In the face of an aging society, an urgent need has emerged to prevent and treat dementia, in general, and Alzheimer’s disease (AD), in particular (Ferri et al., 2006). Despite wide-ranging research efforts to elucidate the putative pathological mechanisms of dementia, available treatments remain limited (Alzheimer’s Association, 2012). Promisingly, however, various factors have been shown to boost the dynamics of relevant neural networks, delaying and even reducing clinical symptoms (Fregni and Pascual-Leone, 2007). Therefore, interventions that enhance neuroplastic responses in early disease stages might be particularly useful to attenuate the rate of cognitive decline in AD patients.

A key model to assess this possibility is afforded by patients with mild cognitive impairment (MCI). This construct has evolved over the past two decades to denote a cognitive state intermediate between normal aging and very early dementia (Petersen and Negash, 2008). Patients with this condition feature objective memory impairment and other cognitive deficits, but their dysfunction is not so severe as to compromise daily activities (Petersen and Negash, 2008). MCI comprises four clinical subtypes: (i) single-domain amnestic MCI (aMCI); (ii) multiple-domain aMCI; (iii) single-domain non-amnestic MCI; and (iv) multiple-domain non-amnestic MCI (Winblad et al., 2004; Anstey et al., 2008; Albert et al., 2011). These subtypes differ in etiology and outcome: whereas both forms of aMCI imply a high chance of progression to AD, non-amnestic varieties involve considerable risk of conversion to non-AD dementia (Winblad et al., 2004; Anstey et al., 2008; Albert et al., 2011).

A second relevant model can be found in individuals featuring subjective cognitive impairment (SCI). This notion encompasses people who report everyday concerns about their cognitive functioning, even in the absence of objectively established impairments. Given its association with subsequent cognitive decline, SCI represents a potential prodromal state of AD occurring even before MCI (Reisberg and Gauthier, 2008).

In sum, aMCI and SCI are likely to anticipate incipient dementia. Despite the high between-patient variability and the uncertainty about their causative physiopathological mechanisms, both conditions offer highly relevant models to implement interventions aimed to boost cognitive performance through modulations of neuronal activity and connectivity. In this sense, non-invasive brain stimulation (NIBS) techniques have emerged as a potentially useful alternative. These tools can selectively enhance adaptive activity patterns, suppress maladaptive patterns, and influence learning and skill acquisition processes by inducing synaptic plasticity and network reorganization (Hummel and Cohen, 2006; Nitsche et al., 2008). In particular, NIBS can have a direct impact on memory mechanisms (including working, episodic, and associative memory) in young adults, elderly adults and patients with neurological dysfunctions (Manenti et al., 2012; Elder and Taylor, 2014).

Building on these antecedentes, a number of studies have assessed the impact of NIBS on MCI patients. Here we aim to critically review such emerging literature. While other reviews have addressed the impact of NIBS in cognitively impaired patients (Freitas et al., 2011; Nardone et al., 2014), here we focus on the key questions and difficulties of setting a treatment protocol. To this end, first we characterize the main features of NIBS. Thereupon, we review available studies examining the impact of NIBS on learning and memory in MCI and SCI. In particular, we describe and discuss in detail the studies’ protocols and parameters, highlight the advantages of combining NIBS with cognitive training. Finally, we outline methodological strategies that could promote major developments in this promising research area.

With NIBS, researchers can transiently influence behavior by altering neural activity (Miniussi et al., 2013). A considerable body of evidence has been forged through the two most commonly used techniques in humans: repetitive transcranial magnetic stimulation (rTMS) and transcranial direct current stimulation (tDCS). Although they differ in several aspects (Table 1), both tools can induce long-term after effects on cortical excitability and neuronal plasticity. Typically, this brings about facilitatory or inhibitory effects (Nitsche and Paulus, 2000), broadly mirroring the workings of long-term potentiation (LTP) and long-term depression (LTD; Dayan et al., 2013). Since such effects can considerably outlast the actual stimulation period, NIBS techniques open valuable avenues for rehabilitation, especially when stimulation protocols are combined with adequate training-based interventions (Zimerman et al., 2013).

TMS has been recently approved in some countries to treat depression and other disorders, subject to approval by local ethical review boards (Rossi et al., 2009). These reservations largely follow from the suboptimal understanding of the putative mechanisms underlying NIBS-related effects. However, a number of contributing factors have been recently identified. A critical one concerns modulation of neurotransmitter levels (Ridding and Ziemann, 2010). Transcranial stimulation may alter the dynamics of excitatory/inhibitory neurotransmitter systems (e.g., GABA and glutamate) affecting the regulation of cortical neuronal activity. For example, rTMS effects decrease if the subjects consume a drug that interferes with NMDA receptors (Reis et al., 2006). Likewise, post-tDCS effects can be blocked by an NMDA-antagonist, whereas a partial NMDA-agonist (d-cycloserine) can selectively extend the duration of motor cortical excitability induced by anodal tDCS (Nitsche et al., 2004). Moreover, MR spectroscopy research (Stagg et al., 2009) shows that anodal tDCS decreases GABAergic transmission, while cathodal tDCS yields similar effects on glutamate concentrations over the motor cortex (MC).

NIBS may also exert long-lasting plastic changes through gene induction. Research based on rTMS protocols (Fritsch et al., 2010) has shown significant enhancements of brain-derived neurotrophic factor (BDNF) mRNA in the hippocampal, parietal, and piriform cortices. For instance, LTP-like effects in adult mice are abolished for specimens with deletion of the BDNF gene, suggesting that this factor could be a critical mediator of direct stimulation effects (Fritsch et al., 2010). Importantly, as a member of the neurotrophin family, BDNF plays an important role in synaptogenesis and synaptic mechanisms related to learning and memory processes.

Additionally, TMS has been successfully used in AD and MCI to track and modulate abnormal mechanisms, including glutamatergic dysfunctions, disruption of synaptic plasticity, and inhibition of LTP (Nardone et al., 2014). For example, TMS research consistently shows that such patients feature a decreased resting motor threshold, which has been interpreted as the electrophysiological correlate of cortical glutamatergic dysfunctions and as a marker of reduced cortical hyperexcitability in AD (Di Lazzaro et al., 2003, 2004; Inghilleri et al., 2006; Trebbastoni et al., 2016). Furthermore, rTMS reduces facilitation of the motor-evoked potential (MEP) in aMCI patients (Trebbastoni et al., 2016), revealing an initial impairment of the mechanisms that underlie glutamate-induced synaptic potentiation (Trebbastoni et al., 2016). On the other hand, a TMS-MRI study indicates that hyperexcitability of the sensorimotor cortex might represent a protective mechanism counteracting the prominent loss of cortical volume in AD and MCI. However, this supposed protective mechanism was found neither on the precuneus or cuneus of AD patients, nor in an MCI group (Niskanen et al., 2011). Furthermore, an assessment of navigated TMS-evoked EEG responses in AD patients showed prominent changes in functional connectivity and reactivity (Julkunen et al., 2008, 2011). In particular, TMS-evoked responses at 30–50 ms decreased significantly in AD over widespread brain regions, suggesting dysfunctions of a large-scale sensorimotor network. Compatibly, a TMS-EEG co-registration study in AD patients offered direct evidence of MC hyperexcitability, extending to the whole sensorimotor system (Ferreri et al., 2016). Although speculative, these changes could be interpreted as a compensatory mechanism allowing for the preservation of sensorimotor programming and execution over a long period, regardless of the disease’s progression. Additional evidence on cortical plasticity in relevant patient samples comes from paired associative stimulation (PAS) and cortical responses to rTMS. Null PAS effects in MCI patients have been reported by Lahr et al. (2016). However, this result should be considered with caution, since the study mixed patients with single- and multiple-domain aMCI which may have clouded effects on a specific subsample. Indeed, previous evidence from Nardone et al. (2012) demonstrated that short-latency afferent inhibition is significantly reduced in patients with multiple-domain amnestic MCI relative to controls, but not in patients with single-domain amnestic MCI or with non-amnestic MCI. Taken together, this incipient evidence suggests that NIBS can directly impinge on the pathological mechanisms underlying some forms of AD and MCI.

Finally, note that the effects of rTMS and tDCS are not restricted to the stimulation site. Depending on stimulation parameters, they can induce changes at distant points through effective modulations of remote, interconnected networks (Plewnia et al., 2003; Hummel and Cohen, 2006). This particular property of NIBS techniques has been assessed in proof-of-principle studies aimed to enhance and decrease excitability of critical brain regions involved in the recovery process (for details see Hummel and Cohen, 2006; Nowak et al., 2010; Schulz et al., 2013). All these tenets illuminate the evidence gathered so far in NIBS studies on MCI patients, which we review below.

Growing evidence indicates that NIBS techniques can significantly modulate various cognitive processes in both elderly neurotypicals and patients with AD. While the impact of NIBS on the latter population has been widely discussed (Hsu et al., 2015), much less attention has been paid to NIBS research on MCI and SCI. Below we offer a joint assessment of this empirical corpus, discussing available results in terms of stimulated brain regions. We deal with TMS studies first, and then consider those employing tDCS (for full details of each study see Table 2).

So far, NIBS studies seeking to improve cognitive function in MCI/SCI patients have yielded inconclusive findings. Although some studies showed favorable results, statistically significant differences in an experimental setting do not necessarily imply clinical applicability. Consequently, the evidence gathered so far, though promising, must be considered extremely preliminary. No definite answer can yet be provided for the ultimate question underlying the field: can NIBS-induced cognitive improvements persist beyond treatment and translate into daily-life benefits? The methodological considerations listed below emerge as critical milestones to properly address this issue.

The present review aimed to summarize research on the effects of NIBS in MCI and to outline the approaches that may improve the efficacy of stimulation protocols. Despite major advances over the past few decades, extant MCI/AD treatments are not completely effective. Although NIBS have not yet been massively applied in SCI/MCI patients, available results hold considerable promise. First, TMS and tDCS are well-suited to explore brain plasticity across the lifespan (Heise et al., 2014). Second, they provide valuable physiologic biomarkers of the state of cortical reactivity, brain network connectivity, dynamics, and mechanisms of brain plasticity. Third, NIBS techniques avoid systemic side-effects and allow for temporally and spatially precise neural interventions that escape the possibilities of pharmacological or complementary therapies. Moreover, discrepancies among available studies may be largely due to differences among tasks, designs, and samples, as opposed to limitations inherent to NIBS. However, many issues remain to be solved and large longitudinal studies are needed to more stringently assess the robustness of available positive results. Furthermore, even if NIBS techniques proved irrelevant for clinical purposes, they could remain highly useful for experimental research. While no definitive conclusions on the efficacy and specificity of NIBS can be advanced at this stage, our review delineates the major challenges to be faced in future studies. In this sense, pre-post neuroimaging designs could reveal putative mechanisms underlying NIBS effects. Also, concurrent use of NIBS techniques with specific tasks targeting impaired domains in individual patients could potentially enhance LTP-like after-effects. In particular, applying NIBS over a dysfunctional neural network while the latter is engaged by a particular behavioral activity could significantly enhance learning- and memory-related effects. Furthermore, the application of NIBS across multiple sessions could render such effects longer lasting. All these considerations should be accompanied by very precise diagnosis of each patient’s MCI subtype, so that subject-specific protocols can be designed and assessed with the implementation of normative databases. Moreover, further studies are required to identify the optimal stimulation parameters yielding robust outcomes for different MCI subtypes as well as early- and late-onset AD. Finally, we emphasize the need to conduct longer longitudinal studies to establish the duration of NIBS-induced benefits. By contemplating these issues, future NIBS studies could shed unprecedented light on functional restoration options for MCI patients, paving the way for new treatments aimed to delay (or event prevent) the development of AD and other dementias.

AB did the literature research and wrote the manuscript. AI and AMG provided insightful thoughts to study concepts, and revised the manuscript. LS and JF assisted with the literature review and revised the final version. MZ wrote, revised critically and approved the final version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.