APOE genotype and cognition in healthy individuals at risk of Alzheimer's disease: A review

APOE-ε4 is best known as a risk factor for Alzheimer's disease (AD). Consequently, there is considerable research interest in understanding whether APOE-ε4 influences cognition in healthy adults. Despite a substantial literature reporting effects of APOE genotype on cognition, findings are inconsistent. In particular, it is challenging to separate whether cognitive deficits in APOE-ε4 carriers reflect the influence of prodromal dementia pathology ("prodromal hypothesis"), or a direct contribution of APOE genotype to individual differences ("phenotype hypothesis"). Variable methodology across studies further complicates the issue. These challenges have limited what can be learnt about the processes underlying cognitive ageing and dementia by studying the influence of APOE genotype on cognition. In this review, we focus on the two compatible neurobiological mechanisms by which APOE genotype may influence cognition in healthy adults (prodromal and phenotype). We summarise the behavioural evidence for the influence of APOE on cognition in non-demented adults and explore key methodological challenges for disentangling the cognitive effects of different neurobiological mechanisms of APOE. Evidence suggests that at least some APOE-ε4 cognitive deficits are due to early AD pathology, whilst sensitive measures of cognition are beginning to reveal subtle cognitive differences between APOE genotypes in mid-adulthood, prior to the onset of the AD prodromal period. We conclude with recommendations for future research to investigate the cognitive consequences of neurobiological processes affected by APOE and maximise the translational potential of this research.


Introduction
Since the discovery of apolipoprotein-E (APOE) ε4 as a genetic risk factor for Alzheimer's disease (AD; Corder et al., 1993;Saunders et al., 1993), researchers have investigated whether healthy ε4-carriers show impaired cognition or more rapid cognitive decline compared to non-carriers, as an early sign of impending dementia (e.g., Bondi et al., 1995;Staehelin, Perrig-Chiello, Mitrache, Miserez, & Perrig, 1999;Wetter et al., 2005), or source of variability in healthy cognitive ageing (e.g., Deary et al., 2003Deary et al., , 2002Greenwood & Parasuraman, 2003;Greenwood, Lambert, Sunderland, & Parasuraman, 2005;Helkala et al., 1996). Understanding the association between APOE genotype and cognition in healthy adults can aid development of preclinical biomarkers for AD, increase knowledge of the mechanisms by which APOE-ε4 increases AD risk, and inform interventions to boost cognition in older adults. However, the reported effects of APOE on cognition are highly inconsistent, and interpretation of these mixed findings is complicated by conceptual and methodological challenges.
A major challenge in understanding the association between APOE and cognition is the difficulty disentangling direct influences of APOE genotype on a 'cognitive phenotype' (Greenwood & Parasuraman, 2003;Greenwood, Lambert, et al., 2005) from the influence of greater prodromal AD pathology in the at-risk group (Smith et al., 1998). In part, this stems from a conceptual challenge where the two possibilities have been framed as competing 'hypotheses' to explain the effect of APOE on cognition (Bunce et al., 2014;Foster et al., 2013;Greenwood & Parasuraman, 2003;Greenwood, Espeseth, Lin, Reinvang, & Parasuraman, 2014;Greenwood, Sunderland, Putnam, Levy, & Parasuraman, 2005;Hofer et al., 2002;Negash et al., 2009;Parasuraman, Greenwood, & Sunderland, 2002), with evidence supporting one 'hypothesis' interpreted as evidence against the other (e.g., Bondi, Salmon, Galasko, Thomas, & Thal, 1999;Foster et al., 2013;Small, Basun, & Backman, 1998;Smith et al., 1998). Unquestionably, the proposed direct versus pathology-mediated cognitive effects related to APOE genotype are underpinned by different neurobiological mechanisms, but these need not be mutually exclusive. Dissecting the relative influence of each neurobiological mechanism on cognitive performance will ultimately elucidate their roles in cognitive ageing and Alzheimer's disease.
In this review, we outline two compatible neurobiological mechanisms by which APOE genotype may influence cognition, and highlight the individual and interacting methodological challenges for separating the cognitive consequences of each mechanism. The main objective is to demonstrate that APOE genotype may affect cognition via multiple neurobiological mechanisms, each with the potential to provide insight into different aspects of cognitive ageing and Alzheimer's disease. We conclude with recommendations for future studies to facilitate interpretation of APOE-related effects and increase the translational value of these findings.
The APOE-ε4 allele is the best known genetic risk factor for sporadic Alzheimer's disease (Corder et al., 1993;Rocchi, Pellegrini, Siciliano, & Murri, 2003). Carriers of the ε4 allele show increased risk and earlier age of onset of AD, compared to non-carriers (Borgaonkar et al., 1997;Corder et al., 1993;Sando et al., 2008). ε4-related AD risk is conferred in a dosedependent manner, with greater risk and earlier age of onset c o r t e x 1 0 4 ( 2 0 1 8 ) 1 0 3 e1 2 3 in ε4-homozygotes compared to ε4-heterozygotes (Blacker et al., 1997;Liu et al., 2013). In contrast, possession of an ε2 allele may provide protection against AD diagnosis (for a review of APOE-ε2, see Suri, Heise, Trachtenberg, & Mackay, 2013). The mechanisms by which APOE-ε4 increases AD risk are not fully understood, but a role of ApoE4 in greater deposition and/or reduced clearance of amyloid-b, resulting in increased aggregation of this hallmark pathological feature of AD, has been proposed (Huang, Weisgraber, Mucke, & Mahley, 2004). Isoform-specific effects on neuroinflammation, neurogenesis, and neuronal toxicity may also contribute to increased risk of AD in ε4-carriers (for a review of APOE mechanisms and AD risk, see Liu et al., 2013). Possession of an APOE-ε4 allele is neither necessary nor sufficient for development of AD (Henderson et al., 1995;Hyman et al., 1996); however, approximately half of ε4-homozygotes will develop AD by age 85, compared to only 10% of non-carriers (Genin et al., 2011).

Phenotype
In contrast to a strong prodromal view, other researchers consider that APOE genotype may exert a direct effect on cognition, independent of future AD diagnosis, via genotypespecific influences on neurobiological processes that affect cognition (Greenwood & Parasuraman, 2003;Parasuraman & Greenwood, 2004). Genetics are an important determinant of cognitive ability (Plomin, DeFries, McClearn, & McGuffin, 2001), and normal allelic variation may contribute to individual differences in cognitive abilities throughout life. This cognitive neurogenetic approach (Greenwood & Parasuraman, 2003;Parasuraman & Greenwood, 2004) proposes a cascade of events linking genotype to cognitive phenotype (also see Deary, Penke, & Johnson, 2010;Meyer-Lindenberg & Weinberger, 2006;Reinvang, Deary, et al., 2010). As the isoform structure of a protein depends on the specific genotype of the coding gene, this will result in isoform-specific alteration of neurobiological endophenotypes, ultimately producing variable cognitive phenotypes.
According to the 'phenotype hypothesis' (Greenwood & Parasuraman, 2003;Greenwood, Lambert, et al., 2005;Parasuraman et al., 2002), the putative roles of ApoE in CNS functions, such as synaptic plasticity and repair, provide an altered neural endophenotype by which APOE genotype may result in a 'cognitive phenotype'. According to this hypothesis, cognitive differences in ε4-carriers are not confined to older adults and are independent of subsequent AD diagnosis (Greenwood, Lambert, et al., 2005). Neuroimaging studies in younger adults (under 60 years) provide corroborative results that suggest APOE-ε4 is associated with differences in the brain prior to the onset of AD pathology, in a manner independent of future AD. Similar patterns have been observed in at-risk ε4-carriers and low-risk ε2-carriers, compared to ε3carriers, in task and resting-state functional MRI, as well as white matter (WM) integrity (e.g., Trachtenberg, Filippini, Cheeseman, et al., 2012;Westlye, Reinvang, Rootwelt, & Espeseth, 2012; for reviews of APOE genotype effects on MRI measures in healthy adults, see Fouquet, Besson, Gonneaud, La Joie, & Ch etelat, 2014;. Whilst the association between APOE and Alzheimer's disease is unusually strong for an individual susceptibility geneedisease association (Genin et al., 2011), a direct effect of APOE on a cognitive phenotype is likely to be subtle. Cognitive ability is influenced by multiple genes, environmental factors, and complex geneeenvironment interactions, so any single gene will exert only a small effect on a cognitive phenotype (Marian, 2012). However, the phenotypic influence of APOE on cognition may be particularly observable in later life. The influence of genetics on cognition is thought to increase with age (e.g., McClearn et al., 1997;McGue & Christensen, 2002), possibly due to neuronal and neurochemical losses with age that highlight genetic influences on cognition (Lindenberger et al., 2008). Normal allelic variation may therefore particularly contribute to increased variability in cognitive c o r t e x 1 0 4 ( 2 0 1 8 ) 1 0 3 e1 2 3 performance in older adults (e.g., Christensen et al., 1999;Wilson, Beckett, et al., 2002).

Literature search
For this comprehensive review of APOE and cognition studies in healthy adults, online searches were performed in PubMed using combinations of the following search terms: 'APOE', 'apolipoprotein', 'cognition', 'cognitive', 'performance', 'cognitive ability', and 'cognitive abilities'. Only studies published from 1993, when APOE genotype was identified as a risk factor for AD (Corder et al., 1993), were considered. Studies that used neuropsychological measures, forming the majority of the literature, were considered for inclusion in Tables 1 and 2. Participants were required to be cognitively healthy and grouped by APOE genotype. Any studies that included individuals with objective cognitive impairment or clinical diagnoses at baseline were not included, unless results were separately reported for non-demented participants. Studies with highly selected samples, e.g., women on hormone replacement therapy, were also not included. Studies that focused on APOE genotype interactions with other variables, including MRI metrics, were not included. Age of the sample (range or mean) and follow-up duration, where applicable, were required to be reported for inclusion. Only studies published in English were considered. As detailed discussion of neuroimaging findings is beyond the scope of this review, functional MRI studies were not included and are reviewed elsewhere (see . Studies with overlapping samples were not excluded (e.g., Bunce, Anstey, Burns, Christensen, & Easteal, 2011;Jorm et al., 2007).
As illustrated in Tables 1 and 2, many studies have investigated the effect of APOE genotype on not only one cognitive test, but on extensive cognitive batteries assessing multiple cognitive domains in the same sample (e.g., Caselli et al., 2001;Duchek et al., 2006;Foster et al., 2013;Levy et al., 2004). In these studies, ε4-related deficits have often been reported in some, but not all, domains, suggesting a domain-specific effect of APOE-ε4 on cognition, with episodic memory particularly affected. However, the pattern of spared versus impaired performance across cognitive domains is inconsistent across studies. Whilst many studies have reported an ε4-related deficit in at least one cognitive domain, the majority of comparisons have resulted in non-significant effects (see Tables 1 and 2). Meta-analyses addressing the domain specificity of APOE effects on cognition, primarily in older samples (over 60 years), suggest APOE-ε4 carriers are specifically impaired on measures of episodic memory, executive function, global cognition, and possibly perceptual speed, but there is no genotype effect on verbal ability, attention, visuospatial function, and primary memory Wisdom et al., 2011). c o r t e x 1 0 4 ( 2 0 1 8 ) 1 0 3 e1 2 3 Interpretation of these heterogeneous findings is challenging. Methodological factors such as age of sample, sample size, and cognitive measures vary substantially among studies, making cross-study comparisons difficult. Inconsistent reporting of the substantial number of potential modifying variables limits the value of meta-analysis to shed light on the sources of heterogeneity in this literature Wisdom et al., 2011). Additionally, since the majority of research in this field is conducted in older samples, it is often unclear whether observed ε4-deficits relate to a direct effect of APOE on cognition or the presence of prodromal AD. In many cases, it is also difficult to resolve whether the absence of significant APOE effects on cognition reflects a genuine lack of genetic effect, or rather a lack of statistical power or insensitive cognitive measures. For example, several studies that reported null effects had large samples (N > 1000) and therefore high statistical power (Bathum et al., 2006;Bunce et al., 2011;Jorm et al., 2007;Liu et al., 2010;Luciano, Gow, Taylor, et al., 2009), but some had young and mid-adulthood samples, so the cognitive measures used may not have been sensitive to an APOE effect in this age group (see 5. Methodological challenges for further discussion). Other studies with similarly large samples, all with older adults, observed significant cognitive deficits in ε4-carriers (Izaks et al., 2011;Laukka et al., 2013;Liu et al., 2010;.
Disentangling the cognitive consequences of the neurobiological mechanisms by which APOE may influence cognition requires methodology that is optimised to addresses specific research questions. Multiple methodological factors must be optimised in concert to improve the ability to interpret findings. Consideration of 'prodromal' and 'phenotypic' as distinct but compatible neurobiological mechanisms, rather than competing hypotheses, will aid the separation of direct and prodromal effects of APOE on cognition. Fig. 1 highlights the importance of separating these cognitive effects of APOE, as they can provide insight into separate neurobiological processes, with applications to distinct aspects of cognitive ageing and Alzheimer's disease. In the next sections, we highlight key methodological challenges for investigating the cognitive effects of different underlying neurobiology of APOE, and make recommendations for future studies.

Age
Age is a major methodological challenge for separating direct effects of APOE on cognition from the cognitive consequence of early AD pathology, as sample age determines the mechanism that can be studied. The long prodromal period of AD makes it difficult to rule out the influence of AD pathology in elderly samples, particularly given the earlier age of AD onset in APOE-ε4 carriers. Significant interactions between age and APOE genotype in older samples (over 60 years), representing an increasing negative influence of APOE-ε4 genotype on cognition with age (e.g., Baxter et al., 2003;Caselli et al., 1999;Laukka et al., 2013;Wisdom et al., 2011), may reflect an increasing presence of AD pathology with age, although other studies have found no significant interaction of age and genotype (Bunce et al., 2011(Bunce et al., , 2014Jorm et al., 2007;Small et al., 2000Small et al., , 2004. Cross-sectional findings in Table 1 are ranked by mean age of sample, but a clear pattern does not emerge. Variable and often broad age ranges between studies make comparison of findings challenging. To minimise inclusion of prodromal AD, most studies use cognitive screening in older samples. However, the method of c o r t e x 1 0 4 ( 2 0 1 8 ) 1 0 3 e1 2 3 cognitive screening used varies greatly between studies (see Table 2), including a single clinician or consensus panel decision, use of clinical diagnostic tools such as the Cambridge Mental Disorders of the Elderly Examination (CAMDEX; Roth et al., 1986) or the Clinical Dementia Rating tool (CDR; Morris, 1993), and varying thresholds of the Mini-Mental State Examination (MMSE; Folstein, Folstein, & McHugh, 1975), whilst other studies report no details of the cognitive screening used, if any. A stringent cognitive screening approach used in a small number of studies is retrospective exclusion of individuals diagnosed with AD during a followup period from the cross-sectional analysis of baseline data. Studies using this approach found significant baseline ε4related deficits in episodic memory, executive function, attention, semantic memory, processing speed, language, and global cognition were no longer significant after incident AD cases were excluded (Bondi et al., 1999;Hayden et al., 2009;Hofer et al., 2002;Knight et al., 2014;Laukka et al., 2013;Praetorius et al., 2013; also see Batterham et al., 2013;Jorm et al., 2007;Schiepers et al., 2012), suggesting the initial ε4-deficits were due to prodromal AD (also see 5.3. Sample size and longitudinal design). However, in some studies, significant ε4-deficits in memory, perceptual speed, and global cognition remained after exclusion of incident AD cases (Hofer et al., 2002;Jonker et al., 1998;Laukka et al., 2013;Praetorius et al., 2013;Winnock et al., 2002;also see B€ ackman et al., 2004;Riley et al., 2000). Even with stringent methods, the possible influence of prodromal AD in older samples is difficult to rule out.
To investigate genetic effects in the absence of prodromal AD pathology, studies have more recently moved to investigate the effect of APOE on cognition in mid-adulthood (30e60 years) samples, before AD pathology is likely to accumulate. Possible antagonistic pleiotropic effects in early life may also be avoided in mid-adulthood samples. Due to the long period of accumulation of AD pathology before symptom onset, individuals in their late 50s may already fall into the prodromal AD pre-dementia stage, meaning that a more restricted age range (e.g., 30e50 years) may be optimal. Without very large samples, the inclusion of participants with a wide age range (Caselli, Dueck, Locke, Hoffman-Snyder, et al., 2011;Wishart et al., 2006) is unlikely to provide clear results.

Cognitive measures
Cognitive measures are key to unpicking the cognitive effects of the prodromal and direct neurobiological mechanisms of APOE. The neuropsychological measures used in most studies, whilst well-validated and quick to administer, typically lack the sensitivity to detect subtle variation in cognition (Houx et al., 2002;Jorm et al., 2007) and lack the specificity to measure variation in isolated cognitive processes (Greenwood et al., 2000, Greenwood, Lambert, et al., 2005Wisdom et al., 2011). Neuropsychological tests therefore have limited use for investigating the direct effect of APOE on a cognitive phenotype in mid-adulthood samples, and a recent metaanalysis found no evidence for an APOE genotype effect on midlife neuropsychological performance . More sensitive 'cognitive assays' that target specific cognitive processes are required to detect subtle direct cognitive effects of APOE genotype (Greenwood, Lambert, et al., 2005). Several studies in older adults have reported selective ε4-related deficits on sensitive cognitive neurosciencebased measures of prospective memory, implicit learning, spatial navigation and working memory (Berteau-Pavy et al., 2007;Driscoll et al., 2005;Duchek et al., 2006;Negash et al., 2007;Reinvang, Winjevoll, Rootwelt, & Espeseth, 2010;Rosen et al., 2002). However, in older samples, such sensitive measures again face the challenge of isolating direct effects of APOE from the potential influence of incipient AD. In a study by Negash et al., 2007, older APOE-ε4 carriers demonstrated deficits in contextual cueing, but no APOE-related effect on serial reaction time performance was observed, interpreted as a selective effect of APOE-ε4 on medial temporal function, but spared frontal function. However, the same pattern of spared and impaired performance was also observed using the same measures in a sample diagnosed with MCI, regardless of APOE status (Negash et al., 2007).
Sensitive cognitive measures have only been used in a small number of studies to investigate APOE-related effects in mid-adulthood samples, although this number is increasing. Many such studies come from Greenwood and colleagues, who, based on observed cognitive deficits in AD patients (Parasuraman et al., 2002), have investigated whether ε4carriers show selective deficits in attentional processes in mid-adulthood. In a series of experiments, mid-adulthood APOE-ε4 carriers demonstrated intact sustained attention, but impaired attentional disengagement in a cued discrimination task, and impaired scaling of attention during a visual search task (Espeseth et al., 2006;Greenwood, Lambert, et al., 2005, Greenwood et al., 2000Negash et al., 2009;although see Lancaster, Forster, Tabet, & Rusted, 2017). Longitudinally, APOE-ε4 homozygotes showed increasingly impaired attentional scaling during visual search, whilst non-carriers improved over 3 years . Additionally, midadulthood ε4-carriers showed cross-sectional selective deficits in retaining visuospatial information in working memory (Greenwood, Lambert, et al., 2005; although see Greenwood et al., 2014). Surprisingly, Greenwood et al., 2014 reported a significant improvement in working memory performance in mid-adulthood ε4-carriers when retested after 3 years, whilst non-carriers showed longitudinal decreases in performance, interpreted by the authors as possible cognitive compensation in APOE-ε4 carriers.
More recently, other researchers have expanded the investigation of APOE effects on mid-adulthood cognition to other cognitive domains. A study by Salvato, Patai, McCloud, & Nobre, 2016 investigated the effect of APOE genotype on longterm memory-guided attention in a mid-adulthood sample (mean age 45.7 years). Whilst APOE genotype had no significant effect on explicit memory or memory-guided attention, ε4-homozygotes showed an impaired correlation between memory and attention, indicating the memories guiding attentional orienting in ε4-homozygotes were not explicitly available (Salvato et al., 2016). The authors suggested these cognitive results might reflect changes in brain regions that underpin long-term memory-guided attention, e.g., medial temporal, parietal and frontal regions, which are also compromised in AD (Salvato et al., 2016). In another study, adults aged 40e51 completed a short-term memory paradigm where participants were required to hold multiple fractals in memory, and after a short delay identify which of 2 fractals had previously been presented, as well as the location in which it was shown (Zokaei et al., 2017). This task was previously shown to be sensitive to memory deficits in patients with medial temporal lobe (MTL) damage (Pertzov et al., 2013). Intriguingly, this paradigm revealed a behavioural advantage in male APOE-ε4 carriers, with less mislocalisation of the remembered item to the location of another item held in memory, interpreted as a possible example of antagonistic pleiotropy in midlife (Zokaei et al., 2017).
The breadth of domains studied with sensitive cognitive assays remains relatively limited to those impaired in patients with AD/MTL damage. Investigation of further neurobiologically relevant domains is warranted. To robustly investigate the direct effect of a gene on behaviour, the measured phenotype should reflect the likely altered neurobiological endophenotype, to provide a plausible and mechanistic association between genotype and phenotype (Goldberg & Weinberger, 2004). Additional cognitive processes that are biologically related to the proposed influence of APOE genotype on CNS function, e.g., synaptic plasticity and neuronal repair (Liu et al., 2013), should therefore be explored. Tasks that assess long-term learning may be particularly sensitive to such effects of APOE genotype on CNS functions (see 6.1.

Recommendations for phenotype effect of APOE on cognition).
Characterisation of the APOE-ε4 cognitive effects related to prodromal AD may also benefit from a more selective focus on measures relevant to the underlying neurobiology. Numerous studies have assessed the effect of APOE genotype on extensive neuropsychological batteries, regardless of the association of each measure with cognitive impairments in early AD. The resulting multiple comparisons problem is inconsistently and sometimes insufficiently controlled for. A more selective assessment of cognitive domains may be advantageous, for example those recommended in a recent consensus statement on cognitive outcomes in preclinical AD based on cognitive correlates of preclinical brain changes, including verbal episodic memory, working memory, and navigation in egocentric space (Ritchie et al., 2017).

Sample size and longitudinal design
The lower frequency of APOE-ε4 in the population, compared to APOE-ε3, has resulted in consistently smaller APOE-ε4 samples in most studies, unless specifically matched between groups (e.g., Caselli et al., 2001Caselli et al., , 1999Salvato et al., 2016;Zokaei et al., 2017). There are also relatively few investigations of the specific effect of APOE-ε2 on cognition, due to the low occurrence of the ε2 allele in the population (although see Bartr es-Faz et al., 1999;Batterham et al., 2013;Blair et al., 2005;Bloss et al., 2010;Bunce et al., 2014;Chey et al., 2000;Chiang, Raptentsetsang, & Jack, 2010;Greenwood et al., 2000;Helkala et al., 1995Helkala et al., , 1996Hofer et al., 2002;Marioni et al., 2016;Pendleton et al., 2002;Quintas et al., 2014;Salo et al., 2001;Staehelin et al., 1999;Ward et al., 2014;Wilson, Bienias, et al., 2002;Lancaster, Forster, et al., 2017). Table 1 also illustrates the variability in sample size across studies. The required sample size may depend on the neurobiological mechanism of APOE under investigation. Longitudinal studies are required to investigate prodromal APOE-ε4 cognitive effects in relation to clinical outcomes, e.g., for biomarker development. Therefore, a large initial sample is required, as cognitive effects will be driven by only a subsample with early AD, and some participants may be lost to follow-up. Sufficient sample size is particularly important if incident AD cases will be retrospectively excluded from baseline analysis, as these cases are more likely to be excluded from the already smaller APOE-ε4 group (e.g., Batterham et al., 2013). Elimination of initially significant baseline APOE effects after exclusion of c o r t e x 1 0 4 ( 2 0 1 8 ) 1 0 3 e1 2 3 incident AD (e.g., Bondi et al., 1999;Hayden et al., 2009;Hofer et al., 2002;Knight et al., 2014;Praetorius et al., 2013) may have resulted from degraded statistical power in some cases. In contrast, a subtle APOE-related cognitive phenotype should be present in all ε4-carriers, therefore smaller samples may be sufficient, as long as the cognitive measures used test appropriate domains and are sufficiently sensitive.

6.
Recommendations for future research In the previous sections, we have outlined the mixed reported effects of APOE on cognition, and discussed how key methodological factors complicate the interpretation of these findings. Some studies have attempted to facilitate interpretation by addressing individual methodological challenges (e.g., larger sample size or under-studied cognitive domain). However, the interactive nature of these methodological factors requires multiple aspects of methodology to be optimised simultaneously to aid the interpretation of APOE and cognition results. Future studies can add to our understanding of the prodromal and direct routes of APOE influence on cognition by employing methodology targeted at specific neurobiological mechanisms, to simplify the interpretation of results.
Here we make recommendations for future studies (summarised in Fig. 1).

Recommendations for phenotype effect of APOE on cognition
As evidence for APOE-related alterations of CNS functions has primarily come from animal literature, evidence of an APOE cognitive phenotype would provide support for an altered neural endophenotype in humans. The applications of this knowledge are two-fold: firstly, to provide greater understanding of the neurobiological processes altered by APOE genotype, and secondly, to inform how genetic variability contributes to individual differences in cognition, providing targets for interventions to maximise cognition in non-neuropathological ageing. The amyloid-independent mechanism(s) that directly influence cognition in ε4-carriers may also reflect a 'susceptibility' factor that contributes to increased risk of AD (Greenwood, Lambert, et al., 2005), and may interact with amyloid-dependent pathways by promoting amyloid deposition (Verghese et al., 2011), or exacerbating the impact of amyloid deposition, possibly via proinflammatory responses (Liu et al., 2013). Greater understanding of the direct effect of APOE genotype on a cognitive phenotype may therefore inform interventions to ameliorate increased risk of AD in ε4-carriers.
Whilst future studies may avoid the influence of prodromal AD when investigating an APOE cognitive phenotype by excluding volunteers positive for preclinical AD biomarkers, e.g., increased amyloid or reduced glucose metabolism on PET imaging, or reduced CSF amyloid (Sperling et al., 2011), limited resources and invasiveness of procedures means such screening is not always feasible. Instead, the use of midadulthood samples (30e50 years) is optimal to investigate a cognitive phenotype prior to the development of AD pathology, whilst also avoiding possible antagonistic pleiotropic effects of APOE on cognition earlier in life. Future research should also examine cognitive domains relevant to the putative neurobiological mechanisms affected by APOE-ε4 using sensitive and specific cognitive measures. A promising initial target is the effect of APOE genotype on long-term learning over days or weeks, e.g., learning a new skill (Scholz, Klein, Behrens, & Johansen-Berg, 2009), rather than minutes or hours tested in neuropsychological assessments, to reflect the effect of disrupted lipid provision on synaptic plasticity. Alternatively, the detrimental effect of APOE-ε4 on response to neuronal injury, via neuronal repair, neuroinflammatory, or neurotoxicity mechanisms (Liu et al., 2013;Verghese et al., 2011), may be associated with more diffuse cognitive deficits, depending on the nature and location of injury (Greenwood, Lambert, et al., 2005).
Furthermore, if APOE-ε4 has a cognitive phenotype independent of future AD diagnosis, APOE-ε2 genotype may also be associated with its own cognitive phenotype, although the supporting neurobiological mechanism is not clear. Amyloid aggregation is an unlikely candidate, as the oldest-old ε2carriers have a similar presence of amyloid pathology compared to ε4-carriers, despite reduced risk of dementia (Berlau, Corrada, Head, & Kawas, 2009). Instead, APOE-ε2 may promote synaptic integrity, suggested by greater dendritic arborisation observed in APOE-ε2 transgenic mice, versus ε3 and ε4 (Dumanis et al., 2009), or facilitate anti-inflammatory processes, as ApoE2 appears to be more effective at protecting cells from oxidative stress-induced cell death compared to ApoE3 and ApoE4 (Miyata & Smith, 1996; see Suri et al., 2013).

6.2.
Recommendations for prodromal effect of APOE on cognition A growing body of work suggests that at least some cognitive deficits in currently healthy older APOE-ε4 carriers are due to prodromal AD. These findings do not rule out the possibility of additional mechanisms of direct APOE influence on cognition, and are important in demonstrating that cognitive deficits relating to future diagnosis of AD can be detected early, in adults currently considered to be healthy. APOE-ε4 genotype alone provides limited prognostic information about whether a person will be diagnosed with AD (Bookheimer & Burggren, 2009), so prodromal AD-related ε4 cognitive effects can be utilised in future studies to develop preclinical cognitive biomarkers to more precisely identify who is most at-risk. Secondary prevention clinical trials enriched with those most at-c o r t e x 1 0 4 ( 2 0 1 8 ) 1 0 3 e1 2 3 risk of AD will reduce the sample size and longitudinal period required to assess treatment efficacy (Holland, McEvoy, Desikan, & Dale, 2012). Cognition can provide cheap, quick and non-invasive preclinical biomarkers to be used alone, or in combination with other biomarkers, such as CSF, PET or MRI metrics, to identify those who will benefit most from treatment and track early treatment efficacy.
Future studies aimed at utilising APOE genotype to characterise the cognitive markers of prodromal AD pathology in ε4-carriers require participants in whom prodromal AD is likely to be present (i.e., over 60s), as well as longitudinal designs to measure clinical outcomes, with sufficient followup period to allow for conversion to AD. The cognitive domains tested should be focused on those affected in early AD to minimise multiple comparisons. Such studies may be able to utilise existing data, for example through initiatives such as the MRC Dementias Platform UK (DPUK; www. dementiasplatform.uk), European Medical Information Framework e Alzheimer's Disease (EMIF; http://www.emif. eu/about/emif-ad) and Alzheimer's Disease Neuroimaging Initiative (ADNI, http://adni.loni.usc.edu/), which pool data from many existing ageing cohorts both locally and internationally.

Conflicts of interest
None.