Mature and immature β-cells both contribute to islet function and insulin release 1

Transcriptionally mature and immature β-cells co-exist within the adult islet. How such 42 diversity contributes to insulin release remains poorly understood. Here we show that 43 differences in β-cell maturity, defined using PDX1 and MAFA expression, are required for 44 proper islet operation. Functional mapping of rodent and human islets containing 45 proportionally more mature β-cells revealed defects in metabolism, ionic fluxes and insulin 46 secretion. At the transcriptomic level, the presence of increased numbers of mature β-cells led 47 to dysregulation of gene pathways involved in metabolic processes. Using a chemogenetic 48 disruption strategy, the islet signalling network was found to contribute to differences in 49 maturity across β-cells. During metabolic stress, islet function could be restored by redressing 50 the balance between immature and mature β-cells. Thus, preserving a balance between 51 immature and mature β-cells might be important for islet engineering efforts and more broadly 52 the treatment of type 1 and type 2 diabetes. 53 54 55 56 to account


INTRODUCTION 57
Type 2 diabetes mellitus (T2DM) occurs when β-cells are unable to release enough insulin to 58 compensate for insulin resistance. At the cellular level, glucose-regulated insulin secretion 59 depends upon generation of ATP/ADP, closure of ATP-sensitive K + (KATP) channels, opening 60 of voltage-dependent Ca 2+ channels (VDCC) and exocytosis of insulin granules 1 . At the 61 multicellular level, insulin release is a tightly controlled process, requiring hundreds of β-cells 62 throughout the islet to coordinate their activities in response to diverse stimuli including 63 glucose, incretins and fatty acids 2,3 . 64 Our current understanding of the mechanisms underlying insulin release is mainly derived 65 from experiments in single β-cells or measures averaged across the entire β-cell complement. 66 However, such studies, which generally view β-cells as a tightly coupled system, are difficult 67 to reconcile with the known heterogeneous nature of β-cell identity and function. Based on 68 transcriptomic 4,5 and protein signatures 6 , marker analyses 7-9 , glucose-responsiveness 10,11 , 69 reporter imaging 12-15 or single molecule hybridization 16 , β-cell subpopulations have been 70 shown to exist with altered maturity states, metabolism, electrical activity, insulin secretion and 71 proliferative capacity (reviewed in 17,18 ). Of note, β-cell subpopulations are highly plastic. 72 During aging and T2DM, β-cells with reduced maturity, metabolism and insulin secretion, but 73 enhanced proliferative capacity, typically increase in proportion in both rodent and human 4,7,8 . 74 At the same time, there is an increase in the number of mature, secretory β-cells that display 75 poorer proliferative capacity 6,7 . Thus, the adult islet houses highly plastic mature and 76 immature β-cell subpopulations whose co-existence might be important for balancing renewal 77 with the need for insulin release. 78 Mature β-cells are generally thought to contribute the most to islet function, since they 79 comprise ~70-90% of the β-cell population, express higher levels of insulin, glucose 80 transporter, glucokinase and maturity genes, and mount normal ATP/ADP and Ca 2+ responses 81 to stimulus (reviewed in 19 ). By contrast, immature β-cells are in the minority, show poor 82 glucose-responsiveness and are less secretory 4,7,8,14,19 . However, β-cell subpopulations that 83 disproportionately influence islet responses to glucose have recently been identified in situ 84 and in vivo [20][21][22] . One of the subpopulations, termed 'hubs', was found to display lowered 85 expression of β-cell maturity markers and insulin, but increased expression of glucose-sensing 86 enzymes, including glucokinase 21,22 . These studies provide the first glimpse that immature 87 cells with similar characteristics might contribute to the regulation of insulin release across the 88 islet. 89 We hypothesized that transcriptionally immature β-cells belong to a highly functional 90 subpopulation, able to overcome their relative deficiencies by interacting with their more 91 mature counterparts to drive insulin release. Using recombinant genetics together with 92 chemogenetic disruption, we therefore set out to alter the balance of immature:mature β-cells, 93 before determining the effect of this manoeuvre on adult islet function. 94

Generation of islets with proportionally more mature β-cells 96
We first generated and validated a novel overexpression model to alter the balance between 97 immature and mature β-cells throughout the population. Here, immature β-cells are 98 operationally defined as expressing low levels of the transcription factors PDX1 and MAFA 99 based upon immunohistochemistry. Islets were transduced with control adenovirus containing 100 PATagRFP (β normal; B-NORM) or a well-characterized polycistronic construct encoding 101 NEUROG3/PDX1/MAFA (Ad-M3C) (β mature; B-MAT). The M3C construct is well-validated 102 23, 24 , a TetO mouse possessing the same construct exists 25 , and driving multiple transcription 103 factors using the same promoter avoids heterogeneous expression profiles. Ad-M3C was able 104 to drive exogenous Neurog3, Pdx1 and Mafa expression (Fig. 1a), expected to occur 105 predominantly in the first two layers of the islet where functional imaging takes place. Native 106 gene expression levels remained unchanged for Neurog3 and Mafa, but ~ 25% lower for Pdx1, 107 consistent with the absence of positive autoregulation seen with Pdx1-fluorophore constructs 108 26 . 109

Analyses of individual β-cells in intact islets showed a non-Gaussian distribution of PDX1 and 110
MAFA protein fluorescence intensities in B-NORM islets, which we arbitrarily define as 111 PDX1 LOW /MAFA LOW and PDX1 HIGH /MAFA HIGH using a 15% cut-off (i.e. the bins spanning 0-15 112 normalized PDX1/MAFA intensity units) ( Fig. 1b-d examined, suggesting that PDX1 LOW and MAFA LOW cells are functionally immature (Fig 1e-g). 119 While very low levels of NEUROG3 could be detected in B-MAT islets (Fig. S2g), a progenitor 120 signature was not detected at the transcriptomic level (see below). A generalized PDX1 121 overexpression across the β-cell population was unlikely given that the mean fluorescence 122 intensity was only slightly (~20%) increased in B-MAT islets ( Preferential overexpression in PDX1 LOW /MAFA LOW (immature) β-cells was confirmed using 125 Pdx1-BFP reporter islets 26 , which read out endogenous Pdx1 levels. Quantification of PDX1 126 and BFP levels in the same cells revealed a strong positive linear correlation in B-NORM 127 islets. However, the correlation was weaker (and slope less steep) in B-MAT islets due to 128 transition of a subpopulation of BFP LOW cells to a PDX1 HIGH state (Fig. 1h). Supporting this 129 finding, BFP LOW cells (prior immature cells) adopted a PDX1 HIGH phenotype in B-MAT islets, 130 while BFP HIGH cells (prior mature) remained PDX1 HIGH ( Fig. 1i and j). These changes were in 131 line with the viral transduction efficiency, which was higher in PDX1 LOW cells ( Fig. S3a and b). 132 While overlap in PDX1 levels in PDX1 LOW and PDX1 HIGH cells in B-NORM islets was observed, 133 this likely reflects variability between experimental replicates, since the values were non-134 normalized. We cannot however exclude the presence of MAFA LOW cells that are not PDX1 LOW . 135 To further understand the sequence of events that occur within the islet following viral 136 transduction, time-course experiments were performed. Notably, a shift in the normalized 137 distribution of PDX1 fluorescence was detected beginning at 24 hrs post-infection, which 138 persisted until 120 hrs (Fig. S3c-f). This change was accompanied by a gradual increase in 139 whole islet PDX1 levels (Fig. S3g), suggesting that, at the low titres used here, immature β-140 cells are more susceptible to viral transduction, and that overexpression increases over time 141 to maintain the same distribution. These data fit with previous reports showing that, while most 142 β-cells are infected with adenovirus, transduction efficiency depends on the capacity of the 143 cell to produce a protein 27 . PDX1 LOW cells are presumably well-placed to ramp-up de novo 144 protein synthesis, since they are also INS LOW (Fig. 1e) and thus unconstrained by higher rates 145 of insulin production. 146 Together, these results show a shift toward proportionally more mature β-cells in B-MAT islets 147 following overexpression, thus validating the model. 148

α-, βand δ-cell identity are maintained in B-MAT islets 149
Further analyses of B-MAT islets detected no differences in the ratios of α-cells or δ-cells with 150 β-cells ( Fig. 1k-n), or numbers of PDX1 + INS + cells ( Fig. 1o and p). Expression levels of the 151 key α-, β-and δ-cell identity markers Arx, Pax6 and Nkx6-1 (Fig. S4a), respectively, were also 152 unaffected. Moreover, we were unable to observe differences in the numbers of PDX1 + GCG + 153 cells (Fig. 1q) or detect bihormonal cells (Fig. S4b), consistent with the lack of viral 154 transduction in non β-cells (Fig. S4c). Indeed, we and others have previously shown that, at 155 the titres used here, adenovirus is highly specific for β-cells due to reduced coxsackie virus 156 receptor expression and low capacity for protein translation in α-cells 27-30 . However, we 157 acknowledge that experiments using a nucleus reporter line would be needed to completely 158 exclude transduction in α-cells. A major effect of PDX1 and MAFA overexpression on cell 159 viability was unlikely, since no changes in expression of genes for ER stress or the unfolded 160 protein response (UPR) were detected between B-NORM and B-MAT islets (Fig. S3d), in line 161 with similar ratios of TUNEL + β-cells (Fig. 1r). 162 Lastly, no differences in proliferation were observed between B-NORM and B-MAT islets (Fig.  163 1s). Thus, mild overexpression of NEUROG3, MAFA and PDX1 leads alters the ratio of 164 immature:mature beta cells without inducing a progenitor-like state, or detectable shifts in 165 proliferation andapoptosis or the proportions of islet endocrine cell types. The schematic in 166 Fig. 1t summarizes the loss of immature β-cell model. 167

PDX LOW /MAFA LOW β-cells are transcriptionally less mature 168
We next investigated whether PDX1 LOW /MAFA LOW cells possess a less mature transcriptional 169 signature. Indeed, β-cell identity and function is maintained by a specific set of transcription 170 factors, which are themselves under the control of a network of β-cell transcription factors (Fig.  171 2a) 31 . Networks of transcription factors regulate gene expression through binding to enhancer 172 clusters in a combinatorial manner 31 . Therefore, changes in expression of β-cell specific 173 transcription factors impact not one, but a network of transcription factors to alter abundance 174 of other key β-cell genes. 175 Analysis of published RNA-seq data showed that transcriptional levels of MAFA and PDX1 176 are highly correlated across islet samples from 64 donors (Fig. 2b), as expected given that 177 they belong to the same co-expression gene network module 32 . This tight correlation was also 178 present for genes located in the same co-regulatory network such as NEUROD1 and NKX6-179 1 (Fig. 2c), but not for those regulated by alternative transcriptional networks such, as GAPDH 180 and GLIS3 (Fig. 2c) 31 . Similar relationships were also captured at the single cell level where 181 human PDX LOW β-cells possess lower RNA abundance of genes present in the same network 182 module, including MAFA, MAFB and NKX6-1 (Fig. 2d) 33 . 183 Together, these co-expression data place PDX1 and MAFA at the heart of the transcription 184 factor network that regulates β-cell identity, suggesting that the lower levels of these two key 185 genes also indicate lower expression levels for other key β-cell transcription factors. 186 Differences in β-cell maturity sustain stimulus-secretion coupling 187 Islets were subjected to detailed functional mapping to understand how differences in β-cell 188 maturity might influence function. Multicellular Ca 2+ imaging experiments on Fluo8-loaded 189 islets (Fig. 3a) revealed reduced Ca 2+ responses to glucose and the generic depolarizing 190 stimulus KCl in B-MAT islets ( Fig. 3b-d), which was consistent between individual islet 191 preparations ( Fig. S5a and  were confirmed using the ratiometric Ca 2+ probe Fura2 (Fig 3f-h), which again was consistent 197 between mouse/preparation ( Fig. S5c and d). Impaired Ca 2+ fluxes in B-MAT islets were 198 associated, but not causally-linked, with a decrease in mRNA expression of the L-type Ca 2+ 199 channel subunits Cacna1d and Cacnb2, but not Cacna1c (Fig. 3i). 200 Suggesting a defect in electrical oscillations, Ca 2+ pulse duration was decreased in B-MAT 201 versus B-NORM islets ( Fig. 3j and k). We therefore explored if the changes in Ca 2+ fluxes 202 observed in B-MAT islets were accompanied by defects in metabolism and amplifying signals. 203 Using the biosensor Perceval, a ~ 2-fold decrease in glucose-stimulated ATP/ADP ratios was 204 apparent ( Fig. 3l and m). Suggestive of altered glucose-sensing, Ca 2+ and ATP/ADP glucose 205 concentration-responses were reduced ( Fig. 3p and q). While mRNA and protein expression 206 levels of glucokinase were not significantly different ( Fig. 3n and o), we note that this does not 207 necessarily correlate with the activity of the enzyme, which is allosterically regulated by 208 glucokinase regulatory protein 34 . Indicating impaired glucose-dependent amplifying signals, 209 cAMP levels were decreased in response to glucose and forskolin ( Fig. 3r and s). No changes 210 in mRNA for the major murine glucose-regulated adenylate cyclase, Adcy8 35 , were detected 211 (Fig. 3t). Potentially unifying the abovementioned metabolic and electrical observations, 212 analysis of PDX1 and Ca 2+ targets in B-MAT islets revealed changes in expression of both 213 G6pc2 and Ascl1 expression 36 , 37 (Fig. 3u). 214 Thus, islets with proportionally more mature β-cells display profound defects in metabolism 215 and stimulus-secretion coupling, including ionic and amplifying signals. 216

Differences in β-cell maturity sustain islet dynamics and insulin secretion 217
Since some β-cell functional subgroups possess an immature or energetic phenotype, we 218 investigated whether loss-of-immaturity would lead to a decline in these subpopulations shown 219 to drive islet dynamics. Fast Ca 2+ recordings (20 Hz) detected cells whose activity preceded 220 and outlasted that of the rest of the population. These cells, algorithmically-identified as 'hubs', 221 comprise ~ 1-10% of the β-cell population, orchestrate islet responses to glucose and show 222 immature traits (PDX1 LOW , NKX6-1 LOW , INS LOW ) 21 . The proportion of hubs was decreased in 223 B-MAT islets (Fig. 4a), most likely due to a reduction in the number of immature cells able to 224 act as hubs combined with decreased expression of Gjd2 (Fig. 4b), which encodes the gap 225 junction protein connexin 36 (Cx36). The loss of hubs was associated with a reduction in 226 indices of coordinated β-cell activity ('connectivity') ( Fig. 4c), typified by a shift toward more 227 stochastic β-cell population responses ( Fig. 4d and e) (Movie S1 and S2). 228 As predicted from the impairments in Ca 2+ fluxes, metabolism, amplifying signals and β-cell-229 β-cell connectivity, glucose-and Exendin-4-stimulated insulin release was markedly 230 decreased in B-MAT islets ( Fig. 4f and g), despite a 2-fold increase in insulin content (Fig. 4h).

231
Insulin secretion was similar in B-NORM and B-MAT islets when uncorrected for content, 232 suggesting that B-MAT islets release only a fraction of their secretory granule pool in response 233 to glucose (Fig. S5e-f). However, fold-change insulin secretion remained significantly 234 decreased in B-MAT islets (Fig.S5g). Super-resolution imaging revealed no differences in 235 insulin granule density at the membrane ( Fig.4i), in line with unchanged expression of mRNA 236 for the exocytotic machinery (e.g. Stx1a, Snap25 and Vamp2) (Fig.4j). Implying the presence 237 of normal insulin gene regulation, Ins1 and Ins2 mRNA levels were unaffected ( Fig. 4k and l). 238 The loss of incretin-responsiveness was surprising given that gut-and islet-derived GLP1 38,39 239 potently upregulates the sensitivity of insulin granules for exocytosis 38 . Further analyses 240 showed a large decrease in glucagon-like peptide-1 receptor (GLP1R) mRNA and protein 241 expression ( Fig. 4m and n), which was accompanied by impairments in Exendin-4-stimulated 242 cAMP ( Fig. 4o-q) and Ca 2+ ( Fig. 4r and s) signals. 243 As such, differences in β-cell maturity contribute to islet Ca 2+ dynamics and insulin release. 244

Differences in β-cell maturity are required for human islet function 245
We next examined whether differences in maturity status of individual β-cells might represent 246 a conserved route for islet function in human islets. As expected, transduction with Ad-M3C 247 (β human mature; B-hMAT) led to increases in exogenous Neurog3, Pdx1 and Mafa mRNA 248 levels (Fig. 5a). Endogenous levels of NEUROGN3, MAFA and PDX1 were unchanged ( Fig.  249 5b). 250 PDX1 fluorescence intensity distribution, visualized using antibodies with cross-reactivity 251 against both human and mouse protein, was bimodal in B-hNORM (β human normal) islets, 252 with peaks corresponding to PDX1 LOW and PDX1 HIGH populations ( Fig. 5c and d), again 253 arbitrarily defined by a 20% cut-off. A similar distribution of PDX1 fluorescence was detected 254 when normalized to DAPI staining ( Fig. S6a and b), or when only PDX1+/INS1+ cells were 255 considered (Fig. S6c). The number of cells occupying the PDX1 LOW range (i.e. immature) was 256 decreased in B-hMAT compared to B-hNORM islets ( Fig. 5c and d), suggesting a shift toward 257 a more homogenous distribution of β-cell maturity. As for mouse islets, PDX1 and INS 258 expression were found to be correlated (Fig. 5e). We were unable to extend findings to MAFA 259 and NEUROG3, since attempts at antibody staining were unsuccessful in the isolated islet. 260 In any case, B-hMAT islets presented with reductions in Ca 2+ responses to glucose or glucose 261 + KCl ( Fig. 5f-i), without alterations in the proportion of responsive cells (Fig. 5j), recorded 262 using the genetically-encoded Ca 2+ indicator, GCaMP6. These defects in Ca 2+ fluxes were 263 associated with significantly lowered expression of mRNA for the L and T-type Ca 2+ channel 264 subunits CACNA1C, CACNA1D and CACNA1G, as well as the Na + channel subunits SCN1B, 265 SCN3A and SCN8A (Fig. 5k). Islet Ca 2+ dynamics were disrupted in B-hMAT islets in general, 266 with decreases in gap junction protein expression (Fig. 5l), proportion of hub cells (Fig. 5m) 267 and β-cell-β-cell coordination ( Fig. 5n and o). Although glucose-stimulated insulin secretion 268 was similar in B-hMAT and B-hNORM islets (Fig. 5p), the former released only a fraction of 269 their granules when corrected for the increase in total insulin ( Fig. 5q and r). Thus, differences 270 in β-cells maturity similarly contribute to mouse and human islet function (Fig. 5s). 271

Increases in the proportion of immature β-cells impairs islet function 272
To investigate whether a balance between mature and immature β-cells is required for normal 273 islet operation, the opposite model was generated by inducing a higher proportion of immature 274 cells across the population. Application of short hairpin RNAs against Pdx1 resulted in a left-275 shift in the distribution of PDX1 protein fluorescence intensities, indicative of loss of PDX1 HIGH 276 cells ( Fig. 6a and b), in-line with downregulation of Pdx1 mRNA (Fig. 6c). MAFA protein 277 fluorescence intensity was also decreased ( Fig. 6a and b), supporting the RNA-seq analysis 278 showing that PDX1 and MAFA belong to the same regulatory network. Immunohistochemical 279 analyses showed no changes in the α-to β-cell ratio, indicating that β-cells were unlikely to 280 be de-differentiating toward an α-cell phenotype (Fig. 6d). B-IMMAT islets presented with 281 lowered insulin content (Fig. 6e), a tendency toward increased basal hormone levels ( Fig. 6f), 282 and absence of glucose-stimulated insulin release that could be restored using Exendin-4 ( Fig.  283 6f and g). Similar to overexpression experiments, glucose-and KCl-stimulated Ca 2+ fluxes 284 were impaired ( Fig. 6h-j), together with decreased expression of the VDCC subunits Cacna1d 285 and Cacnb2 (but not Cacna1c) (Fig. 6k). 286 Together, these experiments demonstrate that increasing the proportion of either immature or 287 mature β-cells results in a similar islet phenotype (i.e. perturbed insulin secretion, ionic fluxes 288 and β-cell population dynamics) (Fig. 6l). 289

Differences in β-cell maturity are encoded by the islet context 290
Since regulated Ca 2+ fluxes are critical for maintaining β-cell differentiation 37 , we wondered 291 whether immature and mature β-cells might help maintain their own phenotype in the islet 292 setting due to differences in their Ca 2+ signals (i.e. through a feedforward mechanism). To test 293 this, we repeated immunohistochemical analyses in dissociated β-cells where cell-cell 294 communications are disrupted, and Ca 2+ dynamics are less pronounced and more stochastic 295 21 . Unexpectedly, the PDX1 and MAFA intensity distributions were right-shifted in dissociated 296 islets, with β-cells in the PDX1 LOW and MAFA LOW range no longer apparent after 24 hr culture 297 ( Fig. 7a-c). A PDX1 LOW subpopulation could still be detected 3 hours after coverslip attachment 298 (Fig. 7d), and PDX1 frequency distribution was similar in scRNA or shGJD2-treated islets ( Fig.  299 7e and f). As such, β-cells likely undergo a gradual adjustment in maturity status following 300 dissociation rather than apoptosis/cell death, these changes occur independently of changes 301 in gap junction signalling (e.g. due to alterations in paracrine input), and should be considered 302 when extrapolating results from studies in dissociated β-cells. 303 To further investigate whether Ca 2+ dynamics might contribute to β-cell maturity directly in the 304 islet setting, we turned to a chemogenetic strategy to precisely control membrane potential. 305 Conditional β-cell silencing was achieved using Ins1Cre animals crossed to a strain harboring 306 stop-floxed alleles for hM4Di, a mutant muscarinic receptor with low affinity for endogenous 307 acetylcholine 40 . Upon administration of designer ligand, the Gi pathway is activated 308 specifically in β-cells, leading to long-lasting electrical silencing via effects on cAMP and G 309 protein-coupled inwardly-rectifying potassium channels 40,41 . We used this manoeuvre to 310 generate D-NORM and D-MAT islets, which possess wild-type (control) or hM4Di alleles, 311 respectively. 312 Specific expression of hM4Di in β-cells was confirmed via expression of a Citrine reporter (Fig.  313 7g). We first tested hM4Di functionality using the second-generation hM4Di agonist J60. As 314 expected, J60 silenced β-cell Ca 2+ spiking activity within 15 mins of application to D-MAT but 315 not D-NORM islets ( Fig. 7h and i) (Movies S3 and S4). No inhibitory effects of hM4Di-alone 316 were detected, with a small but significant increase in basal Ca 2+ levels detected in the 317 presence of the receptor (Fig. 7j). By contrast to J60, the first-generation agonist clozapine N-318 oxide (CNO) decreased Ca 2+ levels ( Fig. 7j) and Ca 2+ oscillation frequency ( Fig. 7k and l) after 319 3 hours, but did not completely suppress β-cell activity. We took advantage of this property to 320 disrupt rather than ablate the β-cell Ca 2+ signaling network. 321 was associated with a shift to more stochastic islet dynamics (Movie S5 and S6), as expected.

328
Gene expression analyses in D-MAT islets showed significant reductions in Cacna1d and 329 Gjd2, with Cacna1c, Cacnb2, Ins1, Ins2, Glp1r and Gck all remaining similar to D-NORM 330 controls (Fig. S7). The phenotype of D-MAT islets was unlikely to be dependent on insulin 331 signaling (or loss thereof), since application of then insulin receptor antagonist S961 to wild-332 type islets increased the proportion of PDX LOW rather than PDX1 HIGH β-cells ( Fig. S6d and e). Moreover, selection by cell death was unlikely to feature in D-MAT islets, since a reduction 334 (but not ablation) in Ca 2+ signaling would be expected to alleviate cell stress 42 . 335 These chemogenetic experiments suggest that either: 1) differences in β-cell maturity are 336 maintained via Ca 2+ signaling patterns encoded by the islet context; or 2) less mature β-cells 337 within the islet represent a ER-stressed or transitory subpopulation, which recovers its identity 338 when rested 42,43 (Fig. 7u). 339

Differences in β-cell maturity influence downstream gene expression 340
To define the transcriptional profile of islets in which immature β-cells are lost, we performed 341 differential gene expression analysis (DGE) on control and B-MAT mouse islets. To increase 342 β-cell maturity throughout the islet, we developed a doxycycline-inducible mouse model for 343 the cistronic expression of PDX1, MAFA and NEUROG3. This was generated by crossing 344 RIP7rtTA mice with those harbouring NEUROG3/PDX1/MAFA/mCherry under the control of 345 a tetracycline response element (Tet-MAT) (Fig. 8a). As expected, Tet-MAT islets displayed 346 increased expression of Pdx1, Mafa and Neurog3 in comparison to control islets (Tet-NORM) 347 (Fig. 8b). This was accompanied by loss of immature β-cells (PDX1 LOW ) ( Fig. 8c and d), as 348 well as impaired Ca 2+ fluxes ( Fig. 8e-g), without evidence of a generalized PDX1 349 overexpression (fluorescence intensity = 11044 ± 1837 AU versus 12679 ± 1813 AU, Tet-350 NORM versus Tet-MAT, respectively; non-significant), Thus, we were able to confirm results 351 in a third independent model, further demonstrating the robustness of the adenoviral 352 transduction model. 353 Doxycycline-treated islets from Tet-NORM and Tet-MAT mice were then subjected to 354 transcriptomic profiling using RNA-seq. Differential gene expression analysis (DGE) revealed 355 83 genes whose expression was significantly altered between Tet-NORM and Tet-MAT islets 356 (at adjusted p-value < 0.05) (Fig. 8h). The majority (94%) of these genes were upregulated in 357 Tet-MAT islets (Fig. 8h). Gene annotation analysis (DAVID) 44 revealed that significantly 358 upregulated genes were enriched for gene ontology clusters related to β-cell function and 359 identity ( Fig. 8i and j), confirming the validity of the model at the transcriptomic level. Gene set 360 enrichment analysis (GSEA) also revealed upregulation of other molecular pathways such as 361 metabolic processes linked to glucose and carbohydrate derivatives (Fig. 8k). Closer 362 inspection of the significantly upregulated genes revealed a number of candidates that might 363 impact insulin secretion including Ucn3, G6pc2, Cox6a2, Rgs4 and Pkib 45-48 , confirmed using 364 RT-qPCR (Fig. 8l). Taken together, these results show that increasing the proportion of mature 365 β-cells in the islet leads to upregulation of key β-cell identity markers, but also results in 366 differential regulation of pathways, such as those involved in cellular nutrient metabolism. 367

Restoring the balance between immature and mature β-cells is protective 368
A decrease in the expression of β-cell identity makers such as NKX6-1, PDX1 and MAFA 369 occurs during metabolic stress 49,50 . This may alter the balance between immature and mature 370 β-cells, with consequences for normal islet function. We therefore examined whether restoring 371 the balance between immature and mature β-cells would prevent islet failure in response to 372 lipotoxic insult. 373 Islets treated for 48 hours with high concentration of the fatty acid palmitate showed a left-shift 374 in the PDX1 fluorescence intensity distribution, primarily due to loss of PDX1 HIGH β-cells ( Fig.  375 9a). Transduction with Ad-M3C reversed this loss ( Fig. 9b and c), with PDX1 expression levels 376 being indistinguishable from BSA-controls. Functional assessment of palmitate-treated islets 377 revealed ~50% lowered Ca 2+ fluxes in response to both glucose and KCl ( Fig. 9d-f). 378 Pertinently, these deficits could be prevented using Ad-M3C ( Fig. 9d-f). From this, it can be 379 inferred that re-establishing the balance between PDX1 LOW and PDX1 HIGH β-cells, and thus 380 restoring differences in β-cell maturity, protects against islet failure during metabolic stress. 381 It is becoming increasingly apparent that β-cells can be grouped into subpopulations according 384 to their transcriptomic and protein signatures. In particular, the existence of immature β-cells 385 in the normal adult islet poses a conundrum, since this subpopulation is generally considered 386 to be poorly functional when viewed in isolation 4,7,8,14 . Despite this, no previous studies have 387 imposed changes on β-cell maturity while examining functional outcomes. Using multiple 388 models, we show here that differences in β-cell maturity are needed across the population for 389 proper islet function. An increase in the proportion of mature β-cells is associated with islet 390 failure due to impaired ionic fluxes, metabolism and cell-cell connectivity (schematic in Fig.  391 9g). Furthermore, redressing the balance between immature and mature β-cells restores islet 392 function under conditions of metabolic stress. Thus, our studies provide direct evidence that 393 both immature and mature β-cells are required for proper islets function and insulin release. 394 Islets with an increased proportion of mature β-cells displayed a large reduction in β-cell-β-395 cell connectivity. This was associated with a decreased number of hubs, immature and 396 energetic cells previously shown to coordinate glucose responsiveness 21 . Indeed, β-cells in 397 B-MAT islets responded more stochastically to glucose, closely resembling the responses 398 seen in islets from ob/ob or Cx36 -/animals 51-53 , as well as following silencing of hubs and their 399 associated cell clusters 21 , or uncoupling of β-cells following dissociation 21 . How might 400 immature β-cells affect β-cell-β-cell coordination so profoundly? We speculate that these cells 401 might be gap junction-coupled as a network within the islet, since mRNA for Cx36 decreased 402 ~50% following their loss, although we acknowledge that dual patch recordings of PDX1 LOW 403 cells would be needed to provide definitive evidence for this. Together with the tendency of 404 PDX1 LOW cells to mount higher amplitude Ca 2+ rises, such preferential communication could 405 allow a subset of β-cells to regulate excitability in neighboring β-cells, as shown by recent 406 modelling approaches 54 . Alternatively, increases in the proportion of mature β-cells might 407 perturb islet function by influencing gene expression or paracrine circuits such as those 408 mediated by somatostatin and GABA. Nonetheless, these results obtained using three 409 different models (viral, DREADD and doxycycline-inducible) confirm our previous optogenetic 410 findings on hub cells 21 , and suggest that a continuum of immature β-cells exists with shared 411 phenotypic and functional features. 412 While raw insulin secretion was unchanged in B-MAT versus B-NORM islets, the proportion 413 of total insulin secreted was reduced. This secretory defect is likely due to a combination of 414 factors reported here, including: 1) reduced glucose-stimulated metabolism (ATP/ADP); 2) 415 decreased Ca 2+ influx, which was refractory to generic depolarizing stimulus; 3) defective β-416 cell-β-cell coordination; and 4) impaired glucose-induced amplifying signals (cAMP), which 417 could not be restored with incretin mimetic or forskolin. Insulin granule density at the 418 membrane and exocytotic marker gene expression were both unchanged. 419 A feature of B-MAT islets was downregulated expression of genes encoding Ca 2+ channels. 420 Given that PDX1 and MAFA are required for β-cell Ca 2+ fluxes, what are the mechanisms 421 involved? One possibility is that Ca 2+ channel expression is higher in immature β-cells due to 422 a fine poise between transcription factor expression and regulation of downstream gene 423 targets ("Goldilocks effect" , and clamping this using overexpression approaches might constrain insulin release. 460 Thirdly, we cannot exclude that PDX1/MAFA LOW -> PDX1/MAFA HIGH cells become senescent 461 or apoptotic, although neither of these possibilities are supported by our transcriptomic 462 analyses. Also, we only looked at islets from 8-12 week-old animals and further studies are 463 required across lifespan, as well as in response to metabolic stressors, especially since 464 senescent β-cells possess transcriptomic signatures of immature cells 13 . Fourthly, NEUROG3 465 was mildly overexpressed, which could feasibly lead to a progenitor-like β-cell state. We think 466 that this is unlikely, as NEUROG3 protein was only weakly detectable, NEUROG3 exists in a 467 dephosphorylated form in the adult islet where it helps to maintain a differentiated state 59,60 , 468 and results were replicated in a chemogenetic model that does not possess NEUROG3 469 activity. In addition, the transcriptomic profile of B-MAT islets did not reveal enrichment for 470 progenitor signatures and classically-defined β-cell identity was apparently normal. 471 Lastly, we acknowledge a number of potential limitations with the overexpression system, 472 quantification and imaging approaches used here: 1) generalized transcription factor 473 overexpression, especially that involving NEUROG3, might lead to impaired islet function and 474 insulin secretion; 2) underestimation of overexpression skewed toward the highest 475 PDX1/MAFA signal intensity bins cannot be excluded (i.e. it might be more difficult to detect 476 overexpression in a cell that already has high levels); 3) the imaging approaches used here 477 could suffer from technical noise, decreasing our ability to accurately quantify PDX1 and 478 MAFA; and 4) exogenous PDX1 might possess different activity or affect different targets 479 compared to endogenous PDX1. In addition, we cannot exclude that intercellular feedback is 480 present whereby when more cells express PDX1 and MAFA, those expressing the highest 481 levels make less, as suggested by the QRT-PCR analyses of endogenous Pdx1 levels.

482
Further studies using novel surface markers and lineage labels, together with scRNA-seq or 483 spatial transcriptomics, will be needed to categorically confirm overexpression specifically in 484 immature β-cells in the intact tissue. 485 In summary, we have performed an in-depth functional interrogation of islets in which 486 proportionally more β-cells have been made mature in terms of PDX1 and MAFA expression 487 levels. These studies suggest that proper islet function is dependent on the co-existence of 488 immature and mature β-cells in the tissue context. Findings from single-cell screening studies 489 or studies in dissociated cells should thus be interpreted carefully in light of differences 490 imparted by the tissue context. Importantly, recreating these subtle differences in β-cell 491 maturity might be pre-requisite for engineering more robust islets from stem cells, as well as 492 preserving insulin release during diabetes and other states of metabolic stress. 493

Mouse models 495
Wild type CD1, Ins1Cre Thor  Male and female 6-12 week-old animals were maintained in a specific pathogen free facility, 518 with free access to food and water. Animal studies were regulated by the Animals (Scientific 519 Procedures) Act 1986 of the U.K., and approval was granted by the University of Birmingham's 520 Animal Welfare and Ethical Review Body. 521

Human donors 522
Human islets were obtained from Canada (Alberta Diabetes Institute, IsletCore) and Italy (San 523 Raffaele, Milan), with necessary local and national ethical permissions, including consent from 524 the next of kin. Studies with human tissue were approved by the University of Birmingham 525 Ethics Committee, as well as the National Research Ethics Committee (REC reference 526 16/NE/0107, Newcastle and North Tyneside, U.K.). Donor characteristics are reported in 527 Table S1. 528

Islet isolation 529
Mice were euthanized by cervical dislocation before inflation of the pancreas via injection of 530 collagenase solution (1mg/ml; Serva NB8) into the bile duct. Pancreata were then digested for 531 12 mins at 37°C in a water bath before purification of islets using a Histopaque or Ficoll 532 gradient. Islets were hand-picked and cultured (5% CO2, 37°C) in RPMI medium containing 533 10% FCS, 100 units/mL penicillin, and 100 μg/mL streptomycin. 534

21
. For primer and probe details, see Table S2. cross-reactivity between antibodies from the same species, sequential staining and re-571 blocking was performed. Samples were mounted on coverslips containing VECTASHIELD 572 HardSet with DAPI (Vector Laboratories Cat# H-1500 performed using a custom routine in ImageJ. Briefly, Gaussian filtered images were subjected 582 to an auto-threshold and binarization step to create a mask, which was then used to identify 583 mean pixel intensity in each PDX1 + or MAFA + cell before construction of a frequency 584 distribution. Glucokinase, insulin, glucagon and somatostatin were quantified using corrected 585 total cell fluorescence (CTCF), according to the following equation: CTCF = integrated density 586 -(area of ROI x mean fluorescence of background). Images were de-noised using a Gaussian 587 smoothing procedure, and linear adjustments to brightness and contrast were made for 588 presentation purposes. 589

Western blotting 613
Samples were collected in urea Laemmli sample buffer (0.2M Tris-HCl, pH 6.8, 40% glycerol, 614 8% SDS, 5% B-ME, 6M Urea, 0.005% Bromophenol Blue) and sonicated (2 x 5 seconds at In all cases, values are normalized against total insulin for each individual experiment to 638 account for differences in βcell proportion with treatment and islet size. 639

Chemogenetics 640
The h4MDi ligands JHU37160 (J60) (Hello Bio Cat# HB6261) and clozapine N-oxide (CNO) 641 (Tocris Cat# 4936/10) were applied to islets at 1 µM for the indicated time points. While P450 642 converts CNO into clozapine, which promiscuously binds endogenous receptors in vivo 67 , this 643 is not expected to be an issue in vitro. In any case, CNO was present under all conditions 644 examined to account for off-target effects. For assessment of intraislet insulin signaling, control 645 islets were treated with 50 nM insulin receptor antagonist S961 (Phoenix Pharmaceuticals, 646 Cat# 051-56) for 48h. 647

Correlation and wavelet analyses 658
Detection of superconnected islet regions was performed using matrix binarization analyses 659 developed in-house, as previously described 68 . Briefly, cells were identified using a region of 660 interest (ROI), intensity over time traces extracted, subjected to Hilbert-Huang empirical mode decomposition to remove noise and baseline trends, and a 20% threshold imposed to binarize 662 cells according to activity status. Co-activity between all cell pair combinations was assessed 663 using the equation Cij = Tij/√TiTj where C is a correlation coefficient, Ti and Tj is the period 664 spent ON for each cell, and Tij is the period both cells spend ON together. Significance was 665 calculated versus the randomized dataset for each cell pair using a permutation step for each 666 binarized data row. This analysis allows identification of cells whose activity repetitively spans 667 that of the rest of the population. Superconnected cells or hubs were defined as cells 668 possessing 60-100% of the correlated links and plotted on functional connectivity maps using 669 the Euclidean coordinates. 670 Wavelet analysis was used to determine the time-localized Ca 2+ oscillation frequency. Spectra 671 were extracted from Ca 2+ traces with a univariate bias-corrected wavelet transform ("biwavelet" 672 package in R), which prevents compression of power as period lengthens. Period was then 673 depicted against time, with a color ramp representing frequency power. 674

Differential gene expression analyses 675
Differential gene expression was obtained using DEseq2 with age-and sex-matched paired 676 Tet-NORM (n = 5) and Tet-MAT samples (n = 5). Differentially expressed genes between 677 control and Tet-MAT islets at adjusted p-value <0.05 were annotated using DAVID BP_FAT 678 44 , with high stringency for clustering. 679 Gene set enrichment analysis (GSEA) was used to interrogate specific gene sets against 680 expression data. GSEA calculates an Enrichment Score (ES) by scanning a ranked-ordered 681 list of genes (according to significance of differential expression (-log10 p-value)), increasing 682 a running-sum statistic when a gene is in the gene set and decreasing it when it is not. The 683 top of this list (red) contains genes upregulated in Tet-MAT islets while the bottom of the list 684 (blue) represents downregulated genes. Each time a gene from the interrogated gene set is 685 found along the list, a vertical black bar is plotted ("hit"). If the "hits" accumulate at the bottom 686 of the list, then this gene set is enriched in downregulated genes (and vice versa). If 687 interrogated genes are distributed homogenously across the rank-ordered list of genes, then 688 that gene set is not enriched in any of the gene expression profiles. We converted human 689 gene sets into homologous mouse gene sets using homologous gene database from MGI. 690

Statistical analyses 691
All analyses were conducted using GraphPad Prism, Igor Pro, R Project or MATLAB software. 692 Unpaired or paired Student's t-test was used for pairwise comparisons. Multiple interactions 693 were determined using normal or repeated measures ANOVA followed by Bonferroni, Sidak 694 or Tukey posthoc testing (accounting for degrees of freedom). Straight lines were fitted with 695 linear regression whilst a polynomial trend was used for multiple regression. Goodness of fit 696 was calculated using R 2 . 697

Data availability 698
The datasets generated during and/or analysed during the current study are available from 699 the corresponding author on reasonable request. 700 Raw read files and processed data files for RNA-seq can be found at the NCBI Gene 701 Expression Omnibus (GEO) database (GSE133798