The prolyl hydroxylase PHD3 maintains β-cell glucose metabolism during fatty acid excess

The alpha ketoglutarate-dependent dioxygenase, prolyl-4-hydroxylase 3 (PHD3), is a hypoxia-inducible factor target that uses molecular oxygen to hydroxylate proline. While PHD3 has been reported to influence cancer cell metabolism and liver insulin sensitivity, relatively little is known about effects of this highly conserved enzyme in insulin-secreting β-cells. Here, we show that deletion of PHD3 specifically in β-cells (βPHD3KO) is associated with impaired glucose homeostasis in mice fed high fat diet. In the early stages of dietary fat excess, βPHD3KO islets energetically rewire, leading to defects in the management of pyruvate fate and a shift away from glycolysis. However, βPHD3KO islets are able to maintain oxidative phosphorylation and insulin secretion by increasing utilization of fatty acids to supply the tricarboxylic acid cycle. This nutrient-sensing switch cannot be sustained and βPHD3KO islets begin to show signs of failure in response to prolonged metabolic stress, including impaired glucose-stimulated ATP/ADP rises, Ca2+ fluxes and insulin secretion. Thus, PHD3 might be a pivotal component of the β-cell glucose metabolism machinery by suppressing the use of fatty acids as a primary fuel source, under obesogenic and insulin resistant states. SIGNIFICANCE STATEMENT Prolyl-4-hydroxylase 3 (PHD3) is involved in the oxygen-dependent regulation of cell phenotype. A number of recent studies have shown that PHD3 might operate at the interface between oxygen availability and metabolism. To understand how PHD3 influences insulin secretion, which depends on intact glucose metabolism, we generated mice lacking PHD3 specifically in pancreatic β-cells. These mice, termed βPHD3KO, are apparently normal until fed high fat diet at which point their β-cells switch to fatty acids as a fuel source. This switch cannot be tolerated and β-cells in βPHD3KO mice eventually fail. Thus, PHD3 maintains glucose-stimulated insulin secretion in β-cells during states of fatty acid excess, such as diabetes and obesity.


INTRODUCTION 56
The prolyl-hydroxylase domain proteins (PHD1-3) encoded for by the Egl-9 homologue 57 (EGLN) genes are alpha ketoglutarate-dependent dioxygenases, which regulate cell function 58 by catalyzing hydroxylation of prolyl residues within various substrates using molecular 59 oxygen (1-4). There are three well-described mammalian isozymes: PHD1, PHD2 and PHD3, 60 which were originally described as hydroxylating the alpha subunit of the transcription factor 61 Hypoxia-Inducible Factor (HIF) under normoxia (4), thus targeting it for polyubiquitylation and 62 proteasomal degradation. When oxygen concentration becomes limited, PHD activity 63 decreases and HIF is stabilized, leading to dimerization with the beta subunit and 64 transcriptional regulation of target genes regulating the cellular response to hypoxia (5). While 65 PHDs are generally regarded to be master HIF regulators, it is becoming increasingly apparent 66 that they target a range of other substrates influencing cell function (6-9). 67 PHD3 is unusual amongst the PHDs: it is transcriptionally regulated by HIF1 during hypoxia 68 (10) although it does not always act to destabilize HIF1 (11, 12). A number of roles for PHD3 69 have been described under conditions of stress or hypoxia, including: macrophage influx and 70 neutrophil survival (13, 14), apoptosis in various cancer models (8, 15, 16), and tumor cell 71 survival (9) (reviewed in (17)). Due to the dependence of PHD3 on alpha-ketoglutarate and 72 oxygen for its activity (18), many of these actions are likely to be mediated through alterations 73 in cell metabolism (19). Indeed, PHD3 increases glucose uptake in cancer cells through 74 interactions with pyruvate kinase M2 (8, 20). In tumors exhibiting mutations in succinate 75 dehydrogenase, fumarate hydratase and isocitrate dehydrogenase 1 and 2 (21-23), PHD3 76 activity is altered by aberrantly high cytosolic concentrations of succinate, fumarate and 2-77 hydroxyglutarate (2-HG), suggesting that inactivation of this enzyme might be involved in the 78 cellular transformation process. PHD3 has more recently been shown to hydroxylate and 79 activate acetyl-CoA carboxylase 2 (ACC2), defined as the fatty acid oxidation gatekeeper, 80 thus decreasing fatty acid breakdown and restraining myeloid cell proliferation during nutrient 81 abundance (24). Together, these studies place PHD3 as a central player in the regulation of 82 glucose and fatty acid utilization with clear implications for metabolic disease risk. 83 Along these lines, PHD3 has been reported to influence insulin sensitivity in the liver (25, 26), 84 as well as maintain glucose-stimulated insulin secretion in a rat β-cell line (27). However, little 85 is known about how PHD3 might contribute to glucose homeostasis and diabetes risk through 86 effects directly in primary pancreatic β-cells. To ensure the appropriate release of insulin, β-87 cells have become well-adapted as glucose sensors. Thus, glucose enters the β-cell by 88 facilitated diffusion through low affinity glucose transporters (28), before conversion into 89 glucose-6-phosphate by glucokinase and subsequent splitting into pyruvate (29). The 90 pyruvate then undergoes oxidative metabolism in the mitochondrial matrix through the 91 tricarboxylic acid (TCA) cycle, driving increases in ATP/ADP ratio and leading to closure of 92 ATP-sensitive K + channels (30). This cascade triggers membrane depolarization, opening of 93 voltage-dependent Ca 2+ channels, influx of Ca 2+ , and Ca 2+ -dependent exocytosis of insulin 94 vesicles through interactions with the SNARE machinery (30). Together with repression of 95 hexokinase, monocarboxylic acid transporter 1 and lactate dehydrogenase A (31, 32), 96 stimulus-secretion coupling prevents the inappropriate release of insulin in response to low 97 glucose, amino acids or lactate. 98 Given its reported roles in dictating fuel preference, we hypothesized that PHD3 might function 99 as a pivotal component of the β-cell glucose-sensing machinery by suppressing the use of 100 4 fatty acids as an energy source (27). To further investigate PHD3-regulated β-cell function in 101 depth, we subjected a model of β-cell-specific Egln3, encoding for PHD3, deletion to extensive 102 in vivo and in vitro characterization, including detailed stable isotope-resolved metabolic 103 tracing. Here, we show that loss of PHD3 causes metabolic remodelling in the early stages of 104 metabolic stress by shifting β-cell fuel source from glucose to fatty acids. However, this 105 metabolic switch is overwhelmed as fatty acids accumulate, ultimately leading to β-cell failure. 106 As such, PHD3 is likely to constitute a fundamental mechanism to restrain fatty acid utilization 107 and maintain glucose-sensing in β-cells during early stages of metabolic stress and insulin 108 resistance. 109 110 5

PHD3 knockout does not induce a hypoxic gene expression phenotype 112
We first generated a model of β-cell PHD3 knockout (βPHD3KO) by crossing the Ins1Cre 113 deleter strain (33) with animals harboring flox'd alleles for Egln3 (34), which encodes PHD3. 114 Recombination efficiency of the Ins1Cre line was verified in-house by crossing to mTmG 115 reporter animals and was found to be >90%. Gene expression analyses showed a 2-fold 116 reduction in Egln3 in βPHD3KO islets ( Figure 1A). Western blotting revealed a similar ~50% 117 knockdown of PHD3 protein in βPHD3KO islets ( Figure 1B), the remainder most likely 118 reflecting the relatively higher levels of Egln3 detected in α-cells, as shown by RNA-seq (35, 119 36). While we attempted immunofluorescence staining, we could not detect a specific signal 120 in β-cells, probably reflecting known sensitivity issues with PHD3 antibodies. Previous studies 121 have shown that PHD3 is highly regulated at the transcriptomic level by hypoxia (10), and in 122 line with this, we also found that Egln3 levels in hypoxic (1% O2) βPHD3CON islets were highly 123 upregulated ( Figure 1C). While Egln3 is expressed at low abundance in sorted β-cells (35,124 36), this is likely to be a result of profound re-oxygenation following dissociation, thus 125 suppressing Egln3 expression (37). 126 To account for HIF-dependent effects on β-cell phenotype in βPHD3KO animals, a number of 127 canonical HIF1α-target genes were assessed. Notably, levels of Bnip3 and Car9 were similar 128 between normoxic (21% O2) βPHD3CON and βPHD3KO islets ( Figure 1D-F). Further 129 suggesting the presence of intact HIF signaling, Bnip3, Car9 were upregulated to similar levels 130 in hypoxic (1% O2) βPHD3CON and βPHD3KO islets ( Figure 1D-F), while Gls did not reliably 131 increase ( Figure 1D-F). Lastly, glucose and KCl-stimulated Ca 2+ fluxes, shown to be sensitive 132 to HIF stabilization (38), were similar in βPHD3CON and βPHD3KO islets exposed to hypoxia 133 ( Figure 1G-J). 134 PHD3 does not contribute to glucose homeostasis and insulin release under normal 135 diet 136 Male and female βPHD3KO mice presented with normal growth curves from 8-18 weeks of 137 age compared to βPHD3CON littermates (Figure 2A and B). Glucose tolerance testing in the 138 same animals showed no abnormalities in glycemia ( Figure 2C and D), which did not change 139 up until the age of 20 weeks ( Figure 2E and F). As expected, βPHD3KO mice displayed similar 140 insulin sensitivity ( Figure 2G), islet size distribution and β-cell mass ( Figure 2H-J) to their 141 βPHD3CON littermates. 142 Isolation of islets for more detailed in vitro workup revealed normal expression of the β-cell 143 transcription factors/differentiation markers Pdx1, Mafa and Nkx6-1 in βPHD3KO islets ( Figure  144 3A-C). Further suggestive of mature β-cell function, live imaging approaches revealed intact 145 glucose-stimulated ATP/ADP ratios ( Figure 3D and E) and Ca 2+ fluxes ( Figure 3F and G) in 146 βPHD3KO islets. While glucose-stimulated insulin secretion was similar in islets isolated from 147 βPHD3CON and βPHD3KO animals, responses to the incretin-mimetic Exendin-4 (Ex4) were 148 blunted ( Figure 3H). This defect in Ex4-potentiated insulin secretion was not due to reductions 149 in Glp1r expression ( Figure 3I) or cAMP responses to the incretin-mimetic ( Figure 3J and K). 150 Moreover, oral glucose tolerance, largely determined by incretin release from the intestine 151 (39), was similar in βPHD3CON and βPHD3KO mice ( Figure 3L). 152 Loss of PHD3 improves insulin secretion at the onset of metabolic stress 153 6 We next examined whether PHD3 might play a more important role in regulating insulin 154 release during metabolic stress. Indeed, the increase in islet size that occurs during insulin 155 resistance is associated with a hypoxic state (40), expected to increase PHD3 levels via HIF1 156 activity (41). Therefore, animals were placed on high fat diet (HFD) to induce obesity and 157 metabolic stress(42). 158 Following 4 weeks HFD, Egln3 was mildly but significantly upregulated in βPHD3CON islets 159 ( Figure 4A). As expected, Egln3 levels remained suppressed in 4 weeks HFD βPHD3KO islets 160 ( Figure 4A). Glucose tolerance testing revealed significantly impaired glucose homestasis in 161 βPHD3KO mice at 4 weeks but not at 72 hours HFD ( Figure 4B and C), despite similar body 162 weight gain compared to βPHD3CON littermates ( Figure 4D). As expected, fasting blood 163 glucose levels were elevated in βPHD3CON mice following 4 weeks HFD ( Figure 4E). There 164 was no effect of Cre or flox'd alleles per se on glucose tolerance following 4 weeks HFD 165 ( Figure 4F). Glucose-stimulated insulin secretion in vivo was however increased in 4 weeks 166 HFD βPHD3KO mice ( Figure 4G), suggesting that glucose intolerance might be associated 167 with hyperinsulinemia (43). These increases in circulating insulin were associated with an 168 almost 2-fold increase in β-cell mass in 4 weeks HFD βPHD3KO mice ( Figure 4H and I), 169 associated with a significant increase in the proportion of larger islets ( Figure 4J). In addition, 170 islets isolated from the same animals secreted significantly more insulin in glucose-stimulated 171 and Ex4-potentiated states ( Figure 4K). Increased islet size or insulin expression induced by 172 HFD were unlikely to account for the overall increase in in vitro insulin secretion, since all 173 measures were corrected for insulin content. Bnip3 and Gls levels remained unchanged 174 ( Figure 4L and M), while Car9 was downregulated ( Figure 4N), suggesting that HIF1α 175 stabilization was unlikely to be a major feature in HFD βPHD3KO islet. 176 Thus, βPHD3KO mice are glucose intolerant on HFD, but their islets are larger and show 177 improved insulin secretion. These data raise the possibility that nutrient-sensing and utilization 178 might be altered in βPHD3KO islets. 179

PHD3 maintains glucose metabolism in β-cells 180
Given the reported roles of PHD3 in glycolysis, we wondered whether the changes in β-cell 181 function observed during the early phases of high fat feeding in βPHD3KO mice might be 182 associated with changes in glucose metabolism. We first looked at glycolytic fluxes using 14 C 183 glucose. While glucose oxidation was not altered at low or high glucose in islets from 4 weeks 184 HFD βPHD3KO mice ( Figure 5A), there was a small but significant decrease in 14 C content in 185 the aqueous phase, indicating a net reduction in tricarboxylic acid (TCA) cycle/other 186 metabolites derived from glycolysis ( Figure 5B). Notably, a 2-fold reduction in incorporation of 187 glucose into the lipid pool (i.e. glucose-driven lipogenesis) was also detected in 4 weeks HFD 188 βPHD3KO islets ( Figure 5C), suggestive of decreased glycolytic flux through the TCA cycle 189 and acetyl-CoA carboxylase 1 (ACC1) (44). 190 To gain a higher resolution analysis of glucose fate, stable isotope-resolved tracing was 191 performed in βPHD3KO islets using 13 C6-glucose. GC-MS-based 13 C6 mass isotopomer 192 distribution analysis showed no differences in glucose incorporation into aspartate, glutamate, 193 malate, fumarate or citrate in either standard chow or 4 weeks HFD βPHD3CON and 194 βPHD3KO islets ( Figure 5D-H). Thus, while the contribution of glucose to aqueous cellular 195 metabolite pools is clearly reduced in 4 weeks HFD βPHD3KO islets ( Figure 5B), there is no 196 net change in the incorporation of glucose into each metabolite i.e. the TCA cycle proceeds 197 7 normally despite lowered glucose fluxes. Islets from animals fed standard chow showed m+2 198 lactate accumulation ( Figure 5I), which is consistent with lactate normally produced as a result 199 of oxidative metabolism of glucose-derived pyruvate. However, during HFD there was a 200 pronounced switch to reduction of pyruvate to lactate (indicated by the m+3 isotopomer) in 201 both genotypes. 202 Further analysis of steady-state lactate levels showed a significant increase in lactate 203 production in islets from HFD-fed βPHD3KO versus βPHD3CON mice ( Figure 5J). Together 204 with the m+2  m+3 switch, this finding confirms initial measures with 14 C glucose indicating 205 reduced fueling of the TCA cycle by glycolysis ( Figure 5K). Furthermore, the tracing data 206 suggest that 4 weeks HFD βPHD3KO islets increase the reduction of pyruvate to support 207 continued glycolysis through regeneration of the cytosolic NAD + pool. The source of the lactate 208 was unlikely to be through increases in expression of the "disallowed" gene lactate 209 dehydrogenase A (Ldha) (31, 32), since Ldha levels were unchanged between βPHD3CON 210 and βPHD3KO islets ( Figure 5L). 211 Together, these data suggest that metabolic stress induces defects in the management of 212 pyruvate fate in βPHD3KO islets, implying that insulin secretion must be maintained and even 213 amplified through mechanisms other than glycolysis in vitro. 214

PHD3 suppresses fatty acid use under metabolic stress 215
We hypothesized that βPHD3KO islets might switch to an alternative energy source to sustain 216 their function, namely beta oxidation of fatty acids, which are present in excess during HFD. 217 Moreover, in cancer cells PHD3 has been shown to increase activity of ACC2, which converts 218 acetyl-CoA  malonyl-CoA, the latter suppressing carnitine palmitoyltransferase I (CPT1), the 219 rate-limiting step in fatty acid oxidation (24). Indicating a profound change in β-cell nutrient 220 preference, supplementation of culture medium with the fatty acid palmitate for 48 hours 221 augmented glucose-stimulated and Ex4-potentiated insulin secretion in 4 weeks HFD 222 βPHD3KO islets ( Figure 6A). By contrast, 4 weeks HFD βPHD3CON islets showed no 223 increase in glucose-stimulated insulin release following culture with palmitate ( Figure 6B), 224 confirming that the fatty acid was unlikely to induce lipotoxicity at the concentration and timing 225 used here. Interestingly, 48 hrs incubation with palmitate increased Ex4-potentiated insulin 226 secretion in 4 weeks HFD βPHD3CON islets ( Figure 6B). 227 Further confirming a switch away from glycolysis, glucose-stimulated ATP/ADP ratios were 228 markedly decreased in 4 weeks HFD βPHD3KO islets ( Figure 6C-E), despite the increased 229 insulin secretion ( Figure 4K). While downstream Ca 2+ fluxes were apparently normal in 4 230 weeks HFD βPHD3KO islets, this was likely due to increased sensitivity of voltage-dependent 231 Ca 2+ channel to membrane depolarization, since responses to KCl were significantly elevated 232 ( Figure 6F and G). Suggesting that CPT1 activity might be upregulated in 4 weeks HFD 233 βPHD3KO islets, experiments were performed at high glucose, normally expected to inhibit 234 CPT1 and fatty acid utilization through generation of malonyl-CoA, and mRNA levels of Cpt1a 235 tended to be increased ( Figure 6H). Moreover, application of the CPT1 inhibitor etomoxir was 236 able to augment ATP/ADP responses to glucose in 4 weeks HFD βPHD3KO but not in 237 βPHD3CON islets ( Figure 6I). In line with this finding, culture with low palmitate concentrations 238 decreased glucose-stimulated Ca 2+ fluxes in 4 weeks HFD βPHD3KO ( Figure 6J) but not in 239 βPHD3CON islets ( Figure 6K), presumably due to increased flux of acetyl-CoA into the TCA 240 cycle. 241 8 Thus, following 4 weeks HFD, βPHD3KO islets become less reliant on glycolysis to fuel 242 ATP/ADP production, are able to sustain oxidative phosphorylation through fatty acid 243 oxidation, and secrete more insulin when both glucose and fatty acids are present. These 244 changes, which are in agreement with our initial in vivo and in vitro phenotyping data ( Figure  245 4), are shown schematically in Figure 6L. 246 ACC1 and ACC2 are differentially regulated at the promoter level in β-cells 247 ACC1 and ACC2 are enzymes that catalyze the carboxylation of acetyl-CoA to malonyl-CoA, 248 the rate-limiting step in fatty acid synthesis. β-cells are thought to predominantly express 249 ACC1 (45, 46), which supplies cytosolic malonyl-CoA to fatty acid synthase for de novo lipid 250 biosynthesis rather than for beta oxidation (44). By contrast, β-cells are reported to express 251 negligible levels of ACC2, which inhibits CPT1 through generation of mitochondrial malonyl-252 CoA to suppress use of fatty acids via beta oxidation (45, 46). 253 To explore potential mechanisms that might underlie the changes occurring in 4 weeks HFD 254 βPHD3KO islets, we decided to interrogate multiple published bulk islet and purified β-cell 255 gene expression datasets (35, 47, 48). Notably, ACACB (encoding ACC2) was found to be 256 present in β-cells, albeit at lower levels than ACACA (encoding ACC1) ( Figure 7A). The 257 expression levels of ACACB were significant, however, reaching similar levels to the β-cell 258 transcription factor HNF1A ( Figure 7A). 259 Closer examination of the promoter of ACACB gene in islets also revealed regulation by a 260 number of established β-cell transcription factors, such as PDX1, MAFB and NKX2-2 ( Figure  261 7B), further confirming the regulated expression of this gene. Unusually, a long non-coding 262 RNA (lncRNA), transcribed antisense to the ACACB gene, was detected ( Figure 7B), 263 consistent with the presence of a negative regulatory mechanism for the expression of this 264 gene in one or more cell types in the islet. 265 Suggesting that any regulation of ACC1/ACC2 by PHD3 is likely to be post-translational, as 266 expected for a hydroxylase, qPCR analyses showed that both Acaca and Acacb expression 267 were similar in 4 weeks HFD βPHD3CON and βPHD3KO islets ( Figure 7C and D). 268 Thus, according to next generation sequencing, ACACB is reproducibly expressed in β-cells, 269 but at levels lower than ACACA. Assuming that protein translation occurs, ACC2 might 270 conceivably contribute to fatty acid oxidation in the absence of PHD3. 271

PHD3 protects against lipotoxicity following prolonged metabolic stress 272
Lastly, we sought to investigate whether islets would eventually decompensate when faced 273 with continued fatty acid/nutrient abundance. Glucose intolerance was still present in 274 βPHD3KO mice following 8 weeks on HFD ( Figure 7E), despite normal insulin sensitivity 275 ( Figure 7F). By this point, however, impaired glucose-stimulated insulin secretion ( Figure 7G) 276 was apparent in isolated βPHD3KO islets. This secretory deficit could be rescued by 277 application of Ex4 to sensitize insulin granules to exocytosis ( Figure 7G). In addition, the 278 amplitude of glucose-stimulated Ca 2+ rises was significantly reduced in 8 weeks HFD 279 βPHD3KO compared to βPHD3CON islets ( Figure 7H and I). Suggesting that profound 280 defects in voltage-dependent Ca 2+ channels might also be present, responses to the generic 281 depolarizing stimulus KCl were markedly blunted in the same islets ( Figure 7H and I). While 282 apoptosis was increased in 8 weeks HFD βPHD3KO islets( Figure 7J and K), α-cell/β-cell ratio 283 ( Figure 7L) and expression of the ER stress markers Ddit3, Xbp1 and Hspa5 ( Figure 7M) 284 remained unchanged. Lastly, the HIF2α-target genes Ccnd1 and Dll4 were found to be either 285 unchanged or downregulated in 8 weeks HFD βPHD3KO islets ( Figure 7N), suggesting that 286 HIF2α stabilization was unlikely to be the sole determinant of phenotype. 287

DISCUSSION 288
In the present study, we build upon previous observations that chemical inhibition of all three 289 PHD enzymes in islets and β-cell lines leads to alterations in glucose-stimulated insulin 290 secretion (27). Specifically, we show that the alpha-ketoglutarate-dependent PHD3 maintains 291 β-cell glucose sensing under states of metabolic stress and insulin resistance associated with 292 fatty acid abundance. Our data suggest that PHD3 is required for ensuring that acetyl-CoA 293 derived from glycolysis preferentially feeds the TCA cycle, linking blood glucose levels with 294 ATP/ADP generation, β-cell electrical activity and insulin secretion. Loss of PHD3 leads to 295 metabolic remodeling under HFD, resulting in a decrease in glycolytic fluxes, an increase in 296 lactate accumulation and utilization of fatty acids as an energy source. This switch cannot be 297 maintained, however, and β-cells eventually fail following prolonged exposure to fatty acids. 298 Thus, PHD3 appears to be a critical component of the β-cell metabolic machinery required for 299 glucose sensing during episodes of nutritional overload. 300

How does PHD3 maintain glucose metabolism in β-cells? Previous studies in cancer cells 301
have shown that PHD3 hydroxylates and activates ACC2, suppressing beta oxidation (24). 302 While β-cells are thought to predominantly express ACC1, the levels of ACACB, which 303 encodes ACC2, were found to be similar to the β-cell transcription factor HNF1A, albeit much 304 lower than for ACACA. Intriguingly, ACACB was enriched for promoter sites suggestive of 305 negative regulation, which is unusual amongst β-cell genes, and Acaca/Acacb were not 306 upregulated in HFD βPHD3CON or βPHD3KO islets. This supports a potential role for post-307 transcriptional modifications in determining ACC2 activity. We thus propose that loss of PHD3 308 might lead to suppression of ACC2 activity, which becomes apparent during HFD when its 309 substrate is present in abundance. Alternatively, PHD3 might hydroxylate and activate ACC1, 310 leading to regulation of CPT1 by malonyl-CoA when fatty acids are supplied in excess, as 311 suggested by glucose oxidation experiments. In any case, experiments with etomoxir strongly 312 infer a role for CPT1 in the negative effects of PHD3 deletion on glucose metabolism. While 313 etomoxir has been shown to target complex I of the electron transport to lower ATP production 314 (49), we don't think this played a major role here, since ATP levels were restored in βPHD3KO 315 islets treated with the inhibitor. 316 dentifying the efficiency of protein translation for ACACB  ACC2, as well as the PHD3 317 hydroxylation sites involved, will be critical. However, currently available antibodies for 318 detection of ACC2, as well its hydroxylated forms, are poor. Moreover, assigning hydroxylation 319 targets using mass spectrometry is controversial: mis-alignment of hydroxylation is commonly 320 associated with the presence of residues in the tryptic fragment that can be artefactually 321 oxidised (50). Thus, studies using animals lacking both PHD3 and ACC2 specifically in β-cells 322 would be required to definitively link the carboxylase with the phenotype here. 323 As normal chow contains a low proportion of calories from fat, metabolic stress was needed 324 to reveal the full in vitro and in vivo phenotype associated with PHD3 loss. These data also 325 support an effect of PHD3 on ACC1/2 and CPT1, since without acyl-CoA derived from 326 exogenous fatty acids, glucose would still constitute the primary fuel source and regulator of 327 insulin release. The lack of phenotype under normal diet is unlikely to reflect the age of the 328 animals, since even at 20 weeks of age, glucose intolerance was still not present in βPHD3KO 329 mice. An alternative explanation is that loss of PHD3 can be compensated under normal 330 conditions, while other mechanisms associated with fatty acid excess and lipogenesis, for 331 example ER stress (51, 52), also contribute to the βPHD3KO phenotype. We feel that this 332 explanation is less likely, however, since we could not detect upregulation of Ddit3, Xbp1 and 333 Hspa5 even following 8 weeks HFD. 334 Suggesting that the phenotype associated with PHD3 loss was not solely due to HIF, no 335 differences in the gene expression of HIF1 targets could be detected in βPHD3KO versus 336 βPHD3CON islets. Indeed, PHD2 is the major hydroxylase that regulates HIF1α stability (11, 337 12), with no changes in activity of the transcription factor detected following PHD3 loss (11, 338 12, 53). Thus, it is perhaps unsurprising that there is a lack of HIF1 transcriptional signature 339 in βPHD3KO islets, in agreement with previous studies in other tissues (53, 54). In addition, 340 glucose-stimulated Ca 2+ fluxes, a sensitive readout of changes in oxygen-dependent 341 regulation (38), were unaffected during hypoxia in βPHD3KO islets. Lastly, no changes in 342 expression of the HIF1-sensitive gene Ldha (55) were detected. Nonetheless, we cannot 343 completely exclude HIF-dependent effects, and as such, studies should either be repeated on 344 a HIF1-and HIF2-null background (i.e. a quadruple transgenic) or using (moderately) specific 345 chemical inhibitors. 346 An intriguing observation was that PHD3 deletion decreased Exendin-4-but not glucose-347 stimulated insulin secretion in islets from animals fed standard chow. Given that Glp1r 348 expression and signaling remained intact in βPHD3KO islets, alterations in cytosolic glutamate 349 accumulation might instead be present, previously shown to prime incretin-responsiveness 350 following its release with insulin from the granule (56). Arguing against this, however, 351 interrogation of the metabolic tracing data showed that steady-state glutamate levels were 352 unchanged between βPHD3CON and βPHD3KO islets, meaning that glucose was still able to 353 enter the malate-aspartate shuttle to produce the neurotransmitter. It will be interesting in the 354 future to pinpoint how PHD3 impinges upon Ex4-potentiated insulin secretion. 355 In summary, PHD3 possesses a conserved role in gating nutrient preference toward glucose 356 and glycolysis during both cell transformation (24) and metabolic stress (here). It will be 357 interesting to now study whether similar effects of PHD3 are present in other cell types 358 involved in glucose-sensing (for example, pancreatic alpha cells, hypothalamic neurons). 359

Study design 362
No data were excluded unless the cells displayed a non-physiological state (i.e. impaired 363 viability). All individual data points are reported. The measurement unit is animal or batch of 364 islets, with experiments replicated independently. Animals and islets were randomly allocated 365 to treatment groups to ensure that all states were represented in the different experiment arms. 366

Study approval 367
Animal studies were regulated by the Animals (Scientific Procedures) Act 1986 of the U.K., 368 and approval was granted by the University of Birmingham's Animal Welfare and Ethical 369 Review Body. 370

Mouse models 371
β-cell-specific PHD3 (βPHD3KO) knockout mice were generated using the Cre-LoxP system. 372 Ins1Cre allele is not associated with any changes in glucose homeostasis in our hands (33, 57). Given 378 recently reported issues with allele-silencing in some Ins1Cre colonies (58), recombination 379 efficiency of our line was regularly monitored and verified to be >90% using ROSA mT/mG 380 reporter animals (59). Animals were maintained on a C57BL/6J background and backcrossed 381 for at least 6 generations following re-derivation into the animal facility. Lines were regularly 382 refreshed by crossing to bought-in C57BL/6J (Charles River). βPHD3CON and βPHD3KO 383 mice were fed standard chow (SC) and/or high fat diet containing 60% fat (HFD), with body 384 weight checked weekly until 18 weeks of age. Animals were maintained in a specific pathogen-385 free facility, with free access to food and water. 386

Intraperitoneal and oral glucose tolerance testing 387
Mice were fasted for 4-6 hours, before intraperitoneal injection of glucose (1-2 g/kg body 388 weight). Blood samples for glucose measurement were taken from the tail vein at 0, 15, 30, 389 60, 90 and 120 min post-challenge. Glucose was measured using a Contour XT glucometer 390 (Bayer, Germany). For mice on SC, intraperitoneal glucose tolerance testing (IPGTT) was 391 performed every 2-4 weeks, between 8-20 weeks of age. HFD-fed mice underwent IPGTT 392 following 4 and 8 weeks of HFD. Oral glucose tolerance testing (OGTT) was performed as for 393 IPGTT, except that 2 g/kg body weight glucose was delivered using an oral gavage tube. 394

Statistics 515
Measurements were performed on discrete samples unless otherwise stated. Data normality 516 was assessed using D'Agostino-Person test. All analyses were conducted using GraphPad 517 Prism software. Pairwise comparisons were made using Student's unpaired or paired t-test, 518 or Mann-Whitney test. Multiple interactions were determined using either Kruskal-Wallis test, 519 one-way ANOVA or two-way ANOVA followed by Tukey's, Dunn's, Dunnett's, Bonferonni's or 520 Sidak's post-hoc tests (accounting for degrees of freedom). 521