (33). Normalized IGF1 and IGF2 percentiles

172 were determined via curve fitting of published reference ranges (Table S1,S2) binned for

173 age and sex (34).

174 Analyses were performed using GraphPad Prism software version 7.0. Data are

175 presented as mean ± standard deviation (SD) unless otherwise specified. Cross-sectional Page 9 of 67 Diabetes

176 IGF and IGFBP levels across cohorts shown as violin plots with median and quartiles

177 overlaid. Multiplicity adjusted p-values < 0.05 were considered significant.

178

179 Data and Resource Availability

180 The datasets presented are available from the corresponding author upon reasonable

181 request. No applicable resources were generated or analyzed during the current study.

182 Diabetes Page 10 of 67

183 RESULTS

184 Total IGF Levels Decrease in Serum of AAb+ Subjects At-Risk for Type 1 Diabetes

185 We investigated the IGF axis in a cross-sectional cohort of N=305 pediatric and

186 adolescent subjects with or at varying degrees of risk for clinical diagnosis of type 1

187 diabetes. There were no significant differences in any of the continuous or categorical

188 demographics between clinical groups (Table 1). Given that some analytes in the IGF

189 axis are directly modulated by growth hormone (GH) (35), which peaks during puberty,

190 we carefully selected samples from age-matched subjects (Table 1). IGF1 levels have

191 been shown to peak around puberty (36), as replicated in our cohort (Fig. 1A), but IGF2

192 levels are not regulated by GH (8).

193 The peak age of type 1 diabetes onset in the UFDI cohort was approximately 10-

194 12 years (Table 1, Fig. 1A). This may be a critical time period for the maintenance of

195 immune tolerance to β-cell antigens, as puberty represents a dynamic period of growth

196 and development that includes metabolic stress (37). Although IGF1 levels peaked

197 around puberty in all groups (Fig. 1A), AAb+ subjects showed significantly lower IGF1

198 levels for their age as compared to AAb- controls (Fig. 1B). Analyzing IGF levels

199 normalized for age and sex (Fig. 1C,D, Table S3) revealed that total serum IGF1 and

200 IGF2 levels were significantly lower in AAb+ subjects compared to AAb- relatives, and

201 both analytes appeared to recover following type 1 diabetes diagnosis, although with

202 different kinetics. Intriguingly, this observation of lower IGF1 and IGF2 levels not only

203 applied to high-risk subjects with multiple AAb (≥2AAb+), but was also noted in subjects

204 with only a single type 1 diabetes-related AAb (1AAb+; Fig. 1E,F, Table S3), which is

205 early in the disease process and prior to detectable loss of β-cell mass or function Page 11 of 67 Diabetes

206 (considered pre-Stage 1 disease) (4). These findings remained statistically significant

207 when analyzing raw (Fig. S1, Table S4) and age-normalized IGF1 and IGF2 data (Fig.

208 1C-F, Table S3). The majority of individuals showed IGF1 and IGF2 levels within normal

209 published reference ranges (34). However, those below the fifth percentile, which is

210 traditionally diagnostic for idiopathic GH deficiency (38), were enriched in 1AAb+ and

211 ≥2AAb+ subjects compared to all other groups (Fig. 1E,F, Table S3,S5), despite a lack

212 of consensus on whether growth is impaired in pre-type 1 diabetes (39).

213 To determine IGF bioavailability, IGFBP levels were measured in the same cohort.

214 We found that circulating IGFBP1 levels were significantly higher in AAb+ versus control

215 subjects. In addition, this increase remained statistically significant in subjects with

216 established type 1 diabetes (Fig. S2A, Table S6). IGFBP1 levels are inversely associated

217 with insulin levels (40), thus, high IGFBP1 may reflect the initiation of metabolic

218 dysregulation in AAb+ subjects. The major component of IGFBP in serum, IGFBP3, was

219 detected at comparable levels across all groups examined (Fig. S2B, Table S6). IGFBP2,

220 IGFBP4, and IGFBP5 were not above the assay LOD in the majority of subjects in this

221 study. IGFBP6 and IGFBP7 levels were decreased in subjects with recent-onset type 1

222 diabetes as compared to AAb- controls or relatives, respectively (Fig. S2C,D, Table S6),

223 while IGFBP7 levels were also significantly lower in AAb+ subjects as compared to AAb-

224 relatives (Fig. S2D, Table S6). Importantly, in contrast to IGF1 and IGF2, there is not a

225 published reference range for this IGFBP laboratory assay. Upon stratifying the cohort by

226 number of AAb, IGFBPs with high affinity for IGFs (i.e., IGFBP1, IGFBP3 and IGFBP6)

227 were not significantly different between AAb-, 1AAb+, and ≥2AAb+ subjects (Fig. S2E-G,

228 Table S6). In contrast, the low affinity IGFBP7 showed significantly lower concentrations Diabetes Page 12 of 67

229 in ≥2AAb+ and recent-onset type 1 diabetes subjects as compared to AAb- subjects (Fig.

230 S2H, Table S6). Together, these data suggest that most high-affinity IGFBPs do not

231 appear to be modulated in pre-type 1 diabetes and that IGFBP measurements do not

232 improve the capacity of IGF1 and IGF2 levels to distinguish type 1 diabetes stages (Table

233 S7).

234

235 IGF1 Decreases with Increased Duration of Type 1 Diabetes

236 Since IGF1 and IGF2 percentiles were significantly decreased during progression to

237 disease, we next asked whether IGF modulation was associated with type 1 diabetes

238 duration. We measured serum IGF levels from a cross-sectional BRI cohort with

239 established type 1 diabetes (N=50), comprised primarily of fasted subjects (Table 1), as

240 compared to UFDI established type 1 diabetes subjects (N=68) reported above. Of note

241 however, disease onset occurred at a significantly older age for the BRI versus the UFDI

242 cohort (Fig. 2A). IGF1 percentiles appeared to drop with longer type 1 diabetes duration

243 in both the primarily pediatric UFDI cohort as well as the BRI cohort comprised of children

244 and adults (Fig. 2B). Interestingly, the slope of IGF1 loss appeared to be steeper in the

245 younger-onset UFDI cohort than the older-onset BRI cohort (Fig. 2B), mirroring the

246 known faster loss of C-peptide in younger-onset subjects (41). IGF2 percentiles, on the

247 other hand, were not associated with disease duration in either cohort (Fig. 2C).

248 The observation of reduced levels of IGF1 with increasing type 1 diabetes duration

249 suggested a potential association between IGF1 levels and remaining β-cell function. In

250 subjects from the BRI cohort, IGF1 showed a trend toward correlation with MMTT-

251 stimulated C-peptide AUC (Fig. S3A), although HbA1c (Fig. S3B) and fasting blood Page 13 of 67 Diabetes

252 glucose levels were not associated with IGF1 (Fig. S3C). As expected, C-peptide AUC

253 significantly decreased with disease duration (Fig. S3D). Therefore, we cannot rule out

254 the possibility that the trending association between IGF1 and C-peptide may be driven

255 by other factors related to disease duration. These data supported the notion that IGF1

256 levels may be associated with residual β-cell function and fueled our investigation of

257 whether IGFs were modulated longitudinally during the development of type 1 diabetes.

258

259 IGF Levels Show Longitudinal Stability in Single AAb+ Subjects

260 We next assessed the longitudinal stability of IGF1 and IGF2 from up to 13 timepoints in

261 the T1DBIT cohort of N=40 pediatric and young adult subjects at-risk for type 1 diabetes

262 (Table 2). Importantly, the groups were well age- and BMI-matched, and subject BMI did

263 not significantly deviate during the study (Table 2). We hypothesized that IGF modulation

264 may mirror the loss of insulin preceding diagnosis and that continued loss of endogenous

265 insulin post-onset may also be associated with a loss of IGF1, in particular. For subjects

266 who entered the study with 1AAb+ and did not progress to disease, IGF1 and IGF2 levels

267 were relatively stable within the individual but variable between subjects (Table 3, Fig.

268 S4,S5), as one would expect given the puberty-adjacent age range of this cohort (Table

269 2). There were no significant differences in intra-subject CVs for IGF1 or IGF2 when

270 comparing those with 1AAb+ at enrollment, ≥2AAb+ at enrollment, and those who

271 developed type 1 diabetes during the study (Table 3).

272

273 Longitudinal IGF1 Levels Decrease Over Time in Multiple AAb+ Subjects Diabetes Page 14 of 67

274 In order to assess grouped IGF trajectories, IGF levels were normalized by their baseline

275 values per subject to minimize the effects of age and sex. Subjects with 1AAb+ at time of

276 enrollment did not show a significant change in IGF1 or IGF2 levels over the course of

277 the study (Fig. 3, Fig. S4,S5). In contrast, a subset of subjects with ≥2AAb+ showed a

278 significant decline in IGF1 levels over time (Fig. 3A, Fig. S6). Indeed, all subjects with

279 longitudinally decreasing IGF1 levels were positive for three or more AAb, although many

280 subjects with three or more AAb had stable IGF1 levels (Table 4). With regard to AAb

281 specificities, all of the subjects with decreasing IGF1 levels possessed IA-2A and ZnT8A

282 (Table 4), which have previously been suggested to serve as an age-independent means

283 of identifying those progressing quickly to disease onset (42). One ≥2AAb+ subject did

284 show a potentially puberty-related increase in IGF1 levels over time (Fig. 3A, Fig. S6);

285 however, we wish to highlight that this subject lost IAA reactivity during the study (Table

286 4) and therefore, may have a lower risk of developing disease (43). Additionally, a subset

287 of ≥2AAb+ subjects showed increased IGF2 levels over time (Fig. 3B, Fig. S7), although

288 it is important to note that a similar trajectory was observed in one of the 1AAb+ subjects

289 (Fig. 3B, Fig. S5). In those who developed type 1 diabetes during the study, IGF1 levels

290 decreased significantly over time (Fig. 3A, Fig. S8), and this decrease was seen

291 immediately prior to diagnosis with most continuing to slowly decrease (Fig. S8). For

292 IGF2, there was more variability in individual levels and no consistency in trend over time

293 (Fig. 3B, Fig. S9).

294

295 Longitudinal IGF1 Levels Decrease Over Time Following Type 1 Diabetes Diagnosis Page 15 of 67 Diabetes

296 For those subjects who developed type 1 diabetes within the timeframe of the T1DBIT

297 study, we compared the trajectory of IGFs pre- and post-diagnosis within the same

298 individual. Here, IGFs were normalized by either their baseline values pre-diagnosis or

299 values at the sample collected nearest to diagnosis, respectively. IGF1 was shown to

300 remain relatively stable pre-diagnosis but to significantly decrease post-diagnosis (Fig.

301 4A, Fig. S8). On the other hand, the pre- and post-diagnosis trajectories for IGF2 were

302 not significantly different (Fig. 4B, Fig. S9). We also observed that longitudinal C-peptide

303 levels significantly decreased post-diagnosis (Fig. 4C), suggesting that IGF1 and C-

304 peptide may be simultaneously modulated. Diabetes Page 16 of 67

306 DISCUSSION

307 The current dogma surrounding the pathogenesis of type 1 diabetes suggests disease

308 occurs from inciting autoimmunity leading to a fundamental metabolic defect. However,

309 the metabolic and growth factor derangements that occur prior to overt hyperglycemia

310 remain incompletely characterized during the natural history of the disease. Therefore,

311 we sought to further characterize the IGF axis in both cross-sectional and longitudinal

312 cohorts representing the various stages of type 1 diabetes. Herein, we report IGF1 and

313 IGF2 defects exist prior to type 1 diabetes onset. In agreement with our results, Peet et

314 al. found IGF1 levels to be significantly lower in AAb+ versus AAb- subjects, but in this

315 previous study, the difference was only apparent at 12 months of age (23). IGFBP3

316 concentrations were not significantly different when comparing AAb+ against AAb-

317 individuals at all ages (23), again in agreement with our cross-sectional data. To our

318 knowledge, our results are the first to report that total IGF2 levels were also significantly

319 decreased in AAb+ as compared to AAb- relatives, analogous to the observations with

320 IGF1. The majority of high-affinity IGFBPs, however, were not significantly altered in pre-

321 type 1 diabetes, suggesting that total IGF levels may accurately reflect IGF bioavailability.

322 As we observed for all three cohorts in this study, others have previously reported

323 that IGF1 significantly decreases post-diagnosis of type 1 diabetes (44, 45). IGF1 levels

324 have also been positively correlated with C-peptide as a measure of residual β-cell

325 function (46). Confirmation of the association between IGF1 levels and disease

326 duration/metabolism provided strong support for evaluating longitudinal IGF trajectories

327 in subjects with pre-type 1 diabetes. We noted that IGF1 and IGF2 levels show

328 longitudinal stability in the T1DBIT cohort, supporting the concept of tracking these Page 17 of 67 Diabetes

329 analytes as potential prognostic biomarkers. In fact, our intra-person CVs for IGF1 in

330 1AAb+ subjects were remarkably similar to those previously reported in healthy adults

331 (47). This report is the first to demonstrate that IGF1 decreased over time in ≥2AAb+

332 subjects but did not change over time in 1AAb+ subjects, suggesting that this change was

333 not related to age and/or puberty– and importantly identifying a new potential explanation

334 for the differing rates of progression among 1AAb+ versus ≥2AAb+ subjects. We found

335 that the subjects with decreasing IGF1 were positive for IA-2A and ZnT8A, AAb that tend

336 to appear closer to clinical diagnosis as a consequence of antigenic spreading (48).

337 These data, in combination with the association between IGF1 and C-peptide changes,

338 suggest that IGF1 and insulin may be lost simultaneously, with these analytes potentially

339 synergizing to dysregulate glucose metabolism prior to clinical diagnosis.

340 In terms of potential study limitations, the UF and T1DBIT samples were drawn

341 from non-fasted subjects, limiting our ability to address the association with glucose levels

342 in these cohorts. We also found that ≥2AAb+ subjects trended toward increasing IGF2

343 levels over time. However, the levels were variable in the cohort and increasing levels

344 were also seen in a 1AAb+ subject which brings this finding under question as potentially

345 an age-related increase rather than strictly a consequence of disease pathogenesis.

346 Clearly, this issue needs to be addressed in future efforts, possibly within a larger study.

347 An additional and potentially important caveat to note is that AAb were measured using

348 different techniques for the T1DBIT and UFDI cohorts. Specifically, for the T1DBIT cohort,

349 AAbs were measured via RIA, but the UFDI cohort used ELISA. The IASP has yet to

350 validate an ELISA with adequate sensitivity and specificity for IAA (49); therefore, IAA

351 were not measured in the cross-sectional UFDI subjects. Since IAA and GADA are often Diabetes Page 18 of 67

352 the first AAb to appear in the natural history of T1D (50), we concede that it is possible

353 that some proportion of the UFDI 1AAb+ subjects may actually possess reactivity to

354 multiple AAb+. This discrepancy may explain why we saw low IGF1 levels cross-

355 sectionally in 1AAb+ and ≥2AAb+ UFDI subjects, but only noted decreasing IGF1 levels

356 in multiple AAb+ subjects from the T1DBIT subjects.

357 While the trajectories of IGF2 were not significantly different pre- and post-

358 diagnosis for the collective T1DBIT cohort, IGF1 was shown to decrease over time post-

359 diagnosis as compared to pre-diagnosis in the same individual. Despite the small sample

360 size for the subjects progressing to type 1 diabetes, using the same subject as a

361 comparison within a short timeframe, before and after diagnosis, provides more

362 confidence that the effect is not driven solely by covariates like age and puberty. Our

363 longitudinal findings mirror those of a recently published study which reported IGF1

364 decreased by two years post-diagnosis; however, this study did not have pre-diabetic

365 timepoints for comparison (45). This study noted a temporary increase in IGF1

366 immediately after type 1 diabetes onset, which they ascribe to the impact of exogenous

367 insulin delivery (45). While we did see this rebound in the cross-sectional UFDI cohort,

368 the T1DBIT cohort do not appear to show this effect. It is important to note that in the

369 cited study, the average HbA1c at diagnosis was 10.5% (45). Recent-onset UFDI subjects

370 were identified from presentation in clinic, so the average HbA1c upon diagnosis was

371 likely higher than in our longitudinal cohort wherein HbA1c was measured monthly,

372 resulting in prompt diagnoses at HbA1c values close to 6.5%. Therefore, some

373 endogenous insulin, potentially acting directly on the liver, may have remained in the Page 19 of 67 Diabetes

374 T1DBIT cohort with initially lower exogenous insulin dosages (51), explaining the

375 discrepancy in IGF1 rebound post-diagnosis in all of the cohorts described.

376 The results presented herein suggest that decreased total IGF levels may occur

377 both before and after the clinical diagnosis of type 1 diabetes (Stage 3 disease). Aberrant

378 modulation of IGFs prior to disease onset could complement existing disease staging

379 efforts in combination with AAb surveillance and glycemic monitoring. Longitudinal

380 studies of longer duration in fasted subjects are needed to assess whether decreasing

381 IGF1 levels in multiple AAb+ subjects would improve disease prediction in place of or in

382 combination with OGTT. These data additionally support the further investigation of IGF

383 modulation as potentially contributing to type 1 diabetes pathogenesis, representing a

384 novel therapeutic target to possibly inhibit autoimmunity or preserve pancreatic health (7)

385 in AAb+ subjects. Diabetes Page 20 of 67

386 ACKNOWLEDGEMENTS

387 We thank Joshua Peterson and Kieran McGrail (University of Florida), Jordan Klaiman

388 and Rachel Hartley (Benaroya Research Institute), and Rachel Hervey (Pacific Northwest

389 Research Institute) for their technical assistance with sample processing, biorepository

390 management, and procurement of demographic data. Special thanks are extended to all

391 study subjects and their families for generously participating. We thank the clinical staff

392 at University of Florida, Nemours Children’s Hospital, Emory University, Benaroya

393 Research Institute (specifically Jenna Snavely and others conducting the BRIDge study

394 of Diabetes), Pacific Northwest Research Institute, and Novo Nordisk Research Center

395 for sample acquisition.

396

397 Funding

398 Project support was provided by grants from the National Institutes of Health (F31

399 DK117548 to M.R.S.; T32 DK108736 to M.A.A.; Clinical and Translational Science Award

400 URL1TR001427 to D.A.S.; P01 AI42288 to M.A.A. and T.M.B.; and R01 DK106191 to

401 T.M.B.); the NIDDK-supported Human Islet Research Network (HIRN,

402 RRID:SCR_014393; https://hirnetwork.org; UC4 DK104216-01 to D.A.S.); the Children’s

403 Miracle Network (M.R.S. and M.J.H.); and JDRF (3-SRA-2016-209-Q-R to C.S.). W.A.H.

404 was supported in this project by a grant from NovoNordisk Inc.

405

406 Duality of Interest

407 J.D.W. and M.v.H. are paid employees of Novo Nordisk. W.A.H. has received research

408 support from Novo Nordisk. Page 21 of 67 Diabetes

409

410 Author Contributions

411 MRS researched and analyzed the data in all figures and tables and wrote the manuscript;

412 CHW conceived the study and reviewed/edited the manuscript; SMG researched the data

413 in figures 1A-C and reviewed/edited the manuscript; ALP contributed to discussion and

414 reviewed/edited the manuscript; RB advised on statistical analysis and reviewed/edited

415 the manuscript; AM recruited subjects; MJH, DAS, JDW, MvH, WAH, and CS designed

416 the respective cohorts and recruited subjects, contributed to discussion, and

417 reviewed/edited the manuscript; MAA and TMB conceived the study and reviewed/edited

418 the manuscript.

419

420 Guarantor Statement

421 As the guarantor of this work, Todd M. Brusko assumes responsibility for ethical

422 completion of the study, integrity of the data, and accuracy of the data analysis reported

423 herein.

424

425 Prior Presentation

426 Parts of this study were presented in abstract form at the 77th Scientific Sessions of the

427 American Diabetes Association, San Diego, California, 9-13 June 2017 and the 78th

428 Scientific Sessions of the American Diabetes Association, Orlando, Florida, 22-26 June

429 2018. Diabetes Page 22 of 67

430 REFERENCES

431 1. Raab J, Haupt F, Scholz M, Matzke C, Warncke K, Lange K, Assfalg R, Weininger 432 K, Wittich S, Löbner S, Beyerlein A, Nennstiel-Ratzel U, Lang M, Laub O, Dunstheimer 433 D, Bonifacio E, Achenbach P, Winkler C, Ziegler AG, Group FdS. Capillary blood islet 434 autoantibody screening for identifying pre-type 1 diabetes in the general population: 435 design and initial results of the Fr1da study. BMJ Open. 2016;6(5):e011144. Epub 436 2016/05/18. doi: 10.1136/bmjopen-2016-011144. PubMed PMID: 27194320; PMCID: 437 PMC4874167. 438 2. Hommel A, Haupt F, Delivani P, Winkler C, Stopsack M, Wimberger P, Nitzsche 439 K, Heinke S, Naeke A, Ceglarek U, Thiery J, Bergert R, Stadthaus D, Groeger K, Heubner 440 G, Schramm U, Dziambor U, Zirkel A, Kiess W, Mueller I, Lange K, Berner R, Bonifacio 441 E, Ziegler AG, Group TFkS. Screening for Type 1 Diabetes Risk in Newborns: The 442 Freder1k Pilot Study in Saxony. Horm Metab Res. 2018;50(1):44-9. Epub 2017/11/09. 443 doi: 10.1055/s-0043-120921. PubMed PMID: 29121687. 444 3. Winkler C, Haupt F, Heigermoser M, Zapardiel-Gonzalo J, Ohli J, Faure T, Kalideri 445 E, Hommel A, Delivani P, Berner R, Kordonouri O, Roloff F, von dem Berge T, Lange K, 446 Oltarzewski M, Glab R, Szypowska A, Snape MD, Vatish M, Todd JA, Larsson HE, 447 Ramelius A, Kördel J, Casteels K, Paulus J, Ziegler AG, Bonifacio E, Group GS. 448 Identification of infants with increased type 1 diabetes genetic risk for enrollment into 449 Primary Prevention Trials-GPPAD-02 study design and first results. Pediatr Diabetes. 450 2019. Epub 2019/06/13. doi: 10.1111/pedi.12870. PubMed PMID: 31192505. 451 4. Insel RA, Dunne JL, Atkinson MA, Chiang JL, Dabelea D, Gottlieb PA, Greenbaum 452 CJ, Herold KC, Krischer JP, Lernmark Å, Ratner RE, Rewers MJ, Schatz DA, Skyler JS, 453 Sosenko JM, Ziegler AG. Staging presymptomatic type 1 diabetes: a scientific statement 454 of JDRF, the Endocrine Society, and the American Diabetes Association. Diabetes Care. 455 2015;38(10):1964-74. doi: 10.2337/dc15-1419. PubMed PMID: 26404926; PMCID: 456 PMC5321245. 457 5. Knip M, Korhonen S, Kulmala P, Veijola R, Reunanen A, Raitakari OT, Viikari J, 458 Akerblom HK. Prediction of type 1 diabetes in the general population. Diabetes Care. 459 2010;33(6):1206-12. doi: 10.2337/dc09-1040. PubMed PMID: 20508230; PMCID: 460 PMC2875424. 461 6. Atkinson MA, Eisenbarth GS, Michels AW. Type 1 diabetes. Lancet. 462 2014;383(9911):69-82. Epub 2013/07/26. doi: 10.1016/S0140-6736(13)60591-7. 463 PubMed PMID: 23890997; PMCID: PMC4380133. 464 7. Shapiro MR, Atkinson MA, Brusko TM. Pleiotropic roles of the insulin-like growth 465 factor axis in type 1 diabetes. Curr Opin Endocrinol Diabetes Obes. 2019. Epub 466 2019/05/27. doi: 10.1097/MED.0000000000000484. PubMed PMID: 31145130. 467 8. Dupont J, Holzenberger M. Biology of insulin-like growth factors in development. 468 Birth Defects Res C Embryo Today. 2003;69(4):257-71. doi: 10.1002/bdrc.10022. 469 PubMed PMID: 14745968. 470 9. Rinderknecht E, Humbel RE. The amino acid sequence of human insulin-like 471 growth factor I and its structural homology with proinsulin. J Biol Chem. 472 1978;253(8):2769-76. PubMed PMID: 632300. 473 10. Rinderknecht E, Humbel RE. Primary structure of human insulin-like growth factor 474 II. FEBS Lett. 1978;89(2):283-6. PubMed PMID: 658418. Page 23 of 67 Diabetes

475 11. Bilbao D, Luciani L, Johannesson B, Piszczek A, Rosenthal N. Insulin-like growth 476 factor-1 stimulates regulatory T cells and suppresses autoimmune disease. EMBO Mol 477 Med. 2014;6(11):1423-35. Epub 2014/10/22. doi: 10.15252/emmm.201303376. PubMed 478 PMID: 25339185; PMCID: PMC4237469. 479 12. Yang G, Geng XR, Song JP, Wu Y, Yan H, Zhan Z, Yang L, He W, Liu ZQ, Qiu S, 480 Liu Z, Yang PC. Insulin-like growth factor 2 enhances regulatory T-cell functions and 481 suppresses food allergy in an experimental model. J Allergy Clin Immunol. 482 2014;133(6):1702-8.e5. Epub 2014/04/01. doi: 10.1016/j.jaci.2014.02.019. PubMed 483 PMID: 24698315. 484 13. Giuliani C, Saji M, Bucci I, Fiore G, Liberatore M, Singer DS, Monaco F, Kohn LD, 485 Napolitano G. Transcriptional regulation of major histocompatibility complex class I gene 486 by insulin and IGF-I in FRTL-5 thyroid cells. J Endocrinol. 2006;189(3):605-15. doi: 487 10.1677/joe.1.06486. PubMed PMID: 16731791. 488 14. Lingohr MK, Dickson LM, McCuaig JF, Hugl SR, Twardzik DR, Rhodes CJ. 489 Activation of IRS-2-mediated signal transduction by IGF-1, but not TGF-alpha or EGF, 490 augments pancreatic beta-cell proliferation. Diabetes. 2002;51(4):966-76. PubMed PMID: 491 11916914. 492 15. Modi H, Jacovetti C, Tarussio D, Metref S, Madsen OD, Zhang FP, Rantakari P, 493 Poutanen M, Nef S, Gorman T, Regazzi R, Thorens B. Autocrine Action of IGF2 494 Regulates Adult β-Cell Mass and Function. Diabetes. 2015;64(12):4148-57. Epub 495 2015/09/17. doi: 10.2337/db14-1735. PubMed PMID: 26384384. 496 16. Takahashi H, Okamura D, Starr ME, Saito H, Evers BM. Age-dependent reduction 497 of the PI3K regulatory subunit p85α suppresses pancreatic acinar cell proliferation. Aging 498 Cell. 2012;11(2):305-14. Epub 2012/02/01. doi: 10.1111/j.1474-9726.2011.00787.x. 499 PubMed PMID: 22212451; PMCID: PMC3408710. 500 17. Kido Y, Nakae J, Hribal ML, Xuan S, Efstratiadis A, Accili D. Effects of mutations 501 in the insulin-like growth factor signaling system on embryonic pancreas development 502 and beta-cell compensation to insulin resistance. J Biol Chem. 2002;277(39):36740-7. 503 Epub 2002/07/05. doi: 10.1074/jbc.M206314200. PubMed PMID: 12101187. 504 18. Campbell-Thompson ML, Kaddis JS, Wasserfall C, Haller MJ, Pugliese A, Schatz 505 DA, Shuster JJ, Atkinson MA. The influence of type 1 diabetes on pancreatic weight. 506 Diabetologia. 2016;59(1):217-21. doi: 10.1007/s00125-015-3752-z. PubMed PMID: 507 26358584; PMCID: PMC4670792. 508 19. Battaglia M, Atkinson MA. The streetlight effect in type 1 diabetes. Diabetes. 509 2015;64(4):1081-90. doi: 10.2337/db14-1208. PubMed PMID: 25805758; PMCID: 510 PMC4375074. 511 20. Holt RI, Simpson HL, Sönksen PH. The role of the growth hormone-insulin-like 512 growth factor axis in glucose homeostasis. Diabet Med. 2003;20(1):3-15. PubMed PMID: 513 12519314. 514 21. Evans-Molina C, Sims EK, DiMeglio LA, Ismail HM, Steck AK, Palmer JP, Krischer 515 JP, Geyer S, Xu P, Sosenko JM, Group TDTS. β Cell dysfunction exists more than 5 516 years before type 1 diabetes diagnosis. JCI Insight. 2018;3(15). Epub 2018/08/09. doi: 517 10.1172/jci.insight.120877. PubMed PMID: 30089716; PMCID: PMC6129118. 518 22. Greenbaum CJ, Beam CA, Boulware D, Gitelman SE, Gottlieb PA, Herold KC, 519 Lachin JM, McGee P, Palmer JP, Pescovitz MD, Krause-Steinrauf H, Skyler JS, Sosenko 520 JM, Group TDTS. Fall in C-peptide during first 2 years from diagnosis: evidence of at Diabetes Page 24 of 67

521 least two distinct phases from composite Type 1 Diabetes TrialNet data. Diabetes. 522 2012;61(8):2066-73. Epub 2012/06/11. doi: 10.2337/db11-1538. PubMed PMID: 523 22688329; PMCID: PMC3402330. 524 23. Peet A, Hämäläinen AM, Kool P, Ilonen J, Knip M, Tillmann V, Group DS. 525 Circulating IGF1 and IGFBP3 in relation to the development of β-cell autoimmunity in 526 young children. Eur J Endocrinol. 2015;173(2):129-37. Epub 2015/05/06. doi: 527 10.1530/EJE-14-1078. PubMed PMID: 25947142. 528 24. Livingstone C. IGF2 and cancer. Endocr Relat Cancer. 2013;20(6):R321-39. Epub 529 2013/10/24. doi: 10.1530/ERC-13-0231. PubMed PMID: 24080445. 530 25. Baxter RC. IGF binding proteins in cancer: mechanistic and clinical insights. Nat 531 Rev Cancer. 2014;14(5):329-41. Epub 2014/04/10. doi: 10.1038/nrc3720. PubMed PMID: 532 24722429. 533 26. Sharma A, Purohit S, Sharma S, Bai S, Zhi W, Ponny SR, Hopkins D, Steed L, 534 Bode B, Anderson SW, She JX. IGF-Binding Proteins in Type-1 Diabetes Are More 535 Severely Altered in the Presence of Complications. Front Endocrinol (Lausanne). 536 2016;7:2. Epub 2016/01/29. doi: 10.3389/fendo.2016.00002. PubMed PMID: 26858687; 537 PMCID: PMC4731488. 538 27. Wasserfall C, Montgomery E, Yu L, Michels A, Gianani R, Pugliese A, Nierras C, 539 Kaddis JS, Schatz DA, Bonifacio E, Atkinson MA. Validation of a rapid type 1 diabetes 540 autoantibody screening assay for community-based screening of organ donors to identify 541 subjects at increased risk for the disease. Clin Exp Immunol. 2016;185(1):33-41. Epub 542 2016/05/04. doi: 10.1111/cei.12797. PubMed PMID: 27029857; PMCID: PMC4908288. 543 28. Woo W, LaGasse JM, Zhou Z, Patel R, Palmer JP, Campus H, Hagopian WA. A 544 novel high-throughput method for accurate, rapid, and economical measurement of 545 multiple type 1 diabetes autoantibodies. J Immunol Methods. 2000;244(1-2):91-103. doi: 546 10.1016/s0022-1759(00)00259-3. PubMed PMID: 11033022. 547 29. Greenbaum CJ, Mandrup-Poulsen T, McGee PF, Battelino T, Haastert B, 548 Ludvigsson J, Pozzilli P, Lachin JM, Kolb H, Group TDTNR, Group EC-PTS. Mixed-meal 549 tolerance test versus glucagon stimulation test for the assessment of beta-cell function in 550 therapeutic trials in type 1 diabetes. Diabetes Care. 2008;31(10):1966-71. Epub 551 2008/07/15. doi: 10.2337/dc07-2451. PubMed PMID: 18628574; PMCID: PMC2551636. 552 30. Association AD. Standards of medical care in diabetes--2014. Diabetes Care. 553 2014;37 Suppl 1:S14-80. doi: 10.2337/dc14-S014. PubMed PMID: 24357209. 554 31. Ranke MB. Diagnostics of endocrine function in children and adolescents. 3rd rev. 555 and extended ed. Basel ; New York: Karger; 2003. xv, 545 p. p. 556 32. Lubin JH, Colt JS, Camann D, Davis S, Cerhan JR, Severson RK, Bernstein L, 557 Hartge P. Epidemiologic evaluation of measurement data in the presence of detection 558 limits. Environ Health Perspect. 2004;112(17):1691-6. PubMed PMID: 15579415; 559 PMCID: PMC1253661. 560 33. Kuczmarski RJ, Ogden CL, Guo SS, Grummer-Strawn LM, Flegal KM, Mei Z, Wei 561 R, Curtin LR, Roche AF, Johnson CL. 2000 CDC Growth Charts for the United States: 562 methods and development. Vital Health Stat 11. 2002(246):1-190. PubMed PMID: 563 12043359. 564 34. Blum W, Schweizer R. Insulin-like growth factors and their binding proteins. Ranke 565 M, editor. Basel: Karger; 2003. 166-99 p. Page 25 of 67 Diabetes

566 35. Clayton PE, Banerjee I, Murray PG, Renehan AG. Growth hormone, the insulin- 567 like growth factor axis, insulin and cancer risk. Nat Rev Endocrinol. 2011;7(1):11-24. Epub 568 2010/10/19. doi: 10.1038/nrendo.2010.171. PubMed PMID: 20956999. 569 36. Goran MI, Gower BA. Longitudinal study on pubertal insulin resistance. Diabetes. 570 2001;50(11):2444-50. PubMed PMID: 11679420. 571 37. Bonner-Weir S, Aguayo-Mazzucato C, Weir GC. Dynamic development of the 572 pancreas from birth to adulthood. Ups J Med Sci. 2016;121(2):155-8. doi: 573 10.3109/03009734.2016.1154906. PubMed PMID: 26998806; PMCID: PMC4900072. 574 38. Bussières L, Souberbielle JC, Pinto G, Adan L, Noel M, Brauner R. The use of 575 insulin-like growth factor 1 reference values for the diagnosis of growth hormone 576 deficiency in prepubertal children. Clin Endocrinol (Oxf). 2000;52(6):735-9. doi: 577 10.1046/j.1365-2265.2000.00999.x. PubMed PMID: 10848878. 578 39. Nambam B, Schatz D. Growth hormone and insulin-like growth factor-I axis in type 579 1 diabetes. Growth Horm IGF Res. 2018;38:49-52. Epub 2017/12/13. doi: 580 10.1016/j.ghir.2017.12.005. PubMed PMID: 29249623. 581 40. Brismar K, Fernqvist-Forbes E, Wahren J, Hall K. Effect of insulin on the hepatic 582 production of insulin-like growth factor-binding protein-1 (IGFBP-1), IGFBP-3, and IGF-I 583 in insulin-dependent diabetes. J Clin Endocrinol Metab. 1994;79(3):872-8. doi: 584 10.1210/jcem.79.3.7521354. PubMed PMID: 7521354. 585 41. Besser REJ, Ludvigsson J, Hindmarsh PC, Cole TJ. Exploring C-peptide loss in 586 type 1 diabetes using growth curve analysis. PLoS One. 2018;13(7):e0199635. Epub 587 2018/07/03. doi: 10.1371/journal.pone.0199635. PubMed PMID: 29969494; PMCID: 588 PMC6029769. 589 42. Gorus FK, Balti EV, Vermeulen I, Demeester S, Van Dalem A, Costa O, Dorchy H, 590 Tenoutasse S, Mouraux T, De Block C, Gillard P, Decochez K, Wenzlau JM, Hutton JC, 591 Pipeleers DG, Weets I, Registry BD. Screening for insulinoma antigen 2 and zinc 592 transporter 8 autoantibodies: a cost-effective and age-independent strategy to identify 593 rapid progressors to clinical onset among relatives of type 1 diabetic patients. Clin Exp 594 Immunol. 2013;171(1):82-90. doi: 10.1111/j.1365-2249.2012.04675.x. PubMed PMID: 595 23199327; PMCID: PMC3530099. 596 43. Endesfelder D, Hagen M, Winkler C, Haupt F, Zillmer S, Knopff A, Bonifacio E, 597 Ziegler AG, Zu Castell W, Achenbach P. A novel approach for the analysis of longitudinal 598 profiles reveals delayed progression to type 1 diabetes in a subgroup of multiple-islet- 599 autoantibody-positive children. Diabetologia. 2016;59(10):2172-80. Epub 2016/07/11. 600 doi: 10.1007/s00125-016-4050-0. PubMed PMID: 27400691. 601 44. Palta M, LeCaire TJ, Sadek-Badawi M, Herrera VM, Danielson KK. The trajectory 602 of IGF-1 across age and duration of type 1 diabetes. Diabetes Metab Res Rev. 603 2014;30(8):777-83. doi: 10.1002/dmrr.2554. PubMed PMID: 24845759; PMCID: 604 PMC4236234. 605 45. Chisalita SI, Ludvigsson J. Insulin-Like Growth Factor-1 at Diagnosis and during 606 Subsequent Years in Adolescents with Type 1 Diabetes. J Diabetes Res. 607 2018;2018:8623560. Epub 2018/03/20. doi: 10.1155/2018/8623560. PubMed PMID: 608 29744370; PMCID: PMC5883934. 609 46. Hedman CA, Frystyk J, Lindström T, Chen JW, Flyvbjerg A, Ørskov H, Arnqvist 610 HJ. Residual beta-cell function more than glycemic control determines abnormalities of Diabetes Page 26 of 67

611 the insulin-like growth factor system in type 1 diabetes. J Clin Endocrinol Metab. 612 2004;89(12):6305-9. doi: 10.1210/jc.2004-0572. PubMed PMID: 15579794. 613 47. Borofsky ND, Vogelman JH, Krajcik RA, Orentreich N. Utility of insulin-like growth 614 factor-1 as a biomarker in epidemiologic studies. Clin Chem. 2002;48(12):2248-51. 615 PubMed PMID: 12446484. 616 48. Wenzlau JM, Juhl K, Yu L, Moua O, Sarkar SA, Gottlieb P, Rewers M, Eisenbarth 617 GS, Jensen J, Davidson HW, Hutton JC. The cation efflux transporter ZnT8 (Slc30A8) is 618 a major autoantigen in human type 1 diabetes. Proc Natl Acad Sci U S A. 619 2007;104(43):17040-5. Epub 2007/10/17. doi: 10.1073/pnas.0705894104. PubMed 620 PMID: 17942684; PMCID: PMC2040407. 621 49. Schlosser M, Mueller PW, Törn C, Bonifacio E, Bingley PJ, Laboratories P. 622 Diabetes Antibody Standardization Program: evaluation of assays for insulin 623 autoantibodies. Diabetologia. 2010;53(12):2611-20. Epub 2010/09/25. doi: 624 10.1007/s00125-010-1915-5. PubMed PMID: 20871974. 625 50. Ziegler AG, Rewers M, Simell O, Simell T, Lempainen J, Steck A, Winkler C, Ilonen 626 J, Veijola R, Knip M, Bonifacio E, Eisenbarth GS. Seroconversion to multiple islet 627 autoantibodies and risk of progression to diabetes in children. JAMA. 2013;309(23):2473- 628 9. doi: 10.1001/jama.2013.6285. PubMed PMID: 23780460; PMCID: PMC4878912. 629 51. Steck AK, Larsson HE, Liu X, Veijola R, Toppari J, Hagopian WA, Haller MJ, 630 Ahmed S, Akolkar B, Lernmark Å, Rewers MJ, Krischer JP, Group atTS. Residual beta- 631 cell function in diabetes children followed and diagnosed in the TEDDY study compared 632 to community controls. Pediatr Diabetes. 2017;18(8):794-802. Epub 2017/01/27. doi: 633 10.1111/pedi.12485. PubMed PMID: 28127835; PMCID: PMC5529265. Page 27 of 67 Diabetes

635 TABLES 636 Cohort UFDI BRI Characteristic AAb- AAb- 1 AAb+ 2-3 Recent Estab- Estab- Control Relative AAb+ Onset lished lished Total Subjects, 69 55 25 27 61 68 50 n Sex, n (%) Male 49 (71) 28 (51) 11 (44) 18 (67) 34 (56) 36 (53) 28 (56) Female 20 (29) 27 (49) 14 (56) 9 (33) 27 (44) 32 (47) 22 (44) Age (years) 12.6 ± 12.0 ± 10.4 ± 13.1 ± 12.1 ± 11.6 ± 22.6 ± 3.4 3.4 4.5 3.9 3.5 3.6 13.6 Height (m)* 1.5 ± 1.5 ± 1.4 ± 1.6 ± 1.5 ± 1.5 ± 1.7 ± 0.2 0.2 0.3 0.2 0.2 0.2 0.1 Weight (kg)* 48.3 ± 49.8 ± 41.3 ± 55.4 ± 49.1 ± 50.8 ± 76.1 ± 24.2 24.0 19.3 19.3 20.6 22.7 13.5 BMI (kg/m2)* 21.0 ± 21.7 ± 20.4 ± 21.2 ± 21.7 ± 22.5 ± 24.3 ± 7.8 7.1 4.8 3.9 5.1 5.9 4.8 BMI Percentile 64.1 ± 66.1 ± 75.7 ± 65.5 ± 73.5 ± 76.3 ± 73.2 ± (%)*,** 27.3 29.2 22.1 24.3 24.3 21.0 21.7 Ethnicity, n (%) Caucasian 37 (54) 37 (67) 16 (64) 21 (78) 43 (70) 37 (54) 44 (88) African-Am 16 (23) 9 (16) 0 (0) 2 (7) 7 (11) 11 (16) 0 (0) Hispanic 8 (12) 7 (13) 4 (16) 4 (15) 8 (13) 14 (21) 1 (2) Asian/Pac-Isl 5 (7) 0 (0) 2 (8) 0 (0) 1 (2) 1 (2) 3 (6) Native-Am 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 2 (3) 0 (0) Other 3 (4) 2 (4) 3 (12) 0 (0) 2 (3) 3 (4) 2 (4) Disease N/A N/A N/A N/A 0.08 ± 4.35 ± 5.1 ± Duration 0.06 3.72 10.7 (years) Metabolic Status, n (%) Fasted 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 38 (76) Random 69 (100) 55 (100) 25 (100) 27 (100) 61 (100) 68 (100) 12 (24) 637 638 Table 1. Demographic information for UFDI cross-sectional cohorts including AAb- 639 controls; AAb- first-degree relatives of a subject with type 1 diabetes; subjects with Diabetes Page 28 of 67

640 1, 2 or 3 AAb; subjects with recent-onset type 1 diabetes diagnosed within the past 641 three months; and subjects with established type 1 diabetes with disease for 642 greater than or equal to three months; demographic information for BRI cross- 643 sectional cohort of subjects with established type 1 diabetes also shown. AAb+ 644 subjects included both relatives of type 1 diabetes subjects as well as unrelated subjects. 645 * Provision of height and weight was voluntary; thus, these data are available for some, 646 but not all study subjects. 647 ** Approximately half of the BRI cohort are adults and are excluded from BMI percentile 648 for children and teenagers. 649 Page 29 of 67 Diabetes

Cohort T1DBIT Characteristic 1 AAb at 2-5 AAb at Developed Enrollment Enrollment T1D During Study Total Subjects, n 14 20 6

Number of Timepoints per 13 (12-13) 13 (11-13) 13 (12-13) Subject, mean (range) Time Between Draws (months) 1.1 ± 0.3 1.2 ± 0.4 1.2 ± 0.4

Sex, n (%) Male 1 (7) 12 (60) 5 (83) Female 13 (93) 8 (40) 1 (17) Age at Enrollment (years) 17.0 ± 8.2 14.2 ± 6.5 15.0 ± 2.8

Height at Enrollment (m)* 1.6 ± 0.1 1.6 ± 0.2 1.7 ± 0.1 Weight at Enrollment (kg)* 59.0 ± 18.0 53.9 ± 22.4 58.3 ± 16.9

BMI at Enrollment (kg/m2)* 23.3 ± 5.9 20.7 ± 4.1 20.7 ± 3.7

BMI Percentile for Children 68.4 ± 22.3 61.1 ± 23.8 52.8 ± 32.3 and Teens at Enrollment (%)*,** Change in BMI Percentile over r = 0.00 r = -0.08 r = -0.02 Study*,** P = 0.97 P = 0.25 P = 0.86 Ethnicity, n (%) Caucasian 10 (72) 15 (75) 4 (67) Hispanic 1 (7) 3 (15) 0 (0) Asian/Pac-Isl 1 (7) 0 (0) 0 (0) Other 2 (14) 2 (10) 2 (33) Disease Duration at N/A N/A N/A Enrollment (years) Metabolic Status, n (%) Fasted 0 (0) 0 (0) 0 (0) Random 14 (100) 20 (100) 6 (100) 650 651 Table 2. Demographic information for longitudinal cohort from Pacific Northwest 652 Research Institute and Novo Nordisk Research Center (T1DBIT) including subjects 653 with 1 AAb at enrollment; subjects with 2-5 AAb at enrollment; and subjects who Diabetes Page 30 of 67

654 developed type 1 diabetes during the study. Subjects were not selected based on 655 family history. Participants underwent clinic or in-home visits with blood draw and 656 questionnaire collection every 4–6 weeks for a total of 13 samples spanning 12 to 18 657 months. Participants who developed T1D during the course of sampling continued on the 658 same frequent sampling schedule at least until completion of their 13th sample visit. At 659 each study visit, non-fasted subjects completed a questionnaire to initiate or update a list 660 of their medical conditions and current medications, if any. Change in BMI reported as 661 Spearman’s correlation results. 662 * Provision of height and weight was voluntary; thus, these data are available for some, 663 but not all study subjects. 664 ** Two subjects from 1 AAb at enrollment and one subject from 2-5 AAb at enrollment 665 groups are adults and therefore are excluded from BMI percentile for children and 666 teenagers. 667 Page 31 of 67 Diabetes

1 AAb at 2-5 AAb at Developed T1D Enrollment Enrollment During Study Coefficient of Intra- Inter- Intra- Inter- Intra- Inter- Variance (CV, %) Subject Subject Subject Subject Subject Subject IGF1 16.3 ± 38.1 ± 18.6 ± 45.5 ± 13.7 ± 22.6 ± 4.4 4.7 7.1 4.3 4.3 5.3 IGF2 11.0 ± 24.6 ± 13.4 ± 23.4 ± 14.3 ± 14.6 ± 3.9 3.6 4.9 5.0 5.6 4.3 668 669 Table 3. Within- and between-subject coefficient of variance (CV) for raw IGF1 and 670 IGF2 measurements in subjects with 1 AAb at enrollment, 2-5 AAb at enrollment, 671 or whom developed type 1 diabetes during the study. CVs were calculated from 11- 672 13 timepoints per subject. 673 Diabetes Page 32 of 67

Subject Age Sex # of ID IGF1 IGF2 AAb GADA IA-2A IA-2A ZnT8A IAA

6 14 F Stable Stable 1 N Y N N N

37 6 F Stable Stable 1 N N N N YN

13 10 F Stable Stable 1 N N N N Y

12 13 F Stable Stable 1 N N N N Y

25 14 F Stable Stable 1 Y N N N NY

19 14 F Stable Stable 1 N N N N Y

32 14 M Stable Stable 1 N N N N Y

34 16 F Stable Stable 1 Y N N N NYN

27 16 F Stable Stable 1 N N N N Y

5 17 F Stable Stable 1 N N N N YN

29 18 F Stable Stable 1 Y N N N N

30 18 F Stable Stable 1 N N N N Y

22 33 F Stable Up 1 N N N N Y

38 37 F Stable Stable 1 N N N N Y

36 12 M Stable Stable 2 Y N N N Y

26 18 M Stable Stable 2 Y N N N Y

3 8 M Stable Stable 2 Y N N YN N

33 6 F Stable Stable 2 N N NY YN Y

35 17 F Stable Stable 2 Y N N Y NYNY

8 4 M Stable Stable 3 Y Y Y N NYN

17 15 F Stable Up 3 Y N N Y YN

31 13 M Stable Stable 4 Y Y Y Y N

2 15 F Stable Stable 4 Y Y Y Y N

4 18 M Stable Up 4 Y Y Y Y N

11 35 F Stable Stable 4 Y Y Y Y N

18 17 M Stable Stable 4 Y Y Y Y NYN

40 5 M Stable Stable 5 Y Y Y Y Y Page 33 of 67 Diabetes

23 14 F Down Up 3 NYNY Y Y Y N

20 14 M Down Stable 3 N Y Y Y NY

39 14 F Down Stable 4 Y Y Y Y N

15 18 F Down Stable 4 Y Y Y Y N

16 15 M Down Up 5 Y Y Y Y Y

10 17 M Down Stable 5 YNY Y Y Y Y

9 11 M Up Stable 3 Y N N Y YN

1* 10 M Stable Stable 5 Y Y Y Y Y

28* 17 M Stable Stable 4 Y Y Y Y NY

7* 14 M Down Up 5 Y Y Y Y YNY

21* 15 M Down Down 3 NY Y Y Y NY

14* 17 M Down Down 4 Y Y Y Y NY

24* 17 F Down Stable 4 Y Y Y Y N 674 675 Table 4. Trajectory of IGF1 and IGF2 in T1DBIT subjects and analysis of AAb 676 positivity. IGF trajectory designated as follows: levels significantly decreased (blue) or 677 significantly increased during the study (red), as determined by Spearman correlation of 678 IGF level versus time. Retaining or gaining positivity for an AAb are shown in green. 679 * These subjects developed type 1 diabetes during the study, therefore, IAA 680 seroconversion may be a consequence of exogenous insulin therapy. 681 Diabetes Page 34 of 67

682 FIGURE LEGENDS

683 Figure 1. Total IGF1 and IGF2 levels are significantly decreased in serum of AAb+

684 subjects at high risk for type 1 diabetes onset. Subjects from UF cross-sectional

685 cohort. (A) IGF1 levels correlate with age for males (Spearman correlation: p < 0.0001, r

686 = 0.60) and females (p < 0.0001, r = 0.38), with an earlier peak in females (circles, red

687 line) compared to males (triangles, blue line). Age at onset of type 1 diabetes in this cohort

688 (includes subjects with recent-onset and established disease) were binned based on

689 frequency and overlaid in gray. (B) Best-fit curves were significantly different for AAb-

690 control (red) vs. AAb+ subjects (blue) (p = 0.0002, Extra sum-of-squares F test). Data are

691 shown as mean (solid line) with 95% CI (dashed lines). Violin plots showing (C) IGF1

692 levels normalized for age and sex are decreased in AAb+ subjects as compared to AAb-

693 controls, AAb- relatives, and subjects with recent-onset type 1 diabetes. (D) IGF2 levels

694 normalized for age and sex decreased in AAb+ subjects and subjects with recent-onset

695 disease as compared to AAb- relatives and established type 1 diabetes. Upon

696 stratification of AAb+ group by number of AAb, (E) decrease in IGF1 and (F) IGF2

697 percentiles remain significant for those with any number of AAb. Kruskal-Wallis with

698 Dunn’s multiple comparisons test: *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001.

699

700 Figure 2. IGF1 decreases with increased duration of type 1 diabetes. (A) Disease

701 onset for BRI cohort is significantly older than UFDI established type 1 diabetes cohort

702 (Mann-Whitney test). (B) IGF1 percentile for age and sex decreases with increasing

703 disease duration in cross-sectional established type 1 diabetes cohorts from UFDI

704 (circles, blue line) and BRI (triangles, red line). (C) IGF2 percentile is not associated with Page 35 of 67 Diabetes

705 disease duration in either cohort. Data are overlaid with best fit lines (solid) and 95% CI

706 (dashed lines). Spearman correlation: *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p <

707 0.0001.

708

709 Figure 3. Longitudinal IGF1 levels are stable in subjects with 1 AAb at enrollment

710 and decrease over time in those with multiple AAb+ at enrollment and subjects who

711 progress to type 1 diabetes. Subjects from longitudinal T1DBIT cohort. (A) IGF1 does

712 not show any correlation with time of longitudinal follow-up in 1 AAb+ subjects, while a

713 subset of those with multiple AAb and who developed type 1 diabetes show significantly

714 decreasing IGF1 over time. (B) IGF2 does not show any correlation with time of

715 longitudinal follow-up in 1 AAb+ subjects, while a subset of those with multiple AAb show

716 significantly increasing IGF2 levels and those who developed type 1 diabetes show

717 variable IGF2 levels over time. Filled circles indicate subjects with significant slopes, while

718 empty circles denote subjects with non-significant slopes in IGF over the course of the

719 study. Kruskal-Wallis with Dunn’s multiple comparisons test: *, p < 0.05; **, p < 0.01.

720

721 Figure 4. Subjects that progress to type 1 diabetes show reduction in IGF1 post-

722 diagnosis. Subjects from longitudinal T1DBIT cohort. Grouped IGF trajectory data for

723 subjects progressing to type 1 diabetes during the study reveals that (A) IGF1 remains

724 stable pre-diagnosis (blue line) and decreases with time post-diagnosis (red line) in the

725 same subjects. (B) IGF2 levels were not significantly different when comparing pre- and

726 post-diagnosis trajectories. (C) Random C-peptide levels are stable pre-diagnosis and Diabetes Page 36 of 67

727 decrease with time post-diagnosis. Data were normalized by baseline levels before or

728 after diagnosis. Best fit lines (solid) shown with 95% CI (dashed). Spearman correlation. Page 37 of 67 Diabetes

Figure 1

134x163mm (300 x 300 DPI) Diabetes Page 38 of 67

Figure 2

178x52mm (600 x 600 DPI) Page 39 of 67 Diabetes

Figure 3

134x55mm (600 x 600 DPI) Diabetes Page 40 of 67

Figure 4

213x61mm (600 x 600 DPI) Page 41 of 67 Diabetes

1 ONLINE SUPPLEMENTARY MATERIALS

Age and B0 B1 B2 B3 B4 B5 RSDR Sex Bin 0-2 y 13.53 -1.894 7.709e-2 -8.257e-4 3.705e-6 -6.048e-9 0.8 2-4 y 24.62 -2.249 6.079e-2 -4.99e-4 1.76e-6 -2.29e-9 0.5 4-6 y 24.18 -1.754 3.812e-2 -2.498e-4 7.028e-7 -7.308e-10 0.5 6-7 y 39.81 -2.254 3.895e-2 -2.238e-4 5.625e-7 -5.278e-10 0.3 7-8 y 55.04 -2.383 3.216e-2 -1.506e-4 3.088e-7 -2.355e-10 0.9 8-9 y, M 71.02 -2.611 3.004e-2 -1.201e-4 2.005e-7 -1.138e-10 0.9 8-9 y, F 48.84 -1.706 1.832e-2 -6.782e-5 1.105e-7 -6.751e-11 0.3 9-10 y, M 77.01 -2.425 2.363e-2 -7.535e-5 8.327e-8 -8.734e-12 1.1 9-10 y, F 69.31 -2.014 1.835e-2 -5.851e-5 7.92e-8 -3.782e-11 1.1 10-11 y, M 87.57 -2.169 1.476e-2 -1.022e-5 -1.003e-7 1.715e-10 1.0 10-11 y, F 71.99 -1.762 1.352e-2 -3.613e-5 4.027e-8 -1.522e-11 1.2 11-12 y, M 99.58 -2.314 1.587e-2 -2.629e-5 -2.422e-8 6.704e-11 1.1 11-12 y, F 55.6 -1.186 7.853e-3 -1.823e-5 1.866e-8 -7.154e-12 0.5 12-13 y, M 85.85 -1.969 1.424e-2 -3.59e-5 3.616e-8 -1.073e-11 1.1 12-13 y, F 56.66 -0.952 4.984e-3 -9.056e-6 7.167e-9 -2.099e-12 0.9 13-14 y, M 85.12 -1.543 8.808e-3 -1.772e-5 1.459e-8 -3.828e-12 0.9 13-14 y, F 89.56 -1.095 4.105e-3 -4.767e-6 1.28e-9 5.209e-13 1.2 14-15 y, M 76.95 -1.102 4.969e-3 -7.889e-6 5.261e-9 -1.204e-12 0.9 14-15 y, F 100.1 -0.9434 2.09e-3 1.872e-6 -7.285e-9 4.388e-12 1.1 15-16 y, M 97.3 -1.074 3.414e-3 -2.157e-6 -2.305e-9 2.202e-12 1.1 15-16 y, F 95.38 -0.9569 2.497e-3 3.219e-7 -5.109e-9 3.342e-12 1.0 16-17 y, M 95.39 -1.009 2.823e-3 1.339e-8 -5.562e-9 3.889e-12 1.0 16-17 y, F 90.46 -0.8734 1.783e-3 3.256e-6 -9.935e-9 6.012e-12 1.1 17-18 y, M 14.11 0.6103 -9.78e-3 4.596e-5 -8.037e-8 4.791e-11 1.1 17-18 y, F 21.3 0.5058 -9.067e-3 4.326e-5 -7.555e-8 4.476e-11 1.2 18-19 y, M 8.195 0.7015 -1.07e-2 5.09e-5 -9.116e-8 5.583e-11 1.2 18-19 y, F -4.779 1.015 -1.349e-2 6.219e-5 -1.116e-7 6.924e-11 1.1 19-20 y -7.826 1.142 -1.567e-2 7.587e-5 -1.438e-7 9.451e-11 1.3 Diabetes Page 42 of 67

20-30 y 82.48 -2.293 1.99e-2 -5.803e-5 6.193e-8 -1.216e-11 1.3 30-40 y 80.52 -2.368 2.17e-2 -6.648e-5 7.413e-8 -1.463e-11 1.2 40-50 y 75.83 -2.359 2.271e-2 -7.135e-5 7.746e-8 -7.634e-12 1.1 50-60 y 80.33 -2.68 2.804e-2 -1.012e-4 1.451e-7 -6.067e-11 1.1 60-70 y 73.41 -2.663 3.031e-2 -1.205e-4 2.002e-7 -1.132e-10 0.9 2 3 Supplementary Table 1. Formulas for IGF1 age and sex-based percentile 4 calculations. Fifth-order polynomial equations were robustly fit over historical IGF1 data 5 (31) in order to solve for Y, IGF1 percentile, using X, raw IGF1 (ng/mL). Substitute 6 appropriate constants according to age and sex bin of subject: Y = B0 + B1*X + B2*X^2 7 + B3*X^3 + B4*X^4 + B5*X^5. Goodness of fit represented as robust standard deviation 8 of the residuals (RSDR). 9 Page 43 of 67 Diabetes

Age Bin Slope Y-intercept Absolute Sum of Squares 1-3 y 235.4 -603.4 1.9e-3 3-5 y 246.4 -636 3.3 5-7 y 246 -640 4.7e-3 7-9 y 248.5 -650 1.9e-2 9-11 y 255.7 -671.4 1.3e-2 11-13 y 256.2 -674.2 2.0e-3 13-15 y 258.6 -682.4 1.1e-3 15-20 y 248.1 -653.8 1.3e-3 20-30 y 233.8 -612 4.9e-4 30-40 y 239.8 -629.3 1.3e-2 40-50 y 221.1 -572 1.2e-3 50-60 y 212.7 -547.6 3.7e-3 60-70 y 210.1 -535.4 1.0e-3 10 11 Supplementary Table 2. Formulas for IGF2 age-based percentile calculations. 12 Semilog lines were fit over historical IGF2 data (31) in order to solve for Y, IGF2 13 percentile, using X, raw IGF2 (ng/mL). Substitute appropriate constants according to age 14 bin of subject: Y = Y-intercept + Slope*log(X). Goodness of fit represented as absolute 15 sum of squares. 16 Diabetes Page 44 of 67

Figure Dunn's multiple comparisons test Mean rank Significant? Adjusted p- diff. value AAb- Control vs. AAb- Relative -5.804 ns >0.9999 AAb- Control vs. AAb+ 84.67 **** <0.0001 AAb- Control vs. Recent Onset 19.77 ns >0.9999 AAb- Control vs. Established 23.58 ns >0.9999 AAb- Relative vs. AAb+ 90.48 **** <0.0001 1C AAb- Relative vs. Recent Onset 25.57 ns >0.9999 AAb- Relative vs. Established 29.39 ns 0.6349 AAb+ vs. Recent Onset -64.91 ** 0.0010 AAb+ vs. Established -61.09 ** 0.0019 Recent Onset vs. Established 3.818 ns >0.9999 AAb- Control vs. AAb- Relative -61.75 ** 0.0010 AAb- Control vs. AAb+ 24.54 ns >0.9999 AAb- Control vs. Recent Onset -0.8724 ns >0.9999 AAb- Control vs. Established -50.33 ** 0.0082 AAb- Relative vs. AAb+ 86.28 **** <0.0001 1D AAb- Relative vs. Recent Onset 60.88 ** 0.0020 AAb- Relative vs. Established 11.41 ns >0.9999 AAb+ vs. Recent Onset -25.41 ns >0.9999 AAb+ vs. Established -74.87 **** <0.0001 Recent Onset vs. Established -49.46 * 0.0145 AAb- Relative vs. 1 AAb+ 50.03 *** 0.0002 AAb- Relative vs. 2-3 AAb+ 48.09 *** 0.0001 AAb- Relative vs. Recent Onset 13.32 ns 0.8028 1E 1 AAb+ vs. 2-3 AAb+ -1.939 ns >0.9999 1 AAb+ vs. Recent Onset -36.72 * 0.0120 2-3 AAb+ vs. Recent Onset -34.78 ** 0.0098 AAb- Relative vs. 1 AAb+ 36.87 ** 0.0099 AAb- Relative vs. 2-3 AAb+ 55.53 **** <0.0001 1F AAb- Relative vs. Recent Onset 31.94 ** 0.0024 1 AAb+ vs. 2-3 AAb+ 18.66 ns 0.9982 Page 45 of 67 Diabetes

1 AAb+ vs. Recent Onset -4.926 ns >0.9999 2-3 AAb+ vs. Recent Onset -23.58 ns 0.2141 17 18 Supplementary Table 3. Multiplicity adjusted p-values for Figure 1.

19 Diabetes Page 46 of 67

Figure Dunn's multiple comparisons test Mean rank Significant? Adjusted p- diff. value AAb- Control vs. AAb- Relative 7.020 ns >0.9999 AAb- Control vs. AAb+ 76.81 **** <0.0001 AAb- Control vs. Recent Onset 28.27 ns 0.6455 AAb- Control vs. Established 40.00 ns 0.0737 AAb- Relative vs. AAb+ 69.79 *** 0.0004 S1A AAb- Relative vs. Recent Onset 21.25 ns >0.9999 AAb- Relative vs. Established 32.98 ns 0.3728 AAb+ vs. Recent Onset -48.54 * 0.0365 AAb+ vs. Established -36.81 ns 0.2447 Recent Onset vs. Established 11.73 ns >0.9999 AAb- Control vs. AAb- Relative -62.82 *** 0.0008 AAb- Control vs. AAb+ 25.00 ns >0.9999 AAb- Control vs. Recent Onset 0.1076 ns >0.9999 AAb- Control vs. Established -47.77 * 0.0152 AAb- Relative vs. AAb+ 87.82 **** <0.0001 S1B AAb- Relative vs. Recent Onset 62.93 ** 0.0012 AAb- Relative vs. Established 15.05 ns >0.9999 AAb+ vs. Recent Onset -24.89 ns >0.9999 AAb+ vs. Established -72.77 **** <0.0001 Recent Onset vs. Established -47.88 * 0.0208 AAb- vs. 1 AAb+ 49.88 *** 0.0002 AAb- vs. 2-3 AAb+ 29.69 * 0.0491 AAb- vs. Recent Onset 11.71 ns >0.9999 S1C 1 AAb+ vs. 2-3 AAb+ -20.19 ns 0.8467 1 AAb+ vs. Recent Onset -38.18 ** 0.0079 2-3 AAb+ vs. Recent Onset -17.98 ns 0.6206 AAb- vs. 1 AAb+ 40.22 ** 0.0036 AAb- vs. 2-3 AAb+ 54.18 **** <0.0001 S1D AAb- vs. Recent Onset 33.02 ** 0.0016 1 AAb+ vs. 2-3 AAb+ 13.96 ns >0.9999 Page 47 of 67 Diabetes

1 AAb+ vs. Recent Onset -7.204 ns >0.9999 2-3 AAb+ vs. Recent Onset -21.16 ns 0.3588 20 21 Supplementary Table 4. Multiplicity adjusted p-values for Supplementary Figure 1.

22 Diabetes Page 48 of 67

IGF1 IGF2 Characteristic n below 5th % of total n below 5th % of total percentile subjects percentile subjects AAb- Control 2 2.90% 6 8.70% AAb- Relative 6 10.91% 0 0.00% 1 AAb+ 8 36.36% 3 12.00% 2-3 AAb+ 9 33.33% 7 25.93% Recent Onset 5 8.20% 1 1.64% Established 7 10.45% 2 2.94% 23 24 Supplementary Table 5. Percentages of IGF1 and IGF2 measurements compared to 25 reference ranges for age and sex, per group. Subjects from UF cross-sectional cohort. 26 IGF1 and IGF2 levels were below the fifth percentile in a higher percentage of 1 AAb+ 27 and 2-3 AAb+ subjects as compared to all other groups. 28 Page 49 of 67 Diabetes

Figure Dunn's multiple comparisons test Mean rank Significant? Adjusted p- diff. value AAb- Control vs. AAb- Relative -8.621 ns >0.9999 AAb- Control vs. AAb+ -48.82 * 0.0205 AAb- Control vs. Recent Onset -18.47 ns >0.9999 AAb- Control vs. Established -50.99 ** 0.0052 AAb- Relative vs. AAb+ -40.20 ns 0.1459 S2A AAb- Relative vs. Recent Onset -9.848 ns >0.9999 AAb- Relative vs. Established -42.37 ns 0.0579 AAb+ vs. Recent Onset 30.36 ns 0.6289 AAb+ vs. Established -2.167 ns >0.9999 Recent Onset vs. Established -32.52 ns 0.3245 AAb- Control vs. AAb- Relative -18.06 ns >0.9999 AAb- Control vs. AAb+ 10.40 ns >0.9999 AAb- Control vs. Recent Onset 19.84 ns >0.9999 AAb- Control vs. Established 0.1971 ns >0.9999 AAb- Relative vs. AAb+ 28.46 ns 0.8714 S2B AAb- Relative vs. Recent Onset 37.90 ns 0.1914 AAb- Relative vs. Established 18.26 ns >0.9999 AAb+ vs. Recent Onset 9.437 ns >0.9999 AAb+ vs. Established -10.20 ns >0.9999 Recent Onset vs. Established -19.64 ns >0.9999 AAb- Control vs. AAb- Relative 18.71 ns >0.9999 AAb- Control vs. AAb+ 22.21 ns >0.9999 AAb- Control vs. Recent Onset 45.92 * 0.0311 AAb- Control vs. Established 32.42 ns 0.2901 AAb- Relative vs. AAb+ 3.499 ns >0.9999 S2C AAb- Relative vs. Recent Onset 27.21 ns 0.9262 AAb- Relative vs. Established 13.71 ns >0.9999 AAb+ vs. Recent Onset 23.71 ns >0.9999 AAb+ vs. Established 10.21 ns >0.9999 Recent Onset vs. Established -13.50 ns >0.9999 Diabetes Page 50 of 67

AAb- Control vs. AAb- Relative -20.97 ns >0.9999 AAb- Control vs. AAb+ 27.17 ns 0.8976 AAb- Control vs. Recent Onset 33.23 ns 0.3237 AAb- Control vs. Established -6.997 ns >0.9999 AAb- Relative vs. AAb+ 48.13 * 0.0382 S2D AAb- Relative vs. Recent Onset 54.20 ** 0.0081 AAb- Relative vs. Established 13.97 ns >0.9999 AAb+ vs. Recent Onset 6.070 ns >0.9999 AAb+ vs. Established -34.16 ns 0.3119 Recent Onset vs. Established -40.23 ns 0.0886 AAb- vs. 1 AAb+ -28.18 ns 0.0808 AAb- vs. 2-3 AAb+ -16.70 ns 0.7643 AAb- vs. Recent Onset -5.574 ns >0.9999 S2E 1 AAb+ vs. 2-3 AAb+ 11.48 ns >0.9999 1 AAb+ vs. Recent Onset 22.60 ns 0.2776 2-3 AAb+ vs. Recent Onset 11.12 ns >0.9999 AAb- vs. 1 AAb+ 11.60 ns >0.9999 AAb- vs. 2-3 AAb+ 17.35 ns 0.7057 AAb- vs. Recent Onset 20.79 ns 0.1185 S2F 1 AAb+ vs. 2-3 AAb+ 5.750 ns >0.9999 1 AAb+ vs. Recent Onset 9.189 ns >0.9999 2-3 AAb+ vs. Recent Onset 3.439 ns >0.9999 AAb- vs. 1 AAb+ 8.928 ns >0.9999 AAb- vs. 2-3 AAb+ -3.700 ns >0.9999 AAb- vs. Recent Onset 14.53 ns 0.6205 S2G 1 AAb+ vs. 2-3 AAb+ -12.63 ns >0.9999 1 AAb+ vs. Recent Onset 5.600 ns >0.9999 2-3 AAb+ vs. Recent Onset 18.23 ns 0.5900 AAb- vs. 1 AAb+ 22.70 ns 0.2955 S2H AAb- vs. 2-3 AAb+ 32.86 * 0.0183 AAb- vs. Recent Onset 30.51 ** 0.0038 Page 51 of 67 Diabetes

1 AAb+ vs. 2-3 AAb+ 10.16 ns >0.9999 1 AAb+ vs. Recent Onset 7.803 ns >0.9999 2-3 AAb+ vs. Recent Onset -2.355 ns >0.9999 29 30 Supplementary Table 6. Multiplicity adjusted p-values for Supplementary Figure 2. 31 Diabetes Page 52 of 67

AAb- Relatives vs. 1 1 AAb+ vs. 2-3 AAb+ 2-3 AAb+ vs. Recent AAb+ Onset Characteristic RRR (95% p value RRR (95% p RRR (95% p value CI) CI) value CI) Log.IGF1 0.06 (0.02 - < 0.0001 1.81 (0.61 – 0.28 4.32 (1.58 – < 0.01 0.20) 5.36) 11.85) Log.IGF2 0.17 (0.04 – < 0.05 0.61 (0.14 – 0.52 2.52 (0.62 – 0.19 0.67) 2.76) 10.19) Age 1.17 (0.98 - 0.09 1.08 (0.90 - 0.39 0.83 (0.71 - < 0.05 1.41) 1.30) 0.96) Log.IGFBP7 0.16 (0.02 – 0.09 0.78 (0.09 – 0.82 1.02 (0.18 – 0.98 1.34) 6.57) 5.67) 32 33 Supplementary Table 7. Circulating IGFs can discriminate type 1 diabetes stages

34 without the need for IGFBP measurements. A multinomial logistic regression model

35 was established using R version 3.4.1. Only individuals with complete IGF1, IGF2, and

36 IGFBP data, as well as reported race/ethnicity were included in the analysis. As height

37 and weight information were voluntarily provided, missing BMI percentile data were

38 imputed with predictive means matching using the MICE package. An automated forward

39 variable selection method from the LEAPS package was used to generate a multinomial

40 logistic regression model evaluating age, BMI percentile, sex, ethnicity, and log-

41 transformed IGF1, IGF2, IGFBP1, IGFBP3, IGFBP6, and IGFBP7 concentrations as

42 explanatory variables for a model to predict the cohort of the subject as the response

43 variable. The variables included in the best additive model are shown here. The relevance

44 of each explanatory variable in modeling differences between groups was determined by

45 calculating relative risk ratios (RRR) with 95% CI, with the logically previous group in

46 progression to disease as the referent.

47 Page 53 of 67 Diabetes

* A B **** **** ** 1000 4000 *** **** * *** * 800 3000 600 2000 400 1000 IGF1, ng/mL 200 IGF2, ng/mL 0 0 l e + t d l e + t d ro v b e e ro v b e e t ti s h t ti s n a A n s n a A n sh o l A O li o l A O li C e t b C e t b - R n a R n a - e t - - t b b c s b b ce s A A e E A A e E A A R A A R

C D ** 1000 4000 **** * ** *** ** 800 3000 600 2000 400 1000

IGF1, ng/mL 200 IGF2, ng/mL 0 0 t t - + + e - + + b b b s b b b se A A A n A A A n A A A O A A A O 1 3 t 1 3 t - n - n 2 e 2 c ce e e 48 R R

49 Supplementary Figure 1. Raw total IGF1 and IGF2 levels are significantly decreased

50 in serum of AAb+ subjects at high risk for type 1 diabetes onset. Subjects from UF

51 cross-sectional cohort. Violin plots showing (A) raw IGF1 levels show decreased IGF1 in

52 AAb+ subjects as compared to AAb- controls, AAb- relatives, and subjects with recent-

53 onset type 1 diabetes. (B) Raw IGF2 levels are decreased in AAb+ subjects and subjects

54 with recent-onset disease as compared to AAb- relatives and established type 1 diabetes.

55 Upon stratification of AAb+ group by number of AAb, (C) raw IGF1 levels and (D) raw

56 IGF2 levels remain significantly decreased for those with any number of AAb. Kruskal- Diabetes Page 54 of 67

57 Wallis with Dunn’s multiple comparisons test: *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****,

58 p < 0.0001.

59 Page 55 of 67 Diabetes

A ** B 30 * 8000 6000 20 4000 10 2000 IGFBP1, ng/mL IGFBP3, ng/mL 0 0 l t l t o e + d o e + d r iv b e e r iv b e e t t A s h t t A s h n la n is n la n is o e A O l o e A O l C R t b C R t b - - n ta - - n ta b b ce s b b ce s A A e E A A e E A A R A A R C D 250 250 ** * 200 * 200 150 150 100 100 50 50 IGFBP6, ng/mL IGFBP7, ng/mL 0 0 l e + t d l e + t d ro v b e e ro v b e e t ti s h t ti s h n a A n s n a A n s o l A O li o l A O li C e t b C e t b - R n a - R n a - e t - e t b b c s b b c s A A e E A A e E A A R A A R E F 30 8000 6000 20 4000

10 2000 IGFBP3, ng/mL IGFBP1, ng/mL 0 0 - t - t + + e b + + e b b b s A b b s A A A n A A A n A A A O A A O 1 3 t 1 -3 t - n 2 n 2 e ce c e e R R G H 250 250 ** 200 200 * 150 150 100 100 50 50 IGFBP6, ng/mL IGFBP7, ng/mL 0 0 - t + + e - t b b b s b + + e A A A n A b b s A A A O A A A n 1 3 t A A O - n 1 3 t 2 e - n c 2 e e c R e 60 R Diabetes Page 56 of 67

61 Supplementary Figure 2. High-affinity IGFBP levels are generally unchanged in

62 serum of AAb+ subjects and subjects with recent-onset type 1 diabetes. Subjects

63 from UF cross-sectional cohort. Violin plots showing (A) IGFBP1 increased in AAb+

64 subjects as compared to AAb- controls. (B) IGFBP3 appears relatively stable during

65 disease pathogenesis. (C) IGFBP6 decreased in subjects with recent-onset disease as

66 compared to AAb- controls. (D) IGFBP7 decreased in subjects with recent-onset type 1

67 diabetes and AAb+ subjects as compared to AAb- relatives. Upon stratification of AAb+

68 group by number of AAb, (E) IGFBP1, (F) IGFBP3 and (G) IGFBP6 levels did not

69 significantly change between at-risk groups. (H) IGFBP7 decreased in subjects with

70 multiple AAb+ as compared to AAb- relatives. Kruskal-Wallis with Dunn’s multiple

71 comparisons test: *, p < 0.05; **, p < 0.01.

72 Page 57 of 67 Diabetes

A 100 B 100

50 50 IGF1 Percentile IGF1 Percentile 0 0 0 2 4 6 4 6 8 10 12 c-peptide AUC (ng/mL) HbA1c (%) R = 0.35 R = -0.08 p = 0.084 p = 0.608

C 100 D 6

4 50 2 IGF1 Percentile 0 0 50 100 150 200 250 300 AUCc-peptide (ng/mL) 0 1 2 3 20 40 60 Blood Glucose (mg/dL) Duration R = 0.09 R = -0.76 p = 0.576 p < 0.0001 73

74 Supplementary Figure 3. IGF1 levels show trends for association with endogenous

75 β-cell function. In previously fasted subjects with type 1 diabetes from BRI, (A) IGF1

76 percentile shows a trend for a positive correlation with C-peptide AUC and does not

77 associate with (B) HbA1c or (C) fasting blood glucose levels. (D) C-peptide AUC is

78 significantly and negatively correlated with disease duration in this cohort. Data are

79 overlaid with best fit lines (solid) and 95% CI (dashed lines). Spearman correlation.

80 Diabetes Page 58 of 67

Patient ID #37 Patient ID #13 Patient ID #12

1000 1 AAb 1000 1 AAb 1000 1 AAb 6 F 10 F 13 F 800 800 800 1-0 AAb 600 600 600

400 400 400 IGF1, ng/mL IGF1, ng/mL IGF1, ng/mL 200 200 200

0 0 0 0 5 10 15 0 5 10 15 0 5 10 15 Months Months Months Patient ID #6 Patient ID #25 Patient ID #19

1000 1 AAb 1000 1 AAb 1000 1 AAb 14 F 14 F 14 F a1c > 5.7 800 800 800

600 600 1-2 AAb 600

400 400 400 IGF1, ng/mL IGF1, ng/mL IGF1, ng/mL 200 200 200

0 0 0 0 5 10 15 0 5 10 15 0 5 15 Months Months Months Patient ID #32 Patient ID #34 Patient ID #27

1000 1000 1 AAb 1000 1 AAb 1 AAb 14 M 16 F 16 F 800 800 800

600 600 600

400 400 400 IGF1, ng/mL IGF1, ng/mL IGF1, ng/mL 200 200 1-2 2-1 200 AAb AAb 0 0 0 0 5 10 15 0 5 10 15 0 5 10 15 Months Months Months Patient ID #5 Patient ID #29 Patient ID #30

1000 1 AAb 1000 1 AAb 1000 1 AAb 18 F 17 F 1-0 18 F 800 800 800 AAb

600 600 600

400 400 400 IGF1, ng/mL IGF1, ng/mL IGF1, ng/mL 200 200 200

0 0 0 0 5 15 0 5 10 15 0 5 10 15 Months Months Months Patient ID #22 Patient ID #38

1000 1 AAb 1000 1 AAb 33 F 37 F 800 800

600 600

400 400 IGF1, ng/mL IGF1, ng/mL 200 200

0 0 0 5 10 15 0 5 10 15 Months Months 81 82 Supplementary Figure 4. Longitudinal follow-up of subjects with a single AAb at

83 enrollment shows IGF1 stability over time. Subjects from longitudinal T1DBIT cohort. Page 59 of 67 Diabetes

84 Raw IGF1 levels of individual T1DBIT subjects shown as solid lines. Subjects arranged

85 in order of increasing age. 5%, 50%, and 95% reference ranges for age and sex are

86 shown per person as horizontal dashed lines. Time of pre-type 1 diabetes (a1c > 5.7) or

87 seroconversion events shown with vertical dashed black lines.

88 Diabetes Page 60 of 67

Patient ID #37 Patient ID #13 Patient ID #12

1250 1250 1 AAb 1 AAb 1250 1 AAb 6 F 10 F 13 F 1000 1000 1000

750 750 750

500 1-0 AAb 500 500 IGF2, ng/mL IGF2, ng/mL IGF2, ng/mL 250 250 250

0 0 0 0 5 10 15 0 5 10 15 0 5 10 15 Months Months Months Patient ID #6 Patient ID #25 Patient ID #19

1250 1 AAb 1250 1 AAb 1250 1 AAb 14 F 14 F 14 F 1000 1000 1000

750 750 750

500 500 500 IGF2, ng/mL IGF2, ng/mL 1-2 AAb IGF2, ng/mL a1c 250 250 250 > 5.7

0 0 0 0 5 10 15 0 5 10 15 0 5 15 Months Months Months Patient ID #32 Patient ID #34 Patient ID #27 1 AAb 1250 1 AAb 1250 1 AAb 1250 16 F 14 M 16 F 1000 1000 1000

750 750 750

500 500 500 IGF2, ng/mL IGF2, ng/mL IGF2, ng/mL 250 250 1-2 2-1 250 AAb AAb 0 0 0 0 5 10 15 0 5 10 15 0 5 10 15 Months Months Months Patient ID #5 Patient ID #29 Patient ID #30

1250 1 AAb 1250 1 AAb 1250 1 AAb 17 F 18 F 18 F 1000 1-0 1000 1000 AAb 750 750 750

500 500 500 IGF2, ng/mL IGF2, ng/mL IGF2, ng/mL 250 250 250

0 0 0 0 5 15 0 5 10 15 0 5 10 15 Months Months Months Patient ID #22 Patient ID #38

1250 1 AAb R = 0.82 1250 1 AAb 33 F p < 0.01 37 F 1000 1000

750 750 Spearman Correlation

500 500 IGF2, ng/mL IGF2, ng/mL 250 250

0 0 0 5 10 15 0 5 10 15 Months Months 89 90 Supplementary Figure 5. Longitudinal follow-up of subjects with a single AAb at

91 enrollment shows IGF2 stability over time. Subjects from longitudinal T1DBIT cohort. Page 61 of 67 Diabetes

92 Raw IGF2 levels of individual T1DBIT subjects shown as solid lines. Subjects arranged

93 in order of increasing age. 5%, 50%, and 95% reference ranges for age and sex are

94 shown per person as horizontal dashed lines. Time of pre-type 1 diabetes (a1c > 5.7) or

95 seroconversion events shown with vertical dashed black lines. Spearman r and p values

96 shown in upper right-hand corner.

97 Diabetes Page 62 of 67

Patient ID #33 Patient ID #3 Patient ID #36

1000 2 AAb 1000 2 AAb 1000 2 AAb 6 F 8 M 12 M 800 800 800

600 2-1 AAb 1-2 AAb 600 600

400 400 2-1 AAb 400 IGF1, ng/mL IGF1, ng/mL IGF1, ng/mL 200 200 200

0 0 0 0 10 15 0 5 10 15 0 5 10 15 Months Months Months Patient ID #35 Patient ID #26 Patient ID #8

1000 2 AAb 1000 2 AAb 1000 3 AAb 17 F 18 M 4 M 800 800 800 2-3 3-2 2-3 3-4 4-3 AAb AAb AAb AAb AAb 600 600 600a1c a1c a1c a1c > < > < 400 400 4005.7 5.7 5.7 5.7 IGF1, ng/mL IGF1, ng/mL IGF1, ng/mL 200 200 200

0 0 0 0 5 10 15 0 5 10 15 5 15 Months Months Months Patient ID #9 Patient ID #20 Patient ID #23

1000 1000 1000 3 AAb R = 0.86 3 AAb R = -0.68 3 AAb R = -0.65 11 M p < 0.001 14 M p < 0.05 14 F p < 0.05 800 800 800 3-4 4-3 3-4 600 3-2 AAb 600 600 AAb AAb AAb

400 400 400 IGF1, ng/mL IGF1, ng/mL IGF1, ng/mL 3-4 AAb 200 200 200

0 0 0 0 5 10 15 0 5 10 15 0 10 15 Months Months Months Patient ID #31 Patient ID #17 Patient ID #39

1000 4 AAb 1000 3 AAb 1000 4 AAb R = -0.68 13 M 15 F 14 F p < 0.05 800 800 800

600 600 600

400 400 400 IGF1, ng/mL IGF1, ng/mL IGF1, ng/mL 200 200 3-2 AAb 200

0 0 0 0 5 10 15 0 5 10 15 0 5 10 15 Months Months Months Patient ID #2 Patient ID #18 Patient ID #4 4 AAb 1000 1000 1000 4 AAb 17 M 4 AAb 15 F 18 M 800 800 800

600 600 600

400 400 400 IGF1, ng/mL IGF1, ng/mL IGF1, ng/mL 200 200 200 4-5 AAb 5-4 AAb 0 0 0 0 5 10 15 0 5 10 15 0 5 10 15 Months Months Months Patient ID #15 Patient ID #11 Patient ID #40

1000 1000 1000 4 AAb R = -0.65 4 AAb 5 AAb p < 0.05 35 F 5 M 800 18 F 800 800

600 600 600

400 400 400 IGF1, ng/mL IGF1, ng/mL IGF1, ng/mL 200 200 200

0 0 0 0 5 10 15 0 5 10 15 0 5 10 15 Months Months Months Patient ID #16 Patient ID #10

1000 5 AAb R = -0.64 1000 5 AAb R = -0.66 15 M p < 0.05 17 M p < 0.05 800 800 5-4 4-5 Spearman Correlation AAb AAb 600 600

400 400 IGF1, ng/mL IGF1, ng/mL 200 200

0 0 0 5 10 15 0 5 10 15 98 Months Months Page 63 of 67 Diabetes

99 Supplementary Figure 6. Longitudinal follow-up of subjects with multiple AAb at

100 enrollment shows IGF1 typically stable or decreasing over time. Subjects from

101 longitudinal T1DBIT cohort. Raw IGF1 levels of individual T1DBIT subjects shown as solid

102 lines. Subjects arranged in order of number of AAb at enrollment, followed by increasing

103 age. 5%, 50%, and 95% reference ranges for age and sex are shown per person as

104 horizontal dashed lines. Time of pre-type 1 diabetes (a1c > 5.7) or seroconversion events

105 shown with vertical dashed black lines. Spearman r and p values shown in upper right-

106 hand corner.

107 Diabetes Page 64 of 67

Patient ID #33 Patient ID #3 Patient ID #36 2 AAb 1250 2 AAb 1250 1250 2 AAb 12 M 6 F 8 M 1000 1000 1000 2-1 AAb 1-2 AAb 750 750 750

500 500 500 IGF2, ng/mL IGF2, ng/mL 2-1 AAb IGF2, ng/mL 250 250 250

0 0 0 0 10 15 0 5 10 15 0 5 10 15 Months Months Months

Patient ID #35 Patient ID #26 Patient ID #8 3 AAb 1250 2 AAb 1250 2 AAb 1250 4 M 17 F 18 M 1000 1000 1000 a1c a1c a1c a1c > < > < 5.7 5.7 5.7 5.7 750 750 750

500 500 500 IGF2, ng/mL IGF2, ng/mL 2-3 3-2 2-3 IGF2, ng/mL 250 AAb AAb AAb 250 250 3-4 4-3 AAb AAb 0 0 0 0 5 10 15 0 5 10 15 5 15 Months Months Months Patient ID #9 Patient ID #20 Patient ID #23

1250 3 AAb 1250 3 AAb 1250 3 AAb R = 0.63 11 M 14 M 14 F p < 0.05 1000 1000 1000 3-4 AAb 3-4 4-3 3-4 3-2 AAb AAb AAb AAb 750 750 750

500 500 500 IGF2, ng/mL IGF2, ng/mL IGF2, ng/mL 250 250 250

0 0 0 0 5 10 15 0 5 10 15 0 10 15 Months Months Months Patient ID #31 Patient ID #17 Patient ID #39

1250 1250 R = 0.67 1250 4 AAb 3 AAb 4 AAb p < 0.05 13 M 15 F 14 F 1000 1000 1000

3-2 AAb 750 750 750

500 500 500 IGF2, ng/mL IGF2, ng/mL IGF2, ng/mL 250 250 250

0 0 0 0 5 10 15 0 5 10 15 0 5 10 15 Months Months Months Patient ID #2 Patient ID #18 Patient ID #4

1250 1250 1250 R = 0.62 4 AAb 4 AAb 4 AAb p < 0.05 15 F 17 M 18 M 1000 1000 1000

750 750 750

500 500 500 IGF2, ng/mL IGF2, ng/mL IGF2, ng/mL 250 250 4-5 AAb 5-4 AAb 250

0 0 0 0 5 10 15 0 5 10 15 0 5 10 15 Months Months Months

Patient ID #15 Patient ID #11 Patient ID #40 4 AAb 1250 1250 4 AAb 1250 5 AAb 35 F 18 F 5 M 1000 1000 1000

750 750 750

500 500 500 IGF2, ng/mL IGF2, ng/mL IGF2, ng/mL 250 250 250

0 0 0 0 5 10 15 0 5 10 15 0 5 10 15 Months Months Months

Patient ID #16 Patient ID #10

1250 5 AAb R = 0.81 1250 5 AAb 15 M p < 0.01 17 M 1000 1000 Spearman Correlation 750 750

500 500 IGF2, ng/mL IGF2, ng/mL 250 250 5-4 4-5 AAb AAb 0 0 0 5 10 15 0 5 10 15 Months 108 Months Page 65 of 67 Diabetes

109 Supplementary Figure 7. Longitudinal follow-up of subjects with multiple AAb at

110 enrollment shows IGF2 typically stable or increasing over time. Subjects from

111 longitudinal T1DBIT cohort. Raw IGF2 levels of individual T1DBIT subjects shown as solid

112 lines. Subjects arranged in order of number of AAb at enrollment, followed by increasing

113 age. 5%, 50%, and 95% reference ranges for age and sex are shown per person as

114 horizontal dashed lines. Time of pre-type 1 diabetes (a1c > 5.7) or seroconversion events

115 shown with vertical dashed black lines. Spearman r and p values shown in upper right-

116 hand corner.

117 Diabetes Page 66 of 67

Patient ID #24 Patient ID #14 Patient ID #21

1000a1c 4 AAb R = -0.70 1000 4 AAb R = -0.71 1000 3 AAb a1c R = -0.89 > 5.7 17 F p < 0.01 17 M p < 0.01 a1c 15 M > a1c p < 0.001 800 800 800 > 6.4 a1c > 5.7 5.7 > 4-5 a1c 6.4Ab 600 600 < 600 5.7 400 400 400 3-4

IGF1, ng/mL AAb IGF1, ng/mL

IGF1, ng/mL 4-5 200 200 AAb 200 a1c > 6.4 0 0 0 5 10 15 10 15 0 10 15 Months Months Months Patient ID #7 Patient ID #28 Patient ID #1 1000 5 AAb R = -0.73 1000 4 AAb 1000 5 AAb 14 M p < 0.01 5-4 17 M 10 M 800 800 800 AAb a1c a1c 4-5 Spearman Correlation a1c > 5.7 > 4-5 600 600 a1c > 5.7 > 6.4 AAb 600 6.4 AAb a1c > 5.7 a1c > 6.4 400 400 400 IGF1, ng/mL IGF1, ng/mL IGF1, ng/mL 200 200 200

0 0 0 5 10 15 0 5 10 15 0 5 10 15 118 Months Months Months 119 Supplementary Figure 8. Longitudinal follow-up of multiple AAb subjects who

120 develop type 1 diabetes during study shows IGF1 decreasing over time. Subjects

121 from longitudinal T1DBIT cohort. Raw IGF1 levels of individual T1DBIT subjects shown

122 as solid lines. Subjects arranged from those diagnosed earlier to later in the study. 5%,

123 50%, and 95% reference ranges for age and sex are shown per person as horizontal

124 dashed lines. Time of diagnosis shown as vertical red line (a1c > 6.4) and pre-type 1

125 diabetes (a1c > 5.7) or seroconversion events shown with vertical dashed black lines.

126 Spearman r and p values shown in upper right-hand corner.

127 Page 67 of 67 Diabetes

Patient ID #24 Patient ID #14 Patient ID #21

1250 1250 4 AAb R = -0.64 1250 R = -0.66 4 AAb a1c > a1c 3 AAb 17 M p < 0.05 17 F 5.7 < a1c 15 M p < 0.05 1000 1000 5.7 1000 a1c a1c > a1c > > 5.7 5.7 > 6.4 6.4 750 750 750

500 500 500 4-5 IGF2, ng/mL IGF2, ng/mL IGF2, ng/mL 3-4 4-5 AAb 250 250 a1c > 6.4 250 AAb AAb

0 0 0 5 10 15 10 15 0 10 15 Months Months Months Patient ID #7 Patient ID #28 Patient ID #1

1250 5 AAb R = 0.76 1250 4 AAb 1250 5 AAb 14 M p < 0.01 17 M 10 M 1000 1000 1000 a1c 4-5 a1c 4-5 a1c > 5.7 a1c > 6.4 a1c > 5.7 > 6.4 AAb a1c > 5.7 > 6.4 AAb 750 750 750 Spearman Correlation

500 500 500 IGF2, ng/mL IGF2, ng/mL IGF2, ng/mL 5-4 250 AAb 250 250

0 0 0 5 10 15 0 5 10 15 0 5 10 15 128 Months Months Months

129 Supplementary Figure 9. Longitudinal follow-up of multiple AAb subjects who

130 develop type 1 diabetes during study shows variable trajectory of IGF2 over time.

131 Subjects from longitudinal T1DBIT cohort. Raw IGF2 levels of individual T1DBIT subjects

132 shown as solid lines. Subjects arranged from those diagnosed earlier to later in the study.

133 5%, 50%, and 95% reference ranges for age and sex are shown per person as horizontal

134 dashed lines. Time of diagnosis shown as vertical red line (a1c > 6.4) and pre-type 1

135 diabetes (a1c > 5.7) or seroconversion events shown with vertical dashed black lines.

136 Spearman r and p values shown in upper right-hand corner.

137