Fasting-refeeding acutely activates hepatic Xbp1s and the UPR.
To define the role of Xbp1s in the postprandial adaptation, we performed a time course study of fasting-refeeding. Xbp1s
mRNA increased more than 10-fold within the first hour of refeeding (Figure , A and B), overlapping with its pattern of protein expression (Figure C). This upregulation was transient. By 3 hours of refeeding, Xbp1s levels dropped to near basal values, only to increase again after 8 hours, when the animals started to eat again (Supplemental Figure 1A; supplemental material available online with this article; doi:
). In addition to Xbp1s, other components of the UPR were also activated during the postprandial state. These include the induction of phosphorylation of eIF2α and the transcriptional activation of Bip, Atf6, CHOP, ERdj4, and Edem1 (Supplemental Figure 1, B–D). The ratio of Xbp1s to Xbp1 does not reflect the absolute increase in Xbp1 splicing, since unspliced Xbp1 is continually transcriptionally replenished. Therefore, even though the Xbp1s/Xbp1 ratio is not changed, there is strong reason to believe that IRE1α activation occurs in the postprandial state as well.
Postprandial activation of hepatic Xbp1s in WT mice and Xbp1s induction in LIXs mice.
Inducible overexpression of Xbp1s in LIXs mice mimics postprandial features in WT mice.
The complex nature of the postprandial hepatic response and transient expression of Xbp1s make it difficult to dissect the role of Xbp1s. We therefore developed a strategy to induce Xbp1s in hepatocytes without the need to expose mice to refeeding (Supplemental Figure 1E). We put the Xbp1s
transgene under the control of a tetracycline-responsive element (TRE) to allow inducible expression by doxycycline (Dox) in the presence of a tetracycline reverse transcriptional activator (rtTA). The rtTA transgene is driven by the Rosa26
promoter with a transcriptional stop cassette flanked by 2 loxP sites upstream of rtTA (11
). Combined with an albumin promoter–driven Cre transgene, we obtained a mouse model with inducible hepato-specific expression of Xbp1s
. When LIXs mice were fed a Dox-containing chow diet, Xbp1s
mRNA was readily detected within 24 hours (Figure D). The endogenous full-length Xbp1 mRNA
level dropped at 24 and 48 hours after induction, consistent with the observation that Xbp1s enhances the splicing of its own full-length Xbp1 precursor (12
). In light of the very short half-life of Xbp1s
), the induction of Xbp1s
is readily reversible. After an overnight fast and with no alternative source of Dox provided, the Xbp1s
mRNA completely disappeared, whereas the full-length endogenous Xbp1
mRNA was restored (Figure D). Upon 24 and 48 hours of induction with Dox, the protein levels of Xbp1s in LIXs mice reached levels about 5-fold greater than those in WT mice after 2 hours’ refeeding (Figure , C and E). The functionality of the transgenic Xbp1s was validated as judged by increased expression of Xbp1s target genes in LIXs mice (Supplemental Figure 1F). The LIXs model therefore allows us to acutely activate and maintain the hepatic expression of Xbp1s without the need to expose mice to a fasting-refeeding routine.
We employed metabolic cage studies to test whether Xbp1s expression effectively mimics the systemic effects of refeeding. In WT mice, the respiratory exchange ratio (RER) increased upon refeeding (Figure , A and B). RER reflects the ratio of CO2 release to O2 consumption, indicating the relative contributions of glycolysis and glucose oxidation versus β-oxidation toward whole-body energy consumption. A higher RER indicates increased glycolysis and glucose oxidation. As expected, refeeding of WT animals led to a rapid rise in RER, suggesting that glucose is the preferred fuel source for CO2 production, consistent with the established transition of fuel usage from fatty acids to glucose upon refeeding. Upon induction, LIXs mice show a significantly higher RER than WT mice during the dark phase (Figure C). Moreover, induction was associated with a trend toward increased food intake in LIXs mice (Figure D) and improved body weight recovery at the end of the metabolic cage study (Supplemental Figure 1G). These results suggest that Xbp1s expression in the liver can effectively mimic the global metabolic characteristics associated with refeeding in WT animals.
Hepatic Xbp1s overexpression mimics the metabolic effects of refeeding.
Xbp1s induction leads to rapid biomass accumulation and glycogen depletion.
Along with the systemic effects on whole-body RER, Xbp1s induction also leads to profound changes in the liver. Wet liver mass increased 1.7-fold in LIXs mice within 48 hours of induction (Table ). Hepatic triglyceride (TG) content increased more dramatically relative to cholesterol and protein content. As a result of the increased TG content, LIXs mice show reduced dry mass content per gram wet liver after lipid extraction. This difference (216.98 – 160.06 = 56.92 mg/g wet liver) was comparable to the increase in TG content (55.83 – 5.8 = 50.03 mg/g wet liver). The weight increase was therefore due to TG and not due to edema formation. Importantly, per total liver, LIXs mice had greater dry mass (226.3 mg in WT vs. 275.4 mg in LIXs), with protein content comparable to controls (mg protein/mg dry liver). Taking the enhanced liver mass into account, this highlights an increase in total liver protein in LIXs mice.
Hepatic parameters after switch to Dox-containing diet for 48 hours
In contrast to the increased hepatic lipid and protein content, hepatic glycogen was severely depleted in LIXs mice after 48 hours of induction (Figure A). This suggests that the rapid accumulation of biomass (1.7-fold in 2 days) results from increased synthesis of macromolecules at the expense of hepatic glycogen/glucose. To test whether the reduced glycogen levels in LIXs mice were due to decreased glucose uptake, we performed immunofluorescence staining of GLUT2, the major glucose transporter in hepatocytes. There was no change upon 24 hours of Dox exposure, whereas LIXs hepatocytes showed an enlarged volume with increased surface staining for GLUT2 signal after 48 hours of induction (Figure B). The enhanced levels of GLUT2 argue against a defect in glucose uptake. In fact, the metabolic cage studies indicate higher glucose utilization in LIXs animals (Figure C). From a more global perspective, this suggests that one of the roles for endogenous Xbp1s in postprandial hepatic remodeling is to coordinate a broad anabolic program in the hepatocyte. Upon Xbp1s overexpression, however, cellular glucose is excessively consumed for these anabolic reactions, and as a result, hepatic glycogen storage is depleted.
Hepatic Xbp1s overexpression triggers hypoglycemia.
Xbp1s induction diminishes hepatic glucose release, which leads to hypoglycemia and triggers lipolysis from adipocytes for hepatic TG accumulation.
Along with the depletion of liver glycogen, the serum glucose levels started to decrease by 72 hours after induction in LIXs mice under fed conditions (Figure C). Moreover, a 6-hour fast caused severe hypoglycemia in LIXs mice within 48 hours after induction. The fed insulin levels started to decrease upon induction and were significantly lower in LIXs mice by 72 hours of induction (Supplemental Figure 2A). A 6-hour fast at that stage still reduced insulin levels in LIXs mice and WT mice to comparable levels. However, 96 hours after induction, fasted insulin levels were even lower in LIXs mice (Supplemental Figure 2B). These results suggest that LIXs mice have impaired adaptation to fasting, presumably due to reduced hepatic glucose release. In WT mice, fasting-induced fatty liver associated with increased TG content, but only a marginal increase in cholesterol (60 mg/g for TG vs. 0.3 mg/g for cholesterol, Supplemental Figure 2C). The fatty liver phenotype observed in LIXs mice (Table ) is thus reminiscent of the changes in hepatic lipid content during fasting. Lipolysis in adipocytes increases as a response to reduced hepatic glucose release and associated low insulin levels. The fasting effects of adipocyte lipolysis on hepatic TG content are further exacerbated in LIXs mice. Indeed, a 6-hour fast significantly increased liver TG content in LIXs mice acutely, within 24 hours of Dox exposure (Figure D). Moreover, we detected a significant increase in serum free fatty acids in the fed state after 72 hours of induction (Figure E). Consistent with the increase in lipolysis, LIXs mice displayed a gradual reduction in the volume of adipose tissue, whereas an increase in hepatic lipid by CT scanning was apparent (Supplemental Figure 2D). As a control, when the mice were switched back to regular chow diet without Dox, the liver steatosis decreased rapidly, and the fat tissue volume increased correspondingly in LIXs mice. No significant changes in liver lipid or fat tissue volume were detected in WT control mice under the same conditions, indicating that the effects are specific for Xbp1s induction rather than indirect through Dox treatment.
Mammals maintain serum glucose levels within a very narrow range. This is mainly achieved through regulation of hepatic glucose release. We therefore examined hepatic gluconeogenesis using pyruvate tolerance tests. Dox exposure for 48 hours caused a reduction in glucose release in LIXs mice (Figure F). Impaired β-oxidation could be a possible mechanism, as it provides essential precursors for gluconeogenesis. Moreover, β-oxidation defects have been considered to be a feature associated with the UPR (14
). The transcriptional levels of β-oxidation and gluconeogenic genes were indeed decreased in LIXs mice (Supplemental Figure 2, E and F). To address whether reduced β-oxidation is a causal factor, we measured the actual β-oxidation potential biochemically in LIXs animals. Surprisingly, we found a similar partitioning of radioactivity in LIXs and control mice for up to 48 hours of Dox induction (Supplemental Figure 2G). This suggests that the β-oxidation capacity in LIXs mice does not substantially differ from that in WT mice. Furthermore, administration of a PPARα agonist (15
) exacerbated rather than rescued the hypoglycemic phenotype (Supplemental Figure 2H), strongly arguing against β-oxidation as a causal factor for the defective glucose release. In fact, G6pc, catalyzing the final step of glucose release, is downregulated in ob/ob
mice by Xbp1s directly and indirectly through FoxO1 (6
). However, the hypoglycemia observed in LIXs mice is unlikely to be solely a consequence of the suppression of G6pc, since hepatic G6pc deficiency is commonly associated with increased
glycogen storage in hepatocytes, not with the reduced
levels that we observe under the conditions examined here (Figure A and ref. 16
Insulin action as the master regulator of hepatic glucose release in ob/ob
mice is enhanced by Xbp1s (7
). This was also observed in the LIXs mice, as LIXs livers showed a faster response to insulin exposure (Supplemental Figure 2I). Furthermore, the mice displayed an improvement in insulin sensitivity as indicated by insulin tolerance tests (Supplemental Figure 2J). In isolated hepatocytes, Xbp1s induction led to an enhanced insulin response, indicating a cell-autonomous effect of Xbp1s (Supplemental Figure 2K). LIXs mice therefore developed severe hypoglycemia along with a significant increase in hepatic biomass and hepatic insulin sensitivity. This enhanced local insulin sensitivity may in fact be a key mediator of the increase in the hepatic anabolic programs. Xbp1s activation therefore mimics a “permanently fed state.”
GalE, regulated by Xbp1s, links the UPR and postprandial remodeling.
Based on the analysis above, we performed a microarray screen for novel target genes of Xbp1s. These genes are expected to show a significant increase in mRNA levels in LIXs mice after induction, and upregulation in WT mice upon refeeding. GalE was one of the most highly upregulated among the genes that fulfilled both of these criteria (Supplemental Table 1).
encodes the enzyme uridine diphosphate galactose-4-epimerase. It is the final and key enzyme of the Leloir pathway and acts in concert with GalK and GalT for utilization of dietary galactose (Supplemental Figure 3A and ref. 9
). GalE catalyzes the interconversion of UDP-galactose (UDP-Gal) and UDP-glucose (UDP-Glc), as well as a pair of larger substrates, UDP-GalNAc and UDP-GlcNAc. GalE is the rate-limiting enzyme for cellular production of UDP-Gal and UPD-GalNAc (10
). The microarray data suggest GalE as a key mediator of Xbp1s action, linking the UPR with cellular glucose metabolism, as UDP-galactosyl and UDP-glucosyl groups are critical substrates for both protein and lipid glycosylation in biosynthetic pathways and regulation of protein folding in the ER (17
Consistent with the microarray analysis, GalE expression was drastically increased in LIXs mice regardless of caloric influx, as assessed by both qPCR and immunoblotting (Figure , A and B). Importantly, this is not a reflection of a general upregulation of the Leloir pathway. GalK, the first enzyme of the pathway, catalyzing the phosphorylation of galactose, remained unchanged in the 24-hour-fed LIXs mice and was significantly downregulated at later time points or under fasting conditions (Figure C), reflecting a compensatory adaptation to the increased GalE expression.
Hepatic Xbp1s overexpression induces GalE.
To assess the relationship between GalE and Xbp1s under normal physiological conditions, we first analyzed the effect of refeeding on GalE levels. We subjected WT animals to fasting of different durations (0, 6, 12, and 24 hours), followed by 2 hours of refeeding. Refeeding induced significant increases in Xbp1s expression in mice fasted for 12 and 24 hours (Figure , A and B). Under these conditions, GalE transcription mirrored the pattern of Xbp1s expression, a phenomenon not observed for galK. We then measured galE expression in LIXs mice after induction by Dox gavage in combination with fasting-refeeding. Xbp1s was robustly increased with a Dox gavage in LIXs mice (Figure , C and D). The regimen of 6-hour fasting followed by 2 hours of refeeding with regular chow increased galE expression significantly in LIXs mice, whereas no difference was observed in WT mice. In line with the greater increase in Xbp1s upon refeeding in 24-hour fasted mice, GalE expression reflected the same pattern (Figure D). Again, this behavior was not observed for GalK. Interestingly, the expression levels of Bip and ERdj4 were comparable in WT and LIXs mice in fasted groups under all conditions, and refeeding led to comparable changes regardless of genotype (Supplemental Figure 3B). These results indicate that the rapid increase in GalE transcripts is unlikely to be a secondary event. No changes in liver TG content, serum glucose levels, or liver acylcarnitine composition were detected between WT and LIXs mice under these conditions (Supplemental Figure 3, C and D), suggesting that the observed GalE upregulation occurs prior to the development of the LIXs phenotypes.
Hepatic GalE expression associates with Xbp1s induction in vivo.
Since ER stress has been implicated in the development of obesity-related metabolic dysfunction (7
), we assessed GalE expression in a high-fat diet–induced (HFD-induced) obese mouse model as well as in leptin-deficient ob/ob
animals. Both conditions were associated with elevated Xbp1s (Figure E). GalE
was upregulated 4-fold in the 10 week HFD-fed mice and doubled in ob/ob
mice (Figure F). Thus, GalE
is upregulated under pathological conditions associated with obesity tied to constitutive Xbp1s activation.
To dissect whether the activation of GalE by Xbp1s was cell autonomous, we infected primary mouse hepatocytes with a lentivirus overexpressing Xbp1s. GalE
was significantly upregulated in Xbp1s-infected cells, whereas the lipogenic genes Fasn
were not altered (Figure A). In another tissue culture model, tunicamycin treatment of Huh7 cells (a human hepatoma cell line) was sufficient to stimulate Xbp1s
but failed to induce GalE
(Figure B). However, when insulin was supplemented to mimic a postprandial setting, GalE
expression was significantly increased. This is consistent with the observation that Xbp1s functions in the postprandial state when insulin is elevated. In fact, WT mice refed after a 24-hour fast showed a greater increase in GalE
expression than LIXs mice refed after a 6-hour fast (Figure D). This is in contrast to the Xbp1s
mRNA expression, as WT mice refed after a 24-hour fast showed a smaller increase in Xbp1s
expression than LIXs mice refed after a 6-hour fast (Figure D). These results reflect an additive effect of insulin on GalE
activation by Xbp1s
, presumably through enhanced nuclear translocation (4
). Indeed, the insulin levels in WT mice refed after a 24-hour fast were much higher than in LIXs mice refed after a 6-hour fast (Figure C).
GalE expression is upregulated by Xbp1s in vitro.
To test whether Xbp1s is not only sufficient but required for GalE activation, we performed siRNA-based knockdown of Xbp1s in isolated primary hepatocytes. GalE expression was significantly reduced under these conditions, indicating that Xbp1s is necessary for galE expression (Supplemental Figure 3E). Although tunicamycin induced Xbp1s expression, Xbp1s-mediated GalE activation was specifically blunted. This is intriguing, as, in fact, tunicamycin is an analog of UDP-GlcNAc. It inhibits the enzyme GlcNAc-1-phosphotransferase, which transfers GlcNAc-1-P from UDP-GlcNAc to membrane-bound Dol-P, forming GlcNAc-P-Dol, the N-glycan precursor synthesized on the cytoplasmic face of the ER. Endogenous UDP-GlcNAc therefore potentially serves as a potent inhibitor of Xbp1s-mediated activation of GalE. In contrast, when thapsigargin, another ER stress inducer, was used, both Xbp1s and GalE increased (Supplemental Figure 3F).
To test whether Xbp1s directly regulates GalE expression, we analyzed the promoter of GalE
and found a highly conserved binding site for Xbp1s
across species: TCCACGTC (Figure A and ref. 18
). A luciferase assay with the GalE
promoter containing the potential binding site showed dose-dependent upregulation by Xbp1s (Figure B). When the potential Xbp1s-binding site was mutated or deleted, the luciferase activity was reduced by 50%, indicating this site is critical for activation by Xbp1s, although other binding sites may also exist (Figure C). By ChIP assays, we found that hepatic Xbp1s is associated with the GalE
promoter in WT mice 2 and 6 hours after refeeding. In contrast, Xbp1s protein was not detected on the GalE
promoter of fasted mice (Figure D). Together, our results indicated a strong temporal correlation between Xbp1s induction and GalE expression both in vivo and in vitro. Importantly, the increase in GalE expression occurs prior to most of the other phenotypic and gene expression changes observed in LIXs mice. This suggests that GalE could be a major mediator for Xbp1s-induced early- and long-term metabolic consequences in LIXs mice. Considering the functional involvement of GalE in the interconversion of galactosyl and glucosyl groups, we propose that GalE plays a critical role in mediating Xbp1s action on both glucose metabolism and the UPR.
In contrast to GalE, the lipogenic genes did not show a significant increase in LIXs mice under fed conditions, and in fact displayed a global decrease with fasting (Supplemental Figure 4A). Acute induction via Dox gavage, illustrated in Figure C, also did not cause a significant increase in the lipogenic program in LIXs mice with refeeding (Supplemental Figure 4B). Importantly, a biochemical assay for de novo lipid synthesis showed that LIXs mice maintained the same rate of fatty acid and cholesterol synthesis as controls (Supplemental Figure 4C). Our data therefore do not support Xbp1s as a direct transcriptional regulator of the lipogenic program under these conditions.
GalE, regulated by fasting-refeeding, provides a novel target for therapeutic intervention for metabolic dysfunction.
Although GalE is highly conserved from bacteria to mammals, little is known about its transcriptional regulation. We found that GalE expression was regulated by fasting-refeeding in WT mice: downregulated during a fast and upregulated upon refeeding (Figure A). Thus, we propose a model to explain the contributions of the Xbp1s/GalE axis in the context of refeeding of WT animals in comparison to the “pseudo-refeeding” status in LIXs mice (Figure B). In WT mice, refeeding activates Xbp1s, which in turn upregulates GalE transcription. This moderate increase in GalE during the first 1–3 hours of refeeding enhances the conversion of UDP-Gal to UDP-Glc to accelerate the restoration of hepatic glycogen by assimilation of diet-derived galactose. Consistent with this model, within 2 hours of refeeding, hepatic glycogen levels were fully restored in overnight-fasted mice (data not shown). After glycogen is replenished, the continuous increase in GalE then interconverts UDP-Glc and UDP-Gal to facilitate protein and lipid biosynthesis, which also competes for intracellular glucose and reduces its release from hepatocytes. However, in the LIXs mice, GalE transcription is maintained at high levels due to sustained Xbp1s activation, which, together with the increased anabolic rate in liver, constantly consumes cellular glucose 6-phosphate (G-6-P). As a result, hepatic glucose release is drastically suppressed, and remaining glycogen is broken down. Responding to the low systemic glucose levels in LIXs mice, adipocyte-mediated lipolysis is enhanced, which leads to the acute accumulation of hepatic TGs at high levels.
Regulation of GalE by fasting-refeeding.
According to this model, GalE is one of the critical regulatory factors for the partitioning of glucose toward biosynthetic activity and away from hepatic glucose release. To test this, we injected an adenoviral preparation to achieve GalE overexpression in ob/ob mice, a model with excessive hepatic glucose production. In an oral glucose tolerance test, GalE adenovirus–infected ob/ob mice showed a significant improvement in glucose tolerance compared with the controls (LacZ adenovirus or no-virus group) (Figure A). The improvement was due to reduced hepatic glucose release, since the GalE adenovirus–infected mice also showed significantly reduced glucose release during a pyruvate tolerance test (Figure B). Similarly, in mice fed a HFD for 10 weeks, GalE adenoviral infection significantly lowered fasting glucose levels (Figure C).
Hepatic GalE overexpression improves insulin sensitivity.
We wanted to examine whether GalE activation causes phenotypic changes in lean WT animals kept on a chow diet (conditions comparable to experiments done in LIXs mice). We therefore injected the adenoviral preparation into WT mice. While an oral glucose tolerance test showed no difference in glucose clearance (Figure D), insulin sensitivity was significantly enhanced in GalE-infected mice (Figure E). This suggests that GalE overexpression alone can potently reduce hepatic glucose release even in WT animals. GalE overexpression, however, is insufficient to cause hypoglycemia in lean mice maintained on a chow diet. This is possibly due to the systemic ability to adjust insulin release, i.e., the GalE-infected group reduced insulin release during the glucose challenge to offset the effect of reduced hepatic glucose release. Indeed, we found that the insulin levels in GalE-infected mice were not only lower at baseline, but also remained so over the entire course of the oral glucose tolerance test (Figure D). Furthermore, we were examining a loss-of-function phenotype in these experiments (loss of hepatic glucose output). In contrast to our genetic Dox-inducible model, which potentially affects every hepatocyte, adenoviral infection is never 100% efficient, even at a high MOI, so that there is always a substantial number of hepatocytes remaining unaffected and responsive of compensation for euglycemia in WT mice.
GalE overexpression increases the rate of interconversion between UDP-glucosyl and UDP-galactosyl groups. If products are constantly removed from the equilibrium, the flux from G-6-P to UDP sugars can be maintained at high rates in hepatocytes. This will ultimately lead to hypoglycemia in LIXs mice. One major pathway for the consumption of UDP sugars is protein glycosylation. We therefore examined the levels of the N-linked glycans on hepatic proteins. LIXs mice showed an increase in N-glycosylation per gram protein in the fed state (Figure A) but no change in the fasted state (data not shown). Moreover, when cellular nucleotide sugars were analyzed, LIXs mice showed increases in cellular levels of UDP-Glc, UDP-Gal, UDP-GlcNAc, and UDP-GalNAc (Figure , B and C). In the fasting state, the LIXs mice retained higher levels of UDP-GlcNAc and UDP-GalNAc than controls, whereas the UDP-Glc and UDP-Gal levels were the same (Supplemental Figure 5A). These data suggest that GalE induction shunts more glucose into the pool of UDP sugars for anabolic reactions. Additionally, these observations also provide a mechanism for the synergistic effects of the activation of β-oxidation by fenofibrate in LIXs mice: the more acetyl-coA generated through β-oxidation, the more rapid the depletion of glucose due to the conversion to acetylated hexosamines.
GalE overexpression enhances biosynthesis of nucleotide sugars and protein glycosylation.
Beyond the increased N
-glycans in fed LIXs mice, we also found an increase in three single-sugar moieties from protein O
-glycosylation, including GlcNAc, Glc, and Man (Supplemental Figure 5B). The LDL receptor is critically dependent on O
-glycosylation for proper folding and function, with both Gal and GalNAc involved (19
). Indeed, we found that the LIXs mice displayed strong hepatic staining for the LDL receptor after a 24-hour Dox exposure (Figure D). Furthermore, siRNA-mediated knockdown of galE
in isolated primary hepatocytes led to a reduction in cellular levels of LDL receptor (data not shown). Consistent with the role of Xbp1s as an anabolic factor, induction of Xbp1s
in primary hepatocytes increased protein synthesis (Figure A). siRNA-mediated knockdown of GalE caused a lower rate of protein synthesis in both WT and LIXs hepatocytes (Figure B). A possible mechanism for this reduced global protein synthesis is an impaired response to insulin. Indeed, a GalE siRNA-mediated knockdown reduced the insulin response in hepatocytes (Figure C). In contrast, overexpression of GalE improved insulin signaling (Figure D). Moreover, to test whether GalE is sufficient to enhance protein synthesis, we examined the effects of GalE overexpression on adiponectin production in a HEK293T cell system. When a GalE expression construct was co-transfected with an adiponectin expression construct, adiponectin protein levels significantly increased in cells (Figure E), with no significant change in mRNA level and adiponectin secretion rate (data not shown). These results suggest that GalE may be a rate-limiting factor for protein synthesis, presumably through enhanced supply of UDP sugars for glycosylation and increased ER capacity for protein folding.
GalE regulates protein synthesis and cellular response to insulin.