Physiology & Behavior 81 (2004) 375 388

Classical conditioning and conditionability of insulin and glucose effects in healthy humans
Ursula Stockhorsta,*, Nicola Mahla, Maren Kruegera, Anja Hueniga, Yolanda Schottenfeld-Naorb, Achim Huebingerb, Hans-Walter Berresheimc, Hans-Joachim Steingruebera, Werner A. Scherbaumb
a

Institute of Medical Psychology, Heinrich-Heine-University Duesseldorf, P.O. Box 101007, D-40001 Duesseldorf, Germany b German Diabetes Center at the Heinrich-Heine-University, Duesseldorf, Germany c Medical Institute of Environmental Hygiene, Heinrich-Heine-University, Duesseldorf, Germany Received 21 February 2003; received in revised form 21 October 2003; accepted 16 December 2003

Abstract We examined whether the effects of intravenously injected insulin and glucose (the physiological endogenous insulin production stimulus) could be classically conditioned in healthy humans. We expected a conditioned blood glucose decrease to a conditioned stimulus (CS) previously paired with insulin and an, albeit lower, blood glucose decrease to a CS paired with glucose injection. In addition, we analyzed glucoregulatory hormone and symptom conditionability. Thirty healthy males were divided into three groups and were given the CS and an intravenous injection of either insulin (0.05 IU/kg) in Group 1, glucose (15%, 0.5 g/kg) in Group 2, or placebo [physiological saline (0.9%)] in Group 3 during the acquisition phase on 4 days. All participants were given the olfactory CS (rosewood peppermint smell) and placebo injection on Day 5 (test). On Day 5, the total blood glucose decrease tended to be higher in Group 1 than in Group 3 ( P < .10), especially at CS presentation ( P < .10) and previous unconditioned hypoglycemia time-point ( P < .05). The conditioned blood glucose decrease was statistically nonsignificant in Group 2, but shortly after CS presentation, insulin level and blood glucose changes were negatively correlated in Groups 1 and 2 in contrast to positive correlation in Group 3. Furthermore, Group 1 showed an increase in noradrenaline ( P < .05), a temporarily delayed increase in growth hormone (GH; P < .05), and an increase of autonomic and neuroglycopenic symptoms, reaching a medium and small effect size, respectively. Group 2 responded with an increase in cortisol ( P < .01) and neuroglycopenic symptoms ( P < .05) at the time-point of the previous unconditioned blood glucose minimum. To conclude, the effects of exogenously applied insulin can be conditioned in a reliable way. In correspondence with the lower intensity of the unconditioned stimulus (US), conditioning effects with glucoseand, thus, endogenously produced insulinare weaker but also reflect the actions of central insulin. Future studies will examine the diverse actions of insulin within the brain further. D 2004 Elsevier Inc. All rights reserved.
Keywords: Insulin; Glucose; Classical conditioning; Humans; Glucoregulatory hormones; Central nervous system

1. Introduction The classical conditioning of drug effects is an important basic research area with practical implications. Drugs usually alter homeostasis. Learning (such as classical conditioning) can be regarded as a necessary control process in the regulation of physiological systems in general [1]. The existence of anticipatory conditioned responses (CRs) helps

* Corresponding author. Tel.: +49-211-81-15349; fax: +49-211-8113015. E-mail address: ursula.stockhorst@uni-duesseldorf.de (U. Stockhorst). 0031-9384/$ see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.physbeh.2003.12.019

the organism to minimize the perturbations such as those induced by drugs or food intake [2]. Classical conditioning always requires the CNS. In this sense, insulin is a very interesting substance for studying the classical conditioning of drug effects: Insulin secretion, in response to increasing blood glucose levels (and other metabolic fuels), is not only a direct action on cells that would occur in vitro, but is also a neurally (vagally) mediated response [1]. Peripheral insulin reaches the brain via a saturable blood-to-brain transport system [3 5] and is detected by central insulin receptors mainly localized in the olfactory bulbs, hippocampus, areas of the hypothalamus, and in the lower brain stem. Insulin therefore fulfills the properties of an unconditioned stimulus

376

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

(US) [1]it is an afferent signal to the brain. The detection of insulin by the brain is the true US, which triggers a vagally (and thus neurally) mediated insulin secretion and blood glucose decrease as the unconditioned response (UR) [1]. The brain also detects peripheral blood glucose level, such as insulin-induced hypoglycemia or glucose-induced hyperglycemia, by central glucose-sensing neurons, mainly localized in the hypothalamus [6], and initiates hormonal counterregulation to restore blood glucose levels to normal [1,6]. However, conditioning insulins effects has so far nearly exclusively been examined in animals [6], using exogenous insulin and measuring the CR in blood glucose only. In our first placebo-controlled experiment with healthy humans [7], we demonstrated a conditioned decrease of blood glucose, an anticipatory insulin increase, as well as conditioning effects in cortisol and neuroglycopenic symptoms after participants had learned an association between an olfactory conditioned stimulus (CS) and exogenous insulin injection (US). A rise of glucose in the blood is the physiological, and thus natural, stimulus for endogenous insulin secretion. Thus, it is interesting to know whether the organism can also acquire a conditioned insulin increase and a subsequent conditioned blood glucose decrease after glucose administration and its resulting endogenous insulin release. So far, two approaches have been used in examining conditioning effects using glucose: first, studies addressing cephalic phase insulin release (CPIR), and second, a few real conditioning studies pairing an arbitrary CS with glucose injection. CPIR is the release of insulin in the preabsorptive phase of a meal after food presentation. CPIR, like the conditioned blood glucose decrease after exogenous insulin application [6], is vagally mediated, but its interpretation as a classically CR is restricted: In most studies, foodassociated stimuli that had been experienced under real-life conditions were tested to evoke a CPIR. A pure conditioning effect can only be demonstrated by using a previously neutral stimulus paired with food or orally presented glucose, examined in a few animal experiments [8 11] and only one conditioning experiment in humans [12]. In that study, a previously neutral stimulus (peppermint flavor and fragrance), as the CS, was paired with a glucose-loaded drink (US) in the experimental group and an aspartamesweetened drink in the control group. In the test, when all participants were administered the CS and the aspartame drink, the experimental group showed a marginal, but significant, conditioned insulin increase, although no conditioned blood glucose decrease. However, to avoid gastrointestinal peptides influencing the insulin secretion, glucose has to be injected (not consumed orally). To our knowledge, only five animal conditioning studies have injected glucose, three with very small samples, and intragroup comparisons only (Sedina, cited in Ref. [13] and Refs. [14,15]) showing some evidence for a blood glucose decrease when saline injections were given in the context (CS) of a previous glucose injection. Of the two well-controlled experimental studies with independent groups [16,17], only one [17]

revealed a conditioned blood glucose decrease after CS glucose pairings but lower than after CS insulin pairings. Consequently, controlled experiments with intravenous glucose were required, especially in humans. We therefore conducted a three-group design where healthy participants in the experimental groups learned a CS US contingency: pairing of CS and intravenous injection of insulin in Group 1 and pairing of CS and intravenous glucose in Group 2. In the Control group 3, CS and placebo were administered (see Method). We hypothesized that previous CS insulin (Group 1) and CS glucose (Group 2) pairings, to a minor extent, would result in a conditioned insulin increase and blood glucose decrease when given CS and placebo injection (test). For exploratory purposes, we also examined the conditionability of glucoregulatory hormones. In Group 1, we expected a conditioned increase in glucagon, catecholamines, cortisol, and growth hormone (GH) to mimick the pattern of the unconditioned (neurally mediated) counterregulation [7]. In Group 2, the (neurally mediated) activation of glucose-lowering mechanisms, mainly via endogenous insulin release, is the actual response. This should initially be accompanied by a decreased secretion of the glucoregulatory hormones, followed by an, albeit low, hormonal increase when blood glucose level approaches its minimum after endogenous insulin rise [18]. Consequently, we predicted an initial decrease and then increase of glucoregulatory hormones as the CR in Group 2. Subjective symptoms should respond in the same way, with a conditioned increase in neuroglycopenic and autonomic symptoms in Group 1, and an initial decrease, then, an increase in Group 2.

2. Method 2.1. Design We used a three-group design conducted under doubleblind conditions. Groups differed by injection substance on conditioning protocol Days 1 to 4 (acquisition phase) and, thus, in CS US pairings. After CS presentation, the participants of Group 1 (CS INS) were injected with insulin intravenously (human insulin Actrapid, NovoNordisk, Mainz, Germany) at 0.05 iU/kg body weight (BW) diluted with NaCl (0.9%, Braun, Melsungen, Germany). Participants of Group 2 (CS GLUC) were given an intravenous injection of 15% glucose solution mixed from 10% and 20% glucose solution (Delta Farm) at 0.5 g/kg BW. Group 3 (CS NaCl) was injected physiological saline intravenously (0.9% NaCl, Braun, Melsungen, Germany). On Day 5, the participants of all three groups received the CS and physiological saline injection. The design was approved by the Ethics Committee of the Medical Faculty of the HeinrichHeine University, Duesseldorf. The participants signed for informed consent. A large injection volume was needed [volume (in ml) = BW (in kg) 3.33] to reach the glucose dose (0.5 g/

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

377

kg) and concentration (15%) required. It was used for all three groups to ensure double-blind conditions. Mean injection volume was 247.8 F 4.4 ml. The main injection volume consisted of physiological saline in Groups 1 and 3 and glucose solution in Group 2. Sessions were scheduled every 48 h (Tuesday, Thursday, Saturday, Monday, and Wednesday). We used the same olfactory CS and application procedure as in our previous insulin conditioning experiment [7]a mixture of 2.1 ml rosewood oil (Oleum Rosea Ligni Brasil) and 1.4 ml peppermint oil (Oleum mentae peperitae, 10%, both distributed by Caelo, Caesar and Loretz, Hilden, Germany) mixed in 50 ml water. So as not to affect the room air, the CS was applied through silicon tubes ending directly in the nostrils, secured by nose goggles (see Ref. [7] for details). The airflow was initialized electronically 1 min before injection, ending 6 min later. 2.2. Participants The participants were 30 healthy 20- to 30-year-old male students (10 participants per group) responding to a student announcement on the university campus for healthy male nonsmokers of normal BW. The mean age ( F S.E.M.) was 25.2 F 0.9 in Group 1, 24.4 F 0.8 in Group 2, and 25.0 F 1.0 in Group 3. Body mass index (BMI) was within normal range in all three groups [Group 1: M F S.E.M. = 22.2 F 0.6 kg/m2; Group 2: 22.3 F 0.5 kg/m2; Group 3: 22.6 F 0.5 kg/ m2). To guarantee a physiological response to the glucose injection in Group 2, the participants had to consume at least 200 g carbohydrates per day as inquired by interview. Smokers [19] and athletes [20] were excluded. Medical exclusion criteria (checked by interview and physical examination) were diabetes (also parents and siblings), endocrine and neurological diseases, cardiovascular disorders, substance abuse, and medication with glucocorticoids. Before participation, the participants were informed about the study, in particular, about the time schedule, the volume of blood taken per session, and possible risks. All participants (future insulin, future glucose and future saline) took part in a single-session prestudy. They were randomly assigned to insulin (0.05 iU/kg) or glucose (15%, 0.5 g/kg) injection in the same environment as the main study, but no olfactory cue was given. The prestudy was done to reduce any putative novelty-stress effect (and thus blood glucose increase) resulting from the unfamiliar environment [21]. Furthermore, we wanted to identify those participants that did not reach an expected blood glucose minimum of at least 50 mg/dl after insulin injection or did not respond with a sufficient insulin secretion after glucose injection, or rejected participation in the main study due to side effects (such as extreme hypoglycemia or venipuncture problems) or personal objections. Of the 53 candidates, 21 were thus excluded. Of the remaining 32 participants, 2 participants dropped out during the main study (due to extreme side effects and personal reasons). Thus, 30 participants (10 participants per group) finally took part. The mean

interval between prescreening and conditioning sessions was 87.5 ( F 8.2) days because the participants endocrine variables (blood glucose, insulin, catecholamines, cortisol, and GH) had to be analyzed before starting the main study. The participants were instructed to fast overnight (at least 10 h, no food, no beverages, except for water or unsweetened herbal tea) and avoid alcohol during all 5 days starting from the evening prior to the first session and physical exercise or sports on the morning of each experimental session. Payment was DM 300 at the end of the last session. 2.3. Experimental procedure Sessions were at 7:45 a.m. for 5 days, separated by 48 h (Tuesday, Thursday, Saturday, Monday, and Wednesday). Fig. 1 summarizes each sessions timeline. Before each session, one in-dwelling catheter (Vasofix, 18 G, 45 mm, diameter 1.2 mm) was inserted into each arm vein, with the dominant arm for injection and the other for blood sampling. The experimental procedure started at 8:00 a.m., with two participants separated by 10 min per session. The sequence of events per session (below) was presented by computer. Each session lasted 128 min (plus approximately 3-min injection time)a 44-min preinjection baseline phase (min 44 to 0 in Fig. 1) and an 84-min postinjection phase. The first 20 min (min 44 to 24) served as the resting period, where participants could read general-interest magazines (e.g., travel magazines). Twentyfour minutes before injection, the first blood sample was taken for analysis of blood glucose, insulin, glucagon, catecholamines, cortisol, and GH. Venous blood samples for glucose analyses were drawn at least every 6 min, with a total of 20 samplings, 4 before (min 24, 18, 12, and 6) and 16 after CS (at min 0 immediately before injection, at min 1, 3, 6, 10, 12, and then every 6 min until min 72). Insulin was sampled nine times (min 24, 0, 1, 3, 10, 24, 30, 48, and 72), and all other hormones four times (min 24, 24, 48, 72) per session. Symptoms were recorded six times (min 24, 12, 24, 36, 48, and 72; [7]). 2.4. Dependent variables The dependent variables were blood glucose, serum insulin, plasma glucagon, plasma adrenaline and noradrenaline, serum cortisol, serum GH, and symptoms. Blood glucose was determined from venous hemolysated blood samples (each 20 Al, filled into a 1.5-ml plastic container with a hemolysis inhibitor) using an Eppendorf EPOS 5060 analyzer and an enzymatic hexokinase method. Blood glucose analysis was performed in duplicate, with the mean of each two measurements as the result. The mean coefficient of retest reliability (average of 20 blood samples per group) ranged between 0.91 and 0.99 for Days 1 to 5. Standard measurement error per day reached mean values (M) between 1.5 and 2.1 mg/dl over the 5 days. Blood samples for the later determination of insulin and counter-

378

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

Fig. 1. Time-line of each session. 20 Al blood sampling every 6 min, more tightly (min 0, 1, 3, 6, and 10) around injection. Larger blood sampling (20 ml) at min 24 ( 1), 24 ( + 1), 48 ( + 2), and 72 ( + 3) for the determination of glucagon, catecholamines, cortisol, and GH. I: blood sampling for the measurement of serum insulin; S: symptom list in the baseline (S 1) and five times after injection (S1 to S5). The CS (smell) started at min 1 and lasted 6 min. Injection was administered in min 0, immediately following the blood sampling in min 0. The participants were run in the morning between 8:00 and 10:08 a.m. (two participants separated by 10-min lag).

regulatory hormones were stored on ice during the session. After each session, the samples were immediately centrifuged (catecholamines: 10 min, 4 jC; all other hormones: 10 min, room temperature) and frozen at 20 jC until assay. Serum insulin, serum glucagon, and plasma catecholamines were determined as previously described [7] with a commercially available microparticle enzyme immunoassay (MEIA, Abbott, Wiesbaden, Germany) to measure human insulin, radioimmunoassay (RIA) for plasma glucagon, and high pressure liquid chromatography (HPLC) for catecholamines in EGTA-plasma using a Beckman System Gold Autosampler 507, a Pharmacia Beckman Electrochemical Detector, and commercial kits (Chromsystems, Munich, Germany). Serum cortisol was determined by enzyme-linked immunoassay (ELISA) from DRG (purchased at BioSource, Ratingen, Germany), with interassay CV < 9.0% for values (M) ranging from 6.3 to 55.6 Ag/dl. Serum GH was measured by an enzyme-amplified sensitivity immunoassay (EASIA, Medgenix Diagnostics, Fleuris, Belgium), with an interassay CV of 6.8% for a mean of 4.1 F 0.28 mU/l and 7.1% for 15.5 F 1.1 mU/l. We used the same symptom list as before [7] for 19 symptoms (headache, palpitation, need to urinate, thirst, hunger, weakness, sweating, fatigue, dizziness, irritation, reduced concentration, anxiety, nervousness, trembling, blurred vision, nausea, salivation, breathing difficulty, and tingling around lips) to be identified and rated by the participant for actual intensity on a five-point numerically graded (0 4) and verbally anchored scale, using the following verbal anchors: 0 = not at all, 1 = hardly, 2 = moderately, 3 = strong, and 4 = very strong, to ensure equidistance between the steps of the rating scale. Symptoms were divided into a neuroglycopenic and an autonomic score, the former covering those five symptoms indicating reduced brain glucose supply (weakness, fatigue, dizziness, blurred vision, reduced concentration), the latter measur-

ing six symptoms indicating autonomic nervous system activation (sweating, tremor, palpitations, hunger, anxiety, nervousness) [22]. 2.5. Data analysis Data are expressed as M F S.E.M. Due to the wide baseline value variations in most endocrine parameters and in accordance with insulin conditioning experiments in animals [3,17] and our previous experiment in humans [7], differences from baseline levels were determined for all parameters [i.e., by calculating the difference between the level at each single post-CS measurement point minus the level at the pre-CS measurement point (min 24)]. Blood glucose is the only parameter that was not only sampled once, but four times in the baseline. To obtain a more stable baseline value, the baseline blood glucose level was calculated as the mean blood glucose of two pre-CS measurement points (i.e., min 12 and 6). Blood glucose difference from baseline level was calculated for each of the 16 post-CS measurement points. To express the total conditioned blood glucose change in a single value [7], the 16 differences were summed yielding the cumulative blood glucose change. To describe the time-course of the blood glucose change, the 16 post-CS measurement points (min 0 to 72) were subdivided into five intervals (Interval 1: mean blood glucose level of min 0, 1, 3, and 6; Interval 2: mean of min 10, 12, and 18; Interval 3: mean of min 24, 30, and 36; Interval 4: mean of min 42, 48, and 54; and Interval 5: mean of min 60, 66, and 72; see also Fig. 3B). This data pooling was based on functional aspects. Interval 1 covers the CS presentation and previous maximal exogenous insulin availability for Group 1 and the maximum endogenous insulin production for Group 2. Interval 3 contains the previous unconditioned hypoglycemia after insulin injection in Group 1, and Interval

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

379

Fig. 2. Blood glucose (mg/dl) on Days 1 to 4 of the acquisition phase. (A) M F S.E.M. for Groups 1 (CS INS) and 3 (CS NaCl). (B) M F S.E.M. for Groups 2 (CS GLUC) and 3 (CS NaCl). Injection immediately after blood sample at min 0.

380

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

5 the relative unconditioned blood glucose minimum after the endogenous insulin rise in Group 2. Nonparametric tests for rank-scaled data were used for comparison calculation to accommodate the small sample size and reduce the impact of outliers. The Mann Whitney U test (SPSS) was used for most of the intergroup comparisons (Group 1 vs. Group 3; Group 2 vs. Group 3) calculating exact tests. Values below the detection limit were obtained for adrenaline and GH, hence, measurements were only known to be smaller than a known number. These left-censored data were analyzed with Gehans test for censored data [23] using Disfree statistical software [24]. Intragroup comparisons were done by Lam-Longneckerss test, a modified Wilcoxon rank-sum test for paired data [25]. Inferences were only made for the cumulative blood glucose change on Day 5 (test). Data analyses for the timecourse of the blood glucose change, serum insulin, counterregulatory hormone, and subjective symptoms are considered exploratory, not confirmative. According to the preliminary

status of these analyses, the latter P values are for descriptive (not inferential) purpose, and were not corrected for multiple comparisons. As this is the first study addressing the conditionability after intravenous glucose in a between-subject design with humans, results with P < .10 are reported for descriptive purpose. We also calculated the effect sizes d of the group means differences as described by Cohen [26,27], with d = 0.2, d = 0.5, and d = 0.8 as indicating small, medium, and large effects, respectively [27]. Based on the effect sizes, the sample size necessary to reach an intended a-level and power for future inferential studies can be easily determined (see Tables in 2.3.1 to 2.3.6 in Ref. [26]).

3. Results For all parameters measured (blood glucose, glucoregulatory hormones, and symptoms), data on acquisition phase (UR, Days 1 4) are described first, followed by test session

Fig. 3. Blood glucose change (mg/dl) relative to the mean baseline level in test (Day 5) in Groups 1 (CS INS), 2 (CS GLUC), and 3 (CS NaCl). (A) Cumulative change over the 16 post-CS measurement points (M F S.E.M.). Mean baseline values (BA) are also listed. (B) Change per measurement point in the five functional intervals (1: min 0 6; 2: min 10 18; 3: min 24 36; 4: min 42 54; and 5: min 60 72). Results of the descriptively conducted Mann Whitney U testsGroup 1 (CS INS) vs. Group 3 (CS NaCl): [*]P < .10 and * P < .05.

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

381

Fig. 4. Serum insulin on Days 1 to 5 in Groups 1, 2, and 3. (A) Serum insulin change (mU/l): Difference from baseline (min 24). Baseline (BA) values (M F S.E.M.) are also listed. See break of y axis on Day 5. (B) Absolute serum insulin level (M F S.E.M.) at all nine measurement points on Day 5. See break of y axis. Results of the descriptively conducted Mann Whitney U tests only for the test sessionGroup 2 (CS GLUC) vs. Group 3 (CS NaCl): [#]P < .10, # P < .05. The results of the descriptively conducted intergroup comparisons on Days 1 to 4 are described in the text.

382

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

Table 1 Day 5 (test)Blood glucose change after CS presentation (5 intervals) correlated with the absolute serum insulin level (8 post-CS measurement points): Spearman rank correlations (rs), P levels per group, and results of the pairwise intergroup comparisons of the correlation coefficients (v2, df=1): Group 1 (CS INS) vs. Group 3 (CS NaCl) and Group 2 (CS GLUC) vs. Group 3 (CS NaCl) Mean blood glucose change per interval Absolute insulin level Groups Group 1 (CS INS) Group 2 (CS GLUC) Group 3 (CS NaCl) rs P rs P rs P Min 0 6 Min 0 .516 * .064 .268[#] .227 .585 .038 Min 1 .567 * .044 .333[#] .173 .524 .060 Min 3 .572** .042 .309# .192 .604 .032 Min 10 18 Min 10 .297 .202 .079 .414 .382 .138 Min 24 36 Min 24 .109 .382 .342 .167 .152 .338 Min 30 .232 .259 .491 .075 .183 .306 Min 42 54 Min 48 .006 .493 .164 .326 .723 .009 Min 60 72 Min 72 .293 .206 .201 .302 .255 .238

* P<.05; Group 1 (CS INS) vs. Group 3 (CS NaCl) ** P<.01; Group 1 (CS INS) vs. Group 3 (CS NaCl) # P<.05; Group 2 (CS GLUC) vs. Group 3 (CS NaCl). [#] P<.10; Group 2 (CS GLUC) vs. Group 3 (CS NaCl).

data on Day 5 (CR). Inference statistics were only calculated for the cumulative blood glucose change in the test session. All other comparisons were considered descriptive only. 3.1. Blood glucose 3.1.1. Unconditioned effects: Days 1 4 Fig. 2A shows the acquisition-phase (Days 1 to 4) blood glucose levels for Groups 1 (CS INS) and 3 (CS NaCl), and Fig. 2B the corresponding data for Groups 2 (CS GLUC) and 3 (CS NaCl). The blood glucose minimum in Group 1 was obtained 24 min after insulin injection, reaching hypoglycemic levels ( V 50 mg/dl) on all 4 days [(M F S.E.M.) Day 1: 31.7 F 2.6; Day 2: 33.3 F 3.1; Day 3: 29.5 F 2.6; Day 4: 28.9 F 2.6 mg/dl]. Group 2 (CS GLUC) blood glucose levels peaked at min 3 after glucose injection on Days 1 to 3 (Day 1: 279.2 F 14.5; Day 2: 289.8 F 11.8; Day 3: 293.8 F 11.7 mg/dl) and min 1 on Day 4 (316.0 F 20.7 mg/dl). Group 3 reached mean values of about 75 mg/dl on Days 1 to 4. 3.1.2. Conditioned effects: Day 5 3.1.2.1. Inferential statistics. Blood glucose baseline on Day 5 reached a comparable level in the three groups, with 77.2 F 2.2 mg/dl in Group 1, 78.8 F 1.8 mg/dl in Group 2, and 77.4 F 1.4 mg/dl in Group 3 (Mann Whitney U test; two-sided: Group 1 vs. Group 3, U = 49.5, P=.971; Group 2 vs. Group 3, U = 43, P=.631). As expected, Group 1 (CS INS) showed a higher cumulative blood glucose decrease from baseline ( 32.4 F 10.4 mg/dl) than Group 3 did ( 14.7 F 9.1 mg/dl; U = 32, P=.095, d = 0.57; Fig. 3A). The cumulative decrease in Group 2 (CS GLUC), with 22.3 F 16.2 mg/dl, was larger

than in Group 3 ( 14.7 F 9.1 mg/dl), but did not attain statistical relevance (U = 48, P=.456, d = 0.18). 3.1.2.2. Exploratory analyses: time-course. Block-wise analysis (Fig. 3B) shows that the cumulative conditioned blood glucose decrease in Group 1 was based on two intervals: min 0 6 (Interval 1) with the CS presentation (M = 1.96 F 0.6 mg/dl in Group 1 vs. M = 0.65 F 0.4 mg/dl in Group 3; U = 28.5, P=.053, d = 0.86) and the previous unconditioned hypoglycemia Interval 3 (min 24 36; M = 1.91 F 0.6 mg/dl in Group 1 vs. M = 0.67 F 0.5 mg/dl in Group 3; U = 23, P=.022; d = 0.66). Group 2 showed a higher decrease than Group 3 did in all five intervals, but with no statistically relevant difference and effect sizes below d = 0.2. 3.2. Serum insulin: exploratory analyses 3.2.1. Unconditioned effects: Days 1 4 The injection of insulin in Group 1 (CS INS) resulted in the pharmacologically increased serum insulin level expected (Fig. 4A). Group 1 insulin levels exceeded that of Group 3, at least until min 24, on each day ( P < .01), then returned to baseline levels. On Days 2 and 3, levels were below baseline by the end of the session (min 48 and 72; P < .05 each). Group 2 glucose injection induced the endogenous insulin production expected, consistently exceeding the Control group 3 (CS NaCl) insulin level at all postinjection measurement points on all days during acquisition ( P V .05, mostly < .001). 3.2.2. Conditioned effects: Day 5 On Day 5 (test), there was no conditioned effect in the deviation from baseline levels in Group 1 compared with

Fig. 5. (A E) Counterregulatory hormones on Days 1 to 5 in Groups 1, 2, and 3. Difference from baseline (min 24). Baseline (BA) values (M F S.E.M.) are shown. (A) Glucagon, (B) adrenaline, (C) noradrenaline, (D) Cortisol, and (E) GH. See break of y axis on Day 5. For Day 5, the difference min 72 minus min 48 was also calculated. Results of the descriptively conducted intergroup comparisonsGroup 1 (CS INS) vs. Group 3 (CS NaCl): [*]P < .10, * P < .05, * * P < .01 and * * * P < .001; Group 2 (CS GLUC) vs. Group 3 (CS NaCl): #P < .05, ##P < .01, and ###P < .001.

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

383

384

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

Group 3 (Fig. 4A). Contrary to our predictions, Group 2 even showed more of a decrease below baseline than did Group 3 at min 3 ( P < .05) and 10 ( P < .10; Fig. 4A). In agreement with the response pattern during acquisition, the baseline deviation scores in Group 1 exceeded that of Group 2 close to CS presentation at min 1, 3, and 10 (at least P < .10). To account for the general effect of an insulin decrease after the high volume injection in all groups, we also analyzed the absolute insulin level. A conditioning effect was obtained (Fig. 4B) in Group 2, with a higher absolute insulin level than that of Group 3, reaching P < .10 at three measurement points (min 0, 24, and 72) and medium effect sizes (min 0: d = 0.68; min 24: d = 0.63; min 72: d = 0.65). Although Day 5 insulin data might indicate that the conditioned blood glucose decrease in Group 1 is not based on conditioned insulin secretion, there is clear support for insulins effect on blood glucose response in Groups 1 and 2 (Table 1). Mainly during CS presentation (i.e., min 0 6), serum insulin level and blood glucose changes were negatively correlated (Spearman rank correlations rs) in experimental Groups 1 (rs about .55) and 2 (rs about .30), but positively correlated (rs about .55) in Group 3. Thus, only in the experimental groups, a greater blood glucose decrease was accompanied by higher serum insulin levels. The correlations in the experimental groups differed significantly from the Control group 3, as expressed in the results of chisquare tests (df = 1). The differences between Groups 1 and 3 were manifested at min 0 (v2 = 5.799) and at min 1 (v2 = 5.666), with a P level < .05, and peaked at min 3 ( v2 = 6.794) at P < .01 (two-sided). The same pattern emerged for Group 2 versus 3, where the intergroup difference was obvious in tendency ( P < .10) at min 0 and 1, and peaked at min 3 (v2 = 3.981, P < .05). 3.3. Counterregulatory hormones: exploratory analyses 3.3.1. Unconditioned effects: Days 1 4 As expected, Group 1 (CS INS) showed the typical insulin-induced counterregulation, affecting glucagon, adrenaline, and noradrenaline in the first phase of counterregulation (min 24), and cortisol and GH in the second phase (min 48). There was a very consistent increase of glucagon compared with Group 3 (CS NaCl) in min 24 on all 4 days of the acquisition phase (of at least P < .01), and, to a minor extent, of adrenaline (Day 2: P < .10; Day 3: P < .001) and noradrenaline (Day 2: P < .10; Day 4: P < .05; Fig. 5A C). Cortisol and GH (Fig. 5D E) reached their maximum 48 min after injection on all four acquisition days and differed from the control group by a P level of at least < .05. Glucose injection in Group 2 (CS GLUC) had minor impact on glucoregulatory hormones. Compared with Group 3 (CS NaCl), Group 2 showed an initial decrease of glucagon in min 24, manifesting on all four acquisition days at a P level of at least < .01 (Fig. 5A). The expected increase in glucoregulatory hormones by the end of the

session became partly obvious for noradrenaline, but differed from Group 3 only on Day 3 ( P < .05; Fig. 5C). 3.3.2. Conditioned effects: Day 5 3.3.2.1. Glucagon. As in our previous experiment [7], a conditioning effect in glucagon after CS insulin (Group 1) was not found, but medium effect sizes were obtained at min 24, the previous maximum UR in Group 1 (d = 0.50). The initial unconditioned glucagon decrease in Group 2 versus Group 3 did not remain with a relevant P level, but reached a small effect size (d = 0.45). 3.3.2.2. Adrenaline. A conditioning effect was not reached in Group 1, but the difference between Groups 1 and 3 reached a moderate effect size in temporal delay to the manifestation of the UR in min 48 (d = 0.54; Fig. 5B). In correspondence with the missing UR, we did not obtain a conditioned effect in Group 2. 3.3.2.3. Noradrenaline. Group 1 (compared with Group 3) showed a conditioned noradrenaline increase, peaking with a delay from the UR at min 48 (U = 25.0, P=.032; d = 0.87; Fig. 5C). In Group 2, a conditioning effect was not obtained, although the response pattern of Day 3 (with a noradrenaline increase by min 72) remained, reaching a small effect size (d = 0.39). 3.3.2.4. Cortisol. Group 1 did not exhibit the previous counterregulation as the CR at min 48, but an interesting delayed CR emerged by the end of the test session. From min 48 to 72, cortisol increased in Group 1 (M= + 0.3 F 0.3 Ag/dl) compared with Group 3 (M = 0.3 F 0.3 Ag/dl; U = 33, P=.109), reaching a medium effect size (d = 0.51; Fig. 5D). This delayed increase was even more obvious in Group 2, with a mean increase of 1.1 ( F 0.4) Ag/dl, compared with the decrease in Group 3 (U = 17, P=.006), reaching a large effect size (d = 1.80). 3.3.3. Growth hormone (GH) As with cortisol, a CR occurred between min 48 and 72 in Group 1 (M= + 1.2 F 0.9 mU/l), exceeding that of Control group 3 (M = 0.4 F 0.2 mU/l; Group 1 vs. 3: W = 50, PU = 0.025, Gehans test, d = 0.77; Fig. 5E). 3.4. Symptoms: exploratory analyses Fig. 6A and B summarizes the subjective symptom change scores divided into neuroglycopenic and autonomic symptoms on Days 1 to 5. Baseline values did not differ between groups on any day. 3.4.1. Unconditioned effects: Days 1 4 The intensity of the neuroglycopenic symptoms (Fig. 6A) increased after insulin injection, peaking at min 24, and yielding relevant differences between Groups 1 and 3 on

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

385

Fig. 6. Symptom scores: Difference from baseline (BA; min 24) on Days 1 to 5 in Groups 1, 2, and 3. BA values (M F S.E.M.) are shown. (A) Neuroglycopenic score, (B) autonomic score. Results of the descriptively conducted Mann Whitney U testsGroup 1 (CS INS) vs. Group 3 (CS NaCl): [*]P < .10, * P < .05, and * * P < .01; Group 2 (CS GLUC) vs. Group 3 (CS NaCl): [#]P < .10 and #P < .05.

Days 2 to 4 of the acquisition phase (Day 2: U = 17.5, P=.006; Day 3: U = 18.5, P=.006; Day 4: U = 18.5, P=.008). The neuroglycopenic symptoms peak in min 24 decreased

from Day 2 to 4 (Z = 1.335, P=.091, Lam-Longnecker test). In Group 2, effects occurred again by the end of the session when blood glucose had reached its relatively lowest

386

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

level after endogenous insulin secretion, manifesting mainly on Days 1 and 4 ( P < .05) and, to a minor extent, on Day 2 ( P < .10). Insulin-induced hypoglycemia (Group 1) led to the autonomic symptoms increase in min 36, differing from Group 3, with P levels of at least < .05 on Days 2 to 4 (Day 2: U = 22.0, P=.013; Day 3: U = 15.0, P=.003; Day 4: U = 27, P=.045). The insulin-induced symptoms peak (min 36) of autonomic symptoms also decreased, but less than that of the neuroglycopenic score (Days 2 to 4: Z = 0.890, P=.187, Lam-Longnecker test). There were no consistent effects of glucose injection in Group 2. Only on Day 4, an initial symptom decrease compared with the control group level (min 12, P < .10) occurred. 3.4.2. Conditioned effects The same response pattern that emerged as the UR in Groups 1 and 2 also emerged as the CR on Day 5. However, the increase of neuroglycopenic symptoms in Group 1 in min 24 did not reach a relevant P level compared with Group 3 in the test session, but an effect size of d = 0.44 was obtained. Corresponding to the UR, neuroglycopenic symptoms in Group 2 increased by the end of the session (min 72), at the time-point of the previous unconditioned bloodglucose minimum after endogenous insulin release, and differed from that of Group 3 (U = 22.5, P=.016; d = 1.13). The pattern of an increase in autonomic symptoms in Group 1 compared with Group 3 remained, now with an earlier manifestation in min 12, i.e., close to CS presentation. It did not attain a significant P level, but had a medium effect size (d = 0.62).

4. Discussion Human participants experiencing a CS with insulin in the acquisition phase of a conditioning protocol react with a conditioned cumulative blood glucose decrease ( P < .10), predominantly manifested during CS presentation ( P < .10) and at the time-point of the previous hypoglycemia ( P < .05). In total, the CR reached about 10% of the UR in the present experiment. As expected, there was also a larger, but statistically nonsignificant, cumulative blood glucose decrease after a previous learning phase with CS glucose in Group 2 compared with Group 3. In this regard, our results agree with the data obtained in animal experiments where conditioned blood glucose decrease after intravenous insulin was reliably demonstrated, whereas a conditioned blood glucose decrease after glucose injection was found by Matysiak and Green [17], but not Woods [16]. Thus, conditioning effects after exogenous insulin were higher and easier to obtain than after endogenously produced insulin [17]. To explain the difference between the conditionability of insulin effects after exogenous injection and glucose-stimulated endogenous production, the intensity of the US, a

parameter affecting classical conditioning per se, becomes relevant. Exogenous injection of insulin resulted in a much higher plasma insulin level (stronger US intensity) than the reactive insulin level after glucose injection. Immediately after the insulin injection (min 1), we measured an insulin level of 232.5 mU/l on Day 1 in Group 1, whereas the reactive, endogenously secreted insulin after glucose injection reached a level of 44.8 mU/l in Group 2. Thus, the insulin level after exogenous application is around five times higher than endogenous production on Days 1 (ratio 5.2:1) and 2 (5.1:1) and about four times higher on Days 3 (3.9:1) and 4 (4.2:1). Although there are indications against a linear relationship between exogenously applied insulin and its transport to the brain [2], higher insulin availability in the brain (the US) in Group 1 than in Group 2 is a reasonable assumption. Despite using a higher insulin dose (0.05 iU/kg BW) in the present experiment than in our previous study 1 (0.035 iU/kg BW) [7], the conditioned blood glucose decrease that reached about 10% of the UR was not higher but lower than in Study 1, where the CR reached 15%. This might be explained by the raised stressfulness of the injection procedure in our present experiment, which needed a large injection volume to realize the conditions in Group 2 (0.5 g/kg of 15% glucose solution), possibly acting as a stressor (with an increase of the sympathetic tone), thus reducing the impact of a conditioned vagal tone increase. We did not find any conditioned insulin increase from baseline on Day 5. In addition, the level even dropped in experimental Group 2 close to CS presentation compared with Control group 3. However, the injection did lead to an insulin reduction in all three groups (especially in Group 2, with the highest insulin levels), probably reflecting the stressfulness of the injection procedure itself. In terms of absolute insulin levels, we found the predicted effect in Group 2, where participants had higher insulin levels than did the control group at three time-pointsclose to CS presentation at min 0, at min 24, and at min 72. However, there is support for the interpretation that the conditioned blood glucose decrease was mediated by the insulin activity (close to CS presentation, blood glucose changes correlated inversely with insulin in both experimental groups) in Groups 1 (CS INS) and 2 (CS GLUC), whereas the controls (CS NaCl) showed the inverse pattern. Thus, only participants in the experimental groups showed decreased blood glucose based on a high insulin level on Day 5. Serum insulin might be not the most sensitive measure for validating any vagal mediation in the conditioned blood glucose decrease. Hence, we also used a noninvasive vagal activity indicator in our experiment, analyzing heart rate variability from continuous ECG recordings by spectral analyses. The high-frequency (HF) heart rate variability band for heart rate changes between 0.15 and 0.4 Hz is regarded as a vagal activity indicator (e.g., Refs. [28 31]). The HF band increased from pre-CS baseline to just after CS presentation in Group 1 and, to a lesser degree, Group 2, with both the

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388

387

UR (Days 1 to 4) and the CR (Day 5). There was no such response in Group 3 [29]. In addition to blood glucose and insulin data, there were interesting effects in glucoregulatory hormones, clearly supporting that the CS triggers a specific CR in both experimental groups. Unconditionally, Group 1 showed a clear counterregulation of hypoglycemia, affecting glucagon, catecholamines (although less consistent) in the firstphase response, and cortisol and GH as the second-phase response. Noradrenaline turned out to be the main hormone for the CR. Noradrenaline, but not adrenaline, conditionability agrees with CNS-mediated blood glucose regulation by noradrenaline and not adrenaline [32]. Furthermore, we found evidence for a conditioned, albeit temporarily delayed, GH increase. The conditioned endocrine effects in Group 1 appear in the same sequence as in the UR, with the first phase affecting the catecholamines and the second phase response in cortisol and GH, but are temporarily delayed. Regarding the conditioning effects in glucoregulatory hormones and symptoms after glucose injection, our preliminary analyses reveal that responses in the regulatory hormones became more conditioned to the consequences of the glucose-induced insulin increase, not to hyperglycemia. On Day 5, in the test session, we found a conditioned increase of cortisol level and an increase of neuroglycopenic symptoms (see below) by the end of the test session. This corresponds to response patterns that was found as the UR after oral glucose in healthy participants [18]: Hyperglycemia affected fewer endocrine regulatory hormones (only a decrease of glucagon) than the subsequent consequences of endogenous insulin secretion, resulting in an increase of glucagon, adrenaline and subjective symptoms when the participants reached their relative glucose minimum (68 F 4 mg/dl). Moreover, this is also in accordance with the result in insulin-dependent diabetic patients: They often remain asymptomatic in a state of hyperglycemia compared with hypoglycemia, and cannot discriminate between a metabolic state of hyper- versus euglycemia [33]. The group symptom differences obtained as a UR remained, although to a minor extent, as CRs [7]. Group differences on Day 5 were observed in autonomic (shortly after CS presentation) and neuroglycopenic (in the previous unconditioned hypoglycemia interval) symptoms in Group 1 compared with Group 3, although only manifesting in small and medium effect size, not reaching relevant P levels. This might be related to the blunted UR over the course of the acquisition phase, probably indicating the development of hypoglycemia unawareness (see Ref. [7] for a similar result) after repeated insulin injection in healthy participants. Regarding glucose effect conditionability (Group 2), we found an increase of neuroglycopenic symptoms ( P < .05) at the time-point of the maximum glucose lowering after endogenous insulin release. More conditioning experiments should address these questions. Alternative conditioning designs may be taken into consideration. Instead of using a placebo control group,

as we did, a differential conditioning protocol can be used, but this will need more sessions. Alternatively, two groups, a CS insulin and a CS glucose group, could be exposed to two test sessions, presenting the CS or not, in a counterbalanced order. Groups could also be divided after acquisition, one half given CS and placebo and the other half only given the placebo. Most of our analyses were done for descriptive purpose. To plan future studies in that field, we calculated effect sizes for the means differences in the test session. As we obtained large and medium effects in most of the analyzed parameters, this encourages to conduct more experiments in that field and with the sample sizes [26,27] needed to obtain the intended a level and power (usually .80) for inferential purpose. Our data also provide hints for suggesting of what is US/ UR and CS/CR in this kind of study. We suggest that there are two afferent signals to the brain after exogenous insulin (Group 1: CS INS): First is the CS-associated increase of insulin (in other words, hyperinsulinemia) detected at central insulin receptors and inducing a vagally mediated insulin release, which is the main signal for conditioning, then, in a second step, the insulin-induced hypoglycemia as a second, temporarily delayed, signal reaching the glucoseresponsive neurons in the brain, initiating a neurally mediated secretion of counterregulatory hormones [6]. Inferring from the response in the test, our data suggest that glucose injection (Group 2: CS GLUC) also consists of a sequence of afferent signals to the brain: First is the glucose-induced hyperglycemia, reaching glucose-responsive neurons and instantly inducing the endogenous insulin increase as the actual, neurally mediated response. Then, endogenously produced insulin becomes an afferent signal on its own, thus explaining that the CR manifests in comparable effects after exogenous and endogenous insulin. We assume, that after both, exogenous insulin administration and glucose injection, the main US for conditioning to occur is the increase of insulin in the brain, detected at the central insulin receptors. Imaging studies might help to answer these questions by analyzing the areas involved when learning a CS insulin and CS glucose association. While the present experiment addresses the endocrine and subjective effects of insulin and glucose, and the conditionability of these effects after peripheral application, examining the effects of central insulin directly without inducing peripheral hypoglycemia would be worthwhile. Animal experiments have shown that insulin injected into the intraventricular space reduced food intake and BW both in lean and normal-weight animals [34,35], as long as peripheral blood glucose level was euglycemic. In addition, learning and memory are improved by central insulin administration in a state of euglycemia [36 39]. For humans, intranasal application sets the conditions for reaching the brain while keeping blood glucose in the euglycemic range [40,41], which provides the chance to examine pure insulin effects.

388

U. Stockhorst et al. / Physiology & Behavior 81 (2004) 375388 [19] Frati AC, Iniestra F, Ariza CR. Acute effect of cigarette smoking on glucose tolerance and other cardiovascular risk factors. Diabetes Care 1996;19:112 8. ` re G, Richard [20] Tremblay A, Pinsard D, Coveney S, Catellier C, Laferrie D, et al. Counterregulatory response to insulin-induced hypoglycemia in trained and nontrained humans. Metabolism 1990;39:1138 43. [21] Flaherty CF, Grigson PS, Brady A. Relative novelty of conditioning context influences directionality of glycemic conditioning. J Exp Psychol Anim Behav Process 1987;13:144 9. [22] Towler DA, Havlin CE, Craft S, Cryer P. Mechanism of awareness of hypoglycemia. Perception of neurogenic (predominantly cholinergic) rather than neuroglycopenic symptoms. Diabetes 1993;42:1791 8. [23] Krauth J. Distribution-free statistics: an application-oriented approach. Amsterdam: Elsevier; 1988. [24] Krauth J. Disfree. Cambridge: Biosoft; 1989. [25] Lam FC, Longnecker MT. A modified Wilcoxon rank sum test for paired data. Biometrika 1983;70:510 3. [26] Cohen J. Statistical power analysis for the behavioral sciences. New York: Academic Press; 1977. [27] Cohen J. A power primer. Psychol Bull 1992;112:155 9. [28] Akselrod S, Gordon D, Ubel FA, Shannon DC, Berger AC, Cohen RJ. Power spectrum analysis of heart rate fluctuation: a quantitative probe of beat-to-beat cardiovascular control. Science 1981;213:220 2. [29] Stockhorst U, Huenig A, Krueger M, Mahl N, Ziegler D, Steingrueber HJ, Scherbaum WA. Effects of insulin and glucose on heart rate variability: unconditioned and conditioned effects. [in preparation]. [30] Jennings JR, McKnight JD. Inferring vagal tone from heart rate variability. Editorial. Psychosom Med 1994;56:194 6. [31] Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Heart rate variability. Circulation 1996;93:1043 65. [32] Smythe GA, Grunstein HS, Bradshaw JE, Nicholson MV, Compton PJ. Relationships between brain noradrenergic activity and blood glucose. Nature 1984;308:65 7. [33] Weinger K, Jacobson AM, Draelos MT, Finkelstein DM, Simonson DC. Blood glucose estimation and symptoms during hyperglycemia and hypoglycemia in patients with insulin-dependent diabetes mellitus. Am J Med 1995;98:22 31. [34] Woods SC, Seeley RJ, Porte Jr D, Schwartz MW. Signals that regulate food intake and energy homeostasis. Science 1998;280:1378 83. [35] Air EL, Benoit SC, Blake Smith KA, Clegg DJ, Woods SC. Acute third ventricular administration of insulin decreases food intake in two paradigms. Pharmacol Biochem Behav 2002;72:423 9. [36] Wickelgren I. Tracking insulin to the mind. Science 1998;280:517 9. [37] Park CR. Cognitive effects of insulin in the central nervous system. Neurosci Biobehav Rev 2001;25:311 23. [38] Biessels GJ, Kamal A, Urban IJA, Spruijt BM, Erkelens DW, Gispen WH. Water maze learning and hippocampal synaptic plasticity in streptozotocin-diabetic rats: effects of insulin treatment. Brain Res 1998;800:125 35. [39] Fro lich L, Blum-Degen D, Bernstein HG, Engelsberger S, Humrich J, Laufer S, et al. Brain insulin and insulin receptors in aging and sporadic Alzheimers disease. J Neural Transm 1998;105:423 38. [40] Kern W, Born J, Schreiber H, Fehm HL. Central nervous system effects of intranasally administered insulin during euglycemia in men. Diabetes 1999;48:557 63. [41] Stockhorst U, de Fries D, Schottenfeld-Naor Y, Huebinger A, Steingrueber HJ, Scherbaum WA. Intranasally administered insulin and its CNS-effects in healthy humans: unconditioned and conditioned responses. [submitted for publication].

Acknowledgements The technical assistance of A. Czekalla, G. Korthaus, and in conducting the biochemical analyses is W. Mohne gratefully acknowledged. Conducted with Deutsche Forschungsgemeinschaft DFG support (Sto 323/1-1).

References
[1] Ramsay DS, Woods SC. Biological consequences of drug administration: implications for acute and chronic tolerance. Psychol Rev 1997;104:170 93. [2] Woods SC, Ramsay DS. Pavlovian influences over food and drug intake. Behav Brain Res 2000;110:175 82. [3] Woods SC. Insulin and the brain: a mutual dependency. Prog Psychobiol Physiol Psychol 1995;16:53 81. [4] Banks WA, Jaspan JB, Huang W, Kastin AJ. Transport of insulin across the blood-brain barrier: saturability at euglycemic doses of insulin. Peptides 1997;18:1423 9. [5] Hopkins DFC, Williams G. Insulin receptors are widely distributed in human brain and bind human and porcine insulin with equal affinity. Diabet Med 1997;14:1044 50. [6] Stockhorst U, Steingru ber HJ, Scherbaum WA. Classically conditioned responses following repeated insulin and glucose administration in humans. Behav Brain Res 2000;110:143 59. [7] Stockhorst U, Gritzmann E, Klopp K, Schottenfeld-Naor Y, Hu binger A, Berresheim HW, et al. Classical conditioning of insulin effects in healthy humans. Psychosom Med 1999;61:424 35. [8] Holmes LJ, Smythe GA, Storlien LH. Monoaminergic activity at the level of the hypothalamus and striatum: relationship to anticipated feeding and pancreatic insulin responses. Brain Res 1989;496: 204 10. [9] Morrell EM, Surwit RS, Kuhn CM, Feinglos MN, Cochrane C. Classically conditioned enhancement of hyperinsulinemia in the ob/ob mouse. Psychosom Med 1988;50:586 90. [10] Roozendaal B, Oldenburger WP, Strubbe JH, Koolhaas JM, Bohus B. The central amygdala is involved in the conditioned but not in the meal-induced cephalic insulin response in the rat. Neurosci Lett 1990;116:210 5. [11] Woods SC, Vasselli JR, Kaestner E, Szakmary GA, Milburn P, Vitiello MV. Conditioned insulin secretion and meal feeding in rats. J Comp Physiol Psychol 1977;91:128 33. [12] Overduin J, Jansen A. Conditioned insulin and blood sugar responses in humans in relation to binge eating. Physiol Behav 1997;61:569 75. [13] Overduin J, Dworkin BR, Jansen A. Introduction and commentary to: M.I. Mityushov (1954) Conditioned reflex secretion of insulin. Integr Physiol Behav Sci 1997;32:228 46. a S. Conditioning of adrenalin anorexia. Nature [14] Russek M, Pin 1962;193:1296 7. [15] Mityushov MI. Conditioned reflex secretion of insulin. Pavlov J High Nerv Act 1954;4:206 12. [16] Woods SC. Conditioned hypoglycemia. J Comp Physiol Psychol 1976;90:1164 8. [17] Matysiak J, Green L. On the directionality of classically-conditioned glycemic responses. Physiol Behav 1984;32:5 9. [18] Genter P, Ipp E. Plasma glucose thresholds for counterregulation after an oral glucose load. Metabolism 1994;43:98 103.