The principal finding is that in people with syndromes associated with frontotemporal lobar degeneration, the neurophysiological responses to the NMDA-R antagonist memantine is conditional on individual GABA concentration. Patients with relatively preserved GABA concentrations have greater enhancement of the mismatch responses (i.e. more negative MMN to memantine vs. placebo). This effect was not explained by regional atrophy. We suggest that in the context of future experimental medicine studies, magnetic resonance spectroscopic quantification of heterogeneity might be useful for stratification. Otherwise, within-group neurochemical heterogeneity is liable to reduce sensitivity to treatment effects and increase type II error in clinical trials.
The selectivity of auditory cortex MMN changes to memantine is not unexpected. For example, in Schizophrenia there is both NMDA-R dysfunction and consistent abnormalities in auditory MMN . Auditory regions are sensitive to memantine in Schizophrenia and controls with drug modulation of neural responses [40–42]. However, frontotemporal lobar degeneration, with bvFTD and PSP, is also associated with prefrontal cortical atrophy and GABA-ergic deficits. Neurophysiological changes can occur prior to atrophy, or in the absence of atrophy. This is in part because of the loss of synapses [78, 79] and reductions in critical neurotransmitters  in bvFTD, PSP [32, 33, 80, 81] and other neurodegenerative disorders [82–84]. Magnetoencephalography, or electroencephalography, may therefore provide sensitive markers of disease progression and drug response. In this study, there was a group-wise reduction in GABA concentrations in patients [8, 26], as expected from post mortem data , but the distribution was wide. This variation in GABA, not atrophy, correlated with the effect of memantine on the cortical MMN response.
The drug response in auditory cortex was conditional on frontal GABA status, two areas that span the hierarchical neurocognitive network for prediction and response in MMN tasks [11, 86]. The auditory cortex is relatively spared by the direct neuropathology of bvFTD and PSP, but within the hierarchical network for prediction and error signalling [87, 88], its response is conditional on backward projections from association cortex. A general feature of sensory processing hierarchies is that feedback and feedforward connections are shaped by laminar specificity in cortical units: feedforward connections project principally from superficial pyramidal cells, while feedback connections arising particularly from deep pyramidal cells and terminate on superficial layers at lower-level regions such as the auditory cortex [14, 87, 88]. Prefrontal GABA regulates the precision of the frontotemporal predictions, and the deep cortical generators of back-projecting beta-oscillations [8, 22]. Indeed, beta power and beta-connectivity are reduced in PSP and several syndromes of frontotemporal dementia [8, 9, 31, 89]. The effect of memantine on mismatch response (MMN) generation, mainly from superficial cortical layers of lower levels of the sensory hierarchy, is thereby moderated by prefrontal GABA. In other words, the mismatch between incoming deviant auditory signals with the predicted standard tone is larger on memantine in the context of (near) normal prefrontal GABA.
An alternative hypothesis is that GABA measurements in the inferior frontal gyrus index widespread GABAergic deficits. However, drug-dependent responses in the auditory cortex are independent of GABA concentrations in occipital cortex, and both occipital lobe and auditory cortex are relatively spared by the direct neuropathology.
The MMN differs across many neurological and psychiatric disorders, including Schizophrenia [77, 90], Alzheimer’s disease , and neurodevelopmental conditions . The MMN provides a robust marker of frontotemporal functioning and is well-tolerated in clinical populations. Our lack of a significant group difference in MMN response differs from previous reports of bvFTD [8–10]. The difference might arise from variations in MMN paradigms or heterogeneity of disease, including variance in GABA concentration, atrophy, and syndrome. Indeed, considerable variation has been reported across Schizophrenic patients , the canonical MMN disorder. We also note the heterogeneity in atrophy across PSP subtypes in a larger cohort . Our planned comparisons pooled PSP and bvFTD patients because of their commonalities in physiology and phenotype noted in other deep phenotyping studies [1, 55], despite the clear differences in underlying molecular pathology. Consistent with the ‘frontal’ cognitive deficits in PSP, the majority of cognitive tests were similarly affected by both groups. Although the auditory MMN did not differ between groups, the supplementary analyses suggests prefrontal MMN differences with particularly blunted response in frontal cortex on placebo in bvFTD. The subgroup sample sizes (n = 10 each group) could be considered relatively small, but bayesian tests indicate strong evidence in favor of a group difference. The absence of a main case-control effect should be interpreted in the light of interactions, such as with the neurochemical variance to which we turn next.
Whilst memantine is an NMDA-R antagonist targeting glutamatergic functioning, the response to drug was conditional on GABA rather than glutamate concentration. There are several possible interpretations. The first is that the MRS-measured glutamate is not exclusively neuronal and available for neurotransmission, but also astrocytic as part of glutamate-glutamine cycling . Second, an interaction between glutamatergic and GABAergic systems reflects a delicate balance between excitatory (E) and inhibitory (I) control of the firing of neuronal ensembles. Neurophysiological proxies of E/I functioning have found changes to this balance with ageing  and neurodegeneration . In this study, we tested the ratio of glutamate/GABA concentrations as a proxy of an individual's E/I balance  and found that patients with relatively normal glutamate to GABA ratio show increased mismatch responses to memantine. Importantly, memantine and another NMDA-R antagonists, ketamine, increase pyramidal output activity through their excitatory inputs to GABA interneurons [47, 50, 51]. We speculate that this interaction between prefrontal GABA and memantine promotes coordinated pyramidal firing in response to deviant tones in the oddball paradigm. Such an effect of memantine on the E/I balance has been proposed in Schizophrenia, via influences on oscillatory dynamics [40, 46] according to the ratio of glutamine to glutamate .
Note that memantine’s effects were moderated by a neurochemical (GABA) that is not its primary target (Glutamate receptors); and in a region (i.e. prefrontal cortex) that is non-overlapping with the dependent measure (i.e. auditory cortex). This has implications for experimental studies, in which stratification may be based on interactions between neurotransmitter systems, and interactions between connected regions. Unfortunately, spectroscopy of the auditory cortex was not available, and the frontal lobe was prioritised. Another consideration for future studies is the degree to which baseline pathology moderates a drug response. Our findings can be interpreted as relatively greater pathology (with less GABA) leading to decreased sensitivity to memantine. This is therefore not a simple restoration of function in those with more severe baseline deficits . While it may be easier to show a drug effect than prove its absence, the patients’ disease severity and heterogeneity will affect the result of analyses of a drug’s group-wise main effect.
There are limitations and caveats to this study, in pharmacology, diagnosis and analysis. First, we note that memantine does not have clinical trials evidence for efficacy, and has been subject to successful (but negative) small phase II trials. We do not advocate its clinical use in either PSP or bvFTD. This study was not a clinical trial. Rather, we used the drug as a well-tolerated psychopharmacological probe of neural systems. We found no differential effect of memantine 10mg on the evoked MMN between groups. In both healthy controls and Schizophrenia a higher 20mg memantine dose changed the mismatch response , whereas 10mg produced no group-wise effect [40, 41]. It’s important to note their age of control (mean = 27.51) and schizophrenic participants (36.44)  is considerably younger than those in the current study. This precludes a direct dose comparison across the studies. Moreover, the memantine effect in schizophrenia was moderated by age . Our 10mg dose was chosen to align with the clinically recommended starting dose in the UK .
Our patient diagnoses were clinical, not genetic or neuropathological. Clinicopathological correlations are very high for both PSP and bvFTD, although we cannot distinguish the Tau versus TDP43 pathology as the basis of the bvFTD cases. In the main analyses, we pooled PSP and bvFTD groups despite the clear differences in underlying molecular pathology, not for power considerations, but because of the commonalities in physiology and phenotype noted in deep phenotyping studies [1, 55]. Consistent with the literature on ‘frontal’ cognitive deficits in PSP, patients with PSP and bvFTD were similarly impaired on many of the same cognitive tests, with limited selectivity of deficits in bvFTD. Statistical power and precision are key considerations, especially with n = 20 per group. We used a crossover design that increases power relative to parallel groups designs for heterogeneous populations and leveraged Bayesian inference to consider the evidence in favour of the null hypothesis, as well as alternate hypotheses. For our principal finding, despite modest numbers, there was sufficient precision to draw inferences, with positive or moderate-to-strong evidence for the association between GABA concentration and change in MMN responses on memantine (BF10 > 7). Only a subgroup of controls completed MRS (n = 12), so we did not attempt correlations with neurotransmitter levels within the control group. For MRS based analysis of patient effects, the ratio of glutamate/GABA concentrations as a proxy of an individual's E/I balance has recently been challenged , in favour of other neurophysiological measures such as the 1/f aperiodic slopes , but the resolution of that debate is beyond the scope of this study.
In conclusion, we have probed neurocognitive systems in two disorders associated with frontotemporal lobar degeneration, combining memantine pharmacological challenge with magnetoencephalography and ultra-high field spectroscopy. Patients’ neurophysiological responses to memantine were proportionate to GABA concentration. It may be possible to de-risk the transition from experimental medicine to clinical trials of heterogeneous populations using neurophysiological outcomes and stratification by spectroscopy.