Superior Glycemic Control With a Glucose-Responsive Insulin Analog: Hepatic and Nonhepatic Impacts

Strict control of glycemia has been demonstrated to have benefits for individuals with type 1 diabetes, but achievement of glycemic goals can increase the frequency of hypoglycemic episodes (1). A number of innovations in insulin design, including flatter pharmacokinetic (PK) profiles of analogs designed for basal dosing and more rapid onset of action for products intended for mealtime use, have improved the treatment of individuals with type 1 diabetes and reduced the rates of adverse effects (reviewed in ref. 2). In spite of these advances in therapeutic management, hypoglycemia continues to be a common complication of type 1 diabetes (3–6). Its occurrence and the fear of its occurrence interfere with achievement of therapeutic goals and impair quality of life (4). Thus, therapeutic strategies that further reduce the risk of hypoglycemia, while successfully responding to hyperglycemia, could significantly improve diabetes care. The development of a glucose-responsive insulin (GRI) analog (i.e., a product that is more effective at higher glycemic levels with curtailed action as blood glucose falls below euglycemia) is an approach that could markedly improve diabetes care.

One approach to creating a GRI is to synthesize insulin analogs that retain capacity to bind to the insulin receptor (IR) and additionally can bind to the mannose receptor (MR) (7–10). Binding of this type of GRI to the MR is enabled by the attachment of saccharides to an insulin backbone; the result of such binding is that the GRI is delivered for lysosomal degradation. With appropriate selection of saccharides, binding to the MR can be modulated by ambient levels of blood glucose, being greatest at low and normal glucose concentrations and correspondingly reduced with higher glycemic levels (7–10). Thus, clearance of the GRI will be higher at low ambient glucose, lowering its availability for IR binding and insulin action. Conversely, these GRIs manifest a lower clearance and higher potential for insulin action under hyperglycemic conditions. Previously reported in vivo studies with GRI analogs in rodent, dog, and minipig models have examined how ambient glucose modulates systemic PK and pharmacodynamic (PD) effects on glucose metabolism (10). A key question that arises is how various tissues shape the glucose-responsive PK and PD of these GRI analogs. It is well understood that the liver is a major organ for clearance of endogenous insulin (11,12) and of insulin analogs (e.g., 13,14), and additionally, it is recognized that a major proportion of systemic capacity for MR-mediated clearance resides within hepatic sinusoidal endothelial cells (HSECs) (15–17). This close juxtaposition of MR on HSECs to adjacent hepatocytes led us to postulate that the hepatic sinusoids have a high fractional extraction of GRI at low-to-normal blood glucose and are a principal site contributing to glucose-responsive PK. It is unclear how hyperglycemia, relative to euglycemia, modulates a GRI’s hepatic PD effects on endogenous glucose production and hepatic as well as nonhepatic glucose disposal. In the studies described in this article, we used tracer and hepatic organ balance techniques to examine hepatic and whole-body extraction of a GRI compared with that of regular insulin under euglycemic and hyperglycemic conditions, as well as to compare their respective effects on hepatic and systemic glucose metabolism in the conscious chronically catheterized dog.

The protocol was reviewed and approved by the Vanderbilt University Institutional Animal Care and Use Committee. Adult male mongrel dogs (n = 8; mean ± SEM 23 ± 1 kg) were purchased from a U.S. Department of Agriculture–licensed vendor and housed and cared for according to Association for Assessment and Accreditation of Laboratory Animal Care guidelines. The dogs were fed once daily a chow and meat diet in amounts calculated to be weight maintaining (18). Each dog underwent two hyperinsulinemic clamp studies. Approximately 16 days before the first study, each dog underwent surgery for placement of sampling catheters in a femoral artery, the hepatic portal vein, and the left common hepatic vein; infusion catheters in a splenic and a jejunal vein; and ultrasonic blood flow probes (Transonic Systems, Ithaca, NY) around the hepatic artery and portal vein, with the distal ends of all catheters and flow probes secured in subcutaneous pockets (19).

The structure of analog 630 (GRI-630) is shown in Supplementary Fig. 1. The attachment of saccharides to insulin can affect in vitro potency at IRs, influenced by the nature of the saccharide, its site of binding on insulin, and the length and composition of the linker. The binding affinities for GRI-630 for IR and MR are presented in Supplementary Tables 1 and 2. Notably, the binding affinity of GRI-630 for the IR potency was reduced ∼20-fold relative to regular human insulin (HI). Structurally, and in terms of IR and MR binding affinities, GRI-630 is highly similar to the GRI analog MK-2640, the PK and PD of which in dog and minipig have been recently reported (10). In screening PK tests in Yucatan minipigs, GRI-630 demonstrated a glucose-dependent change in its clearance similar to that previously shown for MK-2640. In the current studies, one goal of the study design was to use an infusion rate for GRI-630 (hereafter referred to as GRI) that was matched in its PD effect to suppress endogenous glucose production to that of regular insulin during the euglycemic phase of the clamp (clamp period 1 [P1] as described below) and that this PD effect be ∼25% suppression. In pilot studies conducted at Merck Research Laboratories (MRL) (Merck & Co., Kenilworth, NJ), dose-response effects of GRI to stimulate systemic glucose utilization and suppress endogenous glucose production in healthy beagles were characterized to inform selection of the GRI infusion rate to be used in the studies conducted at Vanderbilt. Briefly, these pilot studies used a constant infusion of GRI during a glucose clamp (at euglycemia) as previously described (10) and with use of stable isotope glucose tracer infusion to enable estimation of systemic glucose utilization and endogenous glucose production; findings are presented in Supplementary Fig. 2. All animal procedures were reviewed and approved by the MRL Institutional Animal Care and Use Committee. The Guide for the Care and Use of Laboratory Animals was followed in the conduct of the animal studies. Veterinary care was given to any animals requiring medical attention.

In the studies conducted at Vanderbilt University, the dogs were fasted overnight prior to each study. On the morning of study, the catheters and flow probes were removed from their subcutaneous pockets under local anesthesia. Each dog was placed in a Pavlov harness, peripheral venous access was established in three legs, and a primed, continuous infusion of [3-³H]glucose was begun via a peripheral vein. After a 90-min period of tracer equilibration (−120 to −30 min), there was a 30-min (−30 to 0 min) period of basal sampling, followed by a clamp period (0–300 min). At 0 min, a continuous infusion of somatostatin was started via peripheral vein to suppress endocrine pancreatic secretion, and either analog 630 (range 30–50 and mean ± SEM 36 ± 3 pmol/kg/min) or regular HI (1.8 pmol/kg/min) (referred to as the control study) was infused via peripheral vein, with the order of treatment randomized among the animals. Glucagon was replaced at a basal rate (0.5 ng/kg/min) into the hepatic portal circulation. Glucose (50% dextrose) was infused via peripheral vein as needed to clamp glucose at the target concentrations, 80 mg/dL (on the low end of the euglycemic range) from 0 to 150 min (P1) and 240 mg/dL from 150 to 300 min (clamp period 2 [P2]). The GRI infusion rates varied slightly among the dogs because the infusion rate was adjusted as needed early in the clamp period (0–30 min), if necessary, to achieve a similar rate of fall in glycemia with the GRI and HI. After 30 min, no further adjustments were made in the GRI infusion rates throughout P1 and P2.

At the end of the first study on each dog, the catheters and blood flow probes were returned to their subcutaneous pockets under aseptic conditions. The second study on each animal was ∼2 weeks after the first, to allow plasma clearance of the tritiated glucose from plasma.

Plasma glucose was analyzed on all samples using a GM9 glucose analyzer (Analox, Stourbridge, U.K.). In addition, plasma concentrations of insulin (canine or human, as appropriate), C-peptide, cortisol, glucagon, nonesterified fatty acids (NEFAs), and ³H-glucose were determined as previously described, as were blood lactate, alanine, glycerol, and β-hydroxybutyrate (βOHB) (19). Samples for GRI analysis were collected from the artery and portal and hepatic veins during each dog’s analog experiment; they were analyzed by liquid chromatography–mass spectrometry at MRL.

Hepatic balance calculations were carried out as previously described (19–21) and are briefly outlined in this study. The hepatic substrate load (HL) was calculated as [A] · AF + [P] · PF, where [A] and [P] represent the arterial and portal vein substrate concentrations, and AF and PF are the rates of arterial and portal vein blood or plasma (as appropriate) flow, respectively. Net hepatic substrate balance was calculated as the Load_out − HL, where Load_out is the hepatic vein substrate concentration multiplied by the total hepatic blood or plasma flow (the sum of AF and PF). Hepatic fractional extraction (HFE) (no units) of glucose was the net hepatic substrate balance of glucose divided by its HL. Unidirectional hepatic glucose uptake (HGU) was obtained by multiplying the HFE of [³H]-glucose (³H-HFE) by the HL of [³H]-glucose. Nonhepatic glucose uptake (nonHGU) was calculated as the difference between the glucose infusion rate (GIR) and net hepatic glucose balance, adjusted for change in the size of the glucose pool, as described previously (19). Glucose turnover (rates of endogenous glucose appearance [EndoR_a] and disappearance [R_d]) was calculated according to the circulatory model of Mari (22). ANOVA for repeated measures with post hoc analysis using the Holm-Sidak test was used for comparison of the treatments. A P value <0.05 was accepted as statistically significant. Data are given as mean ± SEM unless otherwise stated.

In the control (HI) studies, endogenous insulin secretion was completely inhibited during P1 and P2, as demonstrated by the suppression of C-peptide concentrations to the limits of detection (Table 1). The arterial plasma insulin concentrations increased from 44 ± 7 pmol/L in the basal period to 112 ± 3 pmol/L during the clamp (Fig. 1A). In contrast, the portal vein concentrations fell from 93 ± 16 (basal) to 82 ± 3 pmol/L in response to peripheral venous insulin delivery, whereas the hepatic vein concentrations remained relatively stable (48 ± 7 [basal] to 45 ± 3 pmol/L [clamp]). There were no significant differences in the HI concentrations between P1 and P2 in any of the blood vessels sampled, and the calculated hepatic sinusoidal HI concentrations remained relatively stable (85 ± 15 pmol/L in the basal period and 89 ± 3 pmol/L during P1 and P2). The HFE of HI did not differ significantly between P1 and P2 in the control studies (0.51 ± 0.04 and 0.48 ± 0.02, respectively; P = 0.18) (Fig. 1C). Likewise, hepatic insulin clearance was similar in P1 and P2 (7.6 ± 0.5 and 8.2 ± 0.5 mL/kg/min; P = 0.06) (Fig. 1D), as was whole-body insulin clearance (16.4 ± 1.1 and 17.2 mL/kg/min, respectively; P = 0.171) (Fig. 1E).

During the studies with the GRI, endogenous insulin secretion (i.e., C-peptide) was again fully suppressed (Table 1). The mean arterial analog concentrations during P2 were ∼20% greater than those during P1 (2.3 ± 0.3 vs. 1.9 ± 0.3 nmol/L, respectively; P < 0.001 between periods) (Fig. 1B), confirming a glucose-responsive PK. The analog concentrations in the portal (2.1 ± 0.3 vs. 1.7 ± 0.3 nmol/L) and hepatic (1.3 ± 0.2 vs. 0.6 ± 0.1 nmol/L) veins were similarly higher in P2 (P < 0.001 between P1 and P2 for both). HFE of the GRI was significantly greater during P1 versus P2 (0.62 ± 0.03 vs. 0.43 ± 0.03, respectively; P < 0.001) (Fig. 1C), and hepatic clearance of the GRI was also greater during P1 versus P2 (8.9 ± 0.5 vs. 7.0 ± 0.3 mL/kg/min, respectively; P < 0.01). Both the HFE and whole-body clearance were significantly greater during P1 with the GRI versus the HI control (Fig. 1C and E). ANOVA did not reveal a significant difference between treatments in hepatic insulin clearance; nevertheless, if the means of the values for all time points during P1 are compared, hepatic insulin clearance was 17% greater (P < 0.01) with the GRI (8.9 ± 0.5 mL/kg/min) than the HI (7.6 ± 0.5 mL/kg/min).

In both studies, plasma glucagon declined slightly from basal during P1, with a further decline during P2; there was no significant difference in the response between the two treatments (Table 1). Arterial plasma cortisol increased modestly (2 to 2.5 × basal) in response to the glycemic decline during P1 of both studies and returned to basal levels during P2; there was no significant difference in the response to the two insulin products (Table 1).

Arterial glucose concentrations did not differ at any time between the treatments (Fig. 2A). The plasma concentrations fell to ∼80 mg/dL by the end of P1. During P2, the arterial plasma concentrations were clamped at 241 ± 2 mg/dL during both the control and the GRI studies. The hepatic glucose loads were indistinguishable between the control and GRI studies (Fig. 2B). Net hepatic glucose output declined modestly from basal during P1 and did not differ between the control and GRI studies (Fig. 2C), therefore achieving the hepatic PD equipoise during P1 that was a goal of the study design. Thus, a potential difference between HI and GRI PD during the hyperglycemic clamp in P2 could be more clearly observed. During the hyperglycemic P2, there was a shift to net HGU (NHGU) with both insulin products, but the rate of NHGU was 54% greater during delivery of the GRI (P < 0.05). Tracer-determined HGU was also enhanced during the GRI versus control study (Fig. 2D) (P < 0.05). The GIR increased markedly during P2 in both the control and GRI studies (Fig. 2E). Early in P2, this was largely attributable to the mass of glucose required to fill the glucose pool, and the GIRs were very similar between treatments. However, the GIR required to maintain the hyperglycemic goal was significantly greater during the last hour of study in the GRI versus the control treatment (10.7 ± 1.0 vs. 8.4 ± 0.9 mg/kg/min; P < 0.05).

The mean rate of nonHGU increased compared with basal conditions during P1 with both treatments (P < 0.05), but even with hepatic PD equipoise during P1, the change from the basal rate of nonHGU was significantly greater in the HI versus the GRI study (Δ0.9 ± 0.3 vs. Δ0.3 ± 0.1 mg/kg/min, respectively; P < 0.001). In contrast, during the final hour of P2, nonHGU was >20% greater with the GRI versus control treatment (P < 0.05), proportionate to the 20% increase of GRI arterial concentration in P2 versus P1, whereas there was not a change in HI arterial concentration across periods. The tracer-determined EndoR_a during the basal period was very similar between insulin treatments; EndoR_a fell 26–28% during P1 with both products (Fig. 3A). During P2, EndoR_a declined further with both treatments, but it plateaued at ∼1.2 ± 0.2 mg/kg/min during the last hour during the HI experiments, whereas it continued to fall with the GRI, reaching a rate of 0.6 ± 0.2 mg/kg/min at the last sampling point (P < 0.05 between treatments). Glucose R_d rose modestly during P1 with both insulin products, averaging 3.6 ± 0.2 and 3.2 ± 0.2 mg/kg/min in the HI and GRI studies, respectively (Fig. 3B). During P2, it tended to be stimulated more with the GRI than HI (mean ± SEM for P2: 11.0 ± 1.1 and 9.6 ± 0.9 mg/kg/min, respectively; P = 0.18). Overall, there was a significant difference in R_d between treatments (P < 0.001). In the HI studies, the rate of glucose clearance increased over the course of P1, fell modestly early in P2, and then climbed slowly, such that the mean rates of glucose clearance were indistinguishable between P1 and P2 (Fig. 3C). In contrast, in the GRI studies, glucose clearance during P1 tended to be reduced in comparison with the HI studies (3.4 ± 0.3 vs. 4.0 ± 0.4 mL/kg/min, respectively; P = 0.15), and, conversely, glucose clearance during P2 tended to be enhanced compared with the HI studies (4.6 ± 0.4 vs. 4.0 ± 0.4 mL/kg/min; P = 0.13). Although the rates of glucose clearance during each of the two treatment periods did not differ between the different insulin products, the overall response was markedly different (P < 0.001).

Arterial concentrations and net hepatic balances of lactate, alanine, βOHB, and NEFA did not differ between insulin treatments at any time (Table 2). However, arterial glycerol concentrations and net hepatic glycerol uptake tended to be lower with the GRI during P1 and were suppressed to a significantly greater extent with the GRI versus the HI treatment during P2. The NEFA values tended to follow those of glycerol.

The development of an insulin analog that could become more available at higher glycemic levels but less available and thus provide less insulin action as glucose levels fall to euglycemia and below has sparked considerable interest (23–32). A glucose-responsive analog as recently described (8), and as was used in the current studies, is one such approach with the potential to markedly reduce the risk of hypoglycemia while maintaining or even bolstering effectiveness in controlling hyperglycemia. In this series of paired studies, we used an insulin chimera with an insulin backbone to which select saccharides were bound by linkers, chosen so that the analog would have the capacity to bind to the MR lectin receptor, with its binding dependent on ambient blood glucose levels (10). At lower glucose concentrations, more of the analog binds to and is cleared by MR receptor, whereas at higher glycemic levels, as MR binding lessens, a higher proportion of the analog is available for interaction with the IR. In the current study, as compared with HI in the same dog, the infusion rate of the GRI needed to achieve bioequivalence at plasma glucose levels of 80 mg/dL was ∼20-fold greater than that of HI, reflecting the lower potency of the GRI, with a substantial proportion of its clearance mediated by MR. The corresponding arterial plasma levels of the GRI analog were 17-fold (P1) to 20-fold (P2) those of HI. Whether clinical development of an analog such as GRI-630, with its considerably reduced potency, is feasible remains to be determined, as is more fully addressed below. The current studies were undertaken as mechanism-of-action studies to examine how the glucose-responsive PK could translate into a glucose-responsive PD effect. During P1, when arterial plasma glucose was allowed to fall to ∼80 mg/dL, HFE of the GRI was significantly greater than that of HI. This was a net effect of GRI clearance, determined by uptake by the MR on HSECs and by the IR on hepatocytes, though their respective contributions cannot be estimated from the current data.

The liver accounted for 42–44% of the GRI’s clearance, whereas gut extraction accounted for 15–18%, but the conditions of study did not allow for the identification of the other tissue(s) contributing to clearance. It is likely that renal clearance played a role, based on its known role in insulin clearance and the presence of the MR in the kidney (33). It is noteworthy that there was a suggestion of a hepatopreferential effect of the analog during P1, because net hepatic glucose output tended to be more suppressed with the GRI versus HI (P = 0.07 between treatments if all time points in P1 are averaged), whereas nonHGU was stimulated to a lesser extent (2.0 ± 0.2 vs. 2.6 ± 0.3 mg/kg/min, respectively; P < 0.05 between treatments) with overall bioequivalent doses of the GRI versus HI. Moreover, the rates of NHGU and tracer-determined HGU during P2 were significantly greater with the GRI. Insulin levels at the liver are one of the primary factors affecting HGU and glycogen storage under hyperglycemic conditions (34–38), and thus, the results in P2, as in P1, are consistent with a greater impact of the analog versus HI on the liver (39).

Despite the tendency toward hepatopreferentiality, it should be noted that the GRI appeared to suppress lipolysis to a slightly but significantly greater extent than HI, based on the relative decline in glycerol concentrations, and this likely contributed to its greater impact on glucose disposal (14). NEFA concentrations and balance data followed the same patterns as those of glycerol but did not reach statistically significant differences. This is likely explained by the fact that glycerol has been shown to be a better indicator of the rate of lipolysis than NEFA, due to the potential for NEFA but not glycerol to undergo a substantial amount of re-esterification within the adipocyte (40,41). There was no significant difference in the concentrations, hepatic extraction, or hepatic and whole-body insulin clearances of HI during P2 versus P1. In contrast, with the GRI, the hepatic extraction and hepatic and whole-body clearance declined significantly in P2 (plasma glucose clamped at 240 mg/dL) compared with P1. Presumably, this decline in hepatic extraction of GRI reflects the effect of hyperglycemia to reduce MR-mediated clearance of GRI, and, as a result, the concentrations of the GRI were significantly higher in P2 versus P1. Moreover, NHGU, nonHGU, GIR, and glucose R_a and R_d were impacted to a greater extent under hyperglycemic conditions with the use of the GRI, consistent with its greater availability. Although the current findings clearly show a beneficial effect of the GRI, it should be noted that only one infusion rate was examined, and thus, further studies are required to determine whether glucose responsiveness is maintained at higher or lower GRI delivery rates, although our previous data indicate that it is (10).

Despite the potentially important and novel proof of pharmacology demonstrated in the current study for GRI-630, it remains to be determined whether clinical development will be feasible for this analog, with its considerably reduced potency. From one point of view, it has been well-established in exploring chemical modifications of insulin analogs that reductions of in vitro potency of ∼10-fold relative to native insulin can nonetheless be associated with in vivo biological activity that is dose-equivalent to that of native insulin (42,43). This dose-equivalent biological activity is related to reduced rates of IR-mediated clearance, in inverse proportion to the extent of reduced in vitro potency. However, these important observations in insulin chemistry are unlikely to extend fully to novel analogs such as GRI-630. GRI-630 and the recently described MK-2640 (8) have an even more reduced in vitro potency than analogs described in the above publications. As designed, they manifest additional clearance mediated by the MR system, and for these analogs, the fractional clearance by MR is considerable. Consequently, it may be that impractically large doses will be required in clinical use of GRI analogs like GRI-630 and MK-2640. Recent efforts in chemistry by our group have identified additional GRI analogs that are more potent toward the IR than GRI-630, by a factor of two- to threefold, while retaining MR binding affinity similar to that of GRI-630 and MK-2640 (R.N., S.L., personal communication). Though in vivo evaluation of these newer analogs has yet to progress as far as that for GRI-630 and MK-2640, good glucose responsiveness has been shown in preclinical assays. Considerably more research will be required to determine if these more potent GRI analogs will have more favorable prospects for clinical development, but the directional improvements for in vitro and in vivo potency may hold promise.

In conclusion, under low-normal glycemic conditions, less of the GRI was available than under hyperglycemic conditions, despite an identical delivery rate. There was greater extraction of the GRI than HI in the presence of low-normal glucose levels, and this difference then shifted during hyperglycemia, as extraction of the GRI decreased considerably, whereas that of HI did not change. Because of the differences in insulin extraction and action, nonHGU and glucose R_d tended to be reduced at the lower glycemic concentrations with the use of GRI-630 versus HI, whereas the GRI was at least as active at the liver as regular insulin, albeit at a higher molar concentration. In contrast, in the presence of hyperglycemia, the GRI was more effective in stimulating glucose uptake by hepatic and nonhepatic tissues than the bioequivalent dose of HI. These data suggest that GRI analogs have the potential to enhance tissue glucose uptake, including that of the liver, under hyperglycemic conditions, thus potentially contributing to improved glycemic control and glycogen storage. In contrast, the decrease in availability of the GRI in the presence of low-normal glycemic levels holds promise for decreasing the risk of hypoglycemia in individuals with type 1 diabetes.

Acknowledgments. The authors thank colleagues from MRL, particularly Carlos Rodriguez, Joel Mane, Andrew Misura, and John Strauss, and appreciate editorial comments from Ester Carballo-Jane (Merck).

Funding. The Vanderbilt Diabetes Research and Training Center Metabolic Physiology Shared Resource and Hormone Assay and Analytical Services Cores (with funding from National Institute of Diabetes and Digestive and Kidney Diseases grants DK-020593 and DK-059637) made important contributions to this work. A.D.C. holds the Jacquelyn A. Turner and Dr. Dorothy J. Turner Chair in Diabetes Research.

Duality of Interest. This work was supported by Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc. (Kenilworth, NJ). D.E.K., R.C.C., P.Z., T.Y., S.L., N.C.K., R.N., T.M.K., M.V.H., S.F.P., and C.M. were full-time employees of Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc. (Kenilworth, NJ), and held stock or stock options in the company during the conduct of these studies. A.D.C. was a consultant to Merck & Co., Inc. at the time these studies were conducted. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. M.C.M. collected and compiled the data and drafted and revised the manuscript. M.C.M., M.S.S., and B.F. performed experiments. M.C.M. and A.D.C. analyzed data and interpreted results of experiments. D.E.K. and A.D.C. conceived and designed the research. R.C.C., P.Z., T.Y., S.L., N.C.K., R.N., T.M.K., M.V.H., S.F.P., and C.M. participated in the development, screening, and selection of GRI-630 and analysis of data. M.S.S. carried out laboratory analyses. B.F. and P.W. conducted all surgical procedures and monitored the animal health. All authors reviewed, had input into, and approved the manuscript. D.E.K. and A.D.C. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in poster form at the 77th Scientific Sessions of the American Diabetes Association, San Diego, CA, 9–13 June 2017.