Effect of the Glucagon-like Peptide-1 Receptor (GLP-1R) Agonists on Autonomic Function in Subjects With Diabetes: a Systematic Review and Meta-analysis.

Background. In addition to the metabolic effects in diabetes, glucagon-like peptide 1 receptor (GLP-1R) agonists lead to a small but substantial increase in heart rate (HR). However, the GLP-1R actions on the autonomic nervous system (ANS) in people with diabetes remain still debated. Therefore, this meta-analysis evaluates the effect of GLP-1R agonist chronic treatment on measures of ANS function in people with diabetes. Methods. According to the Cochrane Collaboration and PRISMA statement, we conducted a meta-analysis considering clinical trials in which the autonomic function was evaluated in people with diabetes chronically treated with GLP-1R agonists. The outcomes were the change of ANS function measured by heart rate variability (HRV) and cardiac autonomic reex tests (CARTs). Results. In the studies enrolled, HR signicantly increased after treatment (p<0.001), whereas low frequency/high frequency ratio did not differ (p=0.410); no changes in other measures of HRV were detected. Considering CARTs, only the 30:15 value derived from lying-to-standing test was signicantly lower after treatment (p=0.002), but only two studies reported this measurement. No differences in other CARTs outcome were observed. Conclusion. The present meta-analysis conrms the HR increase but seems to exclude an alteration of the sympatho-vagal balance due to chronic treatment with GLP-1R agonists in diabetes, considering the available measures of ANS function.


Introduction
The diabetic autonomic neuropathy (DAN) is de ned as a heterogeneous category of disorders of the autonomic nervous system (ANS) in individuals with either diabetes mellitus (DM) or metabolic derangements of pre-diabetes, when other potential causes have been excluded 1 . In particular, the cardiac autonomic neuropathy (CAN) is the manifestation of an ANS imbalance, due to the impairment of autonomic control of the cardiovascular system 1,2 . CAN affects at least 20% of unselected patients, and this incidence raises up to 65% of subjects with either increasing age or long diabetes duration [2][3][4] . However, the actual CAN prevalence varies, depending upon diagnostic criteria, patient cohort, and testing modality 2,5 . According to standard cardiovascular re ex tests (CARTs), the CAN prevalence is reported about 7% in type 2 DM (T2DM) and increases with diabetes duration by 4.6-6% per year 2,5 . It is well known that CAN incidence is in uenced by diabetic disease duration, patient age, glycaemic control, and concomitant metabolic syndrome features 6,7 . The CAN diagnosis in people with diabetes is extremely relevant, in uencing the prognosis for cardiovascular morbidity and predicting the overall cardiovascular risk 8 . Indeed, CAN is related to silent myocardial ischemia, stroke, postural hypotension, exercise intolerance and enhanced intraoperative instability 9 . Therefore, from a clinical perspective, an individual with diabetes and CAN is at higher risk of mortality and of cardiovascular complications with heavy impact on morbidity and prognosis.
Glucagon-like peptide 1 receptor (GLP-1R) agonists represent a relatively new class of anti-hyperglycemic agents, addressing most of the pathophysiological mechanisms involved in the development of T2DM. The main GLP-1R agonist actions are the stimulation of insulin secretion, the inhibition of glucagon secretion, the delay of the gastric emptying time and the stimulation of neogenesis of insulin-secreting cells 10 . Moreover, GLP-1R agonists show favourable effects on body weight and metabolic pro le, with a lower risk of hypoglycaemia. Moreover, the favourable effects on body weight and metabolic pro le, together with a reduction in blood pressure (BP), contribute to reduce the cardiovascular risk. For all these reasons, the GLP1-R agonist use is constantly increasing to treat T2DM.
Among bene cial GLP-1R agonist effects, heart rate (HR) increase has been observed. Despite the resting HR increase could be considered a safety concern 11 , GLP-1R agonist administration is associated to major adverse cardiovascular events (MACE) reduction, including stroke, cardiovascular and all-cause mortality, as suggested by the cardiovascular outcome trials (CVOTs) [12][13][14][15][16][17][18] (Table 1). This GLP-1R agonist favourable effect on cardio-metabolic health is becoming increasingly evident, although the underlying mechanisms remain largely unknown. Improvement of hypertension 19 , endothelial function 20 and a natriuretic GLP-1R agonist-related effect 21 have been proposed. However, while a small but substantial increase in HR by about 3 bpm has been reported by the majority of the available trials 22 , action of GLP-1R agonists on autonomic function in diabetes remains still debated. With this in mind, it is clear that the GLP-1R agonist effects on HR and on ANS need to be reconciled with the favourable cardiovascular outcomes in clinical trials. Thus, we performed this meta-analysis with the main aim to highlight the GLP-1R agonist actions on available measures of autonomic function in diabetic people.

Materials And Methods
This meta-analysis was performed according to the Cochrane Collaboration and PRISMA statement. To ensure originality and transparency of the review process, the meta-analysis was a priori registered in the International Prospective Register of Systematic Reviews (PROSPERO; registration number CRD42020218063). The literature search was performed until November 10th, 2020 considering the following string: (((autonomic function)) OR ((autonomic dysfunction)) OR (cardiac autonomic neuropathy) AND ((((((((GLP-1 receptor agonist)) OR (GLP-1R agonist)) OR (semaglutide)) OR (liraglutide)) OR (exenatide)) OR (lixisenatide) OR (dulaglutide)) OR (albiglutide). Medline, Embase and Cochrane databases were considered.
All available GLP-1R agonists were considered potentially eligible, whether applied in clinical trial aiming at evaluating autonomic function before and after chronic administration in participants with diabetes. Thus, each parameter of autonomic function was considered before and after chronic treatment in the study groups. Moreover, when available, mean parameters were compared between study and control groups after treatment.

Endpoints
CARTs represent the gold standard in autonomic testing 2,23 . CARTs involve measuring autonomic responses through changes in HR and blood pressure (BP) to provocative physiological manoeuvres. The standard CARTs recommended for diagnosis of CAN include: the deep breathing (DB) test (expiration/inspiration ratio, E/I ratio), the lying-to-standing (LS) test (30:15), the Valsalva manoeuvre (VM) and the BP response to standing 2 . E/I index from DB test represents the ratio between the 3 maximum and the 3 minimum RR intervals (the intervals between two consecutive eletrocardiogram R waves) in a cycle of expiration and inspiration. In LS test the maximum/minimum 30:15 value is the ratio of the longest RR interval measured between the 25th and 35th beat after the change of posture and the shortest RR interval measured between the 10th and 20th beat. The Valsalva manoeuvre test indicates the ratio between the longest RR interval after exhalation and the shortest RR interval during exhalation. Other approaches in clinical research are currently available to evaluate CAN such as heart rate variability (HRV), barore ex sensitivity, muscle sympathetic nerve activity, heart sympathetic imaging 24 . Non-invasive and widely used in clinical research, HRV provides key information about autonomic -parasympathetic and sympatheticmodulation of the cardio-vascular system. The measurement of HRV includes two domains. In the frequency domain, the components of the HRV obtained by spectral analysis consist of low frequency (LF) and high frequency (HF) indices. These indices provide information about both the sympathetic and parasympathetic in uences on heart. Thus, the LF/HF ratio is the index of sympathetic-parasympathetic balance. Time domain measures of the normal RR intervals mainly include differences between the longest and the shortest RR intervals, the standard deviations of RR intervals (SDNN), and the square root of the mean squared difference of successive RR intervals (RMSSD) 24 .

Study selection and inclusion criteria
The literature search evaluated all clinical trials with the following inclusion criteria: i) either interventional or observational, ii) in which the autonomic function was evaluated, iii) in people with diabetes, iv) treated with GLP-1R agonists. In particular, change of autonomic nervous system function before and after chronic GLP-1R agonist treatment using measures of heart rate variability was considered the main outcome. Treatment was considered chronic when GLP-1R agonist was administered for more than 4 weeks. Both participants with type 1 and type 2 DM were considered eligible. Randomization and presence of controls were not considered as inclusion criteria.
Data collection process and quality The risk of bias was assessed independently by two investigators (CG and DS), using Cochrane risk-of-bias algorithm. In particular, the following criteria were considered for each included trial, (i) randomization method, (ii) concealment of allocation, (iii) presence or absence of blinding to treatment allocation, (iv) presence or absence of blinding of outcome assessment, (v) potential incomplete data biases and (vi) reporting biases.

Data synthesis and analysis
Using the Review Manager (RevMan) 5.4 Software (Version 5.4.1 Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014), continuous variables were comprehensively evaluated as inverse variance of mean variables. When data were reported in the original manuscript as median or logarithm, they were transformed in mean ± standard deviation. Indeed, mean ± standard deviation is required for the meta-analytic approach. However, since these parameters could be obtained using different approaches, the meta-analyses were performed using standard mean difference. Considering that studies included in the meta-analysis reported different treatment durations and different time-points in which endpoints were evaluated, we considered the last available time point for each trial (median 24 weeks; min 12 -max 72 weeks). The degree of heterogeneity among the studies included in each analysis was examined by inspecting both the scatter in the data points and the overlap in their con dence intervals (CIs), and by performing I 2 statistics. The inverse variance with the xed model was initially chosen, whereas the random effect model was preferred in case of I 2 higher than 60%. Sensitivity analyses were performed, considering the type of diabetes and the GLP-1R agonist used, when possible. Values of p < 0.05 were considered statistically signi cant.
Results 596 manuscripts were identi ed by the literature search (Fig. 1). After abstract evaluations, 16 studies were considered for the full text analysis (Fig. 1). Among these, ten studies were excluded (reasons reported in Fig. 1) and six studies were enrolled [25][26][27][28][29][30] in the nal analysis (Fig. 1). Table 2 summarizes the characteristics of the studies included in the meta-analysis. Moreover, Table 3 shows the characteristics of the patient cohorts included in the analysis: a total of 182 individuals with diabetes were enrolled (112 men and 72 women), with a mean age of 54.68 ± 10.7 years. Only two studies enrolled patients with type 1 diabetes mellitus. Abbreviations: CARTs, cardiovascular re ex tests; DB, deep breathing; E/I ratio, Expiration/inspiration ratio; ER, extended release; HF, high frequency; HR, heart rate; HRV, heart rate variability; LF, low frequency; LS, lying-to-standing; RMSSD, root mean square of successive differences; SDNN, standard deviation of beat-to-beat (NN) intervals; T1D, type 1 diabetes; T2D, type 2 diabetes; VM, Valsalva manoeuvre.

Secondary outcomes
HR was reported in all trials, for a total of 182 patients, evaluated before and after chronic treatment. The HR signi cantly increased after treatment (p < 0.001), with a low heterogeneity rate (6%) (Fig. 3). This signi cant improvement remained considering the use of liraglutide or exenatide separately (Fig. 3). Since Cacciatori et al. did not report a control group 26 , this study was excluded when study and control groups were compared post-treatment. In this analysis, HR was not signi cantly different between study and control groups (mean difference − 0.86; 95%CI: -3.26, 1.55, I 2 = 16%, p = 0.480) (Additional le 1: Supplementary Fig. 3).
Finally, the risk of bias was evaluated (Additional le 1: Supplementary Fig. 10), showing a good quality of the included studies, in terms of reporting and incomplete biases. On the contrary, a variable quality in terms of blinding was detected, re ecting the di culty to perform a completely randomized, double-blind clinical trial in this setting.

Discussion
Here, we con rm that chronic GLP-1R agonist administration increases HR, according to a previously available meta-analysis 22 . Systematically, the HR increase after chronic GLP-1R agonist administration is evident in ve of six included trials. Indeed, only one study found no signi cant changes in HR after GLP-1R agonist administration, but, differently from the others, authors used a short-acting GLP-1R agonist, such as exenatide, for much longer time 28 . Thus, we could speculate that the GLP-1R agonist effect on HR could depend on the molecule used and the duration of the administration. Moreover, together with the HR increase, no signi cant change in other ANS-related parameters is evident in our meta-analysis. This result suggests that the chronic GLP-1R agonist administration may not in uence the sympathetic and parasympathetic functions. Thus, we could speculate that the HR modi cations induced by GLP-1R agonist is not consequence of sympathetic or parasympathetic stimulation, but other mechanisms should be involved.
GLP-1R agonists are increasingly used in clinical practice in diabetes, considering the wide range of positive effects on glucose homeostasis, body weight, BP and the low risk of hypoglycaemia. However, an overall GLP-1R agonist effect on ANS is far from being elucidated, even with the meta-analytic approach. Considering each study separately, interesting results could be extracted. Jaiswal et al. did not detect any change in autonomic function after 18 months of treatment with the short-acting GLP-1R agonist exenatide in patients with T2DM, evaluating either the gold-standard CARTs, such as DB and VM or measures of HRV, such as HR, LF/HF, SDNN, RMSSD 28 . In particular, that study was characterized by different follow-up length (18 months) than other studies included in the analysis (range 12-26 weeks). Therefore, we could not exclude that this difference represents a confounding factor, or a determinant of a kind of mitigation of HR increase over time. On the contrary, Kumarathurai  Accordingly, our ndings show no difference in the LF/HF ratio after treatment, considering both different molecules (exenatide and/or liraglutide) and both type of diabetes (T1DM and/or T2DM). Moreover, others HRV measures, such as SDNN and RSMSSD, do not change after chronic GLP-1R agonist administration. These results are con rmed considering liraglutide and exenatide separately, suggesting no differences between molecules. LF/HF ratio represents an index widely used in clinical practice for CAN evaluation in diabetic patients, providing information about autonomic -parasympathetic and sympatheticmodulation of the cardiovascular system. Even if considered among the methods of investigation for cardiac autonomic dysfunction in human research studies 24 , LF/HF ratio from HRV study con rms to be a measure not accurate and not directly related to sympatho-vagal balance, according to previous studies 32, 33 . Among CARTs measurements, representing the gold standard for the diagnosis of CAN, the only parameter that seems to be in uenced by GLP-1R agonist administration is the 30:15 ratio, which decreases after treatment. However, the strength of this result is limited by the small number of trials reporting this parameter. Thus, our meta-analysis suggests that chronic GLP-1R agonist treatment does not in uence the sympatho-vagal balance in people with diabetes. Hence, the HR increase could depend on different mechanisms.
In animal models, GLP-1 engages GLP-1R in central, peripheral, and autonomic nervous systems, enhancing the sympathetic nervous system activity, and reducing the parasympathetic nervous system activity 34 . In this regard, Baggio et al. suggested that the GLP-1R agonist-related HR increase is the nal effect of direct chronotropic action, which is attenuated by propanolol but not by atropine 34 . Moreover, the in vivo GLP-1R agonist administration induces c-fos expression -a marker of neuronal activity -in the adrenal medulla, activates neurons involved in autonomic control in the brain, and activates tyrosine hydroxylase transcription in brainstem catecholamine neurons 35 . These ndings suggest that the central GLP-1 action could be involved in the regulation of the sympathetic pathway 35 39 . Again, other experimental trials suggested the sympathetic nervous system activation after GLP-1R agonist infusion in healthy individuals 40,41 . All these examples demonstrate that the action of GLP-1R agonist on the sympathetic and parasympathetic systems must be both direct and indirect but this should be further studied with properly designed clinical and experimental trials. Indeed, even the present meta-analytical approach is not able to reach conclusive results, since it is still based on a limited number of studies with small sample size. Thus, the lack of signi cant effects of GLP-1R agonist chronic administration on ANS might be related to the limited amount of data available so far.

Conclusion
In conclusion, the present meta-analysis con rms the HR increase but seems to exclude an alteration of the sympatho-vagal balance due to chronic treatment with GLP-1R agonists in people with diabetes. Indeed, despite the accumulating data linking GLP-1R signalling to autonomic and neuroendocrine responses, the neural pathways underlying these actions are not fully understood. Furthermore, considering some discrepancies in the available preclinical and clinical ndings, it is conceivable to suggest possible species-speci c patterns of GLP-1R, as well as differences among GLP-1R agonists. More information is needed on the mechanisms through which the GLP-1R agonists administration may affect autonomic activity in individuals with diabetes.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.