Ultrasound imaging of pancreatic perfusion dynamics predicts therapeutic prevention of diabetes in preclinical models of type1 diabetes

Abstract

In type1 diabetes (T1D) immune-cell infiltration into islets of Langerhans (insulitis) and β-cell decline occurs years before diabetes presents. There is a lack of validated clinical approaches for detecting insulitis and β-cell decline, to diagnose eventual diabetes and monitor the efficacy of therapeutic interventions. We previously demonstrated contrast-enhanced ultrasound measurements of pancreas perfusion dynamics predicted disease progression in T1D pre-clinical models. Here, we test whether these measurements predict therapeutic prevention of T1D. We performed destruction-reperfusion measurements with size-isolated microbubbles in non-obese diabetic (NOD)-scid mice receiving an adoptive transfer of diabetogenic splenocytes. Mice received vehicle control, or the following treatments: 1) antiCD3 to block T-cell activation, 2) antiCD4 to deplete CD4+ T-cells; 3) Verapamil to reduce β-cell apoptosis or 4) TUDCA to reduce β-cell ER stress. We compared measurements of pancreas perfusion dynamics with subsequent progression to diabetes. AntiCD3, AntiCD4, and verapamil delayed diabetes development. Blood-flow dynamics were significantly altered in treated mice with delayed/absent diabetes development compared to untreated mice. Conversely, blood-flow dynamics in treated mice with unchanged diabetes development was similar to untreated mice. Thus, measuring pancreas perfusion dynamics predicted the successful prevention of diabetes. This strategy may provide a clinically-deployable predictive marker for therapeutic prevention in asymptomatic T1D.

Introduction:

During type1 diabetes (T1D) progression immune infiltration into the islet of Langerhans and β-cell decline occur over many years prior to diabetes onset. This asymptomatic phase presents a window in which preventative therapeutic intervention can be made, while significant β-cell mass remains (Insel et al. 2015). Indeed, recent clinical trials have demonstrated successful delay of diabetes when applied during this asymptomatic phase (Herold et al. 2019). However, responses to therapeutic treatments are often heterogeneous, with only a subset of subjects showing a significant delay in β-cell decline (responders) (Herold et al. 2013). Currently there are no clinically applied indicators to assess whether the trajectory of disease progression is reversed following therapeutic treatments. While there are some clinically applied indicators of asymptomatic T1D progression (Ziegler et al. 2013; Bonifacio 2015; Sosenko et al. 2012), it is unclear if these indicators would be useful for assessing therapeutic efficacy: for example islet-associated autoantibodies are not pathogenic. Thus, new methods to track T1D progression are required.

Ultrasound imaging is a cost-effective, readily deployable, and safe imaging modality. Contrast-enhanced ultrasound (CEUS) utilizes microbubbles (MBs) that consist of a lipid- or protein shell and gas core. As a result of the strong acoustic backscatter of ultrasonic waves and with non-linear signal detection, the MB signal from the vasculature can be well separated from that of the tissue. Destruction-replenishment imaging in combination with CEUS allows for measurements of blood flow perfusion in a tissue. Following MB infusion, a high mechanical index pulse destroys the MBs within a field of view, such as an organ or tissue. The wash in of MBs from elsewhere in circulation can be measured and reflect the kinetics of tissue perfusion (Rim et al. 2001). This approach has been applied previously in animal models and human subjects to measure and quantify blood flow dynamics in the kidneys (Kalantarinia et al. 2009), heart (Bekeredjian, Grayburn, and Shohet 2005), brain (Wiesmann et al. 2004), and the pancreas during pancreatitis (Ripollés et al. 2010) and pancreatic tumors (Dietrich, Braden, et al. 2008).

We previously used contrast-enhanced ultrasound to measure pancreas perfusion dynamics via MB destruction-replenishment imaging. These measurements of pancreas perfusion dynamics could reveal changes in islet blood flow dynamics that result from islet microvascular remodeling during T1D progression (Papaccio 1993; Canzano et al. 2018; Akirav et al. 2011). In preclinical mouse models of T1D, the rate of recovery following microbubble destruction increased with diabetes progression and the amplitude of recovery decreased with diabetes progression. We further demonstrated that each of these measurements could predict the speed of diabetes progression, and could distinguish successful and unsuccessful responses to interventions deigned to prevent T1D (St Clair et al. 2018). In demonstrating that this approach could distinguish subjects in which the underlying T1D progression was halted, prior to diabetes onset, we utilized antiCD4-mediated T-cell depletion as a proof-of-principle intervention (Nicolls et al. 2002). It is unclear whether the response to viable therapies for T1D prevention can also be distinguished by this approach. Furthermore, it is unclear whether this approach allows for predicting the progression to diabetes following therapeutic intervention.

Here we determine the degree to which contrast-enhanced ultrasound measurements of pancreas blood flow dynamics can be used to predict the success of therapies currently being applied clinically for T1D prevention or reversal. This included predicting the action of both immunotherapies and therapies directed against the β-cell. We first confirmed results from our previous study in predicting both T1D progression and the action of antiCD4 proof-of-principle treatment. We then tested whether we could predict the success of antiCD3 (Herold et al. 2002; Herold et al. 1992), verapamil (Ovalle et al. 2018; Xu et al. 2012a) and TUDCA (Engin et al. 2013), using an adoptive-transfer model of T1D (Kupfer et al. 2005).

Materials and Methods:

Ethics Statement-

The study was carried out in compliance with the ARRIVE guidelines. All experiments were performed in compliance with the relevant laws and institutional guidelines, and were approved by the University of Colorado Institutional Biosafety Committee (IBC) and Institutional Animal Care and Use Committee (IACUC, B-95817(05)1D).

Animals-

Female NOD mice were purchased from Jackson Laboratories (Bar Harbor, ME) at age 4 weeks. Female NOD-SCID animals were purchased from Jackson Laboratories at age 10-14 weeks. Throughout the study, animals were monitored weekly for blood glucose concentration utilizing a Contour blood glucose meter (Bayer, Leverkusen, Germany).

Isolation and Adoptive Transfer of Diabetogenic Splenocytes-

Splenocytes were isolated from diabetic female NOD mice (hyperglycemic <1 week), manually dissociated and counted in cold HBSS (without MgCl₂ and CaCl₂). Leukocytes were counted to determine an estimate of cellular density. NOD-SCID mice received a single intraperitoneal (I.P.) dose of 20 x 10⁶ leukocytes resuspended in HBSS. Control animals were injected with an equivalent volume of HBSS without leukocytes.

Contrast-Enhanced Ultrasound (CEUS) imaging-

General anesthesia was established with isoflurane inhalation for a total of 20-25 minutes for all animal imaged. Prior to imaging, a custom made 27G ½” winged infusion set (Terumo BCT, Lakewood, CO.) was attached to a section of polyethylene tubing (0.61 OD x 0.28 ID; PE-10, Warner Instruments) and was inserted in the lateral tail vein and secured with VetBond (3M, Maplewood, MN.). Abdominal fur was removed using depilatory cream, and ultrasound coupling gel placed between the skin and transducer. Foot pad electrodes on the ultrasound machine platform monitored the animal’s electrocardiogram, respiration rate, and body temperature. All animals were constantly monitored throughout the imaging session to maintain body temperature and respiration rate.

A VEVO 2100 small animal high-frequency ultrasound machine (Fujifilm VisualSonics, Toronto, Canada) was used for all experiments. For CEUS imaging a MS250 linear array transducer was used at a frequency of 18 MHz. B-mode imaging (transmit power 100%) was performed prior to microbubble (MB) infusion to identify anatomy of the pancreas body, based on striated texture and location in relation to the spleen, kidney, and stomach (St Clair et al. 2018). Following identification of the pancreas and selection of a region of interest, sub-harmonic contrast mode was initiated. Acquisition settings were set at: transmit power 10%, (MI=0.12), frequency 18 MHz, standard beamwidth, contrast gain of 30 dB, 2D gain of 18 dB, with an acquisition rate of 26 frames per second.

Size-isolated microbubble contrast agent with diameter of 3-4μm (‘SIMB3-4’, Advanced Microbubble Laboratories, Boulder CO.) was injected as a single bolus of ~10 million bubbles in phosphate buffered saline (pH 7.4) into the lateral tail vein via the catheter. SIMBs were allowed to circulate throughout the animal for ~20 seconds to reach a relative steady state of systemic distribution. SIMB destruction was initiated by delivery of a high mechanical index pulse (VEVO2100 burst mode, MI=0.2), to destroy a portion of SIMBs within the imaging plane (Blomley et al. 2001; Ferrara, Pollard, and Borden 2007). Data were acquired for at least 10 seconds following SIMB destruction to adequately measure reperfusion into the tissue.

Each experimental group of animals studied contained control mice, untreated adoptive transfer (AT) mice, and treated-AT mice, which received scans in a random order. Experimental numbers are indicated in relevant figure captions (including mice that were excluded from analysis). The experimental unit of analysis was a mouse.

Data Analysis-

Gating to remove movements as a result of animal breathing was carried out manually or using custom MATLAB scripts (MathWorks, Natick, MA). A region of interest was defined over each pancreas tail region based upon B-mode image, using kidney, spleen and stomach as guide markers, as well by the distinct texture (Ramirez et al. 2020; St Clair et al. 2018). A time-course was generated over each region of interest, excluding pixels in which the contrast was saturated. For analysis of reperfusion kinetics, the background non-linear intensity taken before SIMB infusion was subtracted from the entire time-course. To factor out variability in SIMB infusion and signal attenuation due to depth, each reperfusion time-course was normalized to a 0.5s average of the steady state NL contrast intensity immediately prior to flash destruction. The resultant normalized reperfusion curves were fit in MATLAB using the nonlinear regression function (fitnlm) according to an exponential rise equation, F(t) = C + A(1-e^−kt), where C is the offset from zero, A is the amplitude of the curve and k is the rate of reperfusion. CEUS measurements were excluded from analysis if a poor infusion was recorded (no contrast elevation following SIMB infusion) and/or poor microbubble flash-destruction occurred (and thus poorly defined recovery). Removal was based on the quality of fit, defined as the Coefficient of determination (r²), being less than 0.5. r² was determined from the fitnlm NonLinearModel object output.

Therapeutic drug application-

The dosing regimen for each treatment was based on prior studies. For antiCD4 treatment, NOD-Scid animals two weeks following splenocyte delivery were injected IP with a single dose of 20 mg anti-mouse CD4, clone GK1.5 (BP0003-1; BioXCell, W. Lebanon, NH.) (St Clair et al. 2018). For antiCD3 treatment, NOD-Scid animals two weeks following splenocyte delivery were injected IP with 5 daily doses of 50μg anti-mouse CD3, clone 145-2C11 (BE0001-1FAB; BioXCell) (Herold et al. 1992). For verapamil treatment, NOD-Scid animals two weeks following splenocyte delivery received verapamil (Thermo Fisher Scientific, Waltham, MA.) in the drinking water (1 mg/ml), available continuously ad lib (Xu et al. 2012b). For TUDCA treatment, NOD-Scid animals two weeks following splenocyte delivery were IP injected daily with 300mg/kg body weight TUDCA (Millipore Sigma, Burlington, MA.) dissolved in sterile PBS for 14 days (Engin et al. 2013). AT mice were randomly assigned to treatment or untreated groups.

Statistical Analysis-

All data are presented as means ± SEM. Statistical comparisons were made using paired or unpaired student’s t-tests or ANOVAs where appropriate and as indicated. Outliers, as defined by Grubbs' test, were excluded for the purpose of statistical analysis but retained in figure panels (as noted in figure caption). Statistical significance was taken as p<0.05.

Treatment responders were defined as treated AT mice that developed diabetes after 6 weeks post splenocyte delivery (4 weeks after drug treatment). This time point corresponds to a point at which zero untreated AT mice remain diabetes free. Treatment non-responders were defined as treated AT mice that developed diabetes before or at 6 weeks post splenocyte delivery. Diabetes was defined as showing an adlib blood glucose >250mg/dl, with the measurement time classed as the time point at which diabetes developed. This was validated by a second blood glucose measurement 1 week later.

To test for disease prediction, Receive Operating characteristics (ROC) curves were first generated using rate and amplitude data collected at 4 weeks post splenocyte transfer. For a given rate or amplitude threshold, the sensitivity was defined as the % of untreated AT mice within this threshold and 1-specificity was defined as % of vehicle control mice within this threshold. This process was performed for all Control and untreated AT mice. Maximum likelihood analysis was employed to define an optimum disease prediction threshold for the rate and amplitude, which was used throughout the study. Subsequently, for each group of treated AT mice, this disease prediction threshold was employed to separate treated AT mice at 4 weeks post-transfer into predicted disease positive and predicted disease negative groups, which were subsequently compared. As such the disease prediction threshold is generated independent to the analysis of treated AT mice.

Sample sizes for experimental groups were based on our prior study (St Clair et al. 2018), to provide sufficient statistical power given the measured effect size (reperfusion parameters). When comparing experimental groups, CEUS recordings were not made in a defined order.

Results:

Measuring changes to pancreas blood flow dynamics-

We previously determined that measuring pancreas perfusion dynamics could indicate underlying diabetes progression and predict the progression to diabetes in pre-clinical mouse models of T1D (St Clair et al. 2018). We set out to confirm these findings using an adoptive transfer (AT) model of diabetes. This model shows well-defined T cell mediated autoimmune diabetes that is more consistent and rapid than in NOD mice, allowing more robust assessment of therapeutic reversal (Bendelac et al. 1987). In AT mice, splenocytes from recently diabetic NOD mice were transferred to immune-deficient NOD-scid mice that lack functional T cells. Transferred diabetogenic T cells then drive rapid β-cell decline, leading to diabetes at ~6 weeks following transfer. Measurements of pancreas blood flow dynamics were made prior to adoptive transfer of splenocytes, and at 2 weeks and 4 weeks post-transfer (Fig.1A). To determine whether the impact of therapeutic treatment can be predicted, different treatments were applied at 2 weeks post-transfer. At each imaging time point we performed a ‘destruction-reperfusion’ measurement whereby the reentry of blood-borne microbubble contrast agents, following high mechanical-index destruction, was measured in the pancreas by sub-harmonic contrast enhanced ultrasound (Fig.1B). We fitted the resultant contrast signal time-course following destruction to determine the rate of recovery (reperfusion rate) that can be related to perfusion speed, and the amplitude of recovery (reperfusion amplitude) that can be related to the perfusion volume (Rim et al. 2001) (Fig.1C).

Figure 1 – Overview of ultrasound measurements.

A) Schematic of scan times for the adoptive transfer mouse model. B) Representative images of contrast signal (green) overlaid B-mode ultrasound image of pancreas (P), kidney (K) and spleen (S), before and after microbubble infusion, and immediately following flash destruction and reperfusion. Scale bar represents 5 mm. C) Representative time-course of contrast signal following flash destruction, for a control (above) and AT mouse (below) at 4 weeks following splenocyte transfer. Solid line indicates model fit. D) Schematic of each therapeutic treatment with respect to scan times. Scale bar in B represents 5 mm.

Predicting disease progression-

We first examined the reperfusion kinetics over all control mice (lacking splenocyte transfer) and all untreated AT mice (AT mice that do not receive therapeutic drug) examined throughout this study. Consistent with prior studies, the reperfusion rate progressively increased following splenocyte transfer, (Fig.2A), with a similar changes and variability to that previously observed (St Clair et al. 2018). We combined all control and AT mice in this study and generated an ROC curve using the rate data collected at 4 weeks post splenocyte transfer. We observed very strong separation between Control and AT mice in terms of reperfusion rate (area=0.952, Fig.2B). We then defined a threshold that provided optimal separation between Control and AT mice. Based on this threshold, those animals that showed a greater increase in the reperfusion rate developed diabetes sooner than those animals that showed lower or absent increase in the reperfusion rate (Fig.2C). This threshold was then used throughout the study to separate treated animals predicted to predicted to be disease negative or disease positive. We performed similar analysis considering the reperfusion amplitude. The reperfusion amplitude slightly decreased following splenocyte transfer (Fig.2D) again with a similar variability to that previously observed (St Clair et al. 2018). Consistent with this, upon generating an ROC curves using the amplitude data collected at 4 weeks post splenocyte transfer we observed a moderate separation between Control and AT mice (area=0.778, Fig.2E). We defined a threshold as before that provided optimal separation between Control and AT mice, which again was used throughout this study. Those animals that showed a greater decrease in the reperfusion amplitude also developed diabetes sooner than those that showed lower or absent decrease in the reperfusion amplitude (Fig.2F), albeit with reduced significance and separation compared to analysis using the reperfusion rate.

Figure 2 – predicting disease progression.

A) Reperfusion rate in control animals (black) and Adoptive transfer (AT) animals (red) at baseline (week 0) two weeks and four weeks post splenocyte or vehicle transfer, for all animals within this study. Dashed line represent disease prediction threshold that optimally separates control and untreated AT mice at 4 weeks post splenocyte transfer. B) Receiver Operating Characteristic (ROC) curve for reperfusion rate data at 4 weeks, with Sensitivity= % AT mice within (above) the threshold and 100%-Specificity= %control mice out of (below) the threshold. Area indicates area under ROC curve, with dotted line indicating area of 0.5. Empty circle represents disease prediction threshold with the highest maximum likelihood ratio. C) Survival curves indicating diabetes development of all control and untreated AT mice based on whether measurements of the reperfusion rate at 4 weeks after splenocyte transfer lie above or below a threshold indicated in A (dashed line). D) as in A for reperfusion amplitude. E) as in B for reperfusion amplitude data at 4 weeks post splenocyte transfer. F) As in C for reperfusion amplitude at 4 weeks post splenocyte transfer based on a threshold indicated in D. G) As in A for reperfusion rate in control and AT mice in the Kidney. H) As in D for reperfusion amplitude in control and AT mice in the Kidney. * represents p<0.05, ** represents p<0.01, **** represents p<0.0001 comparing groups indicated, as assed by 2-sided Students t-test (A,D) or Mantel-Cox Logrank test (C,F). Data in A,D represented n=26 AT and n=21 control mice; data in C represented 23 predicted disease + (rate high), 17 predicted disease − (rate low); data in F represented 32 predicted disease + (amplitude high), 8 predicted disease − (amplitude low); (5 AT mice and 2 control mice are excluded in A,D due to low goodness of fit as a result of poor infusion/flash-destruction at 4w).

To test whether the reperfusion dynamics were specific to the site of disease, we analyzed the reperfusion rate and amplitude within the kidney. We did not observe any significant difference between Control and AT mice in terms of reperfusion rate and amplitude (Fig.2G,H). Therefore, the reperfusion rate and to a lesser degree the reperfusion amplitude is predictive of diabetes progression in the AT model of T1D; as we previously demonstrated for the NOD mouse model (St Clair et al. 2018).

Detecting T1D remission achieved by antiCD3 therapy-

We previously determined that measuring the reperfusion rate could predict successful delay or prevention of diabetes in the AT model of diabetes following CD4⁺ T cell depletion (St Clair et al. 2018). We first tested whether we could predict prevention or delay of T1D upon an immunotherapy, antiCD3, which blocks the CD3 T-cell co-receptor that is involved in activation of both CD8+ and CD4+ T cells. AntiCD3 treatment has been demonstrated to successfully delay β-cell decline at onset and prevent T1D prior to onset in clinical studies (Herold et al. 1992; Herold et al. 2002; Herold et al. 2019). In NOD mice antiCD3 induces tolerance in part by allowing expansion of regulatory T cells in secondary lymph nodes (Nishio et al. 2010; Belghith et al. 2003). AT mice were treated at 2 weeks post splenocyte transfer with 5 daily doses antiCD3 (clone 145-2C11) (Fig.1D). This treatment delayed diabetes in ~60% of AT animals and prevented diabetes in ~20% of AT animals (Fig.3A). On average, both untreated AT mice and antiCD3 treated mice showed a significant increase in reperfusion rate at 4w (Fig.3B). Similarly, both untreated AT mice and antiCD3 treated mice showed a significant decrease in reperfusion amplitude at 4w (Fig.3C). In each case no difference was observed between untreated AT mice and antiCD3 treated groups at 2w, prior to treatment. The sub-optimal disease prevention allowed us to compare treatment responders and non-responders. Those antiCD3-treated AT mice that showed a delay or prevention of diabetes (responders) showed a significantly lower reperfusion rate following treatment (Fig.3D). Conversely, those antiCD3-treated AT mice that developed diabetes at a similar time to untreated AT mice (non-responders) showed increased reperfusion rate, similar to that in untreated animals. Responders and non-responders showed similar reperfusion rate prior to treatment at both 0w and 2w. No significant differences were observed in amplitude measurements between responders and non-responders (Fig.3E). Each group also showed similar blood glucose levels at the 4 week time-point (excluding those that had already developed diabetes) (ESM Fig.S1A). Using the threshold previously defined (Fig.2A,B), we separated mice into two groups based on whether they showed a low reperfusion rate (predicted to be disease negative) or a high reperfusion rate (predicted to be disease positive). Consistent with the above observations, those treated AT animals that showed a lower reperfusion rate developed diabetes with a significantly delayed and lower incidence compared to both untreated animals and AT animals with a higher reperfusion rate (Fig.3F). Those animals with a higher reperfusion rate developed diabetes with a similar incidence and speed to untreated AT mice. Thus, the reperfusion rate is predictive of successful or unsuccessful antiCD3-mediated therapeutic delay or prevention of diabetes in the AT model of T1D.

Figure 3 – predicting antiCD3-mediated diabetes prevention.

A) Survival curves indicating diabetes development for control (black), AT (red), and AT+anti-CD3 (blue) animals. B) Reperfusion rate in control animals (black), AT animals (red) or AT+anti-CD3 animals (blue) at baseline (week 0) two weeks and four weeks post splenocyte or vehicle transfer. C) as in B for reperfusion amplitude. D) Average reperfusion rate in anti-CD3 responders (solid diamonds) and non-responders (open diamonds) before (week 0), two and four weeks post splenocyte transfer, together with measurement in untreated AT mice (red). E) As in D for reperfusion amplitude. Individual amplitude and rate measurements for anti-CD3 responders and non-responders can be found in ESM Fig.S3. F) Survival curves indicating diabetes development of untreated AT mice (solid red), together with AT+anti-CD3 mice that are predicted to not respond to anti-CD3 treatment and develop diabetes (Disease +, dashed blue) or that are predicted respond to anti-CD3 treatment and not develop diabetes (Disease −, solid blue), based on measurements of the reperfusion rate at 4 weeks after splenocyte transfer (see Figure 1D). * represents p<0.05, ** represents p<0.01, ‘ns’ indicates p>0.15, comparing groups indicated, as assed by 2-sided Students t-test (B-E) or Mantel-Cox Logrank test (F). Data in A-C represented n=6 controls, n=7 AT and n=15 AT+anti-CD3 mice (4 AT+antiCD3 mice excluded in B,C due to poor infusion/flash-destruction); data in D,E represents 5 responder and 6 non-responders; data in F represents 6 Disease +, 5 Disease −.

We repeated these experiments with proof-of-principle antiCD4 treatment which depletes CD4+ T cells that drive T1D progression. We previously determined that successful delay or prevention of diabetes could be predicted in the AT model of diabetes following antiCD4 intervention (St Clair et al. 2018). AT mice were treated at 2 weeks of age with a single dose of antiCD4 (Fig.1D) which prevented diabetes in ~30% of AT animals (ESM Fig.S2A). The reperfusion rate and reperfusion amplitude showed a high spread in measurements (ESM Fig.S2B,C). Those antiCD4-treated AT mice that developed diabetes at a similar time to untreated AT mice (non-responders) showed increased reperfusion rate, similar to that in untreated animals (ESM Fig.S2D). While those AT mice that showed a delay or prevention of diabetes (responders) showed a lower reperfusion rate following treatment, there were insufficient numbers to judge statistical significance. No difference was observed in amplitude measurements between responders and non-responders (ESM Fig.S2E). Those treated AT animals with a higher reperfusion rate developed diabetes with a similar incidence and speed to untreated AT mice (ESM Fig.S2F). Those treated AT animals with a lower reperfusion rate developed diabetes with a delayed and lower incidence compared to those AT animals with a higher reperfusion rate. These trends are highly consistent with prior finding [10], that indicated the reperfusion rate to be predictive of successful or unsuccessful antiCD4-mediated prevention of diabetes in AT models of T1D.

Detecting T1D remission achieved by β-cell therapies-

An emerging strategy for T1D prevention is to target mechanisms underlying β-cell death/dysfunction (Engin et al. 2013; Xu et al. 2012a). We next tested whether we could predict delay or prevention of T1D upon a β-cell directed therapy, verapamil, that has been demonstrated to successfully delay β-cell decline at disease onset in clinical studies [14]. Verapamil blocks Ca²⁺ channels to inhibit the expression of TXNIP, a protein that promotes apoptosis and is upregulated in the β-cell in diabetes (Xu et al. 2012b). AT mice were continually treated at 2 weeks of age onwards with verapamil in drinking water (Xu et al. 2012a) (Fig.1D). This treatment delayed diabetes in ~40% of AT animals and prevented diabetes in ~25% of AT animals (Fig.4A). Untreated AT mice showed a significant increase in reperfusion rate, but this was not significant, on average, in verapamil-treated AT mice (Fig.4B). Similarly, untreated AT mice showed a significant decrease in reperfusion amplitude, but this was not significant, on average, in verapamil-treated AT mice (Fig.4C). In each case no difference was observed between untreated AT mice and verapamil treated groups at 2w, prior to treatment. Again, the sub-optimal disease prevention allowed us to compare treatment responders and non-responders. Those verapamil-treated AT mice that showed a delay or prevention of diabetes (responders) showed a significantly lower reperfusion rate following treatment (Fig.4D). Conversely, those verapamil-treated AT mice that developed diabetes at a similar time to untreated AT mice (non-responders) showed increased reperfusion rate, similar to that in untreated animals. Responders and non-responders showed similar reperfusion rate prior to treatment at both 0w and 2w. As under antiCD4 or antiCD3 treatment, no significant differences were observed in amplitude measurements between responders and non-responders (Fig.4E). Each group also showed similar blood glucose levels at the 4 week time-point (excluding those that had already developed diabetes) (ESM Fig.S1B). Upon separating mice into two groups, with low reperfusion rate (disease negative) or a high reperfusion rate (disease positive), those verapamil-treated AT animals that showed a lower reperfusion rate developed diabetes with a significantly delayed and lower incidence than those with a higher reperfusion rate (Fig.4F). Those treated animals with a higher reperfusion rate developed diabetes with a similar incidence and speed to untreated AT mice. Thus, the reperfusion rate is predictive of successful or unsuccessful verapamil-mediated therapeutic prevention of diabetes in the AT model of T1D.

Figure 4 – predicting verapamil-mediated diabetes prevention.

A) Survival curves indicating diabetes development for control (black), Adoptive Transfer (AT) (red), and AT+verapamil (green) animals. B) Reperfusion rate in control animals (black), AT animals (red) or AT+ verapamil animals (green) at baseline (week 0) two weeks and four weeks post splenocyte or vehicle transfer. (C) as in B for reperfusion amplitude. (D) Average reperfusion rate in verapamil responders (solid diamonds) and non-responders (open diamonds) before (week 0), two and four weeks post splenocyte transfer, together with measurement in untreated AT mice (red). E) As in D for reperfusion amplitude. Individual amplitude and rate measurements for verapamil responders and non-responders can be found in ESM Fig.S3. F) Survival curves indicating diabetes development of untreated AT mice (solid red), together with AT+verapamil mice that are predicted to not respond to verapamil treatment and develop diabetes (Disease +, dashed green) or that are predicted respond to verapamil treatment and not develop diabetes (Disease −, solid green), based on measurements of the reperfusion rate at 4 weeks after splenocyte transfer (see Figure 1D). * represents p<0.05, ** represents p<0.01, ‘ns’ indicates p>0.15, otherwise p is stated, comparing groups indicated, as assed by 2-sided Students t-test (B-E) or Mantel-Cox Logrank test (F). Data in A-C represented n=6 controls, n=9 AT and n=10 AT+verapamil mice (3 AT+verapamil mice excluded in B,C due to poor infusion/flash-destruction); data in D,E represents 4 responder and 3 non-responders; data in F represents 3 Disease +, 4 Disease −.

Finally, we examined another β-cell directed therapy, Tauroursodeoxycholic Acid (TUDCA), that is being applied in clinical studies (ClinicalTrials.gov Identifier: NCT02218619). TUDCA alleviates ER stress and thus prevents β-cell decline in diabetes (Engin et al. 2013). AT mice were continually treated at 2 weeks post splenocyte transfer onwards with TUDCA via IP injection (Fig.1D). In contrast to verapamil treatment, we did not observe a significant impact on diabetes development by TUDCA

Discussion:

The asymptomatic phase of T1D presents a window in which therapeutic intervention can be made to prevent diabetes. Responses to therapeutic treatments are often heterogeneous, and there is a complete lack of approaches to indicate whether the trajectory of disease progression is reversed upon therapeutic intervention. We previously demonstrated that contrast-enhanced ultrasound measurements of pancreas perfusion dynamics changed during the progression of T1D prior to diabetes onset, with high reproducibility. Importantly, the perfusion dynamics could predict the speed of diabetes progression, and distinguish successful and unsuccessful responses to interventions deigned to prevent T1D. In assessing the responses to therapeutic intervention, proof-of-principle antiCD4-mediated T-cell depletion was used, which is not a viable therapy for human T1D. Therefore it remains to be tested whether responses to relevant therapies for human T1D can be distinguished, and whether the success of these therapies can be predicted prior to diabetes onset.

In this study we sought to test whether changes in contrast-enhanced ultrasound measurement of pancreas perfusion dynamics could be used to predict the therapeutic prevention of type1 diabetes (T1D). We demonstrated that the delay and prevention of diabetes induced by antiCD3 and antiCD4 could be predicted in an adoptive transfer model of T1D. Furthermore, we demonstrated that the delay/prevention of diabetes induced by verapamil-mediated protection of β-cell decline could also be predicted in an adoptive transfer model of T1D. Under conditions where no diabetes was delayed/prevented, during TUDCA treatment, no difference was observed, strengthening how the measurement of pancreas perfusion dynamics strongly correlated with diabetes progression. In all cases the variability in reperfusion rate in control animals and untreated AT animals was low, consistent with our prior study (St Clair et al. 2018). This allowed us to effectively separate animals that progress to diabetes from those that do not, with a high ROC area (>0.95). This provides confidence that the variability in reperfusion rate that we observed in treated AT animals reflects heterogeneity in the success of therapeutic intervention and thus protection from the progression to diabetes. Nevertheless amplitude measurements showed a greater variability and reduced separation between animals that progress to diabetes from those that do not, indicating the need to use the reperfusion rate measurement for tracking disease development. We do also note that that a very small number of animals included in our analysis of non-responder treated AT animals had already developed diabetes. However it has previously been demonstrated that the change in pancreas perfusion dynamics with diabetes is not related to glycemic levels (St Clair et al. 2018).

Our data therefore suggests that contrast-enhanced ultrasound measurement of pancreas perfusion dynamics could be a viable approach to predict successful prevention of diabetes when treated during the asymptomatic phase of T1D. This is critically important, given heterogeneous or incomplete responses to therapeutic reversal (Herold et al. 2013) or prevention (Herold et al. 2019) of T1D in human trials and the lack of any approach presently that can predict the therapeutic reversal of T1D (Garyu et al. 2016). While autoantibodies can be used to predict onset of diabetes (Ziegler et al. 2013), it is unclear whether they can be used to predict the therapeutic prevention of diabetes progression. We further show this approach is applicable to therapies that have been demonstrated to modulate autoimmunity or β-cell viability, indicating it is broadly applicable.

In this study we utilized an AT model of T1D, which shows insulitis and β-cell decline but progresses to diabetes more rapidly and with greater incidence compared to the NOD mouse model. This allowed us to effectively separate mice that responded to therapy from those that did not respond. However given the slower and more heterogeneous progression to diabetes, future work should assess the prediction of diabetes progression following therapeutic intervention in NOD mice. Nevertheless, in the more homogenous model AT model we were still able to effectively separate responder and non-responder mice and predict diabetes progression. While we were able to predict diabetes progression, mice were indistinguishable prior to treatment. Thus prediction of treatment response, prior to treatment is not possible with these measurements. Nevertheless, it is important to note that very few studies to date have demonstrated an ability for imaging approaches to predict diabetes progression or assess therapeutic prevention of diabetes. Furthermore, while circulating islet autoantibodies provide some ability to predict diabetes progression, they are unlikely to provide any assessment of therapeutic prevention. The findings from the present study coupled with the utility or ultrasound imaging modalities show the potential of contrast-enhanced ultrasound measurements to be used to assess the responses to therapeutic interventions designed to prevent T1D.

It is unclear what the specific mechanism is that links insulitis, T1D progression and changes in pancreas perfusion dynamics. We and others have previously demonstrated in models of T1D that changes to the pancreas blood flow reflect changes to islet-specific blood flow dynamics (St Clair et al. 2018; Nyman et al. 2010; Roberts et al. 2017). Islet blood flow dynamics also correlate with remodeling of the islet microvasculature that occurs in both mouse models of T1D and human T1D (Papaccio 1993; Akirav et al. 2011; Canzano et al. 2018; Carlsson, Sandler, and Jansson 1998; Jansson et al. 2016). Altered microvascular organization may result from inflammation and thus lead to altered islet blood flow (Papaccio and Chieffi-Baccari 1992; Papaccio 1993; Papaccio, Frascatore, and Pisanti 1994). However, altered pancreas blood flow, for example as a result of islet decline or upstream arterial blood flow could also lead to islet microvascular reorganization (Roberts et al. 2017; Dai et al. 2013). While we have demonstrated predictive power for pancreas blood flow dynamics, determining the mechanism underlying altered blood flow dynamics in T1D will be important to fully interpret this diagnostic measurement. For example, the factor associated with T1D that determines blood flow dynamics may not fully explain T1D progression or reversal. We also cannot exclude that the treatments applied may directly influence blood flow. For example verapamil regulates smooth muscle contraction and thus could conceivably influence pancreas blood flow. However it is unclear how antiCD3 or antiCD4 would directly influence pancreas blood flow. Nevertheless it is important to note that the change in perfusion dynamics that we observe associated with diabetes development or therapeutic prevention does not imply a change in overall pancreas blood flow with diabetes development or therapeutic prevention. While the rate increase with diabetes indicates increased perfusion velocity, the amplitude decrease indicates a reduced perfusion volume. Therefore the overall blood flow will likely be minimally changed. This is consistent with our prior measurements (St Clair et al. 2018), as well as those made with other modalities that measure overall blood flow (Hirshberg et al. 2009).

The approach we use here is potentially applicable to human subjects. Currently MBs are FDA approved for cardiac imaging, and liver imaging in pediatric populations (Seitz and Strobel 2016). However contrast-enhanced ultrasound has been applied off label, to pancreas imaging for other indications (Dietrich, Ignee, et al. 2008; Napoleon et al. 2010; Ripollés et al. 2010). The islet microvasculature shows remodeling in human T1D, although studies prior to diabetes onset are lacking. However, the mouse and human islet do show important differences in terms of islet blood flow and microvasculature organization (Brissova et al. 2015; Cohrs et al. 2017; Tang et al. 2017). Some therapies such as antiCD3 may also show differing mechanism of action between human and mouse (Bresson and von Herrath 2011). While destruction-replenishment measurements have been applied in human subjects there are differences in the protocol compared to our animal based study, such as contrast agent delivery (bolus vs continual infusion) than could lead to differences in disease prediction (Williams et al. 2011). Therefore, testing this approach in human T1D or with the aid of humanized models for T1D are still needed.

Conclusion:

In summary, we demonstrate that ultrasound measurement of pancreas blood flow dynamics can predict successful therapeutic prevention of diabetes, induced by therapies relevant to treating human T1D. This potentially provides a method to translate to human studies, in which no approach currently exists to assess the therapeutic prevention of T1D.