This randomised, single-centre, patient-blinded, two-period crossover study compared cognitive performance (assessed by working memory performance and reaction time) under hypoglycaemic (aiming for a PG target just below 3.1 mmol/l; the clamp target was defined as 2.8 ± 0.2 [2.6–3.0] mmol/l) and euglycaemic (PG clamp target 5.5 mmol/l ± 10%) conditions in participants with type 1 diabetes mellitus (ESM Fig. 1). Blood was drawn at pre-specified time points to assess counter-regulatory hormone responses. Hypoglycaemia awareness and symptoms during both clamps were also assessed. Participants underwent the sequence of glycaemic conditions in a blind and randomised order determined by sequential enrolment and lowest available number assignment. The two experimental visits were separated by 21–42 days (to avoid effects of counter-regulatory hormone responses or other physiological effects of hypoglycaemia). Female participants attended the two visits at the same stage of their menstrual cycle. The study was conducted from 14 January to 1 December 2013 at the Department of Endocrinology and Department of Nuclear Medicine and PET Center, Aarhus University Hospital, Aarhus, Denmark. Informed written consent was obtained from all participants before any study-related activities. The study was conducted in compliance with International Conference on Harmonisation Good Clinical Practice , the Declaration of Helsinki  and was approved according to local regulations by an independent ethics committee.
Inclusion criteria for participants screened (n = 37) were right-handedness, age 18–64 years, BMI 18.0–28.0 kg/m2, HbA1c ≤9.0% (≤75 mmol/mol), diagnosed with type 1 diabetes mellitus and treated with multiple daily insulin injections or continuous subcutaneous insulin infusion for ≥12 months prior to screening. Key exclusion criteria included known central nervous system abnormalities, structural brain abnormalities (identified by structural MRI scans during screening), severe hypoglycaemia (requiring third party assistance) or ketoacidosis in the last 6 months, clinically defined hypoglycaemic unawareness, and treatment with medications potentially interfering with glucose metabolism. Key experimental visit exclusion criteria included occurrence of a hypoglycaemic event (with PG ≤3.9 mmol/l) within the preceding 48 h. Full inclusion and exclusion criteria are listed in ESM Table 1.
The Wechsler Adult Intelligence Scale (third edition) revised digit symbol substitution test (DSST)  was previously adapted and validated to specifically measure working memory , and has been successfully used with fMRI . Here, we used the stimuli of this adapted DSST paradigm and modified it to our specific PET procedure. As with the fMRI design, we used the modified DSST (mDSST) task to measure working memory. A control task (cDSST) was used for non-task-related brain activation (without a working memory load), including visual and motor cortex activation (for eye sight and movement) and index finger movement.
Subtraction of regional CBF (rCBF) patterns during cDSST from those during mDSST was interpreted as indicative of brain activation patterns exclusively associated with the operation of working memory. The mDSST task consisted of three blocks of 32 randomly presented digit–symbol combinations (ESM Methods: Cognitive tests – mDSST and cDSST task combination blocks and ESM Fig. 2), fitting the specific H2
15O PET design. The cDSST task had the same basic design. Each block began with test instructions shown for 7.8 s and was 0.2 s longer than every 3 min PET acquisition, to ensure participants were engaged in the mDSST or cDSST task during the full 3 min PET acquisition. Tasks were presented using E-prime 2.0 (Psychology Software Tool, Pittsburgh, PA, USA) through audiovisual goggles inside the PET scanner. Correct responses (no response was considered as incorrect) and response time were recorded and analysed.
To test working memory following a prolonged recovery phase (75–90 min after PG of 5–6 mmol/l was restored) following hypoglycaemia or euglycaemia, the paced auditory serial addition task (PASAT) was preferred over the mDSST to avoid bias from habituation [20, 21]. During the PASAT, participants heard a digit and had to add the next digit (presented 3 or 2 s later) and report the sum aloud. Both parts consisted of 60 digits and correct responses were recorded.
Each participant underwent PET imaging at both visits. First, a 6 min transmission scan for attenuation correction was performed. Thereafter, six 3 min tomography sessions with either mDSST or cDSST in a fixed order were performed. Cerebral activity levels were measured as change in brain uptake of radiolabelled water (H2
15O), the retention of which matches the rate of CBF, by means of a high-resolution research tomograph (Siemens/CTI, Knoxville, TN, USA) operating in 3-D mode. Additional detail is provided in ESM Methods: PET imaging.
Experimental visit procedures
Participants attended the study site at approximately 20:00 hours on the day before each experimental procedure, at which point normal insulin treatment was suspended. Participants stayed overnight to ensure stabilisation of PG within the range of 5–8 mmol/l via variable intravenous infusion of insulin (Actrapid®, 100 U/ml) and glucose (20% glucose/dextrose/10 mmol/l KCl), before initiation of experimental procedures at 08:00 hours the following day. On the days of the experimental procedures, cognitive tests were briefly performed ≥1 h before initiation of hypoglycaemia or euglycaemia to prevent practice effects. Each glycaemic condition was preceded by a 60 min run-in period whereby variable intravenous infusions of glucose or human soluble insulin were delivered to obtain a steady-state PG target level of 5.5 mmol/l ± 10%. During the run-in and clamps, the participants’ cannulated hand was placed in a thermoregulated box with their arterialised venous blood sampled for PG measurements using a benchtop glucose analyser (YSI 2300 Stat Plus, Yellow Springs, OH, USA). Euglycaemia was maintained using a glucose clamp (glucose/Actrapid® infusion) for approximately 1 h after which insulin (Actrapid®, 100 U/ml) was given at an infusion rate of 3 mU kg−1 min−1 for 10 min and then reduced to 1.5 mU kg−1 min−1 thereafter. Euglycaemia was maintained or hypoglycaemia induced with glucose infusion rate adjusted accordingly to meet the PG target for approximately 55 min, during which H2
15O PET scans and cDSST and mDSST tasks were performed (Fig. 1). Following glycaemic clamps, participants were brought back to euglycaemia (with glucose infusions to reach a PG target of 5–6 mmol/l) and after approximately 75–90 min the PASAT was conducted. After experimental procedures, participants resumed usual insulin treatment.
Counter-regulatory hormones were measured as a validation that the PG target was sufficient to elicit a counter-regulatory response, and thus, hypoglycaemia. Hormonal responses (noradrenaline [norepinephrine], glucagon, cortisol and growth hormones) were measured 30 min prior to induction of hypoglycaemia or euglycaemia, and 45 min and 150 min (just prior to PASAT test) after induction of hypoglycaemia or euglycaemia.
Hypoglycaemia awareness and symptoms
At screening, hypoglycaemia unawareness was assessed by asking participants ‘Can you feel your hypos?’ and checking their medical files for any indication of unawareness. During the study, hypoglycaemia awareness was assessed both 30 min prior to and 45 min after induction of hypoglycaemia or euglycaemia, by asking participants ‘Do you feel hypo?’. The hypoglycaemia symptoms questionnaire (based on the Edinburgh Condition Scale)  was completed by participants 30 min prior to and 55 min after induction of hypoglycaemia or euglycaemia. It measured autonomic (sweating, palpitations, shaking and hunger), neuroglycopenic (confusion, drowsiness, odd behaviour, speech difficulty and incoordination) and general malaise (headache and nausea) symptoms on a seven point Likert scale.
Cerebral activation was measured as rCBF in 19 pre-specified regions of interest (ROI). Given the hypothesis-driven nature of this study, these regions were selected in accordance with relevant literature for one or more of the following criteria: DSST evoked brain activity patterns in normal conditions, i.e. euglycaemia, without working memory load (precuneus, dorsolateral prefrontal cortex, anterior cingulate gyrus/cortex, posterior cingulate gyrus, posterior supramarginal gyrus and orbitofrontal cortex) [19, 23]; neural substrate of DSST performance (inferior frontal gyrus, superior frontal gyrus, middle frontal gyrus, superior parietal lobe, precuneus, posterior cingulate gyrus/cortex and parahippocampal gyrus) [24, 25]; brain areas showing changes in functional activity in response to hypoglycaemia (medial temporal lobe, hippocampus, parahippocampal gyrus, insula, globus pallidum, striatum, dorsolateral prefrontal cortex, anterior cingulate gyrus, inferior frontal gyrus, superior frontal gyrus, precuneus, posterior cingulate gyrus, posterior supramarginal gyrus and primary visual cortex) [9, 26,27,28,29,30]; and brain areas reported as involved in working memory tasks (dorsolateral prefrontal cortex, inferior frontal gyrus, middle frontal gyrus, superior frontal gyrus, ventromedial prefrontal cortex, orbitofrontal cortex, insula, superior parietal lobe, anterior cingulate gyrus/cortex, hippocampus and thalamus) [31,32,33]. All rCBF measures were normalised to measures in the cerebral cortex, as this region is considered to be less impacted by the duration of type 1 diabetes mellitus . In the present study, there was no significant difference in rCBF in the cerebral cortex between hypoglycaemia and euglycaemia during either cDSST or mDSST tasks.
Three endpoints were used to determine regional cerebral activation: rCBF during mDSST and cDSST performances, and rCBF for working memory. For the cDSST and mDSST endpoints, rCBF was calculated by subtracting the mean of three rCBF values for euglycaemia from the mean of three rCBF values for hypoglycaemia. To isolate changes as a result of working memory, rCBF values during the mDSST test were corrected for the rCBF values during cDSST, by subtracting the mean of three cDSST rCBF values from the mean of three mDSST rCBF values; this correction was conducted for measurements taken during both glycaemic clamps with the totals for euglycaemia subtracted from those for hypoglycaemia ([mean (3 × CBF during mDSST) − mean (3 × CBF during cDSST)] hypoglycaemia − [mean (3 × CBF during mDSST) − mean (3 × CBF during cDSST)] euglycaemia) to isolate changes in working memory during hypoglycaemia.
Endpoints and statistical analyses
The primary objective of the study was to compare cognitive performance during hypoglycaemia with that during euglycaemia. The primary endpoint was the number of correct mDSST scores. For each glycaemic condition, mean mDSST scores, reaction time and PASAT scores were compared using a linear mixed-effect model with glycaemic condition and period as fixed factors and participant as a random factor; mean differences between hypoglycaemia and euglycaemia were estimated from the model and corresponding 95% CI and p values were calculated. The rCBF and predefined ROI (during both cDSST and mDSST), as well as the difference in rCBF between the two tasks, were compared during glycaemic conditions using an analysis of variance with glycaemic condition, period and participant as fixed factors. Because of the hypothesis-driven nature of this trial, no correction for multiplicity was performed with regard to different ROI analyses.
The SD for DSST score between euglycaemia and hypoglycaemia (PG 2.5 mmol/l) was determined in a previous trial (NCT01002768) to be approximately nine. Assuming a similar variability in this trial, using a 5% significance level and two-sided paired t test, a sample size of 25 participants completing both periods was calculated to have 90% power to detect a true difference in DSST score between hypoglycaemia and euglycaemia of approximately six. A total of 28 participants were randomised to ensure at least 25 participants completing both experimental visits.