The Brain Networks Encoding Visceral Sensation

Published Date: 02 Nov 2017

Running Title: Central Reorganization in Diabetes

Dina Lelic, Christina Brock, Magnus Simrén, Jens Brøndum Frøkjær, Eirik Søfteland, Georg Dimcevski, Hans Gregersen6 & Asbjørn Mohr Drewes

Mech-Sense, Department of Gastroenterology and Hepatology, Aalborg Hospital, Aarhus University, Denmark

Institute of Medicine, Department of Internal Medicine, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden

Mech-Sense, Department of Radiology, Aalborg Hospital, Aarhus University Hospital, Aalborg, Denmark

Department of Medicine, Haukeland University Hospital, Bergen, Norway

Institute of Medicine, University of Bergen, Bergen, Norway

GIOME and Sino-Danish Centre for Education and Research, Aarhus, Denmark and Beijing, China

Center for Sensory-Motor Interactions (SMI), Department of Health Science and Technology, Aalborg University, Aalborg, Demark

Corresponding author: Dina Lelic, MSc, PhD

Mech-Sense, Department of Gastroenterology and Hepatology

Mølleparkvej 4, Aalborg University Hospital

DK-9000 Aalborg, Denmark

Telephone: +45 99326247

Fax: +45 99326507

E-Mail: [email protected]

Key words: Diabetes; Operculum; Evoked Potentials; Brain Source Analysis

Funding Source: European Community´s Seventh Framework Programme FP7/2007-2013 under grant agreement no. 223630.

ABSTRACT:

Background: Increasing evidence points to association between long-term diabetes mellitus and abnormal brain processing. The aim of this study was to investigate central changes due to electrical stimulation in oesophagus in patients with upper gastrointestinal symptoms due to diabetic neuropathy.

Methods: 23 diabetes patients with upper gastrointestinal symptoms and 27 healthy controls were included. A standard ambulatory 24-hour electrocardiography was performed. 122-channel evoked brain potentials to electrical stimulation in the oesophagus were recorded. Brain source-connectivity analysis was done. Upper gastrointestinal symptoms were assessed with the Gastroparesis Cardinal Symptom Index and quality of life with SF-36.

Key Results: Diabetes patients with gastrointestinal symptoms showed changes relative to controls in three brain networks: 1) brainstem/operculum/frontal cortex, 2) operculum/cingulate, and 3) mid-cingulate/anterior-cingulate/operculum/deep limbic structures. Operculum brain source in patients was localized deeper and more anterior in all three networks. The shift of operculum source was correlated with the severity of upper gastrointestinal symptoms, decreased heart beat-to-beat interval, and decreased standard deviation of the intervals. The activation of the first network was delayed in patients. Operculum source had higher activity than cingulate in the second network in patients, and this was correlated with decreased physical quality of life. Deep limbic source was localized deeper in patients, which also correlated with decreased physical quality of life.

Conclusion & Inferences: This study indicates involvement of central nervous system in diabetes. Reorganization within opercular cortex was correlated with gastrointestinal symptoms suggesting that operculo-cingulate cortex could contribute to development and maintenance of GI symptoms in diabetes patients.

1. INTRODUCTION

Autonomic nervous system dysfunction is one of the most common complications of long-standing diabetes mellitus (DM) and it seems to play a central role in the development and progression of gastrointestinal (GI) dysfunction and discomfort in these patients (1,2). The symptoms are often severe, leading to reduced quality of life and management of these patients is challenging (3). Therefore, a better understanding of the sensory processing from the GI tract is warranted in order to explore the underlying mechanisms.

In patients with verified diabetic autonomic neuropathy and GI symptoms (such as nausea/vomiting, postprandial fullness, and bloating), Frøkjær and colleagues observed an overall decrease in sensitivity accompanied by an increase in somatic referred pain areas in oesophagus and duodenum to combined mechanical, thermal, and electrical stimulation (4). The referred pain area to visceral pain is a proxy of convergence between somatic and visceral afferents in the central nervous system (CNS). Therefore, while diabetes has long been considered a disease of the peripheral nervous system, these findings also suggest the involvement of CNS in symptom generation (4,5). The peripheral and central mechanisms of diabetic autonomic neuropathy have been investigated in a few studies where evoked potentials (EPs) to visceral electrical stimulation were utilized. As electrical stimulation bypasses receptors, sensory fibres (Aδ and C) and submucosal vagal fibres depolarize, giving rise to evoked potentials. However, these studies were based on selected subjects and a few recording electrodes. Our group has expanded the model to use 62 channel electroencephalographic (EEG) recording to electrical stimulation in the oesophagus in order to investigate the central components of pain in diabetes patients with autonomous neuropathy (6). This design allowed, not only comparison of EP waveforms, but the underlying brain activity as well and evidence of reorganization of the involved brain areas was seen.

Visceral sensation is likely not perceived by sequential activation of singular brain structures but rather by how they communicate with each other. Disruption within these networks at any level could be what differentiates diabetes patients from healthy controls, even when the same brain areas are involved. Thus, studying cross-communication of brain centres is crucial in understanding the underlying mechanisms of visceral sensation and ultimately of diabetes. In the present study, we expanded the model to utilize 122 channel recordings. This gave better scalp coverage and more precise brain source localization. Advanced signal analyses were employed in order to investigate connectivity between different brain centres.

We hypothesized that long-term diabetes resulting in diabetic autonomic neuropathy and severe GI symptoms is associated with modification of afferent input and abnormal processing at both peripheral and central levels of the nervous system. Therefore, the aims of this study were to: 1) investigate differences in amplitudes and latencies of evoked potentials between healthy controls and patients, 2) investigate differences in brain networks underlying the evoked potentials between healthy controls and patients, and 3) in order to investigate the implication of altered brain processing, the changes in brain source locations were correlated to severity of GI symptoms, quality of life (QOL), heart beat-to-beat interval (RR), and standard deviation of heart beat-to-beat intervals (SDNN).

2. MATERIALS AND METHODS

2.1 Subjects

Twenty-three diabetes patients (17 type 1 and 6 type 2) with GI symptoms were included in the study. Thirteen of the patients were recruited at the Department of Internal Medicine at Sahlgrenska University Hospital in Gothenburg, Sweden, and ten patients at the Department of Endocrinology and Gastroenterology at Haukeland University Hospital in Bergen, Norway. All patients (7 males and 16 females, mean age 46.4 ± 11.5 years) had a verified diagnosis and disease duration of 30.5 ± 13 years. Treatment was managed with multiple insulin injection regimens or insulin pumps. Twenty-seven age and gender matched healthy controls (8 males, 19 females, aged 43.9 ± 11.1 years) were included for comparison purposes. Seven of the healthy controls were recruited at the Department of Endocrinology and Gastroenterology at Haukeland University Hospital in Bergen, Norway and twenty-one of the healthy controls were recruited at the Department of Gastroenterology and Hepatology, Aalborg Hospital, Aarhus University Hospital in Aalborg, Denmark. A standard 24-hour ambulatory electrocardiography (Holter monitoring) was performed in all participants, and heart beat-to-beat interval (RR) and standard deviation of heart beat-to-beat intervals (SDNN) were used for analysis as these parameters have been shown to correlate to mortality and disease (7). The local Ethical Committees at all three hospitals approved the study protocol (Gothenburg: 591-09, Bergen: 2010/2562-6, and Aalborg: N20090008).

2.2 Gastroparesis cardinal symptom index

To enable the patients to assess the severity of their symptoms we have used the Patient Assessment of Upper Gastrointestinal Disorder Severity Symptom Index (PAGI-SYM) (8). In its short form, the GCSI; Gastroparesis Cardinal Symptom Index, which consists of the first nine questions of PAGI-SYM, is a reliable and valid tool for measuring symptom severity in patients with gastroparesis (9). The GCSI is based on three subscales: 1) nausea/vomiting (3 items), 2) postprandial fullness/early satiety (4 items), and 3) bloating (2 items). Both patients and healthy controls were asked to rate the severity of the symptoms during the preceding two weeks. The GCSI items range from 0 - no symptoms, to 5 - very severe symptoms. We used a total average GCSI sore (9 items) for analyses.

2.3 Short form questionnaires (SF-36)

The SF-36 is a multi-purpose, short-form health survey with 36 questions. It yields an 8-scale profile of functional health and well-being scores, as well as two summary scores (physical component summary and mental component summary scores) (10,11). It is a generic measure, as opposed to one that targets a specific age, disease, or treatment group. Accordingly, the SF-36 has proven useful in surveys of general and specific populations, comparing the relative burden of diseases, and in differentiating the health benefits produced by a wide range of different treatments. The SF-36 questionnaire enables the patients to estimate their physical and mental health and it covers four weeks.

2.4 Sensory assessment

A modified visual analogue scale (VAS) with anchor words was used to assess the sensation. This scale was used as oesophageal sensation and pain are more diffuse and difficult to score than somatic sensations arising from e.g. the skin. The scale has, in numerous studies, shown to be suitable for the assessment of gastrointestinal sensation/pain intensity (12,13). The anchor words were: 1 = vague perception of mild sensation; 2 = definite perception of mild sensation; 3 = vague perception of moderate sensation; 4 = definite perception of moderate sensation; 5 = pain detection threshold; 6 = slight pain; 7 = moderate pain; 8 = medium pain intensity; 9 = intense pain; and 10 = unbearable pain. During electrical stimulation, the volunteers were asked when the stimulation intensity reached 1, 3, 5, and 7 on the VAS, and the stimulus intensities were noted. Further increase in stimulation intensity could result in severe autonomic reactions, leading to e.g. vomiting (14). The EEG recording was done at the pain detection threshold (VAS 5).

2.5 Electrical stimulation

Following 6 hour fasting, the blood glucose level was adjusted in all subjects to 6 mmol/l by use of a hyperinsulinemic-euglycemic clamp technique (15,16). The oesophageal probe consisted of a 6.2mm outer diameter catheter with eight channels of different sizes and a 5cm long polyurethane bag with a maximum diameter of 8cm without stretching the bag material (Ditens A/S, Aalborg, Denmark). Initially the probe was placed in the stomach. Then the bag was filled with water. The probe was then withdrawn until resistance was met (i.e. the lower oesophageal sphincter was identified). Then the bag was emptied and the probe was positioned in the distal third of the oesophagus (8cm above the lower sphincter) in order to ensure stimulation of primarily visceral afferents. The probe was then taped to the side of the lip and chin in order to keep it in place. Electrical stimulation was delivered by a computer-controlled constant current stimulator (DIGITIMER Ltd., Welwyn Garden City, United Kingdom). Each stimulus was applied as a single pulse of 2ms duration. The current intensity was increased in steps of 1 mA. Intermittent sham stimuli with the same current as in the previous stimulation were randomly delivered to blind the volunteers for the increase in stimulus intensity. Furthermore, the contact to the mucosa was tested with the measurement of the impedance (Aalborg University, Denmark) before stimulation. Impedance <2 kΏ indicated good contact with the mucosa, which was necessary for a controllable electrical stimulation.

2.6 EEG recordings

The surface EEG was recorded from 122 electrodes using an amplifier (Synamp II, Neuroscan, El Paso, USA) and a standard EEG cap (Quick-Cap International, Neuroscan, El Paso, USA) mounted according to the extended international 5-10 system. The reference electrode was placed just above the central electrode (Cz). The impedance of the electrodes was on average kept below 10 kΩ. Fifty stimuli were given corresponding to pain detection threshold (VAS 5), applied with a fixed inter-stimulus-interval of 5 seconds (0.2 Hz) by use of 1000 Hz sampling frequency. The EEG signals were acquired in a dimmed room, with all unnecessary electrical equipment turned off to avoid 50 Hz contamination of the signals. Subjects were instructed to lay calm and relaxed.

2.7 EEG analysis

Offline pre-processing of acquired EEG signals was done as follows: 1) band-pass filtered between 0.5 and 30Hz, 2) epoched from 100 ms before stimulus to 500 ms after stimulus, 3) epochs contaminated by eye movement were discarded, 4) epochs were averaged into EPs, & 5) EPs were average referenced for further analysis.

Latencies and amplitudes of EP peaks at central electrode (Cz) were identified and compared between healthy controls and patients. Then, brain source connectivity analysis was done in order to study the underlying brain networks generating the EPs. The method used for brain source connectivity has been described in detail elsewhere (17). Briefly, this was done in four steps: 1) Multichannel Matching Pursuit (MMP) was utilized to first extract the waveforms of the underlying brain networks. MMP is an adaptive and iterative algorithm. In the first iteration, the MMP searches a dictionary of Gabor atoms for a waveform that is correlated best simultaneously with all the measurement channels. This waveform is the first component. Once this component is found, it is subtracted from the EP signal under analysis, and this creates the first order residuum of the signal. In the second iteration, the MMP algorithm searches the dictionary for a waveform that is correlated best with the first order residuum. This gives the second component, which is then subtracted from the first order residuum to create the second order residuum of data. This procedure of finding the best correlated waveforms with the consecutive residues is repeated until the energy of the last computed residuum is below a chosen limit or a number of desired iterations is reached;

2) similar MMP components between subjects were identified and these were chosen for brain source analysis; 3) brain source localisation was done on the MMP waveforms to find out which brain sources "cross-talked" in a network: sources which have the same waveform (same frequency and phase) and occur in the same time window, are thought to cross-communicate in a network (18). Since MMP components are mono-frequency and have single topographies, each component is either generated by a single source or by a set of sources that operate synchronously (19). Any other type of activity would induce continuous changes in topography. MMP was done in MATLAB (version 7.0.1. The Mathworks, Inc., Natick, MA, USA) and brain source localization was done in Brain Electrical Source Analysis (BESA) (BESA Research 5.3, MEGIS Software GmbH, Gräfelfing, Germany); 4) lastly, percentage of contribution of each brain source to the network was calculated as follows: 1) area under the curve of the waveform for each brain source was calculated. This area represents the source strength; 2) sum of source strengths of all brain sources in the network was calculated, 3) percentage of each source strength contribution to the average was calculated.

2.8 Statistical analysis

Descriptive statistics are reported as mean ± SD. To compare data from the healthy controls with the patients, analysis were done by use of analysis of variance (ANOVA) to study differences in 1) QOL (SF-36), GCSI scores, RR, and SDNN, 2) stimulus intensity, 3) latency and amplitudes of EPs, and 4) source localisations. Pearson’s linear regression analysis was used for correlations between: 1) brain source location and GCSI score, QOL (SF-36), RR interval, and SDNN and 2) percent of contribution of each brain source to the network and the GCSI scores, QOL, RR interval, and SDNN respectively. Correlations were done on deviations from the means in healthy controls and were only done for brain sources where significant differences were seen between healthy controls and patients. The brain areas of interest were pre-hoc defined and between regions of interest, no method for multiple comparisons was used. This approach was used as it limits the likelihood of type II errors which would result in truly important differences being deemed non-significant (20). The software package Sigma Stat v.3.0 (SPSS Inc., Chicago; IL, USA) was used for the statistical analysis and P-values ≤0.05 were considered significant.

3. RESULTS

After analysing quality of EP data, twenty out of twenty-seven healthy controls and fifteen out of twenty-three patients were included in the final analysis. Of the fifteen included patients, four had diabetes type 2. The participants who were excluded had EPs with poor signal-to-noise ratio and the peaks could not be extracted.

3.1 Clinical parameters

Questionnaires and sensory results are reported only for the subjects included in the final analysis. Patient information and questionnaire results are shown in Table I. The SF-36 physical component summary score was 55.62 ± 3.98 in healthy controls and 33.08 ± 10.37 in patients (P<0.001). The SF-36 mental component summary score was 53.05 ± 5.69 in healthy controls and 38.96 ± 12.61 in patients (P=0.003). The average GCSI score was 0.29 ± 0.43 in healthy controls and 2.46 ± 1.34 in patients (P<0.001). The RR interval from resting state 24-hour electrocardiogram was 855.56 ± 95.49 ms in healthy controls and 714.21 ± 117.46 ms in diabetes patients (P<0.001). The SDNN was 153.22 ± 32.17 ms in healthy controls and 99.29 ± 51.04 ms in patients (P<0.001).

Table I. Patient information

ID

Age

Sex

DM

Duration

DM

Type

SDNN

RR

Nausea/

Vomiting

Postprandial Fullness

Bloating

Average GCSI

SF36 PCS

SF36

MCS

1

30

F

17

1

66

700

2.33

2.75

4.50

3.19

33.44

27.11

2

67

F

51

2

102

816

1.00

0.00

0.00

0.33

27.36

62.77

3

49

M

19

2

105

677

2.67

3.00

1.50

2.39

37.61

33.34

4

45

M

25

2

79

691

0.00

0.00

0.00

0.00

23.67

44.98

5

66

M

49

1

49

784

2.00

4.25

4.50

3.58

24.21

29.56

6

24

F

15

1

50

528

4.33

3.00

4.50

3.94

29.63

33.18

7

49

F

46

1

88

595

2.00

5.00

4.00

3.67

30.44

26.65

8

57

M

49

1

4.67

4.75

1.50

3.64

22.30

38.69

9

53

F

40

1

118

729

2.33

0.00

2.00

1.44

38.60

15.68

10

39

F

24

1

71

670

1.00

4.00

4.00

3.00

37.44

42.87

11

25

M

10

2

252

935

1.67

0.50

0.00

0.72

45.51

47.78

12

55

F

44

1

135

695

1.00

2.75

3.00

2.25

42.01

45.77

13

49

M

26

1

112

925

0.33

2.75

3.00

2.03

39.16

57.51

14

33

F

27

1

100

660

0.33

3.00

4.00

2.44

52.76

30.54

15

32

F

29

1

63

594

3.33

4.25

5.00

4.19

12.09

48.02

Abbreviations: DM – Diabetes Mellitus; SDNN - standard deviation of heart beat-to-beat intervals; RR – heart beat-to-beat intervals; GCSI - Gastroparesis Cardinal Symptom Index; PCS – Physical Component Summary; MCS – Mental Component Summary.

3.2 Sensory properties

There were no significant differences in sensory or pain thresholds between healthy controls and patients. Healthy controls required 16.6 ± 5.7 mA to evoke VAS 1 and patients required 22.0 ± 11.3 mA (P=0.1); healthy controls required 22.4 ± 6.2 mA to evoke VAS 3 and patients required 27.8 ± 13.1 mA (P=0.1); healthy controls required 29.8 ± 7.0 mA to evoke VAS 5 and patients required 31.3 ± 12.9 mA (P=0.7); healthy controls required 38.2 ± 7.8 mA to evoke VAS 7 and patients required 36.1 ± 12.7 (P=0.5).

3.3 Latency and amplitude

Grand means of the evoked potential traces in healthy controls and patients are shown in Figure 1. The first peak (P1) occurring at 103.7±15.8 ms in healthy controls and 119.1±27.8 ms in patients tended to be delayed in the patient group (P=0.08). N2 and P2 latencies were increased in patients (P < 0.05). The N2 and P2 peaks in healthy controls occurred at 164.9±12.5 ms and 285.4±33.8 ms, whereas they occurred at 192.9±32.2 ms and 327.1±34.2 ms in patients. The P1-N2 peak to peak amplitude was 5.08±3.1 µV in healthy controls and 4±2.5 µV in patients (P=0.3), whereas N2-P2 peak to peak amplitude was 8.9±4.3 µV in healthy controls and 6.2±2.9 µV in patients (P = 0.02).

Figure 1 approximately here

3.4 Brain source connectivity

There were three dominant brain networks which were present in both healthy controls and patients. These networks were analysed in detail, in order to observe how they were modified in patients. Network details are presented in Table II and visualized in Figures 2a-c.

The first activated network occurred at 8.3±2.5 Hz and was present in 17/20 healthy controls, whereas in patients, it occurred at 7.9±2.0 Hz and was present in 13/15 subjects. The network consisted of a deep source (likely representing brainstem or thalamus), bilateral operculum, and frontal source. Throughout this article, we use the term operculum to represent both insula and secondary somatosensory cortex. As these two areas are anatomically very close, with the spatial resolution of dipoles, they could be indistinguishable. The activation of this network peaked at 109.0±19.0 ms in healthy controls and at 141.4±36.5 ms in patients; there was a significant delay in patients (P=0.001). The overall contribution of the brainstem/thalamus source to the network was significantly lower than in healthy controls (P=0.01). Moreover, the operculum source was localised deeper and more anterior in patients (P≤0.001).

The second network occurred at 3.6±0.6 Hz and was present in 18/20 healthy controls, whereas in patients, it occurred at 3.6±0.5 Hz and was present in 14/15 subjects; the network consisted of cingulate and operculum sources which persisted over the entire time interval and peaked at 198.4±86.3 ms in healthy controls and at 196.3±55.7 ms in patients. Operculum source in this network was also localized deeper and more anterior in patients (P≤0.001). The contribution of the cingulate source (62.6 ± 14.2 %) to this network was much higher than the contribution of operculum (37.4 ± 14.2 %) in healthy controls. In patients, on the other hand, the operculum source had a higher contribution (cingulate: 41.4±19.9 % and operculum: 58.6±19.9 %) (P<0.01).

The third network was a late network occurring at 4.3±1.1 Hz and was present in 11/20 healthy controls, whereas in patients, it occurred at 4.4±1.5 Hz and was present in 11/15 subjects; the network consisted of mid-cingulate (MCC), anterior cingulate (ACC), bilateral operculum, and a deep source (likely representing deep limbic system). The activation of this network peaked at 383.1±51.5 ms in healthy controls and at 379.7±69.6 ms in patients. Operculum source was localized deeper and more anterior, while the deep limbic source was localized deeper in patients relative to healthy controls (P<0.001 and P=0.04).

Figures 2a-c approximately here

Table II. Brain source location coordinates and their overall contributions to the network

HEALTHY CONTROLS

Brain Network 1

Brainstem/Thalamus

Operculum

Right Frontal

Left frontal

RV(%)

X

Y

Z

X

Y

Z

X

Y

Z

X

Y

Z

Mean

7.5

-0.3

-23.5

-27.6

±36.3

-2.1

12.4

30.3

21.0

42.8

-27.3

22.4

44.0

SD

1.2

5.6

6.2

8.2

5.4

6.3

5.0

7.0

9.7

7.0

8.0

8.1

7.7

Overall Contribution

32.5±13.3 %

34.1±9.3 %

33.4±10.2 %

Brain Network 2

MCC

Operculum

RV(%)

X

Y

Z

X

Y

Z

Mean

8.0

-0.8

-2.6

37.0

±36.0

-2.6

12.4

SD

1.3

6.7

11.4

4.5

5.2

5.6

4.6

Overall Contribution

62.6 ± 14.2 %

37.4 ± 14.2 %

Brain Network 3

MCC

ACC

Operculum

Deep Limbic System

RV(%)

X

Y

Z

X

Y

Z

X

Y

Z

X

Y

Z

Mean

8.1

-2.2

-1.0

36.1

-0.4

35.1

7.9

±36.0

-2.8

13.2

±23.2

-11.2

-17.1

SD

1.7

5.3

6.2

2.8

5.3

4.9

3.7

4.0

5.7

4.7

2.5

6.9

1.8

Overall Contribution

23.1±5.4 %

18.1±6.3 %

26.9±6.7 %

31.9±8.7 %

DIABETES PATIENTS

Brain Network 1

Brainstem/Thalamus

Operculum

Right Frontal

Left frontal

RV(%)

X

Y

Z

X

Y*

Z*

X

Y

Z

X

Y

Z

Mean

7.5

-1.7

-26.1

-33.2

±36.4

4.1

1.7

32.4

24.8

39.8

-26.0

20.2

38.9

SD

1.6

4.9

3.9

5.8

4.7

4.8

7.1

6.3

17.5

8.9

7.8

21.5

10.6

Overall Contribution

22.5±11.8 %*

41.4±13.6 %

36.1±10.5 %

Brain Network 2

MCC

Operculum

RV(%)

X

Y

Z

X

Y*

Z*

Mean

8.8

2.5

-4.1

41.3

±37.1

5.1

0.7

SD

1.0

5.6

8.4

5.5

5.0

5.8

6.7

Overall Contribution

41.4±19.9 %*

58.6±19.9 %*

Brain Network 3

MCC

ACC

Operculum

Deep Limbic System

RV(%)

X

Y

Z

X

Y

Z

X

Y*

Z*

X

Y

Z*

Mean

8.9

0.03

-4.1

33.7

0.2

36.0

5.5

±37.9

5.9

-0.3

±25.7

-14.6

-21.5

SD

0.9

4.5

12.9

4.4

3.3

6.2

6.0

3.4

5.6

6.5

2.8

7.1

2.5

Overall Contribution

21.3±10.1 %

18.3±8.3%

29.5±5.9 %

30.9±15.1 %

X coordinate: right-left, Y coordinate: anterior-posterior, Z coordinate: up-down. Significant differences are represented by *. ACC: anterior cingulate cortex; MCC: mid cingulate cortex; RV: residual variance; SD: standard deviation.

3.5 Clinical correlations in patients

Deviations from the healthy mean of the brain sources which were altered in patients were correlated with the GI symptoms, QOL, RR intervals, and SDNN. These results are presented in Figures 3a-c. In all three brain networks, deviations of the y-coordinate (anterior-posterior direction) and z-coordinate (up-down direction) of operculum source were positively correlated with the GCSI score (P<0.001). This means that, the more clinical symptoms the patient had, the deeper and more anterior the operculum source was located. The RR interval and SDNN were negatively correlated to deviations of y and z coordinates of the operculum source in all three networks (P < 0.05), which means that, faster heart rate and lower standard deviation of heart beat-to-beat intervals entail deeper and more anterior operculum source location. In the cingulate-operculum network, deviations of the percentage of operculum and cingulate contributions were negatively correlated with SF-36 physical component summary score (P=0.02) and with SDNN (P=0.02). This means that, as the physical state of the patient gets worse and as the SDNN decreases, the contribution of cingulate source decreases and contribution of operculum source increases. Finally, deviation of the z-coordinate of the deep limbic source in the last network was negatively correlated with SF-36 physical component summary score (P=0.02), which means that, the worse the physical state of the patient is, the deeper the deep limbic source was located.

Figures 3a-c approximately here

4. DISCUSSION

Patients with diabetes and GI symptoms showed changes in three brain networks to painful electrical stimulation in the oesophagus: 1) deep source (likely representing brainstem or thalamus)/operculum/frontal cortex, 2) operculum/cingulate, and 3) mid-cingulate cortex/anterior-cingulate cortex/operculum/deep source (likely representing deep limbic system). The offset of the first network was delayed in patients. In all three networks, the operculum source was localised deeper and more anterior in patients and this was related to severity of upper GI symptoms. The shift of the operculum source was also related to a decreased RR interval and decreased SDNN. Additionally, in the second network, the operculum source had the largest contribution to the network in patients, whereas the cingulate source had the largest contribution in healthy controls. In patients, this operculum dominance over cingulate contribution to the network was associated with a decrease in SF-36 physical component summary scores (reduced QOL) and decreased SDNN. In the last network, deep limbic source was localized deeper in patients and this was also correlated to decreased SF-36 physical component summary scores.

4.1 Methodology

Evoked potentials to electrical stimuli in the oesophagus were investigated in this study. Although electrical stimuli bypass the receptors and are not as physiological as, e.g. mechanical stimuli, they affect the same pathways and result in similar morphology and inter-peak characteristics of the evoked brain response (21). Moreover, electrical stimuli are favoured as most other stimulation paradigms are difficult to control in the gut. The advantage of evaluating brain signals using EPs is their excellent time resolution, being in the order of milliseconds. As the sensory signal reaches the brain within milliseconds, imaging methods such as fMRI and PET, with their time resolution in the order of seconds may give blurred information of brain activity, which may or may not be sensation specific. The disadvantage of EPs, however, is their limited spatial resolution. Furthermore, locations of brain sources obtained by source reconstruction are not precise indicators of activated areas, but represent the so called "center of gravity" where most of the brain activity is occurring. Hence, the major challenge in EP analysis is the brain source reconstruction. However, the latest mathematical advancements in inverse modelling offer the possibility to give a valid estimate of the dominating brain activity with high accuracy (22,23). In this study, we used MMP in combination with source reconstruction in order to study brain networks. This method has been proven to be reliable in presence of different levels of noise and increasing number of brain sources (23).

4.2 Brain processing in healthy volunteers

The evoked potential latencies correspond to what has previously been found due to electrical stimulation in oesophagus in healthy volunteers (24-27). There were three dominant brain networks generating these EPs, consisting of: 1) deep source (likely representing brainstem or thalamus)/operculum/frontal cortex, 2) cingulate/operculum, and 3) MCC/ACC/operculum/deep source (likely representing deep limbic system). The brain networks in this study are consistent with an earlier study where we investigated the visceral pain networks generating EPs to painful oesophageal stimulation in healthy population (17). The early network between brainstem/thalamus, operculum, and frontal cortex reflects the fact that the sensory impulse first reaches brainstem and is directly sent to operculum and frontal cortex via thalamo-cortical projections to initiate pain perception (28). There is much debate in the research community suggesting that the organization of thalamic neurons constitute closed fields making it impossible to record their activity on the scalp. However, there is evidence suggesting thalamus is not a closed field and thus one should be able to find corresponding far field recordings (29). Moreover, a number of studies have consistently found a thalamus source of P14 potential in somatosensory evoked potentials (30,31) and a brainstem source in auditory brainstem evoked potentials (23). Operculo-insular cortex is thought to be the first activated brain region which triggers the pain network in order to give rise to pain experience(32). The cingulate/operculum network persists over the entire time interval, peaking at 198.4±86.3 ms and oscillating at 3.6±0.6 Hz. It has previously been suggested that, depending on the threat level, the anterior insula connects with the MCC already before the stimulus to adjust its sensitivity for the upcoming stimulation (33). This agrees with our finding, where the waveform appeared to start rising, already before the stimulus could be registered and the network persevered over the entire time interval under analysis. The latest activated network between MCC, ACC, operculum, and a deep source (deep limbic system) reflect structures involved in processing of emotional and attentional tasks (34). Interestingly, operculum was present in all three networks. In their review, Augustine et al. discussed that the insula has an important function for integrating visceral sensory and motor activity together with limbic integration and is particularly important in pain perception from the viscera (35). Consistent with this, our finding supports that the opercular cortex plays the main integrative role between the three networks.

4.3 Brain processing in diabetes patients

In this study, prolongation of EPs in diabetes patients was seen, which is consistent with the fact that diabetic autonomic neuropathy induces a decrease in conduction velocity of both peripheral and central Aδ-fibres (36,37). This is seen in the offset of activation of the first brain network, which was delayed in patients, meaning that the pain signal takes longer to reach the brain from periphery. While the decrease in conduction velocity of Aδ-fibres explains the delay in latencies, it does not explain the abnormal brain activation which was seen in diabetes patients. A reduction of EP amplitudes was also seen. This is consistent with previous findings where EPs in diabetes patients had reduced amplitudes (6,38). Although, the same brain networks dominated in diabetes patients as healthy controls, important modifications of brain sources within networks were seen, most likely explaining the reduced amplitudes. The decreased brain source strength of the brainstem/thalamus source in the first network could indicate that the afferent signal strength is dampened either at spinal cord or at the brainstem level, leading to abnormal processing in the brain. However, it could also be related to reduced density/function of peripheral and spinal afferent fibres. In all three networks, the operculum source was located deeper and more anterior in patients. Interestingly, this shift of the operculum source correlated with the patient symptom score, manifested as nausea/vomiting, fullness/early satiety, and bloating. Moreover, as the physical component summary score and SDNN of a patient decreased, the overall contribution of operculum source to the operculo-cingulate network increased. Previous findings have demonstrated that direct electrical stimulation of the anterior insula leads to nausea and vomiting (39) and lesions of insula can cause ictal vomiting (40). These are the symptoms seen in this cohort of diabetes patients, and thus, we propose that the shift of operculum source and its increase in contribution to the operculo-cingulate network contributes to the development and/or maintenance of GI symptoms seen in DM. The RR intervals and SDNN were lower in patients than in healthy controls, and the lower RR intervals and SDNN were correlated to the shift of the operculum source. This correlation of autonomic dysfunction to the shift in operculum source is consistent with the fact that both autonomic regulation and visceral pain/sensation are processed in similar brain areas (41,42). The abnormality in RR interval and SDNN could be indicative of cardiac autonomic neuropathy in diabetes patients, which implies higher mortality risk (7). Moreover, the deep limbic source was localized deeper in patients in the latest activated network, which was related to physical component summary score of the patient. The worse the physical QOL was, the deeper the brain source was localized. Deep limbic system is involved in affective component of pain, mainly dealing with emotion. Diabetes is associated with depression, moderate slowing of mental speed, and diminished mental flexibility (43,44). Although the pathogenesis of these symptoms are clearly multimodal, abnormal activity in deep limbic structures could be a contributing factor.

4.4 Conclusion

This study suggests central nervous system involvement in diabetes patients with GI symptoms. This was revealed by the changes in brain networks relevant for visceral sensations. Moreover, the study confirms the peripheral gut neuropathy involvement in diabetes. These findings might explain some of the pathophysiology behind gastrointestinal symptoms in patients with diabetes and ultimately contribute to better diagnosis and treatment of these patients.

ACKNOWLEDGEMENTS:

The research leading to these results has received funding from the European Community´s Seventh Framework Programme FP7/2007-2013 under grant agreement no. 223630.

No competing interests declared.

FIGURE LEGENDS

Figure 1. Evoked potentials at the central site on the scalp. Latencies (N2 and P2) were delayed in patients and N2-P2 amplitude was lower. The figure is of grand-mean, although the analysis was done on an individual basis. The peak at 0-30 ms represents the stimulus artefact.

Figure 2a. Grand mean plot of brain sources and connections between them (left) and their waveforms (right) in the brainstem/thalamus-operculum-frontal network. The source locations are labelled by the initial of the corresponding brain area: B = brainstem, F = frontal, and O = operculum. The operculum source was localized deeper and more anterior in patients than healthy controls. Moreover, latencies of all brain sources were prolonged and amplitudes were significantly reduced in the brainstem/thalamus source in patients. The figure is of the grand mean, although analysis was done on an individual basis.

Figure 2b. Grand mean plot of brain sources and connections between them (left) and their waveforms (right) in the cingulate-operculum network. The source locations are labelled by the initial(s) of the corresponding brain area: MC= mid-cingulate and O = operculum. It can be seen that the operculum source was localized deeper and more anterior in patients than healthy controls. Moreover, the amplitude of the cingulate source was lower for patients, whereas the amplitude of operculum source was higher.

Figure 2c. Brain sources and connections between them (left) and their waveforms (right) in the MCC – ACC – operculum – deep limbic network. The source locations are labelled by the initial(s) of the corresponding brain area: DL = deep limbic, AC = anterior cingulate, MC = mid-cingulate, and O = operculum. Operculum was localized deeper and more anterior, while deep limbic source was localized deeper in patients.

Figure 3a. Operculum correlations. All correlations were done on patients’ brain source deviations from the healthy mean; for details in the interpretation please see results section. Top: Correlations in brainstem/thalamus-operculum-frontal network. Deviations from the healthy mean of Y coordinate (anterior-posterior direction) and Z coordinate (up-down direction) of opercular source were correlated with the GCSI score, the RR interval, and the SDNN. Middle: Correlations in MCC-operculum network. Deviations from the healthy mean of Y and Z coordinates of the opercular source were correlated with the GCSI score, with the RR interval, and with SDNN. Bottom: MCC-ACC-operculum-deep limbic network. Deviations from the healthy mean of the Y and Z coordinates of the opercular source were correlated with the GCSI score, the RR interval, and SDNN. Abbreviations: PCS: physical component summary, GCSI: gastroparesis cardinal symptom index; MCC: mid-cingulate cortex; ACC: anterior-cingulate cortex.

Figure 3b. Contribution to the brain network correlations in the MCC-operculum network. Left: deviations from the healthy mean of MCC and operculum contributions to the network were negatively correlated with the SF-36 physical component summary score. Right: deviations from the healthy mean of MCC and operculum contributions to the network were negatively correlated with SDNN.

Figure 3c. Deep limbic system correlations in the MCC-ACC-operculum-deep limbic network: deviation from the healthy mean of the Z coordinate of the deep limbic source location was negatively correlated with the SF-36 physical component summary score.