Profound Sensitivity of the Liver to the Direct Effect of Insulin Allows Peripheral Insulin Delivery to Normalize Hepatic but Not Muscle Glucose Uptake in the Healthy Dog

Normally the pancreas secretes insulin directly into the hepatic portal vein, causing an insulin gradient between the liver and the rest of the body. Peripheral insulin delivery eliminates this gradient. As a result, the therapeutic consequence of subcutaneous insulin injection in patients with diabetes is hepatic insulin deficiency and peripheral hyperinsulinemia, relative to normal physiology, a state that contributes to metabolic disease (1–7).

Insulin regulates hepatic glucose production (HGP) and uptake (HGU) via indirect and direct hepatic mechanisms (the former via suppression of lipolysis [8–13] and glucagon secretion [14], and stimulation of CNS insulin signaling [15–18]). Previously we showed that when insulin is delivered via the portal vein its direct effects are dominant over its indirect effects (19) and that direct insulin action alone is sufficient for normal regulation of hepatic glucose metabolism (20,21). This may not hold true when insulin is delivered via the typical therapeutic route, however. With subcutaneous delivery there is much greater exposure of nonhepatic tissues to insulin, and this may cause indirect insulin regulation of the liver to become more important. Furthermore, peripheral insulin delivery could exacerbate the effect of a defect in these mechanisms, leading to further dysregulation of the liver in diabetes. The means by which insulin regulates the liver also has important ramifications for liver-targeted insulin delivery approaches (e.g., hepatopreferential and oral insulin analogs, hepatic-directed vesicle insulin, intraperitoneal insulin delivery, and endogenous insulin secretagogues), which seek to produce superior hepatic glucose control while minimizing peripheral hyperinsulinemia.

We hypothesized that under postprandial-like conditions, peripheral insulin delivery would shift the regulation of HGP and HGU away from insulin’s direct mechanisms of control, making its effects on adipose tissue, pancreas, and brain more critical. For testing this, insulin was delivered peripherally, while insulin’s indirect effects either were allowed to occur or were blocked (triglyceride and glucagon were infused to maintain free fatty acid (FFA) and glucagon levels at baseline, and an increase in hypothalamic insulin signaling was prevented with use of intracerebroventricular delivery of an insulin receptor antagonist).

Studies were carried out on 12 conscious 18-h-fasted mongrel dogs (18–22 kg). To provide baseline control data, we compared molecular readouts with those in liver, skeletal muscle, and adipose samples obtained from an additional three healthy overnight fasted animals. The surgical and animal care facilities met the standards published by the American Association for the Accreditation of Laboratory Animal Care, and diet and housing were provided as previously described (22). The Vanderbilt Institutional Animal Care and Use Committee approved the protocol.

Approximately 17 days before study, the animals underwent surgery for placement of sampling catheters in a femoral artery and the hepatic portal and hepatic veins, and infusion catheters in the splenic and jejunal veins, which drain into the portal vein (22). Ultrasonic flow probes (Transonic Systems, Ithaca, NY) were placed around the hepatic portal vein and the hepatic artery (22). Ten days prior to the study, stereotactic third ventricle cannulation was performed as previously described (18,20,21,23). All dogs were determined to be healthy prior to experimentation, as indicated by 1) leukocyte count <18,000/mm³, 2) hematocrit >35%, and 3) good appetite (consuming at least 75% of the daily ration). On the morning of the experiment, the catheters and flow probe leads were exteriorized from their subcutaneous pockets under local anesthesia. Intravenous (IV) catheters were also inserted into peripheral leg veins for infusion of glucose, hormones, and radioactive tracer, as necessary.

Each experiment (Fig. 1) consisted of a 90-min tracer equilibration period (−120 to −30 min), a 30-min basal sample collection period (−30 to 0 min), and a 4-h experimental period (0–240 min). At −120 min, a primed continuous IV infusion of [3-³H]-glucose (42 µCi prime and 0.35 µCi/min continuous rate; PerkinElmer) was started to measure HGU and HGP. At 0 min, somatostatin (0.8 µg/kg/min; Bachem) was infused IV to suppress pancreatic insulin and glucagon secretion, and insulin (Humulin R; Eli Lilly) was infused into a leg vein (1.35 mU/kg/min). Simultaneously, glucose was infused into the hepatic portal vein (via the splenic and jejunal catheters) at a constant rate in all groups (4 mg/kg/min) to simulate gut glucose absorption and into a peripheral vein as needed to maintain arterial plasma glucose levels at approximately twofold basal (∼210 mg/dL).

In one experimental group (Fig. 1) insulin’s full effects were in play (direct + indirect [D+I]) (n = 6), while in the other, insulin’s indirect effects were blocked (D-only) (n = 6). In the D+I group FFA levels were allowed to fall naturally due to the inhibition of lipolysis by insulin. In addition, the suppression of glucagon by insulin (20) was mimicked through infusing glucagon intraportally in progressively decreasing amounts (0.57 ng/kg/min during the first 30 min and then a 10% decrease every 30 min thereafter, as done previously [21]). Activation of hypothalamic insulin signaling was also allowed (third ventricle artificial cerebrospinal fluid infusion). In the D-only group, with use of a previously established protocol (20,21), IV triglyceride (0.023 mL/kg/min, Intralipid 20%; B. Braun Medical) and heparin (0.495 units/kg/min) were infused at 0 min to prevent a decrease in plasma FFA. In addition, glucagon (GlucaGen; Novo Nordisk) was infused (0.57 ng/kg/min) into the portal vein to maintain its circulating levels at basal. Also, an insulin receptor antagonist (S961; gift from Novo Nordisk [24]) and PI3K inhibitor (LY294002; Sigma-Aldrich [25]) dissolved in artificial cerebrospinal fluid were infused into the third ventricle beginning at −90 min (a strategy shown previously to prevent an increase in hypothalamic insulin signaling and the ensuing PI3K-mediated activation of hypothalamic K_ATP channels [16,17] and transcriptional and metabolic effects at the liver [18,20,21,23]). The lipid emulsion, used frequently to study the effects of FFA on insulin action (10,20,21,26–30), was composed of essential FFA, including linoleic, oleic, palmitic, linolenic, and stearic acids, which comprise the majority of fasting FFA in the circulation (31).

The insulin infusion rate used in this study was chosen to simulate a modest dose of peripherally injected insulin given with a meal. We used less than one-half of the rate that would be secreted to cause the 10-fold rise in endogenous insulin that can occur following a mixed meal (32) because we wanted to minimize the amount of recirculating insulin reaching the liver, thus giving insulin’s indirect effects the most likely opportunity to become evident. Although postprandial glucose absorption is usually complete within 3 h, the experimental period was extended to 4 h to ensure that all of insulin’s acute effects would have sufficient time to manifest.

Plasma glucose, [³H]-glucose, and nonesterified FFA and blood lactate, glycerol, and β-hydroxybutyrate concentrations were determined as previously described (22). Plasma insulin and C-peptide were measured by radioimmunoassay (PI-12K, 100% specificity for canine and human insulin, and CCP-24HK, respectively; MilliporeSigma) and glucagon by ELISA (10-1271-01; Mercodia). Glycogen content was measured in liver and muscle biopsies taken at the end of the study with the method of Keppler and Decker (33).

We calculated unidirectional HGU calculated, as previously described (34), by multiplying hepatic fractional extraction of [³H]-glucose by hepatic glucose load (see Supplementary Material for more detail). This approach allows HGU to be partitioned from HGP, yielding unidirectional uptake (i.e., glucokinase flux) rather than net liver balance. Although net splanchnic glucose balance can be measured in the human, this challenging procedure requires arterial and hepatic vein blood sampling and measurement of liver blood flow, and it reflects the net integration of liver and gut glucose uptake and production. Access to hepatic portal vein blood and blood flow makes the dog a useful model for directly measuring HGU, a parameter that is difficult to assess in other species.

Net hepatic substrate balance was calculated with the arteriovenous difference method (22) to determine substrate load going into and coming out of the liver. Nonhepatic glucose uptake (primarily muscle [35]) equaled the glucose infusion rate minus net hepatic glucose balance (36), where the rate was corrected for changes in the size of the glucose pool, using a pool fraction of 0.65 mL/kg (37), assuming that the volume of distribution for glucose equaled the volume of the extracellular fluid, or ∼22% of the dog’s weight (38). Insulin-independent glucose uptake was estimated to be 67% (∼1 mg/kg/min) of net hepatic glucose output in the basal state and assumed to remain stable throughout the clamp period (36,39).

Glucose turnover, used to estimate HGP and whole-body glucose disposal, was measured with [3-³H]-glucose infusion, based on the GLUTRAN circulatory model of Mari et al. (40). We determined liver glycogen specific activity by dividing [³H]-glycogen (dpm/g liver) by cold glycogen (mg/g liver). We calculated hepatic sinusoidal plasma glucose specific activity by dividing [³H]-glucose (dpm/mL) by unlabeled glucose (mg/mL). The plasma insulin or glucose level entering the hepatic sinusoids was calculated with arterial and portal vein concentrations and the percent contributions of arterial and portal flow to total hepatic blood flow (22).

RNA isolation, cDNA synthesis, quantitative PCR primers and analysis, and Western blotting procedures were performed with standard procedures, details of which are provided in Supplementary Material. Nucleotide sequences of dog-specific PCR primers are shown in Supplementary Table 1, primary antibodies in Supplementary Table 2, and representative Western blot gels in Supplementary Fig. 1.

Statistical comparisons were carried out with SigmaStat (Systat Software, San Jose, CA) using ANOVA for repeated measures with Student-Newman-Keuls post hoc analysis. Statistical significance was accepted when P < 0.05. Data are means ± SEM.

The data sets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request. No applicable resources were generated or analyzed during the current study.

Pancreatic secretion of insulin into the portal vein resulted in hepatic insulin levels that were ∼2.5-fold higher than arterial during the basal period (−30 to 0 min) (Fig. 2A and C). In response to peripheral vein insulin infusion this gradient was abolished, however, falling to 0.8 (portal vein insulin levels were less than arterial during peripheral delivery due to the extraction of insulin in blood perfusing the gut) (0–240 min) (Fig. 2A and B). Plasma C-peptide levels were reduced to near zero during the experimental period, indicating that somatostatin completely suppressed endogenous insulin secretion (Table 1). Arterial insulin increased 8.0-fold during the clamp (Fig. 2A), compared with only 2.8-fold at the liver sinusoid (Fig. 2C), replicating the nonphysiologic distribution of insulin that characterizes therapeutic peripheral insulin delivery. Hepatic plasma insulin fractional extraction (54 ± 2 vs. 58 ± 4%) and arterial insulin clearance (21 ± 1 vs. 22 ± 1 mL/kg/min) were not different in the D+I vs. D-only groups, respectively, during the experimental period.

In accordance with arterial hyperinsulinemia, hypothalamic protein kinase B (Akt) phosphorylation (Fig. 3A) was elevated in the D+I group, but this increase in insulin signaling was blocked by intracerebroventricular infusion of insulin signaling blockers in the D-only group (1.86 ± 0.13-fold vs. 1.16 ± 0.06-fold, respectively; P < 0.05 between groups and vs. baseline). Hepatic sinusoidal plasma glucagon levels (Fig. 3B) were similar during the basal period in the D+I and D-only groups (10 ± 1 and 12 ± 2 pg/mL) but then either fell over time (to 1 ± 1 pg/mL in D+I) or were slightly elevated, but ended close to basal (15 ± 2 pg/mL in D-only; P < 0.05 between groups), due to the intraportal infusion of glucagon. Similarly, arterial glucagon levels fell from 5 ± 1 to 0.1 ± 0.3 pg/mL in D+I but were 6 ± 1 and 8 ± 1 pg/mL in D-only during the basal period and end of the study, respectively. Basal arterial FFA levels (Fig. 3C) were similar in the D+I and D-only groups (930 ± 45 and 1,087 ± 71 µmol/L, respectively) but then either fell due to the effects of insulin on lipolysis (to 20 ± 19 µmol/L in D+I) or were maintained at baseline (1,043 ± 30 µmol/L in D-only; P < 0.05 between groups) due to the infusion of triglyceride. Likewise, net hepatic FFA uptake (Fig. 3D) was similar in the two groups at baseline (2.6 ± 0.2 µmol/kg/min in both) but then was either completely eliminated or maintained close to baseline (2.4 ± 0.3 µmol/kg/min at 240 min in D-only; P < 0.05 between groups). In contrast, arterial β-hydroxybutyrate levels and net hepatic β-hydroxybutyrate output (Table 2) were suppressed during hyperinsulinemia in both groups, although to a greater extent in D+I. These data suggest that hepatic ketogenesis was suppressed both by the direct effect of insulin on the liver (D-only) and due to a lack of substrate (D+I).

Arterial plasma glucose levels doubled during the experimental period (Fig. 4A) (increasing to 206 ± 2 and 211 ± 2 mg/dL in the D+I and D-only groups, respectively]) due to glucose infusion into the portal vein and a leg vein. Hepatic glucose loads and portal to arterial glucose gradients (important determinants of HGU, in addition to insulin) were similar between groups (Fig. 4B and C). Less glucose was required to maintain the same level of hyperglycemia during the last hour of the experiment in the D-only group (Fig. 4D) due to reduced nonhepatic glucose uptake (Fig. 4E), which corresponded to reduced whole-body glucose disposal (Fig. 4F), primarily reflecting differences in muscle glucose uptake.

We hypothesized that peripherally delivered insulin would shift the locus of control of hepatic glucose metabolism away from direct insulin action, creating a circumstance where insulin’s indirect effects would play the predominate role. As such, we expected that HGU would be minimal in the presence of only a small rise in insulin at the liver (D-only) and greater when insulin’s indirect effects were present (D+I), especially when indirect insulin action was amplified by higher arterial insulin levels due to peripheral delivery. In fact, however, HGU did not differ significantly between groups (3.8 ± 0.5 vs. 3.2 ± 0.6 mg/kg/min in D+I and D-only, respectively) (Fig. 5A) (P = 0.5 between groups), with robust uptake occurring despite only a modest rise in insulin at the liver. The area under the curve (AUC) for HGU was 802 ± 110 vs. 666 ± 129 mg/kg/240 min (P = 0.4). Tracer-determined HGP decreased from basal (2.6 ± 0.1 and 2.4 ± 0.2 mg/kg/min in D+I and D-only) to 0.2 ± 0.2 and 0.4 ± 0.2 mg/kg/min by the end of the study, but direct insulin action did so more slowly in the absence of insulin’s indirect effects (Fig. 5B) (P < 0.05). The AUC for the decrease in HGP from basal was 422 ± 81 vs. 221 ± 40 mg/kg/240 min (P < 0.05). Net hepatic glucose balance represents the difference between HGU and HGP (although the three parameters were determined independently). As expected, there was fasting net hepatic glucose output. Then, in response to postprandial-like conditions, there was a rapid switch to net HGU (Fig. 5C) (−3.7 ± 0.6 vs. −3.2 ± 0.8 mg/kg/min, respectively; P = 0.6). As an additional, independent method of calculating HGU, HGP was subtracted from net hepatic glucose balance. Again, there was no measureable indirect effect of insulin on HGU (Fig. 5D) (P = 0.8 between groups).

Liver insulin signaling increased comparably in the two groups, in line with similar hepatic insulin exposure. Insulin receptor and Akt phosphorylation were both elevated (Fig. 6A and B) (P < 0.05 vs. baseline). In response, liver glucokinase (a key insulin-sensitive regulator of HGU [41]) mRNA and protein levels increased (Fig. 6C and D) (P < 0.05 vs. baseline). There was a tendency for glucokinase gene expression to be greater with indirect insulin action, in line with the 35% contribution due to brain insulin action that was observed previously (18,21,23). Glycogen synthase kinase-3β (GSK-3β) phosphorylation also increased (Fig. 6E) (P < 0.05 vs. baseline) and glycogen synthase was activated by dephosphorylation (Fig. 6F) (P < 0.05 vs. baseline). These changes resulted in an increase in liver glycogen content (Fig. 6F) (36 ± 1 mg/g liver at basal, increasing to 67 ± 5 and 62 ± 5 mg/g liver in the D+I and D-only groups, respectively; P < 0.05 vs. baseline) that paralleled the increase in HGU. The ratio of [³H]-liver glycogen specific activity to hepatic sinusoidal [³H]-plasma glucose specific activity, an index of glycogen formed via HGU rather than gluconeogenic flux, did not differ between groups (38 ± 2 vs. 35 ± 3, respectively). It is assumed that glycogen cycling was negligible under the hyperinsulinemic/hyperglycemic study conditions.

Net hepatic gluconeogenic, glycolytic, glycogenolytic, and glycogenesis rates were also determined. Net hepatic alanine uptake (typically the most significant gluconeogenic amino acid) decreased by approximately one-third during the experimental period in both groups, with no significant differences in arterial alanine levels or net hepatic balance between groups (Fig. 7A and B). Hepatic alanine fractional extraction is in part regulated by insulin, and tended to decrease, from 0.27 ± 0.02% to 0.22 ± 0.02% and from 0.27 ± 0.03% to 0.23 ± 0.02% in the D+I and D-only groups, in the basal and experimental periods, respectively. Triglyceride infusion during the fat clamp caused glycerol levels to increase in the D-only group, leading to greater net hepatic glycerol uptake (Fig. 7C and D) (P < 0.05 between groups). Net hepatic lactate output increased at 60 min, which led to an increase in arterial lactate levels (Fig. 7E and F), with a slight tendency toward a greater rise in net hepatic balance in the D+I group.

During the basal period, two-thirds of HGP was derived from net hepatic glycogenolysis, with net hepatic gluconeogenic flux accounting for the remaining amount (Fig. 7G and H). Under postprandial-like conditions there was a switch from net gluconeogenic to net glycolytic flux in both groups, and this effect was greater in the D+I group at 60 min (Fig. 7G) (P < 0.05). This difference was artificially made greater by increased net hepatic glycerol uptake in D-only, which was responsible for 0.2 of the 0.3 mg/kg/min difference in net hepatic gluconeogenic flux at that time point. By the end of study net hepatic gluconeogenic/glycolytic flux was near zero in both groups. The effect on glycogen metabolism was comparatively much larger, with most of the glucose taken up by the liver being stored as glycogen (Fig. 6F), at a rate not different between groups (Fig. 7H) (3.7 ± 0.7 and 3.7 ± 1.2 mg/kg/min in the D+I and D-only groups, respectively, over the experimental period). Liver phosphoenolpyruvate carboxykinase 1 (PCK1) gene expression was reduced by 97 ± 2 and 80 ± 4% (Fig. 8A) (P < 0.05 between groups and vs. baseline), while glucose-6-phosphatase catalytic subunit 1 (G6PC1) expression was reduced by 31 ± 16% in D+I and increased by 66 ± 0.37% in D-only (Fig. 8B) (P < 0.05 for D+I vs. D-only). Hepatic phosphoenolpyruvate carboxykinase (PEPCK) protein content did not differ significantly from baseline but tended be lower in the D+I compared with D-only group (Fig. 8C).

Muscle insulin signaling increased similarly between groups, with Akt phosphorylation elevated by 20 ± 1-fold and 18 ± 1-fold in the D+I and D-only groups, respectively, compared to baseline (Fig. 9A) (P < 0.05 vs. baseline), and glycogen synthase activated by dephosphorylation (Fig. 9B) (reduced by 69 ± 1 and 72 ± 2%; P < 0.05 vs. baseline). At the end of the study, muscle glycogen content was 9.1 ± 0.8, 10.9 ± 0.5, and 10.1 ± 0.9 mg/g muscle in the baseline, D+I, and D-only groups (Fig. 9C). Akt phosphorylation increased in visceral adipose tissue by 4.4 ± 0.1-fold and 4.1 ± 0.2-fold in the two groups (Fig. 9D) (P < 0.05 vs. baseline).

Individuals with diabetes treated with subcutaneous insulin are exposed to an altered distribution of insulin between the liver and the rest of the body. This results in abnormal glucose disposal (36) and peripheral hyperinsulinemia that is associated with metabolic disease (1–7). It may also alter the mechanisms by which insulin controls hepatic glucose metabolism. We hypothesized that peripheral insulin delivery would cause indirect insulin action (via suppression of FFA and glucagon levels and increased hypothalamic insulin signaling) to become the dominate means by which HGP and HGU are regulated. A modest peripheral insulin infusion rate was used, small enough to only increase hepatic insulin 2.7-fold, while it increased 8.0-fold in the rest of the body.

Several key points emerge. First, previous studies have conclusively shown that insulin can regulate hepatic glucose metabolism through multiple mechanisms. Independent of its direct effects on the liver, insulin can indirectly regulate hepatic glucose metabolism by modulating exposure of the liver to FFAs (8–13) and glucagon (14) and via neural input from the brain (15–18). When insulin enters the body via the physiologic route, however, those indirect effects not only redundant are but also are nonessential, since direct insulin action alone can generate a complete metabolic response (21). The current study extends this finding to demonstrate that even with peripheral delivery, insulin’s indirect effects are dispensable to the control of HGU, at least at the dose of insulin that was administered. This is surprising given that the increase in hepatic insulin exposure via recirculation of insulin was small, but it underscores the exquisite sensitivity and rapid response of the liver to direct insulin action. The change in hepatic glucose flux was driven primarily by insulin’s effect on glycogen metabolism, with minimal change in gluconeogenesis. It should be noted that HGU is controlled by three main factors: insulin, hepatic glucose load, and the portal vein to arterial glucose gradient (portal signal) (42), all of which were elevated in this study to create postprandial-like conditions. Studied in isolation, more subtle indirect effects of insulin might have revealed themselves, but when insulin’s indirect actions were combined with the effects of hyperglycemia and the portal glucose signal, a small direct effect of insulin was sufficient to cause the liver to take up and store marked amounts of glucose.

Second, under the conditions of this study, HGU and HGP occurred simultaneously, but unlike for HGU, the regulation of HGP depended on both direct and indirect insulin action, with faster suppression occurring when insulin’s indirect effects were present. Concurrent fluxes likely reflect heterogeneous hepatocyte responsivity, which may be more likely when hepatic insulin levels are submaximal, allowing differential metabolic responses in periportal and perivenous hepatocytes (43). Several possibilities could explain insulin’s indirect effect on HGP. Previous studies have shown that CNS insulin action is time dependent, requiring changes in gluconeogenic gene transcription and protein translation, taking ≥3 h to become apparent (16–18,44). Thus, this mechanism is less likely to have contributed to the greater suppression of HGP that was present between 60 and 150 min in the D+I group. A fall in hepatic FFA uptake causes an increase in glycolytic flux, shunting carbon away from HGP, causing it instead to be released as lactate (10,20,21). The difference in net hepatic lactate output at 60 min can account for only one-third of the difference in HGP between groups, however, and differences in lactate balance and HGP do not correlate after that. Thus, reduced stimulation of HGP by glucagon is the most likely explanation for the faster suppression of HGP in the presence of insulin’s full effects.

Third, human and large animal studies have shown that the capacity (mg/kg/min) for HGU to be stimulated is two to three times greater than the capacity for HGP to be suppressed. Thus, changes in HGU have the potential to have a greater quantitative impact on net hepatic glucose balance. This is especially important in light of the fact that most of a typical day is spent with the liver taking up rather than producing glucose (45). Independent methods were used for measuring HGP, HGU, and net hepatic glucose balance, and two of the methods were combined to provide an additional assessment of HGU. Regardless of how it was measured, indirect insulin action had negligible net impact on liver glucose metabolism. Hyperinsulinemic-euglycemic clamps are often considered the “gold standard” for measuring hepatic insulin sensitivity (4). While generally appropriate for assessing fasting liver responsiveness, the technique has more limited application when it comes to postprandial conditions. First, it is typically used to measure the quantitatively less important HGP rather than HGU, with potential variation in this important parameter thus ignored. Second, during the clamp insulin is commonly administered via a peripheral vein rather than the portal vein. This leads to relative hepatic insulin deficiency and arterial hyperinsulinemia and, thus, assessment of the liver’s sensitivity to exogenous rather than endogenously secreted insulin. Third, euglycemic clamp assessment fails to consider the effects of an increase in the hepatic glucose load and a change in the portal glucose signal, which are both major drivers of HGU during feeding (42). Therefore, on the basis of hyperinsulinemic-euglycemic clamps one is more likely to conclude that insulin’s indirect effects are important to the regulation of hepatic glucose metabolism. Although some studies suggest that hyperinsulinemic clamps fail to assess potentially important metabolic effects of pulsatile pancreatic insulin secretion, in an elegant study Laurenti et al. (46) recently used state-of-the-art measurements to show that insulin pulse characteristics did not correlate with measures of hepatic and peripheral insulin sensitivity in humans without diabetes. Likewise, loss of insulin pulsatility had no impact on the effectiveness of insulin in the regulation of HGU under postprandial-like conditions in the dog (47).

Previously, we examined the effect of portal vein insulin delivery (1.8 mU/kg/min) in a study (21) where all other conditions were matched with those in the present investigation (plasma glucose level, hepatic glucose load, and portal glucose signal). In that study, hepatic sinusoidal insulin levels were 101 ± 10 µU/mL and the AUC for HGU was 791 ± 89 mg/kg/240 min (21). In the current study, with peripheral insulin infusion, the hepatic insulin level was only 54 ± 3 µU/mL but the AUC for HGU was 802 ± 110 mg/kg/240 min, with similar suppression of HGP between the two studies. Thus, the peripheral route of insulin delivery led to liver insulin exposure that was only half as great (54 ± 3 vs. 101 ± 10 µU/mL) but did not impair the liver’s ability to increase glucose uptake or reduce glucose production. In addition, the absence of indirect insulin action had little to no impact on hepatic glucose metabolism in either of those studies. Therefore, under postprandial-like conditions, the healthy liver can respond to even a small direct rise in insulin such that peripherally delivered insulin can produce a full response. On the other hand, in another postprandial-like study with matching glycemic conditions, the effects of a lower insulin dose (1.2 mU/kg/min) were studied (36). There, hepatic insulin levels were either 63 ± 5 or 38 ± 1 µU/mL, depending on whether insulin was given by the portal or peripheral route, respectively (arterial levels were 20 ± 2 vs. 46 ± 2 µU/mL, respectively). In that case, HGU with peripheral insulin delivery was less than one-half that with the portal route. This reaffirms the inability of indirect action to correct for the loss of insulin’s direct effect. In addition, it demonstrates that the liver’s dose response to direct insulin action is steep and narrow, still responsive at 38 ± 1 µU/mL (twofold basal [36]), but saturated at 54 µU/mL (2.8-fold basal; present study). Clearly, insulin’s direct effect plays both a powerful and vital role in the regulation of hepatic glucose metabolism, with profound liver insulin sensitivity able to compensate for the relative hepatic deficiency of insulin caused by peripheral insulin delivery, as long as there is sufficient recirculation of insulin back to the liver.

The direct effect of insulin increased signaling through the hepatic insulin receptor, Akt, GSK-3β, and glycogen synthase, which led to increases in HGU and liver glycogen storage, independent of indirect insulin action. In contrast, as with glucokinase (18,21,23), full control of hepatic gluconeogenic gene expression required indirect input. FFA and glucagon are known to stimulate liver PCK1 and G6PC1 gene expression, while insulin suppresses those effects (48,49), including via CNS insulin action (16,18). Indeed, suppression of gluconeogenic gene expression was impaired when those changes were not allowed to occur. While glucose inhibits PCK1 expression, it increases G6PC1 expression (48,50); thus, G6PC1 mRNA levels increased when hyperglycemia was only opposed by insulin’s direct effect. Despite differential PCK1 expression between groups, PEPCK protein levels did not change appreciably, and there was no difference in gluconeogenic flux. Thus, although insulin clearly plays a role in the transcriptional control of these enzymes through both direct and indirect mechanisms, our data add to the body of evidence demonstrating that gluconeogenic gene expression, enzyme activity, and flux rates do not necessarily correlate, at least over the short-term (18,51,52). Furthermore, it is important to note that it is the effect of insulin on glycogen metabolism rather than gluconeogenic flux that primarily drives insulin-mediated regulation of hepatic glucose metabolism under postprandial-like conditions. Under fasting conditions insulin might have a relatively greater effect on gluconeogenic flux, especially if glycogenolysis is negligible due to liver glycogen depletion as occurs after overnight fasting in rodents (53).

Despite the ability of peripherally delivered insulin to produce a normal response by the liver, a major consequence of this route of delivery is an abnormal distribution of glucose disposal between insulin-sensitive tissues (primarily liver and muscle under postprandial conditions [35]). Normally the liver takes up as much as one-third of the glucose consumed, thus equaling muscle glucose disposal (54–56). Indeed, previously we found that liver and muscle were responsible for equal amounts of glucose uptake when insulin was infused into the portal vein at 1.2 mU/kg/min (36) (Fig. 10, left panel). In contrast, in the current study, when insulin was delivered into a peripheral vein at 1.35 mU/kg/min, almost fivefold more glucose was disposed of by muscle than liver (Fig. 10, D+I group, middle panel). When insulin’s indirect effects were blocked (Fig. 10, D-only group), the need for exogenous glucose during the clamp decreased by a third due to reduced peripheral glucose uptake, most likely due to the effects of FFA on muscle insulin sensitivity (57–61). Although muscle Akt and glycogen synthase phosphorylation did not differ between groups (in line with matched arterial insulin levels), FFA can affect some insulin signaling pathways but not others (62). Muscle glycogen content increased only marginally, despite robust muscle glucose uptake in both groups, but this is to be expected, since resting muscle is glycogen replete. In considering total muscle mass (63), however, we can estimate that muscle glycogen content increased by ∼25.1 and 14.3 g in the D+I and D-only groups, respectively, in line with the differences in nonhepatic glucose uptake between groups (in comparison, total liver glycogen increased ∼15–18 g). Differences in muscle glucose oxidation and lactate release could also account for some of the difference in muscle glucose uptake between groups.

Not surprisingly, patients treated with traditional insulin therapy face greater daily glucose variability associated with insulin dosing errors and significant risk of hypoglycemia due to the delayed onset and sustained effect of insulin on muscle glucose disposal (64,65). On the other hand, the liver has a rapid on-off response to insulin that is ideally suited to meal-related regulation of liver glucose metabolism (64,65). Appropriate liver glucose uptake is required to help to buffer postprandial glucose excursions, and the resulting liver glycogen stores serve as the major defense against hypoglycemia. Furthermore, peripheral hyperinsulinemia is associated with metabolic and cardiovascular disease, is a key contributor to insulin resistance in type 1 diabetes (4,65–68), and is thought to exacerbate and may even initiate the metabolic syndrome and type 2 diabetes (1–3,5–7). Thus, iatrogenic hyperinsulinemia remains a major barrier to restoring normal physiology in people treated with insulin.

The results of the current study emphasize that the primary issue associated with peripheral insulin delivery is not its inadequate direct effect on the liver but, rather, its excessive effect on peripheral tissues. A normal response from the liver may not occur, however, when the liver is metabolically impaired. Whereas HGU plays a major role in limiting overall hyperglycemia in healthy individuals, impaired HGU is an important contributor to hyperglycemia in individuals with diabetes (45,69–74). Likewise, dogs fed a high-fat and -fructose diet rapidly develop resistance to postprandial HGU (75). Further studies are needed for understanding of the impact of the peripheral route of insulin delivery in the insulin-resistant state.

In summary, peripheral insulin infusion did not impair the healthy liver’s ability to take up or store glucose under postprandial-like conditions. This was not due to enhanced indirect insulin action but, instead, was because of the exquisite sensitivity of hepatic glycogen metabolism to insulin’s direct effects. Nevertheless, the metabolic consequences associated with peripheral hyperinsulinemia (1–7) cannot be avoided when insulin is delivered subcutaneously.

Funding. This work was funded by National Institutes of Health (NIH) grant R01DK018243. Hormone analysis was performed by Vanderbilt’s Hormone Assay and Analytical Services Core, supported by NIH grant DK020593. Surgical and experimental expertise was provided by Vanderbilt’s Large Animal Core, supported by DK020593.

Duality of Interest. A.D.C. has the following potential conflicts of interests: Abvance Therapeutics, Biocon, Diakard/Diabetica, Fractyl, Loyal, Metavention, Novo Nordisk, Sekkei Bio, Senda Biosciences, Sensulin Laboratories, Thetis Pharmaceuticals, and vTv Therapeutics. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. G.K., A.D.C., and D.S.E. designed the studies and wrote the manuscript. G.K., K.C.C., M.Sm., B.F., M.Sc., J.H., and D.S.E. conducted the experiments and acquired and analyzed data, and B.F. performed the surgeries. All authors reviewed the manuscript. D.S.E. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.