Remodeling the cortex in memory: Increased use of a learning strategy increases the representational area of relevant acoustic cues

Abstract

Associative learning induces plasticity in the representation of sensory information in sensory cortices. Such high-order associative representational plasticity (HARP) in the primary auditory cortex (A1) is a likely substrate of auditory memory: it is specific, rapidly acquired, long-lasting and consolidates. Because HARP is likely to support the detailed content of memory, it is important to identify the necessary behavioral factors that dictate its induction. Learning strategy is a critical factor for the induction of plasticity (Bieszczad & Weinberger, 2010b). Specifically, use of a strategy that relies on tone onsets induces HARP in A1 in the form of signal-specific decreased threshold and bandwidth. The present study tested the hypothesis that the form and degree of HARP in A1 reflects the amount of use of an “onset strategy”. Adult male rats (n = 7) were trained in a protocol that increased the use of this strategy from ∼20% in prior studies to ∼80%. They developed signal-specific gains in representational area, transcending plasticity in the form of local changes in threshold and bandwidth. Furthermore, the degree of area gain was proportional to the amount of use of the onset strategy. A second complementary experiment demonstrated that use of a learning strategy that specifically did not rely on tone onsets did not produce gains in representational area; but rather produced area loss. Together, the findings indicate that the amount of strategy use is a dominant factor for the induction of learning-induced cortical plasticity along a continuum of both form and degree.

Introduction

That memories are stored in the cerebral cortex is not in dispute. The approaches to identify mnemonic storage in the cortex vary considerably, including techniques as diverse as drawing inferences from brain lesions to functional imaging and electrophysiological recording. Most electrophysiological studies of learning and memory seek correlates, such as the development of physiological plasticity to a particular sensory signal during a learning task. However, as memories have content, i.e., they comprise information about specific events, an alternative approach is needed to determine the extent to which the plasticity constitutes the representation of specific acquired information.

A synthesis of sensory neurophysiology methodology with learning paradigms in hybrid experimental designs has provided such representational information. Instead of determining changes in cortical processing only of signal stimuli, sensory physiological methods provide for the determination of systematic changes in the representation of a stimulus dimension, e.g., by providing receptive fields for a dimension of the sensory signal. This approach is particularly applicable to primary auditory, somatosensory and visual cortical fields because they each contain a topographic representation of one or more stimulus dimensions. It is equally applicable to any cortical field for which the functional and spatial organizations are known. Hence, it is possible to determine if using a particular stimulus value as a signal within the “mapped” dimension produces a specific change in signal processing and representation within that dimension.

This approach has been employed most extensively in studies of acoustic frequency representation in the primary auditory cortex (A1). These experiments first revealed that associative learning actually shifts the tuning of cells in the primary auditory cortex from their original best frequencies (BF) to the frequency of a signal tone (Bakin & Weinberger, 1990). Such associative tuning shifts can produce an increase in the area of representation of the signal-frequency within the tonotopic map of A1 (Hui et al., 2009, Recanzone et al., 1993, Rutkowski and Weinberger, 2005). Receptive field and larger-scale map learning-induced plasticities in the cerebral cortex are referred to collectively as “high-order (cortical) associative representational plasticity” (HARP).

The search for cortical storage underlying memory is further advanced by a comprehensive determination of the attributes of HARP. Thus, most studies of neural correlates demonstrate that they are of associative origin, i.e., reflect the contingency between the conditioned signal stimulus (CS) and a reward or punishment (e.g., Byrne and Berry, 1989, Morrell, 1961, Thompson et al., 1972). However, if the candidate plasticity is part of the substrate of an associative memory, then it should have all of the major attributes of behavioral associative memory. In this regard, the study of acoustic frequency in A1 is currently the only stimulus dimension that has been adequately so characterized. Studies of frequency receptive fields (RFs) (“tuning curves”) have shown that they exhibit all of the major characteristics of associative memory. Signal-specific tuning shifts not only are associative, but also can develop rapidly (within five trials), consolidate (become stronger over hours and days) and exhibit long-term retention (tracked to two months). Additionally, frequency-tuning shifts are discriminative, i.e., shifts are toward reinforced frequencies (CS+) but not toward unreinforced tones (CS−). Finally, HARP in the form of tuning shifts is highly specific, often confining increased response to the signal-frequency plus or minus a small fraction of an octave (reviewed in Weinberger, 2007; see also Calford, 2002, Merzenich et al., 1996, Palmer et al., 1998, Rauschecker, 2003, Syka, 2002).

HARP for acoustic frequency develops in all types of learning studied, including one-tone and two-tone discriminative classical and instrumental conditioning (Bakin et al., 1996, Blake et al., 2002, Edeline and Weinberger, 1992, Edeline and Weinberger, 1993) and with both aversive (e.g., Bakin & Weinberger, 1990) and appetitive conditioning (e.g., Recanzone et al., 1993) including rewarding self-stimulation (Hui et al., 2009, Kisley and Gerstein, 2001). Moreover, HARP for frequency develops in all taxa studied to date: guinea pig [Cavia porcellus] (Bakin & Weinberger, 1990), echolocating big brown bat [Eptesicus fuscus] (Gao and Suga, 1998, Gao and Suga, 2000), cat [Felis catus] (Diamond & Weinberger, 1986), rat [Rattus rattus] (Hui et al., 2009, Kisley and Gerstein, 2001), Mongolian gerbil [Meriones unguiculatus] (Scheich & Zuschratter, 1995) and owl monkey [Aotus trivirgatus boliviensis] (Recanzone et al., 1993). HARP is not limited to non-human animals. The same paradigm of classical conditioning (tone paired with a mildly noxious stimulus) produces concordant CS-specific associative changes in the primary auditory cortex of humans (Molchan et al., 1994, Morris et al., 1998, Schreurs et al., 1997).

HARP has been reported for acoustic dimensions other than sound frequency, including stimulus level (Polley, Heiser, Blake, Schreiner, & Merzenich, 2004), rate of tone pulses (Bao, Chang, Woods, & Merzenich, 2004), envelope of frequency modulated (FM) tones (Beitel, Schreiner, Cheung, Wang, & Merzenich, 2003), direction of FM sweeps (Brechmann & Scheich, 2005), tone sequences (Kilgard & Merzenich, 2002), and auditory localization cues (Kacelnik, Nodal, Parsons, & King, 2006) (for review see Keuroghlian and Knudsen, 2007, Weinberger, 2010). We continue to focus on acoustic frequency because it is by far the best-characterized dimension in sensory associative learning.

However, learning itself is not sufficient to induce specific plasticity. Rather, we discovered that the type of learning strategy employed to solve a task appears to be critical for the formation of HARP in A1 (Berlau & Weinberger, 2008). Learning strategy refers to the complex, multi-dimensional behavioral algorithms that animals employ to solve problems. Specific learning strategies are defined here with terms used as convenient descriptors to highlight the acoustic components of each learning strategy. Thus, they necessarily leave out non-auditory elements like the use of a visual error-cue signal. As memories are probably distributed in the cerebral cortex, it is likely that plasticity for critical components of a learning strategy develop in distributed cortical areas. Here, we focus on how the auditory components of learning strategies dictate HARP in the primary auditory cortex.

In prior studies, groups of rats were trained to bar-press to obtain water rewards contingent on the presence of a signal tone and inhibit bar-presses during silent inter-trial-intervals to avoid error-signaled time-out periods. Although this task appears to be simple, it can be solved in various ways. We identified two learning strategies that animals employed. Subjects could depend on tone onset to initiate bar-pressing and continue until receiving an error signal, while ignoring tone offset. We refer to this strategy as “tone-onset-to-error” or TOTE. Alternatively, subjects could begin responding at tone onset but use the tone offset as a cue to stop bar-pressing, and thus not receive error signals, a “tone-duration” (T-Dur) strategy. The two learning strategies could be distinguished only by analysis of the pattern of bar-presses around and during the presentation of the tone during training trials. Animals using TOTE continue to bar-press immediately after tone offsets while animals using T-Dur do not continue bar-pressing. We found that use of the TOTE learning strategy predicted the development of HARP in the form of signal-specific decreases in threshold and bandwidth. Moreover, the use of the TOTE-strategy was a better predictor of HARP than either the level of correct performance or the degree of motivation. In contrast, use of the T-Dur strategy, regardless of performance or motivation level, never resulted in detectable HARP (Berlau and Weinberger, 2008, Bieszczad and Weinberger, 2010a, Bieszczad and Weinberger, 2010b). These findings support the view that a learning strategy which emphasizes tone onsets, while largely ignoring tone offsets, can be a critical factor in the formation of learning-induced plasticity in the primary auditory cortex.

We hypothesize a continuum on which the degree of tone onset use (or disuse) dictates the amount of HARP in A1: the greater the use of an onset strategy, the greater the magnitude of HARP. Individual subjects may adopt a learning strategy to varying degrees. Thus, the extent to which HARP develops may depend upon the extent to which a tone-onset strategy is employed. We have suggested that frequency-specific local threshold and bandwidth reductions that were previously shown to be dependent on the use of the TOTE learning strategy may be an initial form of HARP in A1 (Bieszczad & Weinberger, 2010b). The next levels of HARP might be local, and ultimately global tuning shifts that underlie specific gains in representational area within the tonotopic map. Thus, if the use of tone onsets in the TOTE-strategy is critical for the induction of HARP, and the amount of TOTE use dictates the degree of HARP, then a greater use of TOTE should produce signal-specific plasticity that surpasses local changes in threshold and bandwidth, to induce gains in representational area. In contrast, without the use of tone onset, the signal would not gain representational area.

Two experiments were used to test the hypothesis that HARP in A1 is dictated by the use of tone onsets. The main goal of the first experiment was to increase animals’ use of tone onsets by increasing their use of the TOTE-strategy. To be specific:

Experiment 1 — If subjects increase their use of the TOTE strategy to solve the problem of obtaining rewards to tones, then they will develop an enhanced form of plasticity, namely signal-specific increases in area of representation within the primary auditory cortex.

The second experiment complements the first in that the goal was to decrease the use of tone onsets. Insofar as previous findings indicated a lack of HARP in A1 when animals use a learning strategy based on beginning to bar-press at tone onset and stopping at tone offset (T-Dur) (Berlau & Weinberger, 2008), we investigated the effect of a learning strategy that relies only on tone offset (R-Off, as explained later), specifically:

Experiment 2 — If subjects rely on the use of tone offset, and not tone onset, to solve the problem of obtaining rewards to tones, then they will not develop signal-specific representational enhancements within the primary auditory cortex, but might instead develop some representational decreases.