Cortical Excitability and Plasticity in Alzheimer’s Disease and Mild Cognitive Impairment: A Systematic Review and Meta-analysis of Transcranial Magnetic Stimulation Studies

Abstract

Background:

Transcranial magnetic stimulation (TMS) is a non-invasive neuromodulation technique. When stimulation is applied over the primary motor cortex and coupled with electromyography measures, TMS can probe functions of cortical excitability and plasticity in vivo. The purpose of this meta-analysis is to evaluate the utility of TMS-derived measures for differentiating patients with Alzheimer’s disease (AD) and mild cognitive impairment (MCI) from cognitively normal older adults (CN).

Methods:

Databases searched included PubMed, Embase, APA PsycInfo, Medline, and CINAHL Plus from inception to July 2021.

Results:

Sixty-one studies with a total of 2,728 participants (1,454 patients with AD, 163 patients with MCI, and 1,111 CN) were included. Patients with AD showed significantly higher cortical excitability, lower cortical inhibition, and impaired cortical plasticity compared to the CN cohorts. Patients with MCI exhibited increased cortical excitability and reduced plasticity compared to the CN cohort. Additionally, lower cognitive performance was significantly associated with higher cortical excitability and lower inhibition. No seizure events due to TMS were reported, and the mild adverse response rate is approximately 3/1000 (i.e., 9/2728).

Conclusions:

Findings of our meta-analysis demonstrate the potential of using TMS-derived cortical excitability and plasticity measures as diagnostic biomarkers and therapeutic targets for AD and MCI.

1. INTORDUCTION

Alzheimer’s disease (AD) is a devastating neurodegenerative disorder that causes a continuous decline in memory, thinking and behavioral skills which ultimately disrupts a person’s ability to function independently. The disease was traditionally defined by and diagnosed according to this relatively heterogenous clinical phenotype. During the past few years, however, there has been a shift to define AD biologically by neuropathological changes or biomarkers, including ß-amyloid deposition (A), pathologic tau (T), and neurodegeneration (N) based on the 2018 NIA-AA Research Framework¹. In this new NIA-AA framework, the hallmark cognitive symptoms of the disease do not factor into the initial diagnosis of AD, but are instead incorporated for staging the clinical severity of AD.

This paradigm shift towards a biological definition of AD coincides with recent advances in research indicating that the pathological onset of the disease can precede the clinical manifestation of AD by many years². This extended preclinical phase may hold particular significance for disease modification, as potential therapies are likely to be most effective in the early, asymptomatic stages of AD. As such, it is important to identify useful and reliable biomarkers that can be used as leading indicators of disease to identify potential cases of preclinical AD before the neurodegenerative disease becomes medically refractory. Therefore, novel biomarkers that can be used to characterize the complementary pathophysiological features within the amyloid-tau-neurodegeneration (ATN) Research Framework¹ are urgently needed.

Transcranial magnetic stimulation (TMS) is a versatile non-invasive brain stimulation tool that may have utility in this respect. TMS applies electromagnetic pulses to the brain with a coil that is carefully placed on the surface of the scalp over a targeted stimulation site³. TMS leverages Faraday’s law of electromagnetic induction whereby a rapidly alternating magnetic field induces an electric current in an adjacent conductive medium – in this case, the cortical tissue. The pulse of generated magnetic field induces a secondary electric current which can trigger action potentials in axons beneath the TMS coil, which propagate to connected cortical neurons³. For example, TMS applied over the primary motor cortex (M1) generates descending volleys that elicit an electromyogenic response in corresponding distal muscles that are referred to as motor evoked potentials (MEPs). When coupled with electromyography (EMG), the TMS-derived MEPs can be used to characterize features of cortical excitability and neural plasticity in vivo^4–6. TMS-derived measures denoting excitatory and inhibitory properties of neurotransmitter systems have been substantially supported by numerous pharmaco-TMS studies^4–6. These various measures of cortical excitability, which are associated with distinct neurophysiological underpinnings, can be obtained via the application of various single-pulse or paired-pulse TMS protocols^6,7. For example, the short latency afferent inhibition (SAI) derived from the paired-pulse TMS paradigm likely involves central cholinergic and GABAergic transmissions⁸ and was found to be altered in AD^9–14. Alternatively, applying rhythmic trains of TMS pulses in a rapid sequence is called repetitive TMS (rTMS), which is known to induce transient after-effects resembling either long-term potentiation (LTP) or long-term depression (LTD) depending on the rTMS protocols applied¹⁵. TMS can be leveraged to non-invasively probe neural plasticity in the human brain by interleaving MEP measures with an rTMS paradigm, whereby the rTMS induced change in MEPs is indicative of transient neural reorganization that reflects LTP- or LTD-like effects.

These TMS-derived measures of cortical excitability and plasticity have theoretical relevance and great potential as novel diagnostic biomarkers along the spectrum of AD. In this systematic review and meta-analysis, we aim to evaluate the capacity of TMS-derived biomarker candidates for differentiating patients with AD and mild cognitive impairment (MCI) from cognitive normal older adults (CN). To this end, we quantitatively examine evidence of cortical excitability and plasticity from TMS studies in AD and MCI, and, furthermore, attempt to provide a more rigorous assessment of the direction and magnitude of the association between cortical excitability, neural plasticity, and cognitive performance in these populations.

2. METHODS

2.1. Search Strategy

Our meta-analysis was conducted in accordance with the preferred reporting items for systematic reviews and meta-analyses (PRISMA) statement¹⁶ (Figure 1), and it is registered with PROSPERO (registration number CRD42020187336). To identify studies for inclusion in this meta-analysis we searched PubMed, Embase, and EBSCOhost (including APA PsycInfo, Medline, and CINAHL Plus), through July 2021. Databases were searched using combinations of the following terms: [Alzheimer or Alzheimers or mild cognitive impairment] and [transcranial magnetic stimulation or TMS] and [excitability or plasticity]. Additionally, we searched reference lists of previous reviews on TMS and cortical excitability for AD/MCI^17–23 to identify additional relevant articles.

Figure 1.

Flow diagram showing the search and selection procedure that was used for this meta-analysis. Diagram adapted from Moher et al. (2009). Abbreviations: AD = Alzheimer’s disease; MCI = mild cognitive impairment; TMS = transcranial magnetic stimulation

2.2. Inclusion Criteria for the Selection of Studies

We included studies that met all the following criteria: 1) TMS studies; 2) reporting quantitative data on motor threshold, afferent inhibition, intracortical inhibition, motor-evoked potential, cortical silent period, intracortical facilitation, and/or central motor conduction time; 3) studies of a cross-sectional design comparing the above measures of cortical excitability and plasticity in AD and/or MCI patients with cognitively normal older adults; and 4) articles written in English. Studies identified through database searches were initially screened on the basis of their title and abstract. They were subsequently excluded if it was clear from the title or abstract that the study was not relevant or did not meet the inclusion criteria. If it remained unclear, the paper was assessed in its entirety. Additionally, studies were excluded if they were conference abstracts/papers.

2.3. Data Extraction

Three authors (YHC, VTT, and JG) independently performed the data extraction, and any disagreements were resolved by joint discussion. Extracted data included sample size, sample characteristics, TMS protocol, statistical data of the cortical excitability and plasticity for effect size estimation, and potential adverse response. When published data were insufficient for data analysis, the original study author(s) were contacted with requests for access to additional data.

2.4. Statistical Analysis

2.4.1. Effect Size Calculation

We used standardized mean difference (SMD, also known as Cohen’s d) to express the effect size of differences in cortical excitability and plasticity of AD or MCI versus cognitively normal older adults. A random-effects model was used to calculate pooled effect sizes and examine whether the averaged effect size was significantly different from zero (p < 0.05, two-tailed). The mean effect was expressed as SMD with 95% confidence intervals. Generally, one effect size was derived from each study. If a study had multiple effect sizes from the same participant group (e.g., different outcome measures) or if the effect sizes were reported from AD and MCI separately within a single study, we obtained one averaged effect size across multiple effect sizes within that study. The specific effects of AD and MCI on various cortical excitability and plasticity measures were further investigated with subgroup analyses (please refer to the 2.4.4. Subgroup Analyses below).

2.4.2. Heterogeneity analysis

We used the Q statistic and the I² index to assess heterogeneity. A probability value less than 0.05 and I² greater than 50% is indicative of heterogeneity between included studies as it exceeds what is expected by chance²⁴.

2.4.3. Publication bias

Publication bias was evaluated with Egger’s Test of Asymmetry²⁵, Duval and Tweedie’s trim and fill²⁶, as well as Orwin’s fail-safe N approaches²⁷. In the absence of publication/selection bias, effect sizes are symmetrically distributed around the overall average effect size since the sampling error is random. Egger’s test evaluated whether the amount of asymmetry is significant. The “trim and fill” method is used to identify and adjust for publication asymmetry in meta-analysis. Additionally, studies that demonstrated lack of differences between groups might not have been submitted and/or accepted for publication. Therefore, we used the Orwin’s fail-safe N to estimate the number of missing studies that would need to be incorporated in our meta-analysis to make the summary effect become trivial.

2.4.4. Subgroup Analyses

Our pre-specified categories for subgroup analyses included patient population (AD vs. CN and MCI vs. CN) and outcome measures (resting motor threshold, active motor threshold, short-latency afferent inhibition, short-latency intracortical inhibition, LTP-like plasticity, LTD-like plasticity, cortical silent period, intracortical facilitation, and central motor conduction time). We would investigate whether group differences on cortical excitability and plasticity depend on severity of cognitive impairment (i.e., AD and MCI) and outcome measures.

2.5. Risk of Bias Assessment in Individual Studies

The risk of bias was assessed using a modified Newcastle-Ottawa Scale (NOS)²⁸ for the non-randomized studies included in the meta-analysis (Supplementary Table 1). Specifically, studies were assessed using a quality score in the following 3 domains: selection of the study groups (maximum 4 points), comparability of the groups (maximum 2 points), and ascertainment of the exposure (maximum 2 points). A higher score indicated better quality. We considered an NOS score of 8 to represent an excellent-quality study, a score of 7 and 6 a good-quality study, a score of 5 a fair-quality study, and a score of ≤ 4 a low-quality study.

3. RESULTS

3.1. Search Results

Our initial search of all databases retrieved 251 studies (Figure 1). After rejecting articles based on the contents of the title and abstract, the full texts of 144 articles were obtained for further examination. Of these, 83 studies were excluded. The remaining 61 studies that met the inclusion criteria were included for this meta-analysis^{9–14,29–83}.

3.2. Study Characteristics

The 61 studies included 2,728 participants (mean age ≅ 69.4 years; ~54% women). Among those participants, 1,454 are patients with AD, 163 are individuals with MCI, and 1,111 are cognitively normal older adults (CN). Overall, the patient group (including both AD and MCI) showed a significant difference across the TMS-derived neurophysiological measures (effect size d = 0.93, p < .0001, number of studies = 61) compared to the CN cohorts. The main characteristics of the included studies are described in Tables 1 and 2, and the distribution of effect sizes is illustrated in Figure 1A. Heterogeneity between the included studies exceeded that expected by chance (Q = 220.59, df(Q) = 60, p < .0001, I² = 72.80), suggesting that the results across the included studies were heterogeneous and subgroup analyses would be needed (please see Section 3.3).

Table 1.

Characteristics of Included Studies: Participants

Study Name	Sample Size: AD/MCI/CN	Mean Age: AD or MCI/CN	%Female: AD or MCI/CN
De Carvalho et al., 1997	14/0/11	68/66	71/73
Pepin et al., 1999	17/0/22	71/69	76/73
Alagona et al., 2001	21/0/18	55-81/52-88a	57/28
Lieport et al., 2001	11/0/10	75/70	82/70
Valls-Sole et al., 2001	8/0/10	54-69/38-83a	NA
Ferreri et al., 2003	16/0/13	75/72	94/62
Alagona et al., 2004	20/0/20	72/69	65/60
Di Lazzaro et al., 2004	28/0/12	72/73	NA
Pierantozzi et al., 2004	12/0/12	65/65	NA
Di Lazzaro et al., 2006	20/0/20	70/72	50/NA
Inghilleri et al., 2006	20/0/20	71/69	NA
Nardone et al., 2006	13/0/15	70/68	46/47
Battaglia et al., 2007	10/0/10	70/68	40/40
Di Lazzaro et al., 2007	10/0/10	72/72	40/NA
Sakuma et al., 2007	12/16/15	NA	NA
Di Lazzaro et al., 2008	12/0/12	69/73	NA
Julkunen et al., 2008	5/5/4	73/74/78b	40/40/75b
Martorana et al., 2008	11/0/12	73/68	NA
Nardone et al., 2008	17/0/22	68/70	41/45
Martorana et al., 2009	10/0/10	73/72	NA
Olazaran et al., 2010	11/0/12	77/77	55/50
Casarotto et al., 2011	9/0/9	72/72	56/56
Khedr et al., 2011	45/0/37	68/66	64/65
Koch et al., 2011a	20/0/10	71/NA	50/NA
Koch et al., 2011b	10/0/10	73/72	50/NA
Niskanen et al., 2011	15/18/21	74/72/72b	67/50/52b
Hoeppner et al., 2012	19/0/19	72/69	63/53
Koch et al., 2012	14/0/14	NA	36/NA
Marra et al., 2012	18/0/10	72/72	39/50
Martorana et al., 2012	19/0/10	68/72	68/NA
Trebbastoni et al., 2012	11/0/11	78/75	73/64
Bonni et al., 2013	15/0/12	75/NA	NA
Brem et al., 2013	16/0/13	70/68	69/54
Di Lorenzo et al., 2013	12/0/12	70/72	NA
Martorana et al., 2013	17/0/8	69/72	NA
Terranova et al., 2013	10/0/14	80/77	40/36
Balla et al., 2014	13/0/13	73/72	54/46
Nardone et al., 2014	8/0/8	73/73	NA
Yang et al., 2014	40/0/45	66/67	58/56
Chandra et al., 2016	17/0/17	62/NA	36/NA
Di Lorenzo et al., 2016	54/0/24	68/66	48/50
Ferreri et al., 2016	12/0/12	72/69	58/50
Koch et al., 2016	40/0/24	71/69	43/50
Lahr et al., 2016	0/24/24	74/69	54/67
Trebbastoni et al., 2016	0/35/20	74/71	37/55
Benussi et al., 2017	79/0/32	71/62	49/56
Koch et al., 2017	41/0/20	71/70	51/50
Kumar et al., 2017	32/0/16	76/76	53/50
Benussi et al., 2018	63/0/39	72/69	51/67
Di Lorenzo et al., 2018	15/0/10	70/71	53/50
Motta et al., 2018	60/0/30	68/66	46/54
Yildiz et al., 2018	5/0/9	81/53	40/44
Di Lorenzo et al., 2019	75/0/0	NA	NA
Minkova et al., 2019	0/19/22	72/70	47/36
Benussi et al., 2020	273/0/147	71/58	51/58
Brem et al., 2020	34/0/13	69/66	59/54
Buss et al., 2020	0/17/10	70/67	47/60
Di Lorenzo et al., 2020	15/0/12	70/71	47/50
Khedr et al., 2020	15/0/25	66/61	60/48
Colella et al., 2021	0/14/16	75/71	29/50
Meder et al., 2021	15/15/23	73/69/67b	33/27/36b

Note.

^aage range;

^bAD/MCI/CN;

AD = patients with Alzheimer’s disease; CN = cognitively normal older adults; MCI = patients with mild cognitive impairment; NA = data not available.

Table 2.

Characteristics of Included Studies: TMS-Derived Neurophysiological Measures Included in the Meta-Analysis

Study Name	RMT	AMT	SAI	SICI	Plasticity	CSP	ICF	CMCT
de Carvalho et al., 1997	+							+
Pepin et al., 1999	+	+		+
Alagona et al., 2001	+
Lieppert et al., 2001	+			+		+
Valls-Sole et al., 2001	+
Ferreri et al., 2003	+
Alagona et al., 2004	+					+
Di Lazzaro et al., 2004	+		+	+
Pierantozzi et al., 2004	+	+		+
Di Lazzaro et al., 2006	+		+
Inghilleri et al., 2006	+				+	+
Nardone et al., 2006	+	+	+	+			+	+
Battaglia et al., 2007	+				+
Di Lazzaro et al., 2007	+	+	+	+
Sakuma et al., 2007	+		+
Di Lazzaro et al., 2008	+	+	+	+
Julkunen et al., 2008	+
Martorana et al., 2008	+	+		+
Nardone et al., 2008	+	+	+	+			+	+
Martorana et al., 2009	+		+
Olazaran et al., 2010	+
Casarotto et al., 2011	+
Khedr et al., 2011	+	+				+
Koch et al., 2011a					+
Koch et al., 2011b	+
Niskanen et al., 2011	+
Hoeppner et al., 2011	+			+		+	+
Koch et al., 2012		+			+
Marra et al., 2012			+
Martorana et al., 2012	+		+
Trebbastoni et al., 2012	+				+
Bonni et al., 2013	+
Brem et al., 2013	+	+		+	+		+
Di Lorenzo et al., 2013	+	+	+	+			+
Martorana et al., 2013	+
Terranova et al., 2013	+				+
Balla et al., 2014		+				+
Nardone et al., 2014	+		+	+			+
Yang et al., 2014	+
Chandra et al., 2016	+					+		+
Di Lorenzo et al., 2016	+		+		+
Ferreri et al., 2016	+
Koch et al., 2016	+		+		+
Lahr et al., 2016					+
Trebbastoni et al., 2016	+				+
Benussi et al., 2017	+		+
Koch et al., 2017	+				+
Kumar et al., 2017	+
Benussi et al., 2018	+		+	+			+
Di Lorenzo et al., 2018					+
Motta et al., 2018	+		+	+	+		+
Yildiz et al., 2018			+
Di Lorenzo et al., 2019					+
Minkova et al., 2019					+
Benussi et al., 2020	+		+	+			+
Brem et al., 2020	+	+		+	+		+
Di Lorenzo et al., 2020	+	+
Khedr et al., 2020	+	+				+
Colella et al., 2021	+	+			+
Meder et al., 2021	+

Note. + = TMS-derived physiological measures included in the meta-analysis; AMT = active motor threshold; CMCT = central motor conduction time; CSP = cortical silent period; ICF = intracortical facilitation; Plasticity = long-term-potentiation-like and long-term-depression-like plasticity; RMT = resting motor threshold; SAI = short latency afferent inhibition; SICI = short latency intracortical inhibition.

3.3. Subgroup Analyses

3.3.1. Patient population

Fifty-five studies included patients with AD and 10 studies included patients with MCI. Among them, 4 studies composed of both patients with AD and MCI (Table 1). Our subgroup analysis revealed a significant difference across the TMS-derived neurophysiological measures between AD and CN cohorts (effect size d = 0.99, p < .0001; Figure 1B) as well as between MCI and CN (effect size d = 0.37, p < .05; Figure 1C).

3.3.2. TMS-derived neurophysiological measures

The TMS-derived neurophysiological measures include cortical excitability (i.e., resting motor threshold, active motor threshold, short-latency intracortical inhibition, cortical silent period, intracortical facilitation, short-latency afferent inhibition, and central motor conduction time) and cortical plasticity (i.e., LTP-like and LTD-like plasticity). Subgroup analyses of each individual measure are reported below and summarized in Tables 2 and 3.

Table 3.

Results of Subgroup Analyses on TMS-Derived Neurophysiological Measures

	Patients with Alzheimer’s disease vs. Cognitively normal older adults		Patients with mild cognitive impairment vs. Cognitively normal older adults
Resting motor threshold (RMT)	↓	n = 48	↓	n = 8
Active motor threshold (AMT)	↓	n = 15	Non-significant	n = 2
Short latency afferent inhibition (SAI)	↓	n = 19	Non-significant	n = 1
Short latency intracortical inhibition (SICI)	↓	n = 17	No data available	n = 0
LTP-like plasticity	↓	n = 13	↓	n = 5
LTD-like plasticity	Non-significant	n = 6	No data available	n = 0
Cortical silent period	Non-significant	n = 8	No data available	n = 0
Intracortical facilitation (ICF)	Non-significant	n = 10	No data available	n = 0
Central motor conduction time (CMCT)	Non-significant	n = 4	No data available	n = 0

Note. n = number of studies; LTD = long-term depression; LTP = long-term potentiation; ↓ indicates decreased value of TMS-derived neurophysiological measures in patients with Alzheimer’s disease or mild cognitive impairment

3.3.2.1. Resting motor threshold (RMT).

RMT refers to the minimum TMS intensity required to evoke motor-evoked potentials (MEPs) with peak-to-peak amplitude of ≥ 50 μV in 50% of trials when ten consecutive single TMS pulses are applied over the primary motor cortex (i.e., hot spot of the target muscle at rest)⁸⁴. Fifty-two studies included in this meta-analysis reported the RMT data. Our subgroup analysis showed that patients with AD and patients with MCI exhibited significantly lower RMT compared to the CN (AD: number of studies = 48, effect size d = 1.05, p < .0001; MCI: number of studies = 8, effect size d = 0.39, p < .005), suggesting hyper-excitability in patients with AD and MCI compared to cognitively normal older adults (Table 3).

3.3.2.2. Active motor threshold (AMT).

AMT is defined as the minimum TMS intensity needed to produce MEPs with peak-to-peak amplitude of 200 μV in 50% of trials while participants maintain a voluntary contraction of the target muscle. Subgroup analysis revealed that patients with AD showed significantly lower AMT compared to the CN (number of studies = 15, effect size d = 0.77, p < .0001), suggesting hyper-excitability in AD (Table 3). The difference in AMT was not significant between MCI and CN (number of studies = 2, effect size d = 0.33, p > .05).

3.3.2.3. Short latency afferent inhibition (SAI).

In this paired-pulse TMS paradigm, the TMS-elicited amplitude of MEPs is typically reduced by adding non-invasive peripheral median nerve stimulation 19-50 ms before the TMS over the primary motor cortex (i.e., hot spot of the target muscle)⁸⁵. The degree of inhibition is referred to as SAI and, as mentioned in the introduction, it is a neurophysiological measure that can reveal cholinergic function in the central nervous system in vivo (i.e., acetylcholine facilitates the reduced MEP amplitude in this protocol)⁸⁶. Subgroup analysis indicated that patients with AD exhibited significantly reduced SAI relative to the CN (number of studies = 19, effect size d = 1.89, p < .0001), suggesting reduced inhibition and impaired cholinergic function in AD (Table 3). The difference in SAI was not significant between MCI and CN (number of studies = 1, effect size d = 0.07, p > .05).

3.3.2.4. Short latency intracortical inhibition (SICI).

In this paired-pulse TMS protocol, an initial subthreshold (conditioning) pulse and a subsequent suprathreshold (test) pulse are delivered at short inter-stimulus intervals of 1-6 ms through the same TMS coil^87–90. The conditioning TMS pulse leads to inhibition of the test amplitudes of MEPs, and this MEP reduction is a potential surrogate marker of GABA_A receptor-mediated postsynaptic inhibition of corticospinal neurons⁸⁹. Subgroup analysis showed that patients with AD exhibited significantly reduced SICI relative to the CN (number of studies = 17, effect size d = 0.68, p < .01), suggesting reduced inhibition associated with GABA_A in AD (Table 3). No SICI data were available in MCI.

3.3.2.5. LTP-like and LTD-like plasticity.

The LTP- and LTD-like plasticity of the primary motor cortex can typically be enhanced by applying excitatory and inhibitory rTMS, respectively. The changes in MEPs following rTMS are usually used to denote integrity of cortical plasticity. Our subgroup analysis showed that, for the LTP-like plasticity, both patients with AD and MCI exhibited significantly reduced response to excitatory rTMS compared to the CN (AD: number of studies = 13, effect size d = 1.20, p < .0001; MCI: number of studies = 5, effect size d = 0.86, p < .05; Table 3). For the LTD-like plasticity, AD and CN did not show significant differences in response to inhibitory rTMS (number of studies = 6, effect size d = 0.28, p > .05). No LTD-like plasticity data were available in MCI. These findings suggest impaired LTP-like plasticity in both AD and MCI.

3.3.2.6. Cortical silent period, intracortical facilitation, and central motor conduction time.

Our subgroup analyses did not show significant differences in the following outcome measures between AD and CN: cortical silent period (number of studies = 8, effect size d = 0.67, p > .05), intracortical facilitation (number of studies = 10, effect size d = 0.20, p > .05), and central motor conduction time (number of studies = 4, effect size d = 0.05, p > .05). No data were available in MCI.

3.4. Correlations Between Cognitive Function and TMS-Derived Measures

Fourteen studies reported correlations between cognitive function and TMS-derived neurophysiological measures. Overall, individuals with worse cognitive performance showed higher cortical excitability as measured by RMT or AMT (number of studies = 6, effect size d = 0.81, p < .0001) and lower cortical inhibition as measured by CSP, SICI or SAI (number of studies = 7, effect size d = 0.88, p < .01). However, the correlations between cognitive function and cortical plasticity measures were not significant (number of studies = 4, effect size d = 0.53, p > .05).

3.5. Adverse Events

No incidents of seizure were reported in any of the included studies. Seven studies assessed the incidence of other minor adverse responses related to the application of TMS^{36,53,66,70,73,79,80}. Among them, 7 patients with AD and 2 cognitively normal older adults experienced a mild, but completely reversible nausea^66,73. The minor response did not interfere with the participants’ ability to complete their TMS protocols. The mild adverse response rate is approximately 3/1000 (i.e., 9/2728) in patients with AD or MCI and cognitively normal older adults.

3.6. Publication Bias

Publication bias was evaluated using the Egger’s Test of Asymmetry, Duval and Tweedie’s trim and fill as well as Orwin fail-safe N approaches. The Egger test did not reveal significant asymmetry across included studies for all outcome measures except for the RMT in AD. We applied the trim and fill method for the RMT in AD and the analysis showed that no studies should be excluded. The Orwin fail-safe N analysis revealed that 3,853 studies with a mean effect size of 0 would be needed to offset the conclusion that we are able to draw from the 52 studies included in this analysis of RMT in AD (i.e., to bring p-value greater than 0.05). Similarly, other significant group differences identified in this meta-analysis would require at least 25 unpublished studies to bring the p-values > 0.05 (i.e., 2,291 studies for SAI in AD; 416 studies for LTP in AD; 217 studies for AMT in AD; 164 studies for SICI in AD; and 25 studies for LTP in MCI). The analysis of publication bias suggests that our findings are robust, and that the probability of potential publication bias is very low.

3.7. Risk-of-Bias Assessment in Individual Studies

Among all the included studies, 6 studies are of excellent methodologic quality (NOS score 8 of 8), 52 studies are of good methodologic quality (NOS score = 6 or 7), and 2 studies are of fair methodologic quality (NOS score = 5). Overall, our quantitative analysis indicated that most of the included studies exhibited low risk of bias. The assessment of risk of bias for all included studies is summarized in the Supplementary Table 1.

4. DISCUSSION

This meta-analysis, which included 61 studies with 2,728 participants, found that M1-TMS can be an effective tool to characterize and detect pathophysiological alterations along the spectrum of AD. Specifically, we report that patients with AD showed higher cortical excitability (i.e., lower resting and active motor thresholds), lower cortical inhibition (i.e., lower short latency afferent and intracortical inhibitions), and impaired cortical plasticity (i.e., decreased LTP-like plasticity) compared to the CN cohort. This meta-analysis also showed increased cortical excitability (i.e., lower resting motor threshold) and reduced cortical plasticity (i.e., decreased LTP-like plasticity) in patients with MCI compared to the CN cohorts. Furthermore, when reported, lower cognitive performance was significantly associated with higher cortical excitability and lower cortical inhibition.

4.1. Hyperexcitability in mild cognitive impairment and Alzheimer’s disease

The findings of our meta-analysis suggest that patients with MCI and AD exhibit significantly higher cortical excitability than cognitively normal older adults. Our findings are in line with network hyperexcitability and hypersynchrony reported from previous fMRI and EEG studies^91–94 and are in accordance with an increased incidence of seizure in AD^95–97. Findings of hyperexcitability have also been reported in MCI patients. For example, hippocampal hyperactivity has been reported in MCI patients during memory task-fMRI; and anti-epileptic medications like levetiracetam have been shown to both attenuate this task-related hyperactivity and improve memory performance^98,99. Similar task-fMRI findings have been reported in young asymptomatic adults who are carriers of the APOE e4 allele, which is the strongest known genetic risk factor for sporadic late-onset AD¹⁰⁰.

Notably, synaptic excitation and inhibition are inseparable events⁹⁹. Everything from the simplest activity generated by sensory stimuli to the most complex information transfer across neural networks requires a tight balance between excitatory and inhibitory drive. Glutamatergic neurons facilitate excitatory neurotransmission, while a dense network of GABAergic interneurons provides tight reciprocal control. Information processing in the brain relies on the precise temporal and spatial control of neural transmission throughout neural networks, and GABAergic inhibition enables this precision within and between brain regions¹⁰¹ Imbalances in excitatory/inhibitory drive can be catastrophic not only because of the direct effects of this network/circuit desynchrony on the intricate processes underlying complex human behavior, but also because of the pathophysiological consequences of neuronal hyperexcitability^102,103.

Several potential mechanisms might explain the hyperexcitability observed in MCI and AD. Though they are discussed separately below, it is important to emphasize the high degree to which these mechanisms are intertwined.

4.1.1. GABAergic Considerations

Cortical hyperexcitability might be due directly to the degeneration of inhibitory GABAergic interneurons^104–107. This is supported by significant differences between AD and CN cohorts in the SICI (section 3.3.2.4), which is a surrogate for GABAergic function reported in prior pharmaco-TMS studies¹⁰⁸.

GABAergic inhibitory interneurons have synaptic input onto multiple glutamatergic neurons to generate precise oscillatory rhythms that coordinate the timing of pyramidal cell firing¹⁰⁹. These specialized GABAergic interneurons, such as parvalbumin-positive cells, generate action potentials at high frequency and demand a large amount of energy utilization¹¹⁰. These highly energized fast-spiking GABAergic interneurons could be vulnerable to aging and AD when energy supply becomes compromised¹¹⁰. For example, in amyloid precursor protein (APP) 23 – presenilin (PS) 45 mice, hyperactivity of cortical neurons was found to be associated with decreased GABAergic inhibition¹¹¹. Additionally, reduced GABAergic terminals on the membrane surfaces of glutamatergic neurons proximal to Aβ plaques were observed in AD patients and APP-PS1 transgenic mice, indicating that the loss of GABAergic terminals may lead to the hyperactivity of the neurons in contact with Aβ plaques¹¹². Furthermore, GABA administration (e.g., benzodiazepines) has been shown to enhance inhibition, reduce hyperexcitability and improve memory function in several AD animal models^{98,107,111,113}. The above evidence together with increased cortical excitability and reduced cortical inhibition reported in our TMS meta-analysis support the hypothesis that cortical hyperexcitability in AD and MCI may be due to the degeneration of GABAergic interneurons that results in loss of control of glutamatergic neuron activity^104–107. These GABAergic neurons that degenerate early in AD and animal models of AD could result in neuronal network hyperexcitability and may partially explain the increased seizure incidence in AD¹⁰⁴.

4.1.2. Cholinergic Considerations

As a potent neuromodulator, acetylcholine (ACh) is known to exert a wide range of influence over neural signaling. Emerging experimental animal models provide strong evidence in support of ACh’s modulation of neuronal excitability, specifically through action on GABAergic activity. For example, one recent study reported the novel findings of co-transmission of ACh and GABA¹¹⁴. Specifically, the authors report that all cholinergic terminals in the hippocampus effectively establish new GABAergic synapses. These findings suggest that deterioration of the cholinergic system in the basal forebrain indirectly contributes to the GABAergic deficits that promote hyperexcitability in AD¹¹⁴. Similar work from a distinct experimental model reports consistent findings that asynchronous and reduced ACh release ultimately dampens GABAergic tone¹¹⁵. Taken together, this suggests that the well documented cholinergic deterioration in AD may be indirectly promoting states of hyperexcitability through a reduction in GABAergic tone.

4.1.3. Ion Channel Considerations

This hyperexcitability could also be derivative of dysfunctional ion channel function, which can produce aberrantly elevated membrane excitability. Decades of pharmaco-TMS research reliably support the utility of RMT as a surrogate marker for membrane excitability¹⁰⁸. From a series of pharmaco-TMS studies, RMT is consistently found to be modulated by drugs acting on voltage-gated cation channels but is unchanged in the presence of GABA-ergic (e.g., benzodiazepines), glutamatergic (e.g., NMDAR antagonists), or cholinergic (e.g., Achetylcholinesterase inhibitors) pharmaceutical agents. Therefore, the finding that RMT is reduced among AD patients in this meta-analysis is suggestive of increased membrane excitability.

There are a multitude of mechanisms that might contribute to this supposed increase in membrane excitability along the continuum of AD – many of which involve the amyloid and tau pathogenic proteins that formally characterize the disease. First, hydrophobic amyloid oligomers may insert into neuronal membrane, perturb typical membrane structure, and create large conductance pores that disrupt ion homeostasis¹¹⁶. Neurofibrillary tau tangles can also form conducting ion pores in the lipid bilayers of neuronal membranes that non-selectively allow the passive diffusion of ions¹¹⁷. Alternatively, amyloid oligomers may indirectly increase membrane excitability by altering the function of voltage-gated sodium channels (VGSCs)¹¹⁸. BACE1 and presenilin (PS)/γ-secretases are proteolytic enzymes that are primarily responsible for the generation of pathogenic amyloid oligomers. Irrespective of amyloid oligomers, these secretases can also independently regulate the surface expression of VGSCs which modulates sodium currents and neuronal excitability¹¹⁹. Amyloid precursor protein (APP), which is also involved in the generation of pathogenic amyloid deposition, has similarly been shown to increase VGSC expression and heighten membrane excitability¹²⁰. Lastly, in addition to its action on VGSC expression, amyloid can also modulate activity of the sodium potassium pump to further disrupt ion homeostasis and membrane excitability¹²¹.

Another well documented contribution to ion disruption and hyperexcitability in AD is calcium mishandling¹²². For example, the largest known ion channel, the ryanodine receptor (RyR), is also known to be implicated in AD¹²³. RyR-mediated calcium release is critical to sustain life; it facilitates all of the body’s muscular contractions as well as the brain’s neurotransmissions. Numerous animal models highlight the role of RyR-mediated calcium leak in the pathogenesis of AD and subsequent excitotoxic neurodegeneration¹²⁴. Notably, RyR alterations have also been reported in post-mortem samples of human brain tissue in patients along the AD continuum. Specifically, changes in RyR expression that are consistent with vulnerability to increased neuronal excitability have been reported in the medial temporal lobe of MCI patients¹²⁵.

4.1.4. Glial Considerations

Cortical hyperexcitability in MCI and AD could also be due to the impairment of microglia-driven neurosuppresion¹²⁶. Microglia, accounting for 10-15% of all cells found within the brain, are known to react to potential threats, control the neuro-inflammatory response, prune non-functional synapses, and produce ligands that support neuronal survival¹²⁷. In addition to these capacities, recent studies revealed that microglia also attenuate brain network hyperexcitability by suppressing neuronal activity^126,128. Through varied mechanisms, this neuroprotective function of microglia can be conceptualized as “brakes” on excessive neuronal activity¹²⁸. During neuronal activation, both neurons and astrocytes release ATP^129–132. Microglia can detect synaptic release of ATP¹³³, and an increased number of microglia protrusions would be recruited to the activated synapses¹²⁶. ATP is then converted into AMP by microglial ATP/ADP hydrolyzing ectoenzyme^134,135. This is then followed by conversion of AMP to adenosine, which suppresses neuronal activity¹³⁶. These findings suggest that microglia-driven neurosuppresion might play a complementary role in restricting excessive neuronal activation that cannot be sufficiently suppressed by inhibitory neurons alone¹²⁶. Microglial dysfunction is a well characterized pathological trait of AD, which may have an underappreciated role in the subsequent neuronal hyperexcitability¹³⁷.

4.2. Dysfunction of Central Cholinergic System in Alzheimer’s disease

Another important finding of our meta-analysis is that patients with AD exhibited significantly reduced short-latency afferent inhibition (SAI) compared to cognitively normal older adults. As mentioned in Section 3.3.2.3, SAI is a TMS-induced neurophysiological measure that has great potential to probe major cholinergic sources in the central nervous system in vivo⁸⁶. Using a paired-pulse TMS paradigm, the TMS-elicited amplitude of motor output is reduced by adding non-invasive peripheral median nerve stimulation 19-50 ms before the TMS over the primary motor cortex⁸⁵. In other words, the amplitude of motor output is decreased in the presence of a peripheral conditioning pulse because the somatosensory afferent input inhibits the corticospinal output from the subsequent TMS pulse at the primary motor cortex. The SAI is believed to result from direct thalamo-cortical projection to the primary motor cortex via cholinergic paramedical thalamic nuclei⁸⁶. There is strong evidence from pharmacological TMS studies to support the involvement of ACh in the generation of SAI⁸⁶. For example, intravenous injection of scopolamine, an antagonist of ACh at muscarinic receptors, reduces SAI and produces amnestic effects in healthy younger adults¹³⁸. Administration of acetylcholinesterase inhibitors (e.g., donepezil, FDA-approved drug to treat AD) increase SAI^139,140. Our meta-analytic results together with previous findings suggest that the SAI derived from the paired-pulse TMS paradigm may serve as a useful surrogate marker to identify AD patients. Future work is needed to assess the utility of SAI in predicting the progression of AD.

4.3. Reduced cortical plasticity in Alzheimer’s disease and mild cognitive impairment

In addition to hyperexcitability, our meta-analysis also showed reduced cortical plasticity in both patients with MCI and AD. Impaired synaptic function and diminished neural plasticity are early features of AD and MCI^141–144. In addition to its early emergence, clinical data suggests that synapse loss is the strongest pathophysiological correlate of cognitive decline^145–147. A meta-analysis including publications of postmortem examination of neural tissue from AD patients highlights the widespread nature of synapse loss in multiple brain regions including the hippocampus, frontal cortex, cingulate gyrus, entorhinal cortex, temporal cortex, and amygdala¹⁴⁸. Human in vivo measures of cortical plasticity are only becoming available recently with the 11C-UCB-J^149–151 or 18F-UCB-H^152,153 positron emission tomography (PET) targeting the synaptic vesicle proteins 2A (SV2A). However, the SV2A PET has not been commonly used for patients with MCI or AD due to the cost of PET scans and the requirement of on-site cyclotron production of the 11C (half-life = 20.4 mins) and 18F (half-life = 110 mins) tracers.

Yet, rTMS in combination of electromyography (EMG), functional magnetic resonance imaging (fMRI), and/or electroencephalography (EEG) has become an established method capable of providing useful surrogate markers for in vivo measures of cortical plasticity. The rhythmic trains of TMS pulses have been used to induce LTP-like or LTD-like effects following single or multiple rTMS sessions in both animal^154–157 and human studies^158–164. For example, rTMS delivered to lightly anesthetized rats induced ribosomal protein S6 phosphorylation in large numbers of neurons indicating the activation of molecular pathways critical for plasticity¹⁵⁵. In another study of rats, rTMS was found to induce structural alterations of synaptic plasticity, increase the expression of synaptic protein markers (e.g., synaptophysin) and increase the expression of Ca²⁺/calmodulin-dependent protein II¹⁵⁷. In a separate study of mice, synaptic structural plasticity was enhanced and both the pathway of brain-derived neurotrophic factor and phosphorylated cyclic adenosine monophosphate response element binding protein was activated by rTMS¹⁵⁶. In human studies, descending corticospinal activity evoked by rTMS of the motor cortex has been recorded directly from the high cervical epidural space of conscious participants who has electrodes inserted for control of intractable dorso-lumbar pain^162,165. These epidural studies showed that rTMS led to a pronounced increase in the excitability of cortical circuits, suggesting the involvement of inter-neuronal network in rTMS-induced plasticity^162,165. Accumulating evidence has demonstrated that the degree of rTMS-induced plasticity can be measured non-invasively by interleaving EMG^{15,160,164,166–169}, fMRI^170–175 or EEG^{161,172,176–179} with rTMS to examine changes in response to stimulation. Lastly, pharmacologic rTMS-EMG studies have reported that LTP-like after-effects from rTMS are NMDA receptor dependent¹⁶⁸. This multimodal approach may provide important insights into functions of cortical plasticity and serve as a useful tool for clinical diagnosis of MCI and AD.

4.4. Limitations and Future Directions

Although our meta-analysis has demonstrated the promising role of TMS in measuring cortical excitability and plasticity in vivo for individuals with AD and MCI, it also highlights several important key areas for further research based off both the key findings and the limitations inherent in some of the studies included in this meta-analysis.

First, very few TMS studies have been conducted in individuals with MCI. MCI represents a heterogeneous clinical condition both in its etiology and in risk of conversion to AD. The amyloid-tau-neurodegeneration (ATN) framework was proposed for researchers to better capture this heterogeneity across the spectrum of AD, and to define the disease based upon its diverse neurobiological underpinnings. In this framework, Amyloid (A) and Tau (T) represent the specific proteinopathies characteristic of AD and a biological definition of AD necessitates the presence of both. Alternatively, while not necessary for an AD diagnosis, the neurodegeneration (N) component is a non-specific category that is incorporated to further qualify, stage, and subtype AD¹. Several studies have examined TMS parameters and their relationship with ATN biomarkers in MCI and AD^{39,48,58,59,68}. For example, Koch et al.^58,59 found that higher cerebrospinal fluid (CSF) levels of tau proteins in patients with AD were associated with more impaired LTP- and LTD-like cortical plasticity as indexed by changes in MEP amplitude following theta burst stimulation protocols. From the same research group, Martorana et al.⁶⁸ reported that higher dysfunction of central cholinergic activity as measured by SAI was correlated with both lower levels of CSF amyloid and higher levels of phosphorylated tau. When evaluated in isolation, other studies have reported that a combination of TMS parameters could successfully detect distinct pathophysiological processes to improve the differential diagnosis of dementias²⁰, and, furthermore, that the diagnostic confidence by including TMS based biomarkers to routine assessment in dementia is comparable to the more well-established amyloidosis biomarkers¹⁸⁰. TMS based measures like SAI can also reportedly predict which patients will respond to pharmacologic therapies for AD¹⁴⁰, which may have utility as an (N) biomarker to subtype and/or stage AD patients. These encouraging findings suggest that future research should develop prediction models incorporating TMS data (e.g., SAI) and ATN biomarkers to evaluate the complementary contribution of TMS parameters to the ATN framework. These TMS parameters, which can be obtained at relatively low costs and with less invasive procedures, may provide additive value to 1) improve the differential diagnosis of AD, 2) enable earlier diagnosis of AD, 3) improve subtyping of AD, and 4) better predict disease progression.

Second, several factors (e.g., genetics, drugs, endogenous brain oscillations, history of synaptic activity, attention, age, time of day, aerobic exercise, and sex) that influence the induction of cortical plasticity have been identified in TMS studies^181,182. For example, concurrent TMS with electroencephalography (EEG) has been used to characterize alterations of brain oscillations in AD and MCI^{51,54,183–185}, track disease progression¹⁸⁶, and measure response to therapy¹⁸⁷. The combination of TMS and EEG has great potential for non-invasive, real-time assessment of cortical reactivity and large-scale network dynamics in AD and MCI. Future research should consider multi-modal approaches and take these above-mentioned factors into account when examining differences in case-control studies.

5. CONCLUSIONS

These TMS-derived measures of cortical excitability and plasticity described herein advance our understanding of the pathophysiology of AD and MCI in the context of the aging brain and have great potential as diagnostic biomarkers for identification of individuals with AD and MCI. It would be important for future studies to examine the relationships between these TMS-derived neurophysiological measures and the amyloid-tau-neurodegeneration (ATN) biomarkers¹. This area of investigation will provide a more complete view on potential advantages that TMS might exhibit in comparisons to currently available biomarkers, and ultimately, provide necessary knowledge to develop more effective and refined strategies for prevention, diagnosis, and management of AD in the incipient stage of the disease.

Figure 2.

Forest plots. Individual and pooled effect sizes of comparisons: (A) between patients with Alzheimer’s disease (AD) or mild cognitive impairment (MCI) and cognitively normal older adults (CN); (B) between AD and CN; and (C) between MCI and CN.