A predictive coding account of MMN reduction in schizophrenia
Introduction
Schizophrenia is a complex neuropsychiatric disorder affecting about 1% if the population worldwide. It is characterized by a large range of positive and negative symptoms. In order to develop new drugs and to better understand this complex disease, it is necessary to use neurophysiological markers of the pathology (Javitt, 2009, Javitt, 2015). In the last 25 years, the mismatch negativity (MMN) has become one of the most popular markers of schizophrenia (Todd, Michie, Schall, Ward, & Catts, 2012).
The MMN is thought to index deviance detection in the auditory domain (Näätänen et al., 2005Näätänen, Jacobsen, & Winkler, 2005) but could reflect a more general cortical computation as counterparts have been observed in other modalities (Krauel, Schott, Sojka, Pause, & Ferstl, 1999; Pazo-alvarez, Cadaveira, & Amenedo, 2003; Shinozaki, Yabe, Sutoh, Hiruma, & Kaneko, 1998; Tales, Newton, Troscianko, & Butler, 1999). It is most commonly elicited by the oddball paradigm: a stream of repetitive (standard) sounds is rarely interrupted by the occurrence of a different oddball sound. When the response evoked by a given tone presented as a standard is subtracted from the response evoked by the same sound presented as a deviant, a negative component can be observed between 100 ms and 200 ms after the onset of the tone over the frontocentral electrodes in EEG. The MMN can also be detected over the temporal sensors in magneto-encephalography (MEG). These topographies correspond to sources in the Heschl’s gyrus or the auditory cortex. Crucially, the MMN is elicited even when subjects do not pay attention to the sounds and engage in other tasks like reading or watching a silent movie. The fact that MMN can be elicited independently of the attentional state of the subject makes it a reliable tool for both clinical use and translational investigation in animal models.
The reduction of MMN amplitude in schizophrenia is a robust finding (Umbricht & Krljes, 2005) in both medicated and unmedicated patients that has been reported in over 120 papers since the first publication by (Shelley et al., 1991). In some cases, the reduction of MMN amplitude in patients correlates with the intensity of the cognitive symptoms (Baldeweg, Klugman, Gruzelier, & Hirsch, 2004). Interestingly, the MMN amplitude is reduced by NMDA receptor (NMDA-R) antagonists like PCP and ketamine in normal subjects and in animals (Javitt, Steinschneider, Schroeder, & Arezzo, 1996; Umbricht et al., 2000). These drugs are also known to elicit psychotic symptoms similar to schizophrenia in normal subjects and to increase psychotic symptoms is schizophrenic patients. For this reason, the NMDA-R deficiency hypothesis is one of the leading theories to explain schizophrenic symptoms.
The predictive coding framework (Rao & Ballard, 1999; Friston, 2005) is attracting more and more attention to explain both MMN generation (Bendixen, Schröger, & Winkler, 2009; Friston, 2005; Lieder, Stephan, Daunizeau, Garrido, & Friston, 2013) and schizophrenic symptoms (Adams, Stephan, Brown, Frith, & Friston, 2013; Fletcher & Frith, 2009). In a previous paper (Wacongne, Changeux, & Dehaene, 2012), we proposed a neuronal model implementing predictive coding principles in the cortical microcircuit and showing that it could reproduce many properties of the MMN as observed in normal subjects. In this paper, we investigate whether the attenuation of MMN amplitude observed in schizophrenic patients can be reproduced by modifying NMDA-R dependent currents in this model.
The MMN was originally reported to be the earliest potential sensitive to deviance. More recent data showed that earlier potentials can also show sensitivity to deviance (Escera & Malmierca, 2014; Grimm, Escera, Slabu, & Costa-Faidella, 2011). Although some argue that MMN could reflect of simple synaptic adaptation (May & Tiitinen, 2009), recent data and models strongly support the contribution of an independent active predictive process in the generation of the MMN that can be dissociated from repetition suppression or synaptic adaptation (Parmentier, Elsley, Andrés, & Barceló, 2011; Strauss et al., 2015; Todorovic & de Lange, 2012; Wacongne et al., 2012). The interaction between passive synaptic adaptation and active prediction is investigated in “normal” and “NMDA-R impaired” models and the implications for the clinical research are discussed.
Section snippets
Network architecture
This network was developed in a previous paper (Wacongne et al., 2012). The proposed neuronal network aims at modeling the response of primary auditory cortex to incoming sounds. Fig. 1 shows an implementation of the model for an input composed of two pure tones, hereafter called A and B. Each column of the network represents a cortical column with its thalamic input responding maximally to one of the two frequencies of the input. The two frequencies A and B are supposed to be different enough
Results
The author refers the reader to (Wacongne et al., 2012) for a detailed account of MMN generation by the predictive coding model in normal subjects. In a nutshell, we previously showed that the key proprieties of MMN in normal subjects could be reproduced by a neuronal model based on predictive coding principles that use NMDA-R dependent spike timing dependent plasticity (STDP) rules to learn to predict upcoming tones based on statistical regularities in the past inputs (Fig. 1). Specifically,
Discussion
This study builds on a previous neuronal model of MMN generation and examined the effect of NMDA-R dysfunction on MMN amplitude. MMN amplitude was found to be reduced consistently with the observations made in humans and monkeys following the administration of NMDA-R antagonists like PCP or ketamine. It brings further support to the hypothesis of an impairment in predictive processes as a cause for schizophrenic symptoms. The interaction of NMDA dependent learning with upstream synaptic
Conclusion
This study showed that the reduction of MMN amplitude observed in schizophrenic patients in the oddball paradigm is consistent with an impairment of predictive processes mediated by NMDA dependent plasticity. It also stresses the challenges when disentangling with certainty which of the different components of MMN generation may be affected using the oddball paradigm. A better understanding of the neuronal basis of the disease can only emerge from more diagnostic paradigms.
Acknowledgments
I thank Stanislas Dehaene and Jean-Pierre Changeux for useful discussions. This work was supported by the College de France funding to the author.
References (40)
- et al.
Processing of auditory stimuli during visual attention in patients with schizophrenia
Biological Psychiatry
(1998) - et al.
Mismatch negativity potentials and cognitive impairment in schizophrenia
Schizophrenia Research
(2004) - et al.
Canonical microcircuits for predictive coding
Neuron
(2012) - et al.
How the human auditory system treats repetition amongst change
Neuroscience Letters
(2004) - et al.
Responses of human auditory association cortex to the omission of an expected acoustic event
Neuroimage
(2001) - et al.
Impaired sensory processing in male patients with schizophrenia: a magnetoencephalographic study of auditory mismatch detection
Schizophrenia Research
(1999) - et al.
Why are auditory novels distracting? Contrasting the roles of novelty, violation of expectation and stimulus change
Cognition
(2011) - et al.
MMN in the visual modality: a review
Biological Psychology
(2003) - et al.
Human auditory cortex is activated by omissions of auditory stimuli
Brain Research
(1997) - et al.
Mismatch negativity: an index of a preattentive processing deficit in schizophrenia
Biological Psychiatry
(1991)