Targeting impaired nutrient sensing with repurposed therapeutics to prevent or treat age-related cognitive decline and dementia: A systematic review

BACKGROUND
Dementia is a debilitating syndrome that significantly impacts individuals over the age of 65 years. There are currently no disease-modifying treatments for dementia. Impairment of nutrient sensing pathways has been implicated in the pathogenesis of dementia, and may offer a novel treatment approach for dementia.


AIMS
This systematic review collates all available evidence for Food and Drug Administration (FDA)-approved therapeutics that modify nutrient sensing in the context of preventing cognitive decline or improving cognition in ageing, mild cognitive impairment (MCI), and dementia populations.


METHODS
PubMed, Embase and Web of Science databases were searched using key search terms focusing on available therapeutics such as 'metformin', 'GLP1', 'insulin' and the dementias including 'Alzheimer's disease' and 'Parkinson's disease'. Articles were screened using Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia). The risk of bias was assessed using the Cochrane Risk of Bias tool v 2.0 for human studies and SYRCLE's risk of bias tool for animal studies.


RESULTS
Out of 2619 articles, 114 were included describing 31 different 'modulation of nutrient sensing pathway' therapeutics, 13 of which specifically were utilised in human interventional trials for normal ageing or dementia. Growth hormone secretagogues improved cognitive outcomes in human mild cognitive impairment, and potentially normal ageing populations. In animals, all investigated therapeutic classes exhibited some cognitive benefits in dementia models. While the risk of bias was relatively low in human studies, this risk in animal studies was largely unclear.


CONCLUSIONS
Modulation of nutrient sensing pathway therapeutics, particularly growth hormone secretagogues, have the potential to improve cognitive outcomes. Overall, there is a clear lack of translation from animal models to human populations.


Introduction
Dementia is a syndrome affecting more than 5% of the world's population over 60 years of age (Organisation, 2017). Although Alzheimer's disease (AD) is the most common subtype of dementia, many other subtypes exist including; Vascular Dementia (VD), dementia with Lewy Bodies (LBD), Parkinson's dementia (PD) and Frontotemporal Dementia (FTD) (Organisation, 2017). Ageing is the major risk factor for the development of all dementia subtypes (Society, 2016). Mechanistically, ageing is defined as an accumulation of molecular and cellular damage leading to a gradual decrease in physiological reserves (Organization, 2017). As such it has been hypothesized that targeting ageing pathways may be a viable therapeutic option for treating dementia. One such pathway is the nutrient sensing pathway (Lopez-Otin et al., 2013), which relates to the detection of extracellular nutrients, for instance glucose-sensing via the highly evolutionarily-conserved insulin and IGF-1 signaling (IIS) pathway (Lopez-Otin et al., 2013). Other key inter-related effectors of the nutrient sensing pathway include the Mammalian Target of Rapamycin (mTOR), which detects high amino acid concentrations, AMPK and sirtuins, which detect low energy states, and transcription factors known as Forkhead box O proteins (FOXO) (Lopez-Otin et al., 2013;Mc Auley et al., 2017).
Deregulated nutrient sensing is increasingly thought to play a role in the pathophysiology of neurodegenerative diseases such as AD (Fluegge, 2019;Liu and Sabatini, 2020;Shafei et al., 2017). One of the most fundamental pathological mechanisms shared by subtypes of dementia is neurodegeneration (Moya-Alvarado et al., 2016). This process is often accompanied by impaired neurogenesis , and abnormal protein aggregations (Dugger and Dickson, 2017), which might be driven by dysfunctional autophagy (Wong and Cuervo, 2010). Nutrient sensing is increasingly emerging both as a key modulator of the neurogenesis process (Fidaleo et al., 2017), and, via mTOR, of the autophagic process (Jahrling and Laberge, 2015). In mice, higher mTOR signaling has been associated with Aβ accumulation, whilst decreasing mTOR signaling has been shown to reduce Aβ levels (Caccamo et al., 2010). Human post-mortem studies have found higher levels of activated mTOR (Griffin et al., 2005;Li et al., 2005) and its downstream effectors (Tramutola et al., 2015) in affected brain regions of AD and MCI patients compared to controls. The broad role that deregulated nutrient sensing may play in dementia offers a possible therapeutic pathway, as many nutrient sensing-modulating therapeutics, such as metformin, already exist (Rena et al., 2017). To date, however, the therapeutic indications of these medications do not include neurodegenerative diseases (Australian Medicines Handbook Pty Ltd., 2021).
The aim of this systematic review is to summarize the evidence in human and animal populations for the use of Food and Drug Administration (FDA) approved nutrient sensing-modifying therapeutics to prevent age-related cognitive decline or improve cognition in ageing, mild cognitive impairment (MCI), and dementia populations.

Selection of articles
The protocol of this systematic review was registered at PROSPERO International prospective register of systematic reviews (Reg #: CRD42018091645). PubMed, Web of Science and Embase databases were searched until the 29th March 2019. The search strategy (Heard et al., 2018) focused on key search terms for the dementias, such as; AD, VD, PD, LBD, and ageing. The strategy also included key terms for nutrient modulating proteins and therapeutics, such as; insulin, mTOR, glycogen synthase kinase-3 (GSK-3), metformin, dipeptidyl peptidase-4 (DPP-4) and glucagon-like peptide-1 (GLP-1). The complete search strategy has been previously published (Heard et al., 2018). In addition snowballing was used to search references within included articles.

Eligibility criteria
Articles included in this review met the following inclusion criteria: 1) Populationanimals or humans; normal ageing or neurodegenerative disease, such as dementia (AD, VD, PD, or LBD). Populations likely to have a higher pace of ageing such as Type 2 Diabetes Mellitus (T2DM), insulin-resistant, or obesity, were also included. In animals, normal ageing was defined as a strain not at a greater propensity to develop dementia and not manipulated to mimic dementia. Dementia models were defined as strains at a greater propensity to develop dementia compared to normal ageing strains or being modified to become more likely to develop dementia. In humans, normal ageing was defined as a population not suffering from dementia or mild-cognitive impairment.
2) Study designinterventional studies with comparators; including randomized or quasi-randomised controlled trials, cohort studies, and pre/post studies. 3) Intervention -FDA approved therapeutics known to influence the nutrient sensing pathway. 4) Outcomecognitive function measured using neuropsychological tests. In animals, using mice as an example, neuropsychological tests may include spatial memory tests (Morris water maze [MWM], radial arm water maze [RAWM]), associative learning tasks (passive avoidance), recognition memory tasks, and others (Rodriguiz and Wetsel, 2006) (Lin et al., 2013).
Articles were excluded if they met one of the following exclusion criteria: 1) exercise as the sole intervention, 2) in vitro data only, 3) conference abstract, review, editorial, or letter to the editor, 4) ≤5 population size for human studies, 5) intraperitoneal/intravenous streptozotocin-induced diabetes models unless specifically stated as recapitulating a T2DM phenotype, due to its otherwise inappropriateness in mimicking the pathogenesis of age-related dementia following the onset of T2DM diabetes, 6) intracerebroventricular streptozotocininduced models without reporting a desired cognitive endpoint of either cognitive decline or dementia, due to their inappropriateness in mimicking the pathogenesis of human AD (Grieb, 2016), and 7) published in a language other than English.

Article selection and data extraction
Three reviewers (DH, CT and TR) independently screened the titles and abstracts for inclusion. Full text articles were then screened by two independent reviewers (DH and TR) to isolate articles of interest. A fourth reviewer (ABM) resolved any disagreements between the reviewers. Articles were screened using Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia). Included studies were separated into the following four groups for data extraction: 1) proteostasis -repurposed therapeutic (please refer to (Heard et al., 2018)), 2) proteostasis -novel intervention (defined as a novel molecule, botanical extract, or dietary manipulation), 3) deregulated nutrient sensing -repurposed therapeutic (reported here) and 4) deregulated nutrient sensingnovel intervention. Where an intervention is thought to modify both pathways (for example the mTOR inhibitor, rapamycin) it was included in the loss of proteostasis group. Therapeutics that were investigated in articles that passed this selection process were then screened using the FDA website to ensure that they met with FDA approval. This systematic review followed the PRISMA guidelines (Supplementary Table F) and focuses on deregulated nutrient sensing and repurposed therapeutics.
The following variables were independently extracted for all articles by two reviewers (BK and DH): author, year of publication, study design, animal model/population (dementia subtype or normal ageing), metabolic status (T2DM, insulin-resistance or normal), sample size, age, sex, baseline cognition/stage of disease, duration of intervention, cognitive outcome, therapeutic, comparator group, and hallmark(s) of ageing targeted by the intervention. For articles using animal models, the following additional variables were extracted: species, method of dementia induction, and method of metabolic disease induction.
For binary outcomes the number of events and total number in groups, percentage of events or ratios with confidence intervals, were extracted; for continuous outcomes, mean or median, with standard deviation, standard error, confidence intervals or interquartile range, and number of participants, were extracted, along with other reported results such as mean difference, p-values, or F-statistic for overall measures of cognitive function.

Data analysis
A semi-quantitative data analysis was performed on extracted outcome data. Dementia models were considered preventative if the intervention was administered prior to the onset of dementia, and were considered to be therapeutic if the intervention was administered after dementia onset. For animal and human studies, an overall positive effect of the administered therapeutic on cognitive performance was defined where a positive primary cognitive outcome was reported, or >50% of the cognitive tests demonstrated a statistically significant improvement in the treatment group compared to the comparator group. A moderately-positive result was defined where ≥ 20% of positive cognitive outcomes were observed, or a single cognitive composite test (e.g. MMSE), demonstrated a statistically significant improvement in the treatment group compared to the comparator group. A finding was considered negative where < 20% of the cognitive outcomes were positive in the treatment group compared to the comparator group. Finally, a detrimental treatment outcome was defined as a statistically significant decrease in cognitive performance for the treatment group compared to the comparator group.
Extracted outcome data for each reported population in all therapeutic classes was then stratified according to cohort (normal ageing or MCI/dementia). In both animals and human studies, a therapeutic class was considered overall to have a beneficial effect on cognition if ≥50% of all reported populations (of species) within a given cohort reported an overall positive effect of the administered therapeutic on cognitive performance. The size of each population was also taken into account, whereby a study with larger populations showing no effect of an administered therapeutic on cognition may warrant reconsideration of a therapeutic class' overall efficacy if studies with smaller populations reported a beneficial cognitive effect.
To investigate for the presence of any correlates that may be found within therapeutic classes (including T2DM status, sample size, population age etc.), populations were ranked in order of: effect of administered therapeutic on cognition (positive, moderately-positive, negative, detrimental effect), followed by population cohort (AD, PD, VD, MCI, dementia, normal ageing), method of induction of dementia/ metabolic disease (for animals), duration of administered therapeutic, dose of administered therapeutic, and were visually examined for any indications of bias.
Articles within therapeutic classes were compared for their amenability to meta-analysis, including aspects such as: cohort, cognitive outcome measure, test condition, unit of test, and reported comparative method by which measures of significance were made within the article (Stone and Rosopa, 2017).

Registered human trials
To provide an overview of the progress in the field of repurposing therapeutics in humans to prevent the onset of age-related cognitive decline or treat mild cognitive impairment (MCI) and dementia, clinical trials registered before 4th of July 2020 that have not yet provided results, were summarized by searching clinicaltrials.gov. The conditions: aging; mild cognitive impairment; Alzheimer disease; vascular dementia; Parkinson disease; Lewy Body disease were key words that were searched for each of the therapeutic classes, for studies of all phases utilizing adult participants.

Risk of bias
The risk of bias was assessed by two reviewers (BK, DH) using the Cochrane Risk of Bias tool v 2.0 for human studies (Higgins et al., 2011) and SYRCLE's risk of bias tool for animal studies (Hooijmans et al., 2014). The Cochrane Risk of Bias tool v 2.0 analyses bias using six key sources of bias, specifically; sequence generation, allocation concealment, blinding of participants and personnel, random outcome assessment, incomplete outcome data, and selective outcome reporting. Each of these was denoted 'green' for 'low risk' if this aspect was reported and deemed to mitigate bias, 'orange' for 'some concerns' if anything less than an absolute mitigation of bias was reported, and 'red' for 'high risk' if this aspect was reported and deemed to encourage bias. SYRCLE's risk of bias tool for animal studies analyses bias using the above categories but also includes baseline characteristics, and random housing. Each of these categories was denoted 'green' for 'low risk' if this aspect was reported and deemed to mitigate bias, 'blue' for 'unclear' if this aspect was not reported, and 'red' for 'high risk' if this aspect was reported and deemed to encourage bias. Overall, a given human or animal study was classified as low risk of bias if <2 sources of bias was deemed to have 'some concerns', and no source of bias was deemed to have high risk of bias. Possible financial conflict of interest was assessed by evaluating disclosed affiliations to a known pharmaceutical company.

Study selection and characteristics
Overall, 114 articles were analyzed (animals n = 91 articles and human n = 23 articles), of which 81/91 articles focused on a dementia model in animals (58/91 mice as the experimental model), while 17/23 articles focused on MCI/dementia pathology in humans (Fig. 1). Table 1 provides an overview of the population demographics for all articles, and the domains of cognition tested. For a detailed description of all articles please refer to Supplementary Tables A.1 and A.2. Overall 32 nutrient sensing-modifying therapeutics, and seven combinations, have been tested for their effect on cognitive outcomes in either animal or human subjects. Of these, in animal models, GLP-1 agonists were the most tested therapeutic (13/91), followed by glitazones (10/91). In human studies, intranasal insulin was the most tested therapeutic (7/ 23), followed by glitazones (5/23). Glitazones, GLP-1 agonists, growth hormone, growth hormone secretagogues, metformin, and intranasal insulin, have been assessed in both animal and human subjects. For a detailed description of all reported cognitive outcomes please refer to Supplementary Tables B.1 and B.2. Fig. 2 provides a summary of the results of all interventions tested. Fig. 3 shows the status of interventional trials registered by 4th July 2020 on clinicaltrials.gov that have not yet released their results, investigating the influence of nutrient sensingmodifying therapeutics on cognitive outcomes in human populations. For a detailed description of the status of these registered interventional trials please refer to Supplementary Table C.

Growth hormone
Whilst growth hormone (GH) receptors are present in many organs, the many metabolic and growth-related effects of GH are accomplished both directly via receptor-signaling, and indirectly via insulin-like growth factor (IGF) (Brooks and Waters, 2010). GH use is approved by the FDA for the treatment of GH deficiency, hypopituitarism, AIDS wasting syndrome, and short bowel syndrome (Sigalos and Pastuszak, 2018). Overall, 2/114 studies (1 animal, 1 human) each investigated the effect of GH on cognition in one normal ageing population. GH may have a beneficial effect on cognition in normal ageing animals (Ramsey et al., 2004), and a possible positive effect in normal ageing humans (Papadakis et al., 1996) (Fig. 2).

Growth hormone secretagogues
Growth hormone secretagogues, which include growth hormone releasing hormone agonists, and growth hormone secretagogue receptor agonists (the natural ligand of which is ghrelin), have been evaluated for use in growth retardation, altered body composition, and gastrointestinal dysfunction, some of which have been approved by the FDA (Ishida et al., 2020). Overall, 5/114 studies (2 animal, 3 human) investigated the effect of growth hormone secretagogues on cognition in normal ageing (3/5 studies) or dementia (4/5 studies) populations. These studies sometimes investigated more than one disease population. In animals, across 2 studies, growth hormone secretagogues have a significant beneficial effect on cognition in animal therapeutic dementia populations (Fig. 2) (Kunath et al., 2015;Madhavadas et al., 2014). In humans, across 3 studies, three normal ageing populations and one mixed normal ageing/MCI population Friedman et al., 2013;Vitiello et al., 2006), and two MCI populations Friedman et al., 2013) were investigated for changes in cognitive outcomes. Growth hormone secretagogues likely have a significant beneficial effect on cognition in human MCI populations, and may have a beneficial effect on cognition in human normal ageing populations (Fig. 2). In humans, this effect was independent of administered therapeutic, dose, duration, cognitive domain assessed, quality of the article, or number of participants ( Fig. 4b and Supplementary Table B.2). The number of participants in each population varied between the two studies examining MCI/dementia cohorts (eight (Friedman et al., 2013) to 31  participants receiving treatment) and the three studies examining normal ageing cohorts (six (Friedman et al., 2013) to 67 in the mixed normal ageing/MCI  receiving

Metformin
Metformin is a biguanide which lowers blood glucose levels by decreasing intestinal absorption, decreasing production of hepatic glucose, and increasing insulin sensitivity, and is the preferred approved first-line therapeutic for use in T2DM (Collins and Costello, 2019;Corcoran and Jacobs, 2018). Overall, 9/114 studies (8 animal, 1 human) investigated the effect of metformin on cognition in normal ageing (4/9 studies) or dementia (7/9 studies) populations. These studies sometimes investigated more than one disease population. In animals, across 8 studies, five normal ageing populations (Ahmed et al., 2017;McNeilly et al., 2012;Thangthaeng et al., 2017) and eight dementia populations (Ahmed et al., 2017;Allard et al., 2016;Chen et al., 2019a, b;McNeilly et al., 2012;Mostafa et al., 2016;Ou et al., 2018) have been assessed. Metformin shows a beneficial effect on cognition in animal therapeutic dementia populations, but not normal ageing populations (Fig. 2). This effect was independent of the method of disease induction (T2DM or otherwise), dose, duration, sample size, cognitive domain assessed or quality of the article (Fig. 4a and Supplementary Table B.1). In humans, in a single study, one normal ageing population with insulin resistance Fig. 2. Outcome of trials in animals and humans investigating nutrient sensing-modifying therapeutics on cognitive performance. A, n = number of articles, DPP-4 = dipeptidyl peptidase-4, GH = growth hormone, GLP-1 = glucagon-like peptide 1, II = insulin infusion, NMDA = N-methyl -Daspartate, PDE = phosphodiesterase, RA = receptor antagonist, SSRI = selective serotonin reuptake inhibitor. P = Preventative model, with therapeutic being given prior to the cognitive onset of dementia. Ф = These two populations were mixed, with ~55% being Normal Ageing and ~45% being MCI. Ω = Statistical run-in plots suggested that this 'improvement' in ADAS-Cog could have followed a chance worsening and represent regression to the mean.

Intranasal insulin
Insulin receptors are widely distributed throughout the brain, and intranasal delivery of insulin has been shown to achieve excellent penetration into the brainpotentially augmenting roles played by endogenous insulin such as the regulation of neuronal processes of metabolism, plasticity, growth, cholinergic function, and survival (de la Monte, 2013). Overall, 13/114 studies (6 animal, 7 human) investigated the effect of intranasal insulin on cognition in populations of normal ageing (4/13 studies) or dementia (10/13 studies). These studies sometimes investigated more than one disease population. In animals, across 6 studies, nine normal ageing populations (Anderson et al., 2017a;Bell and Fadool, 2017;Maimaiti et al., 2016) and eleven dementia populations Mao et al., 2016;Salameh et al., 2015) have been assessed. Intranasal insulin has an overall significant beneficial effect on cognition in animal therapeutic dementia populations, but not normal ageing populations (Fig. 2), which was not related to the method of disease induction (T2DM or otherwise), dose, duration, sample size, cognitive domain assessed or quality of the article (Fig. 4a and Supplementary Table B.1). In humans, across 7 studies, two normal ageing populations (Reger et al., 2006) and 34 MCI/dementia populations (Claxton et al., 2015(Claxton et al., , 2013Craft et al., 2012Craft et al., , 2017Reger et al., 2006;Rosenbloom et al., 2014;Stein et al., 2011) were investigated for changes in cognitive outcomes. Overall, intranasal insulin does not have a positive effect on cognition in human normal ageing populations or MCI/dementia populations (Fig. 2).

Insulin infusion
All levels of human physiology are influenced by insulin, which signals through the insulin receptor glycoprotein present on the surface of many different tissues (Akintola and van Heemst, 2015). In clinical settings it is currently utilized for the treatment of T1DM and T2DM (George and Woollett, 2019). Overall 4/114 studies (4 human) investigated the effect of insulin infusion on cognition in normal ageing (4/4 studies) or dementia (2/4 studies) populations. These studies sometimes investigated more than one disease population. In humans, across 4 studies, 15 normal ageing populations (Kern et al., 2001;Morris et al., 2016;Watson et al., 2009Watson et al., , 2003 and ten MCI/dementia populations (Morris et al., 2016;Watson et al., 2009) were investigated for changes in cognitive outcomes. One of these studies (Watson et al., 2009) examined the effect of insulin infusion, octreotide + insulin infusion, or octreotide infusion, on cognition in normal ageing or dementia populations. Insulin infusion may have a beneficial effect on cognition in human MCI/dementia populations (Watson et al., 2009), but worse cognitive outcomes have been reported in one study (Morris et al., 2016). The number of participants in each population was similar between the two studies examining MCI/dementia cohorts, ranging from seven receiving treatment in the Apolipoprotein E4 positive population, to 16 receiving treatment in the full population of the same study (Watson et al., 2009). In normal ageing populations, the majority of studies showed negative results (Fig. 2) (Kern et al., 2001;Watson et al., 2009).

Octreotide + insulin infusion, and octreotide infusion
Octreotide is a somatostatin analogue which inhibits the release of hormones from the anterior pituitary and hormones of the gastroenteropancreatic system (such as glucagon and insulin), and is primarily approved for the treatment of thyrotropinomas and acromegaly (Debnath and Cheriyath, 2019). Overall 1/114 studies (1 human) investigated the effect of octreotide + insulin infusion, and octreotide infusion alone on cognition in normal ageing and dementia populations. This study (Watson et al., 2009) investigated more than one disease population -examining one normal ageing population, and three dementia MCI/dementia populations, for changes in cognitive outcomes. Octreotide infusion may have a significant beneficial effect on cognition in human normal ageing populations (Fig. 2). Neither octreotide + insulin infusion nor octreotide infusion alone had a significant effect on cognition in human MCI/dementia populations (Fig. 2).
Five combinations of nutrient sensing therapeutics were investigated for their effect on cognitive outcomes in 11 animal dementia populations (Bahramian et al., 2016;Chen et al., 2019b;Gad et al., 2016;Infante-Garcia et al., 2018;Khalaf et al., 2019); DPP-4 inhibitor/memantine (Khalaf et al., 2019), glitazone/GLP-1 agonist (Gad et al., 2016), ICV insulin/memantine (Bahramian et al., 2016), metformin/statin/aspirin/angiotensin-converting enzyme inhibitor (Infante-Garcia et al., 2018), and metformin/GLP-1 agonist (Chen et al., 2019b). Combinations of nutrient sensing therapeutics may have a significant beneficial effect on cognition in animal dementia populations (Fig. 2), although each combination was not explored in more than one study. Fig. 4a shows the SYRCLE risk of bias ratings for animal studies. The majority of animal studies had an unclear risk of bias, as none reported on random outcome assessment, and the majority of studies did not report on the blinding of personnel, random housing, allocation concealment, or baseline characteristics. The risk of bias was similar across animal models of normal ageing, therapeutic dementia models, and preventative dementia models. The method of sequence generation (e.g. randomized) was reported in 43/91 articles. Overall, 91/91 studies had a low risk of bias for selective outcome reporting, and most studies provided all cognitive outcome information, although some did not provide detailed information in cases where cognitive outcome findings were reported to be non-significant. Overall the risk of bias was similar across animal studies regardless of the therapeutic being tested. Fig. 4b shows the Cochrane risk of bias rating for human studies. Overall, 18/23 studies were classified as having an overall low risk bias. Furthermore, 20/23 studies utilized randomized sequence generation, 17/23 studies were double-blinded, and 18/23 studies blinded the outcome assessment. With regards to incomplete outcome data, 18/23 studies were of low concern. Overall the risk of bias was similar across human studies regardless of the therapeutic being tested, with the exceptions of intranasal insulin, and insulin infusion, in which more than half of the studies raised some concerns for bias potential in one or more categories. Financial conflict of interest is an important possible source of bias that is not taken into account with the Cochrane risk of bias tool. Financial interests may be a potential bias in 7/23 studies Harrington et al., 2011;Kern et al., 2001;Luchsinger et al., 2017;Tzimopoulou et al., 2010;Vitiello et al., 2006;Watson et al., 2019), which either disclosed affiliations to a known pharmaceutical company or did not provide a statement reporting any conflicts of interest.

Risk of bias across studies
For a detailed description of each article's risk of bias refer to Supplementary

Discussion
Nutrient sensing-modifying therapeutics may improve ameliorate cognitive decline in MCI or dementia populations but there is limited evidence supporting their preventative effect for ageing populations. Specifically, GLP-1 agonists, growth hormone secretagogues, metformin, DPP-4 inhibitors and PDE-inhibitors were identified as the therapeutics with the most promising cognitive improvements in dementia populations.
It has been reported that greater than 80% of patients with AD have either T2DM or impaired fasting glucose (Janson et al., 2004), and has been estimated to increase an individual's risk of developing dementia by 50% (Biessels et al., 2014), though the precise mechanism is likely to be multi-aetiological (Verdile et al., 2015). Whilst many animal studies utilized insulin resistant/T2DM models, this review did not note an association between T2DM status and cognitive outcomes in animal studies. In human studies however, with the exception of (Luchsinger et al., 2017), no human cohorts were reported as having insulin resistance or T2DM.
GLP-1 agonists and DPP-4 inhibitors, also known as incretin mimetics, frequently showed an improvement in cognitive outcomes in experimental dementia models. Incretin mimetics work by mimicking the functions of the natural incretin hormones. It has been demonstrated that incretins can act directly upon the brain, as peripherally secreted GLP-1 can cross the blood brain barrier (Groeneveld et al., 2016). DPP-4 inhibitors may additionally target the brain by impacting its vasculature (Groeneveld et al., 2016). Furthermore, several in vitro studies have demonstrated that incretins have neurotrophic (Faivre et al., 2011;Liu et al., 2006;Nyberg et al., 2005;Perry et al., 2002) and neuroprotective (Kimura et al., 2009;Liu et al., 2007Liu et al., , 2009Perry et al., 2002) properties in the brain. Importantly, studies have also demonstrated that incretins may influence synaptic plasticitylong term potentiation (LTP) and cognition (Gault and Hölscher, 2008a, b;McClean et al., 2010). As such the positive effects of GLP-1 agonists and DPP-4 inhibitors observed in this review may be explained by a combination of mechanisms that are neurotrophic, neuroprotective, and modulatory of synaptic plasticity. Furthermore, whether a combination of GLP-1 agonist/DPP-4 inhibitor could provide a synergistic increase in GLP-1 levels and improve cognition has not yet been studied.
Similar to the incretin mimetics, growth hormones, specifically, growth hormone secretagogues, were also likely to improve cognitive outcomes in MCI, and dementia models. Currently there is uncertainty around whether growth hormone elicits this cognitive effect directly, indirectly via IGF, as a combination of the two, or through a disparate mechanism (Devesa et al., 2018). Growth hormone may also be able to induce and function through the local expression of IGF-1 in the brain (Pathipati et al., 2011). There are two isoforms of IGF -IGF-1 and IGF-2 -both of which are further structurally related to proinsulin (Lewitt and Boyd, 2019). Additionally, growth hormone secretagogues have been shown to effectively increase IGF levels, a key constituent of the IIS pathway (Lopez-Otin et al., 2013), whilst potentially mitigating the side effects of direct GH administration (Sigalos and Pastuszak, 2018). The relationship between cognition and IGF is yet to be fully understood, although the activity of the IGF axis is well understood to decrease with age (Williams et al., 2018). A study of healthy older men of mean age 69.1 (range 65-76) showed a significant correlation between higher circulating levels of IGF-1 and better performance on measures of perceptual-motor and processing speed (Aleman et al., 1999), whilst an eight year follow-up study of 286 men of mean age 67.2 (range 48-88) found that higher IGF-1 levels at baseline are associated with worse future cognitive function in processing capacity and the MMSE (a measure of global cognition) (Tumati et al., 2016). Historically, IGF-2 has only been considered to play a significant role in the central nervous system during embryonic development (Lewitt and Boyd, 2019), due largely to its limited expression in the brain compared to IGF-1 (Cianfarani, 2012;Lewitt and Boyd, 2019). However, whilst its roles and molecular mechanisms through which it functions in the central nervous system and on metabolism are still largely unknown (Cianfarani, 2012), a prospective study exploring the associations between the various IGF's, their binding proteins, and cognitive function found higher circulating levels of IGF-2 are associated with better cognitive function (Green et al., 2014). In rats and mice, IGF-2 has been found to play a critical role in memory (Chen et al., 2011). It may be that only optimum levels of IGF-1 in conjunction with higher circulating IGF-2 are associated with improved long-term cognitive functionwith levels of IGF-1 too high or too low, or a deficiency of IGF-2 leading to detriment. Additionally, ghrelin -the natural ligand of growth hormone secretagogue receptor agonistshas been shown to increase when fasting (Ariyasu et al., 2001). The increase in both lifespan and healthspan that results from caloric restriction has been clearly elucidated across all studied organisms (although not yet proven in humans) (Anderson et al., 2017b;Gems and Partridge, 2013;Mattison et al., 2017), and despite being less studied, there is growing evidence also indicating a positive effect of intermittent fasting on ageing (Hwangbo et al., 2020). Because a fundamental component of the beneficial effects of fasting are believed to be achieved through the suppression of mTOR (Papadopoli et al., 2019), it is also possible that the mechanism through which the potential cognitive benefits of growth hormone secretagogues are achieved is through mimicking the effects of fasting/caloric restriction. Suppression of the mTOR pathway might be in general a more effective way to reduce Aβ levels (Caccamo et al., 2010), as it is presumably over-activated in AD patients (Zhao and Townsend, 2009).
The results of our review suggest that growth hormone secretagogues may have beneficial impacts on cognition in MCI populations, and potentially normal ageing populations. Yet, the relationship between IGF levels and dementia is also contentious; a meta-analysis of nine studies identified no significant association between serum IGF-1 levels and AD (Ostrowski et al., 2016), and a study of British men found no association between baseline circulating IGF-1, IGF-2 and dementia risk after 17 years of follow-up (Green et al., 2014). Furthermore, the relationship between circulating levels of IGF-binding proteins in AD patients, which may reduce the amount of bioactive IGF for a given total serum IGF level, is poorly understood (Bonham et al., 2018;Galle et al., 2019). It may be that the efficacy of growth hormone secretagogues is related to the progression of dementiathat administration of these therapeutics to individuals with MCI, before significant dampening of IGF-signaling has occurred, could retain IGF-1 and IGF-2 levels in the optimum range and prevent further cognitive decline Friedman et al., 2013).
Overall metformin may have some cognitive benefit in animal models of dementia, but this did not translate to humans or normal ageing animal models. The interplay between metformin, T2DM, and neurodegenerative disease is complex and studies examining metformin in these populations are often inconsistent . There is evidence that metformin given in clinical settings with diabetic populations reduces the risk of dementia (Campbell et al., 2018;Chin-Hsiao, 2019), and cognitive impairment (Ng et al., 2014). It has been hypothesized metformin may act via modulation of tau, and studies examining metformin's neuroprotective effects have largely focused on tau levels and Aβ production . While there are inconsistencies in its effect on Aβ production (Hettich et al., 2014;Picone et al., 2015), metformin may decrease total tau levels and phosphorylated tau (Kickstein et al., 2010;Li et al., 2012), and may further instigate neuroprotection through the activation of AMPK . AMPK activation may enable neuroprotection through the induction of autophagy, angiogenesis, and neurogenesis (Jiang et al., 2014;Jin et al., 2014;Poels et al., 2009;Venna et al., 2014).
The current available evidence does not support glitazones as a beneficial therapeutic for MCI and dementia populations. However, a longitudinal observational study (Lu et al., 2018) demonstrated a lower incidence of dementia in populations taking pioglitazone and metformin, than with other metformin-based dual therapies. This may indicate that glitazones are effective in preventing dementia only in combination with another therapeutic. Similarly, administration of insulin did not appear to benefit cognitive performance. Although intranasal insulin, which effectively bypasses the systemic circulation and blood-brain barrier to directly enter the cerebrospinal fluid, may have some role in improving cognition in dementia patients with shorter-term use (Dubey et al., 2020).
In contrast to glucose-lowering therapeutics, PDE-inhibitors exhibit their likely beneficial cognitive effect on experimental dementia models through increasing levels of cAMP and/or cGMP (García-Osta et al., 2012). PDE-inhibitors activate the cAMP response element-binding, which may promote gene transcription (Impey et al., 1996;Lu et al., 1999) that has been implicated in long-term memory formation and persistent long-term potentiation (Tully, 1997;Yin and Tully, 1996). This may involve the formation of new synaptic connections in the hippocampus (Ran et al., 2012;Tully et al., 2003), and it has been suggested that this can mitigate the cognitive impacts of dementia by enhancing synaptic function (García-Osta et al., 2012). Further mechanisms may be cognitive vasodilatory properties, and/or as a consequence of emotional arousal (Reneerkens et al., 2009). As elicited by articles examining normal ageing animals in this review, cognitive-enhancing effects of PDE-inhibitors have been observed in a number of different normal healthy animal species (Richter et al., 2013). Currently two phase IV clinical trials (one AD cohort, one VD cohort) examining the effect of PDE-inhibition of cognition in humans have been completedno improvement in cognitive outcomes were reported for the AD cohort (Lee et al., 2019), and the results of the VD cohort remain unpublished.
It is possible that combinations of therapeutics may provide synergistic improvements in cognitive outcomes. In this review, our search elicited studies examining a number of therapeutic combinations in animals. All, except the metformin/GLP-1 agonist combination, demonstrated improvements in cognitive outcomes in dementia models. Metformin may have a benefit in combination with glitazones (Lu et al., 2018) and sulfonylureas (Hsu et al., 2011). Mechanistically, targeting multiple pathways would seem to be necessary for treatment of dementia, which is a complex multi-aetiological pathological entity. Currently approved agents are limited to cholinesterase inhibitors, memantine, or a combination of these agents, but many other therapeutics are currently in clinical studies as add-on therapies to the standard of care (Cummings et al., 2019).
Therapeutics classified as 'Other' were not investigated in any human population and were examined in <3 studies in animals, thus making it difficult to draw tenable conclusions relating to their efficacy in human normal ageing and dementia populations. Some of the more promising 'other' therapeutics included in this review are amylin & analogues, caffeine, ICV insulin, and subcutaneous insulin. Our search of registered human trials found a number of 'other' therapeutics currently being investigated, including melatonin, statins, adrenoreceptor agonists, and angiotensin receptor blockers.
There has consistently been a poor translation of successful therapeutics of pre-clinical animal dementia models to successful interventions in human dementia clinical trials (Franco and Cedazo-Minguez, 2014). Aspects such as the wide range of animal dementia models available, the questionable accuracy of these in mimicking human age-related dementia, and differences in study design between animal and human studies, must presumably all contribute to this lack of translation. The majority of animal dementia models utilized by studies in this review were transgenic, overexpressing or producing mutant products of human genes such as amyloid precursor protein (APP), tau, and presenilin 1 (PS1). However, a wide variety of other dementia models were also utilized, such as administration of ICV Aβ or streptozotocin. It remains contentious how extrapolatable findings are from animal models such as these, which are at best incomplete representations of a complex multi-aetiological disease processan example being that transgenic models do not fully recapitulate neuronal loss (Elder et al., 2010) which is a fundamental pathological mechanism of dementia (Moya-Alvarado et al., 2016). Although human interventional trials are often carried out for a much shorter portion of time relative to the individual's lifespan than in animal trials, the longer duration of animal trials are often carried out using a much smaller number of animals. In the future, animal studies may more clearly define the dementia model's baseline level of cognition, and their own aims of exploring interventions as being either neuroprotective, cognitive enhancing, or disease-modifying agent, to help discriminate the various modes by which successful pre-clinical interventions may be having their effect.

Limitations
Our search strategy was based primarily on key terms related to the main nutrient sensing pathways, with the addition of a selection of therapeutics well-known to modulate these processes. Therefore, our search may have missed studies that examine drugs that modulate nutrient sensing not named in our search and not mentioning nutrient sensing pathway or related key terms. Secondly, whilst the therapeutics included in this review have had prior FDA approval, certain specific administration methods of these therapeutics have not; namely intranasal insulin, insulin infusion, octreotide infusion, and combination insulin infusion + octreotide infusion. Thirdly, we cannot exclude publication bias, particularly in animal studies which are unlikely to be registered and may be less likely to be published if results are negative. Fourth, we did not perform any formal statistical analysis and results are based on reported p-values. The significance of p-values is influenced by sample size. This is likely to have impacted our findings as the majority of animal studies included in this review had a low sample size (n≤10-15, with n = 5 for some mouse studies), and 11/23 human studies utilized human populations with <15 individuals. The small sample size of these studies specifically, the animal studies is a widespread structural problem within animal research. Studies with small sample sizes are likely to produce ambiguous or misleading results as smaller numbers can inflate the effect size. Ensuring experiments use appropriate sample sizes are critical for reproducibility of findings and future animal research should focus on utilizing power calculation to ensure the appropriate number of animals are included in all treatment arms. Fifth, due to the variation in reported data -different; animal models, FDA therapeutics, dosages, duration of treatment, length of study and control groups -a meta-analysis reporting overall effects on cognition could not be performed. Sixth, whilst the age of animals was largely consistent, mice utilized as preventative and therapeutic dementia models may have been slightly younger than normal ageing models. Finally, this review has also identified some reporting omissions within this research field. Specifically, the use of the SYRCLE bias tool has highlighted large gaps in reporting of the trial design in animal studies. Aspects such as random outcome assessment, blinding of personnel, random housing, allocation concealment, baseline characteristics, and sequence generation, was often unclear. Future animal trials should consider following the SYRCLE guidelines (Hooijmans et al., 2014).

Conclusions
The results of this review indicate that nutrient sensing-modifying therapeutics have the potential to alter cognitive outcomes in MCI or dementia populations. Overall, the translation of therapeutic efficacy from animal models to human populations is limited. Further studies are required to fully elucidate the potential of GLP-1 agonists, growth hormone secretagogues, metformin, DPP-4 inhibitors, and PDE-inhibitors in dementia.