A systematic review of preclinical studies exploring the role of insulin signalling in executive function and memory

Beside its involvement in somatic dysfunctions


Introduction
Whilst impaired insulin signalling has been traditionally associated with metabolic dysfunctions like type 1 (T1DM) and type 2 diabetes mellitus (T2DM), metabolic syndrome (MS), and obesity (Klimova et al., 2018;Landau and Pinhas-Hamiel, 2019; van de Vondervoort et al., 2016), recent evidence indicates that it is also associated with impairments in cognitive capabilities (Moheet et al., 2015).Both T1DM and T2DM are associated with mental and motor slowing and decrements in attention and executive functioning (McCrimmon et al., 2012).Memory deficits are frequently reported in patients with T2DM (Zhang et al., 2015); moreover, anhedonia and impulse control disorders (eating disorders and addiction) are often associated with diabetes (De Jonge et al., 2014).These symptoms constitute hallmarks of specific mental disorders in which, accordingly, alterations in insulin signalling have been observed: attention deficit hyperactivity disorder (ADHD) (Landau and Pinhas-Hamiel, 2019), Alzheimer's disease (AD) (Burillo et al., 2021), obsessive-compulsive disorder (OCD) (Grassi et al., 2022), depression (Lyra e Silva et al., 2019), and drug addiction (Brambilla et al., 1976).The socioeconomic costs of these diseases are huge and continue to rise.For example, the International Diabetes Federation estimated that while 537 million adults (20-79 years) are currently diabetic, these numbers are projected to increase steadily by 2045, when 783 million adults will live with this condition.The yearly associated healthcare economic burden of diabetes per se accounts for approximately 1.2 trillion US dollars (da Saúde, 2017).Should altered insulin signalling also represent a risk factor for mental disorders, these costs would further increase.For example, although the contribution of insulin signalling to e.g.AD still needs to be elucidated, approximately 35 million people worldwide currently have this form of dementia (World Health Organization, 2022).Predictably, these numbers are projected to increase at an unsettling rate (approximately 85 million AD patients by 2050) (World Health Organization, 2022).The aforementioned estimations are only the tip of the iceberg whereby they do not account for other insulin-related mental comorbidities like ADHD and OCD.Thus, exploring the role of insulin signalling in executive functions and memory may beget remarkable advantages in terms of public health and its associated costs.
How insulin regulates somatic functions has been the key question of countless scientific studies.Most early investigations focussed on its role in mediating general metabolism (Samson and Garber, 2014).These studies contributed to understanding how insulin regulates glucose homeostasis and energy balance (Boden, 2001;Brown and Walker, 2016;Huang, 2009) and how derailments in these processes result in metabolic disorders like T1DM and T2DM (DeFronzo, 2004), and MS (Banday et al., 2020).The primary deficit in T2DM is insulin resistance (Brunton, 2016), characterised by a reduced insulin sensitivity of cells in peripheral tissues (e.g., muscles, liver, and adipose tissue).This generates hyperfunction of the β-cells of the pancreas which ultimately elicit hyperinsulinemia: an increased production of insulin aimed at maintaining normal blood glucose concentrations.This process gradually impairs β-cells functionality and causes insulin deficiency and hyperglycaemia, with fasting plasma glucose concentrations > 110 mg/dL (Banday et al., 2020) representing the symptomatic threshold for T2DM.
Recently, the interest for the role of insulin has started to extend beyond energy metabolism to encompass the central nervous system (CNS) (Banks et al., 2012;Clarke et al., 1986).Evidence for a role for insulin signalling in the CNS is related to its widely-expressed (Chiu et al., 2008) receptors (e.g.Insulin Receptor, IR, and Insulin Growth Factor-1 receptor IGF-1R) in the brain.The presence of insulin in the brain derives from two main paths: from the periphery as it can cross the blood brain barrier or via direct synthesis by neurons (Creo et al., 2021;Fanelli et al., 2022).Accordingly, beside its role in glucose metabolism, recent evidence indicates that insulin contributes to several cognitive functions, such as learning, memory, integration of sensory information, and modulation of synaptic plasticity (Nisticò et al., 2012).
Genetic, clinical and preclinical (Biessels and Despa, 2018;Blázquez et al., 2014;Koekkoek et al., 2015) studies support the evidence that altered insulin signalling is involved in mental function and disease.For example, several studies reported a correlation between T2DM and AD and observed that insulin signalling may represent a common pathophysiological risk factor (Burillo et al., 2021;Pardeshi et al., 2017).Beyond AD, T2DM patients are at increased risk of milder forms of cognitive decline other than memory, such as processing speed and executive functions (Monette et al., 2014;Palta et al., 2014).These may occur during pre-diabetic stages and slowly worsen over time (Biessels et al., 2014).Importantly, impairments in impulse control, as a characteristic of ADHD, have also been observed in a large cohort of obese patients (Sinclair et al., 2000) further strengthening the potential association between metabolic syndrome and cognitive impairments.Furthermore, genetic and genomic studies reported that dysregulated insulin-dependent signalling cascades are associated with OCD (Bralten et al., 2020;van de Vondervoort et al., 2016).Accordingly, clinical investigations consistently reported that anti-diabetic drugs have beneficial effects on cognitive impairments in both AD, OCD, and other forms of cognitive decline (Fink et al., 2018;Munõz-Jiménez et al., 2020).
Although the historical analysis of this literature suggests that the interest in the somatic function of glucose preceded the interest in its role in the brain, a very early account looked at this relationship from the opposing side.Thus, even before the discovery of insulin, Kooy (Kooy, 1919) hypothesised that mental disorders triggered the emergence of hyperglycaemia.Ultimately, the view that insulin signalling may be involved in both somatic and mental disturbances is now consolidated.Yet, it is unclear whether mental disturbances are secondary to somatic alterations, whether the latter are consequence of the former, or whether they are due to diverse insulin-related mechanisms acting independently in the periphery and the CNS.
The fundamental mechanisms underlying the comorbidity between T2DM and cognitive decline have been investigated in preclinical studies by means of animal models.For example, several authors reported that a consolidated experimental model of AD (transgenic mice expressing human amyloid precursor protein and presenilin 1, APP/ PS1) exhibited impaired cognitive capabilities associated with poor glycaemic control (Denver et al., 2018).Similarly, insulin receptor β-subunit deficient mice exhibit impaired memory capabilities associated with altered long-term potentiation, the latter representing a form of synaptic plasticity (Nisticò et al., 2012).Finally, van de Vondervoort and collaborators (van de Vondervoort et al., 2019) reported increased compulsivity and anxiety in an experimental model (TALLYHO/JngJ mice) recapitulating most of the metabolic abnormalities observed in T2DM patients: insulin resistance, hyperglycaemia, hyperinsulinemia, and obesity.
Based on these considerations, we aimed to further detail the role of insulin signalling in the comorbidity between mental and somatic disturbances by systematically analysing the available rodent literature in rodents.To this aim, we propose a qualitative description of available preclinical studiesconducted in adult mice and rats exhibiting hyperglycaemiainvestigating the role of impaired glucose metabolism in the comorbidity between somatic and mental impairments (limited to working memory, spatial memory and/or attention).

Review protocol
The systematic search was conducted in accordance with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines (Moher et al., 2015).The protocol (SYRCLE's protocol; Supplementary item 1) for animal intervention studies (de Vries et al., 2015) was submitted to the PROSPERO registry on May 16th, 2022 and registered on June 12th, 2022 (registration number: CRD42022331458).

Literature search and study identification
A comprehensive systematic literature search was conductedon April 7th, 2022 -in three online databases (PubMed, Scopus, Web of Science).The search strategy involved, as issues of interest, altered insulin signalling (with hyperglycaemia representing its proxy) and the investigation of the behavioural phenotypes isomorphic to specific symptoms of mental disturbances (i.e., executive function and memory); the search was limited to studies conducted in rats and mice.The complete search strategies used in each database were: • TITLE-ABS-KEY (hyperglycaemia OR hyperglycemia) AND TITLE-ABS-KEY ((spatial AND memory) OR attention) AND TITLE-ABS-KEY (mouse OR mice OR rat OR rats)) for Scopus database; • "hyperglycaemia"[Title/Abstract] OR "hyperglycemia" [Title/Abstract] AND "spatial"[Title/Abstract] AND "memory"[Title/Abstract] OR "attention"[Title/Abstract] AND "mouse"[Title/Abstract] OR "mice"[Title/Abstract] OR "rat"[Title/Abstract] OR "rats"[Title/Abstract] for Pubmed; • TS= ((hyperglycaemia OR hyperglycemia) AND ((spatial AND memory) OR attention) AND (mouse OR mice OR rat OR rats) for Web of Science.
During the first phase of examination (i.e., screening of titles and abstracts), the following prioritization of exclusion criteria was used: (1) language other than English; (2) non-original researches (e.g., reviews, commentaries, editorials, book chapters); (3) no full-text articles (e.g., meeting abstracts); (4) studies in vitro, studies in humans, studies in non-human animals other than rats and mice; (5) outcome measures other than working memory, spatial memory and/or attention.Two observers (MP and AMO) independently screened the articles of the first phase.The additional exclusion criteria in the second phase of full-text screening of the eligible articles were: (5) outcome measures other than working memory, spatial memory and/or attention; (6) experimental manipulations not resulting in hyperglycaemia; (7) other control conditions (e.g., low-fat diet used as control instead of standard diet, etc.).The data were independently extracted by two reviewers per site (MP and AMO; AOL and DAS; MS and JCG, respectively) and discrepancies were resolved by principal investigators of each site (SM, JCG, and DAS).

Data extraction and synthesis
The study characteristics extracted from the full-text articles eligible for qualitative data included the following categories: (i) bibliographic details (DOI, title, authors, publication year, journal); (ii) study design characteristics (number of experimental groups, number of subjects per group, type of study design); (iii) animal model characteristics (species, strain, sex, age and/or weight at the beginning of the study, type of test used to evaluate spatial memory, working memory and/or attention); (iv) intervention characteristics (type of experimental manipulation adopted to induce hyperglycaemia, details regarding the experimental manipulation, type of non-hyperglycaemic control, details on the assessment of hyperglycaemia, e.g. higher blood glucose concentrations compared to controls, and or to a predefined threshold); (v) outcome measure (direction of the variation of the behavioural phenotypes isomorphic to working memory, spatial memory and/or attention in experimental subjects exhibiting hyperglycaemia and in nonhyperglycaemic controls).If available, data on the variation of glucose metabolism-/insulin signalling related parameters (obtained after the original induction of hyperglycaemia, for example through glucose tolerance, insulin resistance, etc.) were collected.

Assessment of the risk of bias
To evaluate the methodological quality and validity of the included studies, we used the SYRCLE's Risk of Bias (RoB) tool for animal studies, developed by Hooijmans and co-authors (Hooijmans et al., 2014) by adjusting the Cochrane's RoB tool (Higgins et al., 2011).The RoB tool for animal studies is divided into 10 items (for Selection bias: sequence generation, baseline characteristics, allocation concealment; for Performance bias: random housing, blinding; for Detection bias: random outcome assessment, blinding; for Attrition bias: incomplete outcome data; for Reporting bias: selective outcome reporting; for Other: other sources of bias).With specific reference to the Reporting bias, we note that adequate tools to pre-register the experiments (and thus allow a systematic evaluation of the consistency between the planned and the reported studies) are available only since 2021 (Olevska et al., 2021).(Moher et al., 2009).Diagram of the identification, screening, eligibility, and inclusion of the literature search.

A.M. Ottomana et al.
Thus, for studies preceding this date, we deemed the Reporting bias as not applicable.Two reviewers per affiliation site (MP and EP; AOL and DAS; MS and JCG) performed the quality assessment of each article by independently assessing the aforementioned criteria.

Study selection
The search strategy described above resulted in 851 bibliographic records.The process of selection is summarized in Fig. 1 by using the PRISMA flow diagram.References were exported to Microsoft Excel and, after duplicates were removed, 447 studies remained.The first selection phase (for language other than English, non-original research, no full text article) resulted in 347 studies; the second selection phase (i.e., fulltext articles screening) resulted in 91 studies eligible for inclusion in the systematic review, while a total of 256 articles were excluded for: assessment of phenotypes other than attention, spatial memory, and working memory (n = 188), species other than mouse/rat (n = 44), experimental manipulations not resulting in hyperglycaemia (n = 19), and lack of an appropriate control group (n = 6).

Study characteristics
The characteristics of the 91 studies included in the present review are summarized in Fig. 2.
Whilst the behavioural phenotypes in association with hyperglycaemia included working and/or spatial memory, and/or attention, only one article (Moreira et al., 2007) simultaneously evaluated all of them using the same lever-press test to discriminate the three different domains (the position of the levers served as a spatial and operant stimulus at the same time and the ability to change a discrimination, once learned, was considered an attentional task).The most evaluated phenotype was spatial memory: of all the studies considered, only in 11 of them was memory not investigated.The most widely adopted procedure to investigate this domain was the Morris water maze, which has Fig. 2. The radial graph summarizes the studies assessed in the present review as a function of experimental subjects, experimental manipulations, methodologies adopted to confirm the presence of hyperglycaemia, and behavioural outcomes.Accordingly, each 90-degree quadrant describes all of the 91 studies contemplated in the review as a function of a different aspect.Moving from the centre to the periphery of the radial graph, each category is further subdivided into granular details.STZ: streptozotocin, HFD: high fat diet, MSG: monosodium glutamate, NIC: nicotinamide.been applied in 68 of the 91 studies (n: 1,4,5,7,8,14-16,18,20-24,26,29, 30,33-46,50-62,64-66,68-70,72-77,79,80,82,83,85-91).Some of these studies also used the elevated plus-maze learning task (Bhutada et al., 2010), the novel-object recognition test (Taylor et al., 2015) and/or the Barnes maze test (Jin et al., 2018;Momeni et al., 2021) to retest the same spatial memory phenotype.The Barnes maze test was used exclusively (and not as a restest) by three other authors (Li et al., 2012;Madhavadas et al., 2016;Madhavadas and Subramanian, 2015) while three articles used a variation of the novel-object recognition test: object location (Braga et al., 2021;Van Der Kooij et al., 2018) and place-recognition test ( de Senna et al., 2017).Moreover, the radial arm water maze test (Malone et al., 2008;Rababa'h et al., 2019), the T-maze (Joshi et al., 2021;Tanokashira et al., 2021;W. H. Wang et al., 2019;W. Wang et al., 2019), and the Y-maze (n: 15,25,47,49) were employed as spatial memory tasks.
Although the Morris water maze test is currently used for spatial memory, it has been originally devised to dissociate "spatial mapping" and "working-memory" theories of hippocampal function; accordingly, five articles of the present review (n: 15,34,64,73,74) used this test to assess working memory.The latter was also examined in 17 studies, which assessed spontaneous alternation as a proxy for working memory: of these, five articles used a T-maze test (n: 2,11,12,63,84), either in its original dry version or in a water incarnation (Tanokashira et al., 2018;W. H. Wang et al., 2019;W. Wang et al., 2019), nine used the Y-maze test (n: 13,19,27,31,49,67,78,81,89), while only one (Choeiri et al., 2005) employed the four-arm maze test.The remaining two articles used the passive avoidance test (Georgy et al., 2013) and the step-down inhibitory avoidance task (Remor et al., 2019) to assess aversive associative memory.With respect to these tasks, we note that, as also described by the authors themselves (Georgy et al., 2013;Remor et al., 2019), they are predominantly used to investigate emotional memory rather than working memory.Yet, the same authors also specified that they considered these cortical-dependent tasks related to working memory, whereby the latter affects the impact of aversive stimuli.
Regarding attention, only two articles analysed it through a rewarded lever press task (Moreira et al., 2007) (the ability to change the discrimination of the rewarded lever, once it had been learned, was considered an attentional task) and the nest construction (Yeh et al., 2015), that according to the authors, is a task involving a broad network of brain regions and has previously been used to evaluate attention in mice (Filali and Lalonde, 2009).
Herein, we evaluated the different experimental manipulations adopted to induce hyperglycaemia, the details on the assessment of hyperglycaemia, and the type of non-hyperglycaemic control (Table 1).All articles with no appropriate control condition (e.g., low-fat diet used as control instead of standard diet, control group not exposed to vehicle, etc.) were excluded.Therefore, all control conditions belong to the following categories: (1) control strains, be them inbred or outbred, specific for those models that spontaneously exhibited variations in glucose metabolism/insulin signalling; (2) a corresponding control condition for the environmental manipulation (e.g.subjects exposed to standard diet for models based on dietary interventions); (3) wild-type or hemizygous mice as controls for genetically-engineered mice; (4) vehicle-treated animals for the pharmacological modulation.In nine manuscripts (n: 18,21,44,58,60,64,67,68,70), the information on the type of non-hyperglycaemic control was unclear; we took this limitation into consideration during the assessment of the risk of bias.
Since blood glucose concentrations fluctuate as a function of the time elapsed between the last meal and its measurement, we deemed it relevant to evaluate whether and how fasting has been considered in the relevant articles.While in 39 articles glucose concentrations have been measured following a fasting period, in seven studies (n: 44,70,78,79, 87,88,90) it was assessed without fasting.The remaining articles did not provide details regarding this parameter.
Only eight articles used female subjects (n: 16,26,35,48,54, 59,62,63), three of which (Collison et al., 2012;Patel and Udayabanu, 2014;Yeh et al., 2015) entailed a pool of males and females.All experimental subjects were adult at the time of behavioural and metabolic phenotyping.Only in three articles did the treatment to induce hyperglycaemia begin at a neonatal stage (Jin et al., 2018;Madhavadas et al., 2016;Madhavadas and Subramanian, 2015).Given that our aim was the comparison between the hyperglycaemic group and the relative controls, all the studies had a between-subjects study design; in one of them (Rodríguez et al., 2016), the behavioural and metabolic tests were performed in two independent groups with the same experimental treatment.

Hyperglycaemia and behavioural outcomes
The effects of hyperglycaemia on the behavioural parameters of interest (within each test) are illustrated in Table 2.The tables include, under separate headings, the direction of the variation of the behavioural phenotypes isomorphic to working memory, spatial memory, and attention, respectively.Most of the studies observed an impairment in the behavioural phenotype in the comparison between hyperglycaemic subjects and their relative controls; only one study (Dharavath et al., 2019a), conducted in female rats, reported an improvement in the spatial memory domain after 16 and 20 weeks of a HFD treatment lasting 24 weeks (although, at the end of treatment, they reported an impairment), and 11 papers described no significant difference between the two groups of interest for spatial (n: 7,8,12,17,22,39,42,43,46) and working memory (n: 13,46,63).

Other considerations
Although possible treatment therapies of hyperglycaemia were not the purpose of our review, some of the presented studies have also analysed the effects of several treatments.Therefore, while we are not able to provide a systematic review of this aspect, we believe that these considerations may help analysing the predictive validity of animal models of diabetes (i.e., whether treatments used in our species are also effective in experimental models), in terms of changes in blood glucose concentrations and cognition.Although metformin is a first-line therapy for the treatment of diabetes, only in three articles (Delkhosh-Kasmaie et al., 2018;Li et al., 2012;Tanokashira et al., 2018) has it been used as a treatment for diabetes-associated dysregulation.In all of them, metformin ameliorates diabetes-associated decline in hippocampal neurogenesis, learning and memory.Several authors used alternative approaches such as herbal medicines (Mao et al., 2008;Mirshekar et al., 2011;Tabatabaei et al., 2016;Wu et al., 2012), dark chocolate (Madhavadas et al., 2016), prebiotic and probiotic (de Cossío et al., 2017;T. H. J. Liu et al., 2020;T.H. Liu et al., 2020) and physical exercise or a backward switch from HFD to a regular diet (Braga et al., 2021;de Senna et al., 2017).All these treatments ameliorated metabolic and cognitive dysfunctions related to the hyperglycaemic condition in the animal model.While metformin is a validated anti-diabetic drug, these alternative treatments clearly deserve special consideration on whether results can be translated to humans or not.
Sex differences constitute an additional important aspect that would warrant a systematic approach.While this aspect was not among the primary scope of our study, we can nonetheless provide some preliminary considerations.Of all the 91 studies considered, only eight included female subjects.In two of them, the authors pooled subjects of both sexes thus limiting the possibility to discern between males and females (see Table 1, n: 54,59).One study (see Table 1, n: 35) reported a direct comparison between males and females and observed genderspecific effects, with males more affected by hyperglycaemia than females.The remaining five articles (see Table 1,n: 16,26,48,62,63) presented data on female subjects only.One of them (see Table 2, n: 26) constitutes the only instance in which hyperglycaemia resulted in a temporary cognitive improvement: i.e. hyperglycaemia resulted in improved spatial memory 16 and 20 weeks after the beginning of a high fat diet, and in impaired spatial memory four weeks later.The other four studies either reported a general impairment (see Table 2,n: 16,48,62) or lack of differences (see Table 2, n: 63) as a function of hyperglycaemia.While these articles partly reverberate the results observed in the studies conducted in males, their scant number poses some caveats as to whether the findings reported in the majority of studies (malebiased) may translate to females.

Risk of bias (RoB)
The risk of bias assessment of all included studies is shown in Fig. 3.The assessment of RoB included all the final 91 articles, for which often the experimental details were only partly reported (Avey et al., 2016).This resulted in an overall unclear risk of bias (52.09%).Yet, when data were correctly reported, there was a generally low risk of bias based on SYRCLE's RoB tool (33.19%), with a limited percentage of high risk of bias (5.27%).The judgement "not applicable" resulted in a 9.45% overall risk of bias but it was only influenced by the "Reporting bias" (Hooijmans et al., 2014).

General discussion
The primary purpose of the present systematic review was to identify whether experimental rats and mice characterized by hyperglycaemia also exhibit behavioural abnormalities in the domains of working and spatial memory, and attention.The studies of interest are characterized by hyperglycaemia, which has been induced via different methodologies, be them pharmacological interventions, environmental modulations, transgenic approaches, naturally occurring mutations based on strain differences, or a combination.All of them allow the analysis of mechanisms related to diabetes and are important to understand the pathogenesis and progression of the disease as well as to evaluate potential therapeutic strategies with an elevated translational value.Accordingly, an animal model relevant for the study of diabetes, should mirror the pathophysiology and natural course of diabetes, and/or develop complications of the disease with an aetiology similar to the human condition (Varga et al., 2015).These considerations are particularly relevant to diabetes, which is characterised by multiple facets and different main diagnoses: T1DM is an autoimmune disease in which pancreatic β-cells are targeted to be destroyed by antibodies produced by immune cells (Gillespie, 2006).In contrast, the pancreatic β-cells are active in T2DM and synthesize insulin but at dysregulated level and/or not sufficiently efficiently (Brunton, 2016).Chronically elevated blood glucose concentrations represent a commonality in T1DM or T2DM.Since T1DM is characterized by the deficiency of insulin production, the deficit is achieved in experimental animals through chemical destruction of pancreatic β-cells or through breeding of rodents that spontaneously develop diabetes.Although the endpoint of β-cell destruction is similar to T1DM in humans, the mechanism for the β-cell destruction is not autoimmune, therefore the aetiology differs from the human condition.On the other hand, T2DM animal models should recapitulate insulin resistance, a certain degree of β-cell failure, and obesity.It

Table 2
Behavioural test for each cognitive domain of interest and relative outcomes; bold text refers to an improvement and italic text refers to a no change in that phenotype, otherwise there was an impairment of the domain investigated.

N
Ref.
OUTCOME and TEST to evaluate working memory OUTCOME and TEST to evaluate spatial memory OUTCOME and TEST to evaluate attention (Georgy et al., 2013) Passive avoidance test Morris water maze test n/a (Arnold et al., 2014) T-maze test n/a n/a (Joshi et al., 2021) n/a T-maze test n/a (Remor et al., 2019) Step-down inhibitory avoidance task Morris water maze test n/a (Wu et al., 2014) n/a Morris water maze test n/a (Moreira et al., 2007) lever press task (total presses) lever press task (left versus right lever discrimination) lever press task (active presses in FR2, FR3, FR5 and PR) (Rodríguez et al., 2016) n/a Morris water maze test n/a (Biessels et al., 1996) n/a Morris water maze test n/a (Madhavadas et al., 2016) n/a Barnes maze test n/a (Madhavadas and Subramanian, 2015) n/a Barnes maze test n/a (Tanokashira et al., 2018) Water T-maze n/a n/a (W.H. Wang et al., 2019 Y-maze test n/a n/a (Malone et al., 2008) n/a Radial water maze test n/a (Treviño et al., 2015) n/a Morris water maze test n/a (Nurdiana et al., 2017) n/a Morris water maze test n/a (Hardigan et al., 2017) Y-maze test n/a n/a ( de Senna et al., 2017) n/a Place recognition test n/a (Wu et al., 2012) n/a Morris water maze test n/a (Lin et al., 2018) Morris water maze test Morris water maze test n/a (Collison et al., 2012) n/a Morris water maze test n/a (Zhou et al., 2015b) n/a Morris water maze test n/a (Zhou et al., 2017) n/a Morris water maze test n/a (Huang et al., 2012) n/a Morris water maze test n/a (Huang et al., 2007) n/a Morris water maze test n/a (Huang et al., 2019) n/a Morris water maze test n/a (Lin et al., 2017) n/a Morris water maze test n/a (He et al., 2020) n/a Morris water maze test n/a (Zhou et al., 2018) n/a Morris water maze test n/a (Sibiya and Mabandla, 2017) n/a Morris water maze test n/a (Kumar and Maqbool, 2020) n/a Morris water maze test n/a (Choeiri et al., 2005) Four-arm maze test Morris water maze test n/a (Marissal- Arvy et al., 2018) n/a Y-maze test n/a (Braga et al., 2021) n/a Object location test n/a (Van Der Kooij et al., 2018a) n/a Y-maze test and Object location task n/a (Bhutada et al., 2010) n/a Morris water maze test and Elevated plus maze learning task n/a (Pathan et al., 2008) n/a Morris water maze test n/a (Esmaeili et al., 2017) n/a Morris water maze test n/a (Ren et al., 2013) n/a Morris water maze test n/a (Yeh et al., 2015) n/a Morris water maze test Nest construction (Pei and Sun, 2018) n/a Morris water maze test n/a (Ye et al., 2018) n/a Morris water maze test n/a (Mao et al., 2008) n/a Morris water maze test n/a (T.H.J. Liu et al., 2020;T.H. Liu et al., 2020) n/a Morris water maze test n/a (Patel and Udayabanu, 2014) n/a Morris water maze test (J.T.H. Liu et al., 2020;J. Liu et al., 2020) n/a Morris water maze test n/a (Jin et al., 2018) n/a Barnes maze test and Morris water maze test n/a (H.W. Wang et al., 2019;H. Wang et al., 2019) n/a Morris water maze test n/a (Babic et al., 2018) T appears that no single animal model involves all of these characteristics, but some of them could provide very similar traits in one or more aspects of diabetes (T1DM and/or T2DM) in humans.Although we selected the relevant studies based on hyperglycaemia, our primary interest was the role of insulin signalling.Therefore, an important prerequisite is that hyperglycaemia constitutes a valid proxy of altered insulin signalling.To assess this prerequisite, we first discuss the extent to which the experimental models considered in this review adequately mimic diabetes.Following this examination, we proceed with the evaluation of the association between diabetes and cognitive alterations.Finally, we interpret the observed results as a function of their risk of bias.

Pharmacological induction of hyperglycaemia
Some experimental models utilise a chemical approach to induce diabetes, particularly in the form of diabetogenic agents such as streptozotocin and alloxan.Alloxan is a toxic glucose analogue whose accumulation in pancreatic β-cells inhibits insulin secretion and induces reactive oxygen species formation that are ultimately responsible for the death of the cells (Cefalu, 2006).Streptozotocin is a highly selective pancreatic islet β-cell-cytotoxic agent and inhibits insulin secretion causing a state of insulin-dependent diabetes mellitus (Lenzen, 2008).It is often administered at a single high dose to produce an immediate blood glucose concentration > 500 mg/dL and to cause β-cell total necrosis (Cefalu, 2006).However, lower doses of STZ administered multiple times, are capable to delay the onset of hyperglycaemia with a partial damage of pancreatic islets.This slower process triggers an inflammatory process that causes an additional loss of β-cells, which, in turn, results in insulin deficiency, hyperglycaemia, polydipsia, and polyuria (Radenković et al., 2016).This response more closely resembles T1DM in pathogenesis and morphologic changes than the single, high-dose of STZ (Furman, 2015).Another protocol of diabetes induction entails the concurrent administration of STZ and nicotinamide, wherein the latter partially protects the β-cells damage induced by the former (Fukaya et al., 2013;Szkudelski, 2012).This combination Fig. 3. Risk of bias assessment, score (%) per risk of bias item.The RoB tool for animal studies is divided into 10 items."Random outcome assessment" bias was considered "Unclear" because the details on the sequence of animal testing were never reported."Reporting bias" was judged as "not applicable" for all the studies published before 2021.In this respect it should be noted that the "Reporting bias" item was prospectively included in the SYRCLE's tool (in agreement with the Cochrane's tool) although at present it is difficult to assess, as protocols for animal studies are not yet mandatorily registered in central, publicly accessible databases (Hooijmans et al., 2014;Olevska et al., 2021).
produces a model of insulin-deficient, but not insulin-resistant T2DM that is a major feature of most human cases.It is characterized by stable, moderate hyperglycaemia associated with an approximately 60% loss of β-cell function.
Although most of the reviewed articles used models of diabetes (see Table 1), our interest was directed towards any type of manipulation capable of inducing hyperglycaemia.Some scholars used different drugs to achieve this goal or to study the behavioural effects of glucose dysregulation.One of these studies reported that isoflurane and sevoflurane (Wu et al., 2014) may impair glucose tolerance by decreasing insulin secretion and glucose utilization.An additional pharmacological approach was constituted by the use of monosodium glutamate (MSG), that is frequently used as a flavour enhancer in the food industry.Some evidence indicate that MSG treatment in animals, during the first day of life, leads to the development of obesity and hyperglycaemia in adulthood (Bahadoran et al., 2019;Dolnikoff et al., 2001).Although these observations suggest that MSG may constitute a useful agent for the induction of T2DM-like abnormalities in animals, the underlying mechanisms are not completely understood and thus may be secondary.As one of the reviewed articles showed, chronic dexamethasone administration may induce diabetes-related metabolic dysfunctions in animal models (Patel and Udayabanu, 2014).Dexamethasone represents a first-line anti-inflammatory drug.Unfortunately, long-term therapy is associated with metabolic side effects, including hyperglycaemia, hypertension, and hepatic steatosis that contribute to insulin resistance and diabetes (Cefalu, 2006).Additionally, one manuscript used an antipsychotic drug, clozapine.While it generally improves the cognitive symptoms of schizophrenia, it can cause serious metabolic side-effects (Siskind et al., 2016), likely mediated by its impact on glucagon-like peptide (Siskind et al., 2019).Usually, all these pharmacological models are invaluable when studying the mechanisms by which hyperglycaemia may contribute to microvascular complications such as neuropathy, nephropathy, and retinopathy.However, because they may be toxic to organs and tissues other than the pancreatic islet β-cells, these models do not precisely mimic the human condition (Zhang et al., 2008).Chemically-induced models present a phenotype that closely resembles that observed in T1DM patients; yet, the mechanisms underlying β-cell damage are different from the human disease.Thus, while the construct validity of these models may be limited, their predictive validity is highly relevant.

Environmental induction of hyperglycaemia
To mimic the nutritional determinants of T2DM, several authors adopted a differential strategy entailing the administration of HFD.In these instances, experimental subjects are exposed to an HFD nutritional regime to induce insulin resistance.This regime has been sometimes associated with the concurrent administration of a moderate dose of STZ to reduce β-cell capacity (Reed et al., 2000), a combination resulting in hyperglycaemia, hyperinsulinemia and insulin resistance (Yorek, 2016).The use of HFD to induce insulin resistance and to produce mild or moderate insulin deficiency may represent a valid experimental model of T2DM, whereby it can provide relevant information regarding many of the complications associated with human diabetes.HFD has also been utilized to model chronic inflammation, which is an important pathogenic mechanism of T2DM.Chronic overfeeding triggers inflammation, which leads to alterations in peripheral insulin receptor-associated signalling and thus reduces the sensitivity to insulin-mediated glucose clearance.These events ultimately result in elevated fasting glucose and insulin concentrations as well as in a reduction in glucose tolerance, all of which constitute relevant indicators of insulin resistance (Heydemann, 2016;Nagy and Einwallner, 2018).Moreover, a long-term HFD induces metabolic disorders, oxidation, inflammation, changes in islet size, and irregular secretory functions in the pancreas (Wu et al., 2022;Zhao et al., 2022).Strong evidence supporting HFD as a valid methodology to reproduce the complications associated with T2DM is the fact that it induces telencephalic insulin resistance associated with systemic hyperglycaemia (Cefalu, 2006).This is often associated with a chronic hyperactivation of cortical and hippocampal neurons which may ultimately predispose toward cognitive impairments.
An alternative manipulation adopted to modulate metabolic functions capitalised upon psychosocial stress as a strategy to induce hyperglycaemia.Specifically, chronic psychosocial stress in mice has been reported to increase peripheral and central glucose concentrations and, subsequently, to relate to the emergence of stress-induced cognitive impairments (Van Der Kooij et al., 2018b).It has been suggested that psychosocial and metabolic stress share common underlying mechanisms with glucose dysregulation having a central role.These findings have been associated with the metabolic consequences of environmental stressors in our species, wherein chronic stressors, low socioeconomic status, severe mental health problems, or aggressive behaviour have been shown to increase the risk of T2DM (Hackett and Steptoe, 2016;Winchester et al., 2016).Accordingly, animal models have shown that exposure to experimental stressors may anticipate the onset of chronic subclinical inflammation (Black, 2003).Additionally, animal models that mimic T2DM and T2DM-related metabolic conditions suggest that insulin resistance may lead to chronic inflammation, which may in turn induce cognitive decline (Kelly and Ismail, 2015).It remains to be determined, however, whether central glucose dysregulation is linked to stress-induced cognitive impairments or whether abnormal glucose metabolism may contribute to individual susceptibility to the adverse consequences of chronic stress.

Genetic determinants of hyperglycaemia
Animal models with naturally occurring mutations have traditionally constituted a unique resource potentially mimicking the construct validity of the disease whereby: (i) they become spontaneously diabetic; and (ii) the course of the disease is markedly influenced by genetic background.The Lepr db/db mice represent one of the most widely studied genetically-induced experimental models of diabetes.They are homozygous for the spontaneous mutation of the leptin receptor (Lepr) that is involved in food intake, energy expenditure, and body weight (Berger et al., 2021).Mice carrying this spontaneous mutation exhibit obesity, chronic hyperglycaemia, pancreatic β-cell atrophy, and hypoinsulinemia.Mutations in leptin receptor have been shown to cause early-onset severe obesity and insulin resistance in mice and humans (Wang et al., 2014).While, in humans, it is difficult to disentangle whether insulin resistance precedes or is secondary to the development of obesity, in mice, the temporal association of these symptoms can be prospectively investigated.Studies conducted in Lepr db/db mice seem to suggest that the onset of insulin resistance anticipates the onset of obesity.Thus, Lepr db/db mice have a natural history of the disease similar to that observed in humans whereby they become hyperinsulinemic early in life (within 2 weeks of age, i.e. before consuming chow) and develop obesity by 3-4 weeks.Hyperglycaemia, associated with a β-cell failure, becomes manifest at age 4-8 weeks (Bates et al., 2005); this is followed by a compensatory hyperplasia of the islet of Langerhans which keeps being associated with hyperinsulinemia until the latest stages of life (18-20 months)("97 -B6 db Strain Details, 0006").Another genetically engineered diabetic and obese model is the KKay mouse.This mouse was generated by transferring the Ay gene (conferring these mice an unusual yellow coat colour) onto a glucose-intolerant mouse strain (KK).While mice of the KK strain develop diabetes of polygenic origin, the Ay mutation leads to obesity as a function of the agouti protein being expressed in incorrect locations ("68 -Strain Details, 0024").By approximately two months of age, due to insulin resistance, KKay mice develop the following diabetes-like abnormalities: hyperglycaemia, hyperinsulinemia, glucose intolerance, and obesity.The KKAy mice have excessively large pancreatic islets and degranulated pancreatic β-cells.Their obesity is partially due to fat cell hypertrophy, caused by a drop in dopamine and noradrenaline in the hypothalamus ("The Characteristics of KKAy Mice -Maze Engineers").Due to these features, these mice resemble both the early (β-cells impairment in the pancreas and hyperglycaemia) and the late stages of diabetes (β-cells can no longer release insulin and insulin replacement therapy is required).
The Akita strain ("48 -Akita Strain Details, 0035") constitutes an additional experimental model of spontaneous diabetes.These mice are characterised by a mutation at the level of the Ins2 gene, one of the two genes encoding for insulin in mice (the second one being Ins1).A mutation in the Ins2 gene leads to incorrect folding of the insulin protein, which in turn results in toxicity in pancreatic β-cells, reducing β-cell mass and insulin secretion.Heterozygous Ins2Akita mice develop insulin dependent diabetes, including hyperglycaemia, hypoinsulinemia, polydipsia, and polyuria within four weeks of age, thus representing a valid experimental model of T1DM.Just as in humans men are more often affected than women, so also in rodents diabetes-like abnormalities appear more frequent in males than in females (Tramunt et al., 2020).Accordingly, in Akita mice, the phenotype is more severe in males than females ("48 -Akita Strain Details, 0035").
Other authors engineered mouse models based on different components of the insulin signalling pathways.Specifically, Tanokashira and collaborators (Tanokashira et al., 2021) capitalised upon the role of the insulin receptor substrate-2 (Irs-2), which plays a fundamental role in metabolism and growth of every tissue.Irs2 knockout mice exhibit a progressive development of a T2DM-like phenotype: while they show hyperglycaemia as early as three days of age, they become diabetic by 10 weeks when they exhibit reduced β-cell mass and insulin resistance in skeletal muscle and liver ("21 -IRS-2 KO Strain Detail, 0044").
Diabetic rats also represent an important research tool.For example, the Zucker diabetic fatty rat (ZDF) is usually used as a model for the study of T2DM associated with obesity ("Zucker Rat/Charles River").Like db/db mice, ZDF rats present a mutation on the leptin receptor, which induces obesity and hyperglycaemia within the first few months of age.The diabetic like features exhibited by ZDF rats appear to be associated with an inability to increase β-cell mass; this results in an insufficient insulin secretion, which ultimately fails to compensate for the obesity-dependent insulin resistance (Clark et al., 1983).Thus, although there are similarities between the human condition and the abnormalities exhibited by ZDF rats, the latter may be characterised by limited degree of construct validity whereby humans with T2DM do not have inadequate β-cell proliferation in early life (Garnett et al., 2005).The Goto-Katazaki (GK) (Guest, 2019) rat is another model used for the study of diabetes.The GK rat is non-obese, has a decreased β-cell mass, and is characterised by liver and skeletal muscle-insulin resistance.Due to impaired insulin secretion, fasting blood glucose concentrations are also slightly increased.Disease progression of this rat has been associated with chronic inflammation and hence utilized in the study of pathophysiology and therapeutic studies of diabetes (Xue et al., 2011).An excess adipose tissue is linked to chronic inflammation as a consequence of the attraction of macrophages in the adipose tissue.It is this increased macrophage infiltration that in part suggests a link between obesity, inflammation and the development of diabetes (Surmi and Hasty, 2008).

Hyperglycaemia as a comorbidity in experimental models of cognitive disturbances
While most of the reviewed studies exploited experimental models of metabolic disturbances induced via alterations at some level of the insulin signalling pathways, some others capitalised upon the known association between T2DM and cognitive impairments and late-onset AD (Hamzé et al., 2022;Kandimalla et al., 2017;Pardeshi et al., 2017;Watanabe et al., 2015).These pathologies have been reported to share several pathophysiological features and common risk factors (Devi et al., 2012;Niedowicz et al., 2014).In mice without pre-existing AD-like symptoms, the induction of diabetes (genetically, pharmacologically or by diet) is associated with an hyperphosphorylation of tau protein (Li et al., 2007;Son et al., 2012).Complementarily, some of the reviewed articles used transgenic mice with mutations linked to Alzheimer's disease such as APP/PS1 (Yeh et al., 2015), 3 ×Tg-AD (Huang et al., 2019) and PS19 (Nakaoku et al., 2019) and observed that they became hyperglycaemic.
Ultimately, although there is not a single animal model recapitulating all the causative factors and associated phenotypic abnormalities observed in our species, the comprehensive consideration of available literature strongly supports the notion that preclinical studies may beget relevant information in the understanding and therapy of insulin-related somatic and mental abnormalities.

Test paradigms to confirm the presence of hyperglycaemia in experimental models
In all of the screened studies, the presence of hyperglycaemia has been assessed via multiple methodologies that differed in terms of reference values, control groups, timing of evaluation, sample collection, and presence of fasting.Therefore, while these differential approaches captured a welcome heterogeneity in the study of complex phenomena, they nonetheless reduced the capability to cross compare different studies.Within this realm, some considerations may be sensible: for example, the use of the same approach to test blood glucose concentrations for both control and treated groups (e.g., either a nonfasting state or a fasting state) shall benefit this field of investigation; likewise, a clear definition of hyperglycaemia shall be useful.Indeed, while the threshold of hyperglycaemia in our review has not always been constant and unique, the spectrum of thresholds used is close to what happens in our species.Generally, the blood glucose concentration threshold was > 250 mg/dL in non-fasting conditions and > 150 mg/dL in fasting conditions.One important benefit of a standardized reference value is the fact that it may align and adjust the experimental techniques across laboratories thus favouring the reproducibility and external validity of these studies.

Test paradigms to confirm the presence of cognitive impairments in experimental models
A different consideration pertains to the test paradigms required to assess individual cognitive capabilities, for which different methodologies exist and that for which a univocal standard is neither feasible nor necessarily advisable.As mentioned above, the main aim of this review was to assess whether hyperglycaemic rodents exhibited alterations in working and spatial memory, and attention.Of these parameters, spatial memory has received the highest level of interest by this scientific community.This is substantiated by the fact that of the 91 studies, nearly all investigated this phenotype and all of them reported consistent impairments.It is important to emphasize that spatial memory has been addressed via different experimental paradigms (i.e.Morris water maze, Barnes maze, novel-object recognition, T-Y-and radial-armmaze).The convergence of results adopting different methodologies, both in terms of inducing hyperglycaemia and the behavioural readout employed, strengthens the association between hyperglycaemia and spatial memory, whereby it has been observed under heterogeneous conditions (Richter et al., 2010).As discussed elsewhere, heterogenized experiments have been reported to yield more stable results and to be characterized by a reduced number of false positives compared to homogenized experiments (Macrì and Richter, 2015).These findings support the view that alternative experimental strategies may indeed enhance the reproducibility and translational value of preclinical animal research.

Insulin signalling and cognition: candidate biological determinants
As we reviewed, hyperglycaemiaas a proxy of insulin-related metabolic dysfunctionsled to consistent impairments in the cognitive domains of interest.There are strong preclinical, epidemiological and clinical evidence (Biessels et al., 2008;Rom et al., 2019;Van Den Berg et al., 2010;Zhang et al., 2021) in support of the association between diabetes and cognitive dysfunctions, which may concern one or different domains, including processing speed, executive function, learning and memory (Arnoriaga- Rodríguez et al., 2020;Backeström et al., 2021;Omladič et al., 2020;Sadanand et al., 2016;Sattar et al., 2017).Diabetes can be viewed as a metabolic disorder resulting in accelerated cognitive ageing in terms of dementia and cognitive decline.Ultimately, diabetes-related cognitive impairments may be viewed as another long-term complication of diabetes.There is also evidence suggesting that hyperglycaemia per se has detrimental effects on cognitive function, whereby acute hyperglycaemia has been associated with poor cognitive outcomes, potentially as a function of accumulation of reactive oxygen species in the brain.
People with long-standing diabetes and no other diagnosed diabetesrelated complications have poorer working memory (Awad et al., 2017;Gallardo-Moreno et al., 2022).This finding has been related to the fact that hyperglycaemia per se may induce structural abnormalities in the prefrontal cortex (Lyoo et al., 2013), a brain region involved in working memory.Additionally, hyperglycaemia has been associated with changes in different regions of the brain, including the hippocampus (Nevo-Shenker and Shalitin, 2021).Accordingly, hyperglycaemia, in young patients with T1DM, has been shown to influence long-delay spatial memory months after that first diagnosis (Semenkovich et al., 2016).Likewise, changes in hippocampal synaptic plasticity and subsequent impairments in spatial memory, are well documented in animal models of T1DM and T2DM (Soares et al., 2013).Whereas numerous clinical studies (Broadley et al., 2017;Pappas et al., 2019;Redondo et al., 2016;Sattar et al., 2017;Zhao et al., 2020) have demonstrated a correlation between executive functions and diabetes, only in two articles of the present review has this domain been investigated.Thus, further preclinical studies are needed to elucidate the mechanism underlying the association between diabetes and executive function.
A number of possible mechanisms to explain the association between dysglycaemia, hyperglycaemia and cognitive dysfunction have been suggested.As reported above, one hypothesis is that chronic exposure to elevated glucose concentrations may accelerate cognitive decline (Awad et al., 2004;Messier et al., 2011).Moreover, impaired insulin signalling in the brain may represent a highly promising research avenue.Specifically, while having an important function in glucose transport, the highly abundant insulin receptors in the brain have been implicated in cognitive processes.Several observations suggest that cognitive decline is a consequence of insufficient insulin action in the brain, either due to insulin resistance, insulin deficiency or both.Insulin signalling in the brain has important roles in brain physiology and cognition (Biessels and Reagan, 2015).

Caveats associated with risk of bias
In order to evaluate the degree of confidence with which the conclusions of our work can be extrapolated on a large scale, we carefully assessed the RoB associated with several methodological aspects.Had we identified a widespread elevated RoB, our considerations would be substantially devalued.Yet, the elevated RoB was observed only in a fraction of the domains relevant to the scopes of our review (detailed in the following lines).Specifically, according to SYRCLE's protocol, the overall high RoB identified in the studies considered in the present review was 5.27%.This value was computed as the average of the instances (percent) in which we identified an elevated RoB for a given category.In particular, we observed instances of elevated RoB in the following items: (1) "Baseline characteristics": while we considered hyperglycaemia as a baseline value of interest, two articles of the review did not explicitly report whether this prerequisite had been met but rather referred to the already available evidence that the animal model of interest was characterised by hyperglycaemia; (2) "Performance bias: blinding" and "Detection bias: blinding": in few instances we observed that blinding could not be guaranteed whereby the individual who planned the study also performed the experiments and analysed the data; (3) "Attrition bias: incomplete outcome data": with respect to this item, we observed that few studies failed to report in the analysis all the experimental subjects used in the experiment without providing an explicit explanation for these attrition rates; (4) "Other source of bias": we took into account studies in which the solution administered to the control group (vehicle) was not specified; (5) "Reporting bias": with respect to this parameter, we attempted to identify whether the experiments reported in the published experiments matched those that were officially planned/registered.In this specific case, the analysis was limited to the articles published after 2021, when the Animal Study Registry (Olevska et al., 2021) a platform wherein animal studies can be registered) has been launched.We note, however, that this parameter is difficult to assess whereby protocols for animal studies are not yet mandatorily registered in central, publicly accessible database (Olevska et al., 2021).As a consequence, be it due to the absence of enforcement and/or limited availability of repositories, none of the studies considered in the present review had been previously registered.We believe that this aspect warrants particular consideration by the preclinical scientific community.As a matter of fact, while on the one hand we deem this aspect sufficiently relevant to be included in the SYRCLE's protocol, on the other hand, we apparently have limited interest and tools to implement it on a large scale.
An additional warning to our community can also be derived from the unclear overall risk of bias, which in the present review attained a value of 52.09%.While an unclear risk of bias does not directly denote inappropriate study planning/execution/reporting, it nonetheless hinders the possibility to thoroughly grasp its fundamental details, especially in light of the current reproducibility crisis.We posit that the reporting of methodological details in animal studies shall considerably improve, and that the promotion of high-quality standards for registration and reporting of animal studies shall represent a desired goal in preclinical research.

Conclusions
Our systematic review strongly supports the view that hyperglycaemia in experimental models of metabolic dysfunctions is associated with cognitive impairments.It is thus plausible that hyperglycaemia, as a core feature of diabetes, disturbing insulin signalling and favouring insulin resistance, not only affects systemic metabolism, but also directly impacts the brain, by disturbing cerebral insulin pathways and the associated cognitive functions.Complementarily, although this was not the core aim of our study, hyperglycaemia has been observed in experimental subjects that were originally selected based on their known cognitive impairments and not on their metabolism.Therefore, just as hyperglycaemia may constitute a risk factor for cognitive impairments, so also the latter can influence the former, ultimately binding diabetes and cognition in a recurrent cycle.

Fig. 1 .
Fig. 1.PRISMA flow diagram for preclinical studies(Moher et al., 2009).Diagram of the identification, screening, eligibility, and inclusion of the literature search.

Table 1
Details regarding the experimental manipulation to induce hyperglycaemia and the experimental subjects (f: female; m: male).
(continued on next page) A.M. Ottomana et al. (continued on next page) A.M. Ottomana et al. (continued on next page) A.M. Ottomana et al.