When DOX is administrated to mice via drinking water, the total DOX intake is in principle directly proportional to the DOX concentration and average daily water consumption before treatment. However, the drinking behavior of mice may vary during the indicated treatment period because DOX gradually induced hyperglycemia. To test this, we measured the mean accumulated water intake (Figure 1A) and dosage consumed (Figure 1B) for 1 week following treatment of 50 and 200 µg/mL DOX, respectively. We found that the mean volume intake and dosage consumed were higher in the DOX-treated groups (both 50 and 200 µg/mL) than the estimated rate (baseline) since day 5. We also found that the mean volume intake and dosage consumed increased more rapidly in the 200 than 50 µg/mL group compared to their respective baseline, indicating that polydipsia was more severe in the 200 than 50 µg/mL group. Our results suggest that polydipsia developed during the course of treatment possibly increased the daily water consumption leading to an increased DOX intake.

Although polydipsia caused a positive feedback in water and, therefore, DOX intake, there was still a strong positive correlation (r = 0.8912, p = 0.0005) between the increased fed glucose level and the accumulated DOX intake at any time point for both groups (Figure 1C). Moreover, the 200 µg/mL group also demonstrated a significantly higher mean fed glucose level than the 50 µg/mL group at the indicated time points even at 7 days post DOX withdrawal (Figure 1D). We also performed glucose tolerance test (GTT) at day 7 post DOX withdrawal to confirm a significantly more impaired glucose tolerance in the 200 than 50 µg/mL group (Figure 1E). Altogether, our results show that polydipsia developed during the course of treatment led to discrepancy between actual and expected DOX intake based on stable daily water consumption; but the degree of impaired β cell function was still correlated to the total DOX intake in a dose-dependent manner.

Since adult β cells regenerate by replication from preexisting β cells (17, 21), severe β cell death could delay or inhibit β cell regeneration after injury. To determine the optimal dose of DOX treatment in long-term β cell regeneration studies, DOX was administrated in drinking water at 50, 200, 500, 1,000, and 2,000 µg/mL, respectively, for 7 days to induce β cell lesion in βDTA mice. At D7 posttreatment, immunostaining for INSULIN was performed (Figure 2A), and we found that the β cell mass was significantly reduced by at least 80% in all doses compared to the untreated control (Figure 2B). Although the size of individual pancreatic islets appeared to be smaller in the higher dosage groups (200–2,000 µg/mL, Figure 2A), there were more numbers of small islets in the higher dosage groups, so the overall β cell mass did not show significant difference in the range of 50–500 µg/mL (Figure 2B).

Interestingly, only the 50 µg/mL group showed a significantly lower fed glucose level than the higher dosage groups; whereas the higher dosage groups showed similar high fed glucose levels at D7 posttreatment (Figure 2C). A similar trend was also observed in long-term monitoring of fed glucose level, in which the 50 µg/mL group demonstrated recovery at D21 posttreatment with fed glucose remained stable afterward; whereas the higher dosage groups showed no decline in fed glucose levels (Figure 2D). Our GTT data also confirmed that the 50 µg/mL group demonstrated a continued improvement in glucose tolerance from D28 to D56 posttreatment (Figure 2E); whereas the higher dosage groups revealed sustained impaired glucose tolerance at D63 posttreatment (Figure 2F). In fact, the glycemic control in the higher dosage groups was so impaired that even the fasting glucose levels remained elevated for 63 days (Figure S1 in Supplementary Material). Therefore, a treatment of 50 µg/mL DOX for 7 days was sufficient to induce significant β cell loss yet retain their regenerative capability. Our results also indicated that well-preserved β cell morphology was essential for β cell regeneration.

Having optimized the DOX treatment protocol, we then tested if transient expression of VEGF_A by local injection of modRNA could facilitate β cell regeneration with functional recovery. Schematic diagram of experiments was illustrated in Figure 3A. We first confirmed efficient translation of modRNA by transfecting 1 µg eGFP modRNA into hESCs in vitro (Figure S1 in Supplementary Material); and then confirmed hVEGF_A protein secretion by transfecting 1 µg control eGFP or hVEGF_A modRNA into hESCs in vitro. We collected supernatant 24−48 h post-transfection, and the secreted VEGF_A protein was examined by Western blot (Figure 3B). We then performed in vivo transfection in which βDTA mice were treated with 50 µg/mL DOX in drinking water for 7 days followed by intrapancreatic injection of 100 µg control luciferase (Luc) or hVEGF_A modRNA. Hyperglycemia was confirmed with the fed glucose level ≥15 mM or 270 mg/dL 7 days after DOX withdrawal. Mice were then assigned into the Luc or hVEGF_A group with each group had a similar mean fed glucose levels. Seven days after modRNA injection, we observed increased vascularization in the pancreas (Figure 3C) with expression levels of both the human and mouse VEGF_A mRNA significantly increased, respectively, in the hVEGF_A modRNA-treated group compared to that of the Luc control group (Figure 3D). Moreover, immunostaining for CD31 also showed that there was increased pancreatic vascularization in the hVEGF_A modRNA group 56 days post-transfection, and CD31⁺ blood vessels were found in close proximity to the pancreatic islets (Figure 3E). Our results showed that intrapancreatic injection of hVEGF_A modRNA induced transient expression of the hVEGF_A protein that promoted angiogenesis in the pancreas.

To examine the morphology of pancreatic islets at day 56 post-transfection, we performed immunostaining for INSULIN (INS) and GLUCAGON (GCG) and found that the islets of the hVEGF_A modRNA-treated group restored their typical morphology with the INS⁺ β cells surrounded by the GCG⁺ α cells in the periphery (Figure 4A). Moreover, the hVEGF_A modRNA-treated group had increased number of islets (Figure 4A), improved β cell mass (Figure 4B), and increased insulin gene expression as determined by qPCR (Figure 4C) compared to that of the Luc modRNA-treated group. Furthermore, we performed GTT at days 14 and 56 post-transfection (Figure 4D), our results showed that glucose tolerance of the hVEGF_A-treated group was significantly improved from D14 to D56; whereas the control group did not show any improvement during the same period (Figures 4D,E).

Despite treatment with a single dose of hVEGF_A modRNA improved glucose tolerance two months after β cell death (Figure 4E), there was a large variation in the recovery of β cell mass among individuals within the same group (Figure 4B). Since we found a strong positive correlation between the change in fed blood glucose level and accumulated DOX intake (Figure 1C), we hypothesized that polydipsia developed during the course of DOX treatment amplified the initial variation in daily water consumption and, therefore, the regenerative capacity among individuals within the same group. In fact, a large variation in the fasting blood glucose levels was also observed in βDTA mice treated with higher doses of DOX (200–1,000 µg/mL). Such a variation also increased with time after DOX withdrawal, suggesting a differential β cell regenerative capacity among individuals treated with the same dose of DOX (Figures S2B–D in Supplementary Material).

To reveal the influence of drinking habit on individual’s β cell regenerative capacity, we monitored the long-term β cell recovery for two groups of mice with subtle difference in their mean daily water consumed before treatment of 50 µg/mL DOX: the mean daily water consumption for these two groups were 3 mL (group/G1, n = 5) and 5 mL (group/G2, n = 5) per mouse, respectively; and both were within the normal physiological range. During DOX treatment, polydipsia developed in individual mouse at a different speed according to individual’s initial daily water consumption, so the difference in accumulated DOX intake between the two groups was dramatically amplified (as illustrated by the dashed and solid lines representing expected and actual intake, respectively, in Figure 5A). At the end of 7-day treatment, the mean accumulated DOX intake for individuals of G1 was 1.3 mg and of G2 was 3.6 mg, respectively. Furthermore, despite the two groups showed very similar fed (Figure 5B, day −7) and fasting (Figure 5C, day −7) blood glucose levels before treatment, a significant difference in their fed glucose level was observed since day 14 after DOX withdrawal (Figure 5B). Therefore, a slight difference in the drinking habit among individuals treated with the same concentration of DOX water led to significant difference in the accumulated DOX intake that was evidenced by neither the fed (Figure 5B, day 0) nor fasting (Figure 5C, day 0) blood glucose level at the time when DOX was withdrawn.

In fact, we also observed a significant difference in β cell functional recovery with respect to accumulated DOX intake even within the range of 1.3−3.6 mg per mouse. β Cell function had been recovering in G1 mice as evidenced by the continued decrease in both fed (Figure 5B) and fasting (Figure 5C) blood glucose levels with time; and the reduced individual variation in both fed and fasting blood glucose levels with time. Moreover, the G1 group also demonstrated a significantly improved glucose tolerance by comparing results of GTT at D56 and D14 post DOX treatment (Figure 5D). Immunostaining for INS and GCG also showed that the islet morphology of the G1 was better than the G2 group at D56 post DOX treatment (Figure 5E). There were also more INS⁺ β cells (Figure 5E), significantly greater β cell mass (Figure 5F) and higher Insulin gene expression (Figure 5G) in the G1 than G2 group. Together, our results showed that long-term recovery of β cell mass and glucose tolerance was dependent on the accumulated DOX intake during the initial treatment phase; and individual variation in the rate of β cell regeneration was significantly influenced by one’s drinking habit even if the same concentration of DOX was given to the group in the same bottle of drinking water.

Because hVEGF_A modRNA demonstrated some improvement in glucose tolerance following DOX-induced β cell death; yet there was large variation in β cell mass among individuals within the same group (Figure 4), and we found that the difference in accumulated DOX intake also contributed to variation in spontaneous β cell recovery (Figure 5), we hypothesized that the effect of hVEGF_A modRNA treatment might also be dependent on the accumulated DOX intake. To address this, two cohorts with slightly different drinking habit were treated with 50 µg/mL DOX in drinking water for 7 days, and the mean accumulated DOX intake by each mouse at the end of treatment was 1.5 and 3.0 mg, respectively. In each cohort, mice were assigned to balanced control (Luc modRNA) and treatment (hVEGF_A modRNA) groups based on comparable fed blood glucose levels at day 0. Experiments were conducted as outlined in Figure 3A.

After injection of 100 µg modRNA, the fed (Figure 6A) and fasting (Figure 6B) blood glucose level of the hVEGF_A modRNA-treated group decreased more rapidly than that of the Luc group in the cohort received 1.5 mg DOX accumulatively from day −7 to 0. However, a reverse trend was observed in the cohort received 3.0 mg DOX (Figures 6A,B). It is of noted that the DOX dose 1.5 mg did not result in impaired fasting but fed blood glucose level throughout the experiment; it was likely that less insulin is needed to maintain fasting than fed blood glucose levels. Furthermore, our GTT results also demonstrated that the hVEGF_A modRNA treatment improved glucose tolerance in the cohort received 1.5 mg DOX (Figures 6C,D); while such effect was not observed in the cohort received 3.0 mg DOX (Figures 6C,D). Therefore, the benefit of hVEGF_A modRNA in β cell regeneration was observed only in individuals with less severe β cell loss as evidenced by better improved glucose tolerance in βDTA mice received lower (1.5 mg) but not higher (3.0 mg) amount of DOX.

The βDTA mice [Tg(Ins2-rtTA)2Efr Tg(teto-DTA)1Gfi/J] were purchased from The Jackson Laboratory and maintained in the Laboratory Animal Center of The Chinese University of Hong Kong (CUHK). All experimental procedures were approved by the Animal Experimentation Ethics Committee of CUHK and performed in compliance with “Guide for the Care and Use of Laboratory Animals” (8th edition, 2011) established by National Institutes of Health. Male mice of 8−12 weeks were used for experiments, and there was no hyperglycemia, glucose intolerance, or polydipsia observed before DOX induction. To induce β cell death, three to five mice were housed per cage with free access to drinking water contained 50−2,000 μg/mL DOX (Sigma, D9891) for 7 days. Water bottles were protected from light, and DOX water was frequently checked for replacement during treatment. One week before DOX treatment, the average daily water consumption per cage over 7 days was measured by marking the calibration on water bottles, and the baseline mean daily water consumption per mouse in a cage was estimated based on the number of mice housed. During DOX treatment, the mean accumulative DOX intake per mouse within a period was estimated as (total water consumption per cage × doxycycline concentration)/number of mice housed.

Blood samples were collected from tail vein and measured by a commercial glucometer (Contour^® TS, Bayer). The fed blood glucose levels were measured regularly at 9:00−12:00; and the fasting blood glucose levels were measured after an overnight fasting for 14−16 h. For GTT, mice were injected intraperitoneally (i.p.) with 2 g glucose per kilogram body weight after taking the fasting blood glucose levels; and the blood glucose levels were monitored at 15, 30, 60, and 120 min after glucose injection. The blood glucose readings were then plotted against time after glucose injection, and the area under curve (AUC) was calculated using GraphPad Prism 5.0. The baseline adjusted AUC was used sometimes if the fasting blood glucose levels between groups were significantly different.

Chemically modRNA for eGFP, Luc and hVEGF_A were synthesized as previously described (15, 16, 31). After DOX withdrawal, 100 µg hVEGF_A or Luc modRNA was injected into the mouse pancreatic tail directly in 60 µL mixture containing 10 µL modRNA, 10 µL Opti-MEM I (Thermo Fischer, 31985), and 40 µL Lipofectamine RNAiMAX (Thermo Fischer, 56532), using a 29G insulin syringe (BD, 320431) as previously described (32).

The human embryonic stem cell (ESC) line was maintained as previously described (15). 1 µg hVEGF_A or eGFP modRNA was transfected per well of cells cultured on 6-well plates using Lipofectamine RNAiMAX as previously described (15, 31). Transfection medium was replaced by conventional culture medium 6 h after transfection. During 24–48 h post-transfection, cells were cultured in fresh serum free medium. The culture supernatant was collected afterward, and proteins were precipitated by the trichloroacetic acid method. The presence of hVEGF_A protein was confirmed by Western blot (mouse anti-hVEGF_A antibody, 1/100, BD Catalog no. 554359). The membrane was lightly stained by Ponceau S as loading control before blotting with antibody.

The mouse pancreata were dissociated in TRIzol reagent, and total RNA was isolated according to manufacturer’s instruction (Thermo Fischer, 15596018). cDNA was synthesized using iScript cDNA synthesis kit (BioRad, 170-8891), and qPCR was performed using iTaq universal SYBR Green supermix (BioRad, 172-5120) on QuantStudio 12K Flex Real-Time PCR system (Thermo Fischer). The relative gene expression level of each sample was calculated by 2^−ΔΔCT method using mouse β-actin or human GAPDH as internal control. hESC cDNA and pancreatic cDNA of age-matched male βDTA mice (untreated) or mouse ESC cDNA were used as calibrator to normalize gene expression for hESC samples and mouse pancreatic samples, respectively. Primers used for mouse genes: Vegfa (5′-GCTTCCTACAGCACAGCAGA-3′; 5′-AATGCTTTCTCCGCTCTGAA-3′), Ins2 (5′-GGAGCGTGGCTTCTTCTACA-3′; 5′-CAGTGCCAAGGTCTGAAGGT-3′), Pdx1 (5′-GATGAAATCCACCAAAGCTCACGC-3′; 5′-AATTCCTTCTCCAGCTCCAGCAGC-3′), and β-actin (5′-TTTGCAGCTCCTTCGTTGCCG-3′; 5′-TTTGCACATGCCGGAGCCGTT-3′). For human genes: VEGFA (5′-AAGGAGGAGGGCAGAATCAT-3′; 5′-CCAGGCCCTCGTCATTG-3′) and GAPDH (5′-TGTTGCCATCAATGACCCCTT-3′; 5′-CTCCACGACGTACTCAGCG-3′).

The wet weights of mouse pancreata were taken immediately after dissection. Tissues were fixed in 4% PFA at 4°C overnight, washed three times by PBS, and infiltrated by 30% sucrose at 4°C. The pancreata were frozen in OCT blocks and sectioned at 8 µm thickness. Discontinued sections with 15 sections apart were mounted onto a slide. For quantification of each marker, 6 discontinued sections were stained and sampled per mouse pancreas. Immunofluorescence was performed using the following primary antibodies: guinea pig anti-insulin (1/400, Dako, A0564), rabbit anti-glucagon (1/200, Dako, A0565), and rat anti-CD31 (1/100, BioLegend, 102501). AlexaFluor-488 or 546 linked goat secondary antibodies were used (1/500, Thermo Fischer), and nuclei were stained by Hoechst 33342 (1 µg/mL, Thermo Fischer, H3570) before mounted with Aqua-Poly/Mount (Polysciences, 18606). For image acquisition, each entire pancreas section was scanned, and images were automatically stitched using Leica fluorescence microscope with LAS AF software (Leica DM4000, Germany). Quantitative image analysis was performed using Image J (NIH). The β cell mass was calculated as followed (13, 17): pancreas wet weight × (the sum of INSULIN⁺ stain area of 6 discontinued sections)/(the sum of total nucleated area of 6 discontinued sections) × 100%. For comparison, the β cell mass of each mouse in the experimental groups was normalized to the mean value of β cell mass of age-matched male βDTA mice (without any treatment) and presented in percentage.

Data analysis was conducted using GraphPad Prism 5.0. To compare two independent groups, independent t-test was performed, and one-way ANOVA followed by Bonferroni’s or Dunnett’s multiple comparison was used for more than two independent groups. Paired t-test was used to compare the same group at the indicated time points for longitudinal study. Pearson correlation coefficient r was calculated to measure correlation. p Values were specified for each test, and a p value < 0.05 was considered statistically significant.