Gamma band plasticity in sensory cortex is a signature of the strongest memory rather than memory of the training stimulus

Highlights

• Gamma waves index coordinated neuronal activity, more effective than uncoordinated.

• Tone plus stimulation of the nucleus basalis implanted associative memory in rats.

• Post-training generalization gradients revealed individual differences in tone memory.

• Gamma increase in primary auditory cortex was linked to the strongest tone in memory.

• Thus, increased neural coordination may be a substrate for specific memory contents.

Abstract

Gamma oscillations (∼30–120 Hz) are considered to be a reflection of coordinated neuronal activity, linked to processes underlying synaptic integration and plasticity. Increases in gamma power within the cerebral cortex have been found during many cognitive processes such as attention, learning, memory and problem solving in both humans and animals. However, the specificity of gamma to the detailed contents of memory remains largely unknown. We investigated the relationship between learning-induced increased gamma power in the primary auditory cortex (A1) and the strength of memory for acoustic frequency. Adult male rats (n = 16) received three days (200 trials each) of pairing a tone (3.66 kHz) with stimulation of the nucleus basalis, which implanted a memory for acoustic frequency as assessed by associatively-induced disruption of ongoing behavior, viz., respiration. Post-training frequency generalization gradients (FGGs) revealed peaks at non-CS frequencies in 11/16 cases, likely reflecting normal variation in pre-training acoustic experiences. A stronger relationship was found between increased gamma power and the frequency with the strongest memory (peak of the difference between individual post- and pre-training FGGs) vs. behavioral responses to the CS training frequency. No such relationship was found for the theta/alpha band (4–15 Hz). These findings indicate that the strength of specific increased neuronal synchronization within primary sensory cortical fields can determine the specific contents of memory.

Introduction

A major advance in broadening our understanding of the neural substrates of learning and memory has involved a shift in emphasis from linear stimulus–response circuits to neural networks. Following Hebb’s insights (Hebb, 1949), it is now generally accepted that coordinated neuronal activity forms during learning to represent and store relevant information, serve cognition and ultimately behavioral action. Gamma frequency oscillations (∼30–120 Hz) are thought to reflect the synchronous activity of neurons both within and across cortical fields (Buzsáki & Wang, 2012). The timescale of gamma oscillations is appropriate for synaptic integration (Salinas and Sejnowski, 2000, Volgushev et al., 1998) and spike timing dependent plasticity (STDP) (Bi and Poo, 1998, Isaac et al., 2009, Wespatat et al., 2004). Particularly relevant to the domain of learning and memory, increased gamma activity has been linked to processes such as attention (Börgers, Epstein, & Kopell, 2008) and short-term memory (Lutzenberger et al., 2002, Pesaran et al., 2002). Moreover, the level of gamma activity at the time of encoding can predict the degree of later recall (Fell and Axmacher, 2011, Osipova et al., 2006, Sederberg et al., 2006, Sederberg et al., 2007). An increase in cortical gamma power also develops during simple associative auditory classical conditioning in humans (Miltner, Braun, Arnold, Witte, & Taub, 1999), underscoring its ubiquity and the potential applicability of appropriate animal models to mechanisms of human learning.

Animal models of associative learning have identified candidate neural substrates for the representation and storage of signal stimuli in the cerebral cortex. For example when a tone is paired with a reinforcer, receptive fields (RF) in the primary auditory cortex (A1) shift to emphasize the frequency of the conditioned stimulus (CS) (Bakin and Weinberger, 1990, Edeline and Weinberger, 1993, Gao and Suga, 2000, Kisley and Gerstein, 2001). Such representational plasticity has the main attributes of associative memory: associativity, specificity, consolidation and long-term retention (reviewed in Weinberger, 2007). Gamma activity may play a critical role in the development of cortically based associative learning. For example, an increase in gamma power within A1 predicts both specific CS-directed cortical plasticity and also behaviorally validated learning 24 h later, but only during initial acquisition, not during maintenance of the memory (Headley and Weinberger, 2011, Weinberger et al., 2006).

Heretofore, it has been assumed tacitly that enhanced gamma activity induced by a signal stimulus during learning reflects the increased salience or behavioral relevance of that stimulus, e.g., the CS in simple associative learning. However, there is an alternative possibility. It is well known that even when perceptual, acquisition and storage processes are functioning optimally, the content of the resultant memory can differ from the actual experience. Subjects trained identically do not all acquire the exact same content (Bieszczad and Weinberger, 2010b, Bieszczad and Weinberger, 2010a, Bieszczad and Weinberger, 2012, Ohl et al., 2001, Polley et al., 2006). Therefore, the greatest increase in gamma activity may actually reflect the stimulus that has gained the greatest strength through learning, rather than the training stimulus. The relative strength of memory for different stimuli along a sensory dimension cannot be determined during training, but rather depends on obtaining post-training stimulus generalization gradients (Bouton, 2007, Mostofsky, 1965). Discrimination learning (i.e., reinforced CS+ with non-reinforced CS−) is a well-documented example. The peak of the post-training generalization gradient is generally not at the CS+ but is displaced to a stimulus value that is farther away from the CS− (Purtle, 1973). Such “peak shift” has been thought to reflect the summation of an excitatory neural gradient centered on the CS+ and an inhibitory neural gradient centered on the CS− (Spence, 1937).

Recently, we found such shifted generalization peaks in simple associative conditioning, due to pre-training exposure to various tones that induced an inhibitory neural gradient in primary auditory cortex (A1) (Miasnikov & Weinberger, 2012). This disjunction between the training frequency and the peak of the generalization gradient provides a unique opportunity to determine whether enhanced gamma activity during learning is tied to the CS frequency or to the peak of the generalization gradient, i.e., to the tone that is most strongly represented in memory. If increased gamma activity reflects increased neural synchrony that is part of the substrate of auditory frequency memory, then the greatest increase in gamma should be tightly linked to the strongest memory, regardless of the training frequency. We report here the analysis of changes in gamma activity that had been recorded during the previous study. The same changes in gamma were analyzed two ways: based on the CS training frequency and based on the peak of the generalization gradient.