Separate memory-related processing for auditory frequency and patterns

Transcription

1 Psychophysiology, 36 ~1999!, Cambridge University Press. Printed in the USA. Copyright 1999 Society for Psychophysiological Research Separate memory-related processing for auditory frequency and patterns CLAUDE ALAIN, a ANDRE ACHIM, b and DAVID L. WOODS c a Rotman Research Institute, Baycrest Centre for Geriatric Care & Department of Psychology, University of Toronto, Canada b Laboratoire de Neuroscience de la Cognition, Université du Québec à Montréal, Canada c Department of Neurology and Center for Neuroscience, University of California at Davis, Northern California System of Clinics, Martinez, California, USA Abstract Detecting deviant, and potentially meaningful, auditory events depends on transient representations of preceding stimuli. Here, we examined whether the neural circuitry underlying deviance detection system varied as a function of deviance type. In different blocks of trials, participants were presented with a sequence that included standard and deviant tones differing in frequency or a sequence of tones that alternated regularly in frequency with occasional deviant repetitions. Both frequency- and pattern-deviant stimuli elicited a mismatch negativity ~MMN! peaking between 120 and 175 ms poststimulus. The MMN amplitude distribution was more frontal for frequency-deviant than for patterndeviant stimuli. There are two possible explanations for these results. Both frequency- and pattern-deviation MMNs might arise in the same set of generators whose relative strength of activation varies. Alternatively, frequency- and pattern-deviation MMNs could originate in different generators. These alternatives were investigated using principal component analysis and signal identification methods. These methods revealed that no common signal space could account for both of the MMNs, indicating different generator sources for the analysis of frequency and pattern deviance. The results suggest separate memory-related processing for auditory frequency and patterns and indicate that the neural circuit of deviance detection varies as a function of the perceptual context. Descriptors: Mismatch negativity, Deviance detection, Frequency deviance, Pattern deviance, Sensory memory Sensory memory refers to the transient representations of stimuli for subsequent integration with previously presented stimuli or recalled information. This memory is important in audition because the perception of complex signals such as speech and music requires the integration of successive stimulus features that occurs over short intervals. The representation of sounds in sensory memory has been divided into two separate phases: preperceptual and perceptual or synthesized auditory memory ~Massaro, 1972!. The preperceptual stage is also called after-image or echoic memory and lasts several hundred milliseconds, whereas perceptual memory can last up to 20 s depending on the study and the complexity of the stimuli ~for a review see Cowan, 1984; Cowan, 1995!. An important difference between the content of these sensory memories is found in the coding of complex sound patterns. The perceptual or synthesized memory retains the temporal structure of complex sound ~Massaro, 1972!, including their sequential order whereas the preperceptual memory holds only a spectral This study was supported by an FRSQ postdoctoral fellowship to C.A. and by NINDS grant NS32893 to D.L.W. We thank Leun Otten, Lori Bernstein, and Cyma Van Petten for helpful comments on previous versions of this manuscript. Address reprint requests to: Claude Alain, Rotman Research Institute, Baycrest Centre for Geriatric Care, 3560 Bathurst Street, North York, Ontario, M6A 2E1, Canada. calain@rotman-baycrest.on.ca. average of the consecutive components weighted in favor of the most recent element ~Cowan, 1995; Massaro, 1972!. In recent years, the mismatch negativity ~MMN! component of human event-related brain potential ~ERP!, has emerged as a tool to investigate the neurophysiological basis of auditory sensory memory. The MMN is elicited by infrequent ~deviant! changes within a sequence of homogeneous ~standard! stimuli. The deviant sounds may differ in such dimensions as frequency, duration, intensity, spatial location, or spectral pattern ~Kraus, McGee, & Koch, 1998; Näätänen, 1992; Näätänen et al., 1997!. Deviations from simple auditory patterns such as a sequence of tones that alternate ~Alain & Woods, 1997; Alain, Woods, & Ogawa, 1994; Nordby, Roth, & Pfefferbaum, 1988; Tervaniemi, Saarinen, Paavilainen, Danilova, & Näätänen, 1994! or decrease ~Tervaniemi, Maury, & Näätänen, 1994! regularly in frequency can also generate an MMN. The MMN is thought to reflect a neural mismatch between the incoming deviant stimulus and a memory of the standard stimuli, providing a direct and noninvasive measure of sensory memory ~Näätänen, Paavilainen, Alho, Reinikainen, & Sams, 1989; Ritter, Deacon, Comes, Javitt, & Vaughan, 1995; Winkler & Näätänen, 1992, 1995!. For instance, deviant stimuli presented without the standard sounds do not generate an MMN ~Näätänen et al., 1989! and backward masking reduces MMN amplitudes in a way that corresponds to performance decrements in perceptual judgment tasks ~Winkler & Näätänen, 1992, 1995!. 737

2 738 C. Alain, A. Achim, and D.L. Woods The nature of auditory representations underlying the MMN generation has attracted considerable attention over the past decade. Originally, the auditory trace associated with the MMN was thought to include only sensory information about standard stimuli such as frequency, intensity and duration. Gomes, Bermstein, Ritter, Vaughan, and Miller ~1997! provided evidence suggesting that the MMN generation may depend on a unified or Gestalt representation of the stimuli. In their study, participants were presented with three standard stimuli that differed in frequency and intensity. Deviant stimuli were made by combining the frequency of one standard stimulus with the intensity of another standard stimulus. The wrong combination of features elicited an MMN, consistent with the idea that MMN generation depends on a unified representation of the standard stimuli. However, there is also evidence suggesting that various features of stimuli can be stored independently. Giard et al. ~1995! presented subjects with sequences of stimuli that included rare deviant sounds differing in intensity, frequency, or duration. They found different MMN topographies as a function of the deviant type and proposed that frequency, intensity, and duration have separate neural representations ~for different results see Paavilainen, Alho, Reinikainen, Sams, & Näätänen, 1991!. Using a similar paradigm, Levanen, Ahonen, Hari, McEvoy, and Sams ~1996! found different mismatch field ~MMF, the magnetic counterpart of the MMN! topographies for changes in frequency, duration, or stimulus onset asynchrony ~SOA!. The amplitude distribution of MMF to frequency deviance was more anterior than the MMFs elicited by duration or SOA deviants. In addition, the equivalent current dipole location for the SOA deviants was located superior to the MMF to duration or frequency. Together, these findings suggest that the MMN may rely upon multiple traces, each representing a feature of the stimulus ~e.g., frequency, intensity, and duration! located in different regions of the auditory cortex. Lastly, differences in MMF topographies elicited by deviant tones, deviant chords, or deviant patterns have also been reported ~Alho et al., 1996!, suggesting that processing of stimulus deviance in simple and complex sound patterns may also involve distinct neural systems. However, the conclusion that different topographies reflect distinct neural substrates relies on the implicit assumption that each MMN originates from a single cerebral source. If the MMNs were generated by two or more sources with distinct distributions, then the same generators activated in different proportions could produce MMNs with distinct scalp distributions. One way to evaluate whether the neural systems underlying different MMNs vary as a function of the deviant type is to test whether they share a common signal space. This test can be carried out in two stages. First, the number of principal components required to account for the signal in each condition must be determined. This process determines the minimum number of sources underlying each MMN. Second, the maximum number of principal components is extracted from the two conditions pooled together. If different MMN topographies reflect different proportions of activity in the same generators, then the same principal components should account for the signals in both conditions. In contrast, if different MMN topographies reflect distinct neural substrates, then the common signal space would include more generators than either condition alone and several principal components would be required to account for the signal in both conditions. This approach requires a method of detecting signals in the residuals after filtering out one or more spatial principal components. For this purpose, we used a method called PC1 ~Achim, 1995!, which is summarized in the method section. In the present study, we examined whether the MMNs elicited by frequency- and pattern-deviant stimuli have different neural generators. Participants were presented with two different sequences: ~1! a sequence of identical tones that included occasional tones differing in frequency and ~2! a sequence of tones alternating regularly in frequency with rare breaks in the alternating pattern ~see Figure 1!. If the MMN to frequency- and pattern-deviant stimuli involve the same generator or set of generators, then they should have a common signal space. In contrast, if different MMN topographies reflect distinct neural substrates, then the common signal space would include more generators than either condition alone and several principal components would be required to account for the signal in both conditions. Method Participants Twenty-two participants ~12 men; aged years, mean 30.4 years! volunteered their time or participated for pay. All were right handed, had normal hearing, and gave informed consent according to the guidelines of the Veterans Administration Medical Center and the University of California, Davis. Stimuli and Task Participants were presented with two different auditory sequences in a sound-attenuated chamber. In one, repeated standard tones ~1000 Hz! were presented with rare deviant tones differing in frequency ~1122 or 1414 Hz, probability 3% each!. In the other sequence, a pattern of tones that alternated regularly in frequency ~500 and 2000 Hz! was presented with occasional deviant repetitions ~probability 3% for both frequencies!. Stimuli were delivered monaurally ~100 ms in duration, 10 ms rise0fall times, 85 db SPL! at a rapid rate with fixed 200-ms intervals between consecutive stimuli. Broadband noise ~65 db SPL! was presented in the opposite ear to mask environmental noise and prevent crosshearing. Participants were instructed to ignore the auditory stimuli. To control for attention, participants performed a continuous visual discrimination task. The visual stimuli consisted of five thick ~prob- Figure 1. Schemata of the stimuli used. Top: Sequence of identical stimuli mixed with rare frequency deviant stimuli. Bottom: Sequence of tones alternating regularly in frequency with occasional repetition breaking the alternation. In both sequences, deviant stimuli are indicated by an asterisk.

3 Mismatch negativity to frequency- and pattern-deviant stimuli 739 ability 80%! or five thin ~probability 20%! vertical bars subtending visual angles of or , respectively. The bars were presented at central fixation on a monitor subtending at a distance of 160 cm from the participant, with a luminance of 26.9 cd0m 2 over a gray background ~6.2 cd0m 2!. Stimulus duration was 57 ms. Interstimulus intervals ~ISIs! between the visual stimuli ranged from 450 to 1450 ms ~mean 850 ms! according to a rectangular distribution. Participants were asked to attend to the visual stimuli and to press a button as fast as possible whenever the thin vertical bars ~i.e., the targets! appeared. Reaction times ~RTs! were analyzed for correct trials only, that is, those trials with RTs occurring between 200 and 1,000 ms after the thin bars were presented. Responses at other times were classified as false alarms ~FAs!. All participants were presented with four conditions defined by the combination of the two ears and the two sequence types. Trials were blocked by condition and the order of conditions was counterbalanced across subjects. Participants were presented with 12 blocks of trials. Each stimulus block contained 220 visual stimuli including 44 targets ~probability 20%! and 734 auditory stimuli including 44 deviant sounds ~probability 6%!. In total, 2,640 visual and 8,800 auditory stimuli were presented to each participant. Electrophysiological Recording and Data Analysis The electroencephalogram ~EEG; bandpass Hz; 256 Hz sampling rate! was recorded from 28 electrodes over the scalp ~FP1, FPz, FP2, nose, left preauricular, right preauricular, T1, F7, F3, Fz, F4, F8, T2, left mastoid, T3, C3, Cz, C4, T4, right mastoid, T5, P3, Pz, P4, T6, O1, Oz, and O2!. Vertical and horizontal eye movements were recorded from electrodes lateral and below the left eye. All electrodes were referenced to four interconnected electrodes at the base of the neck, balanced through a potentiometer to cancel electrocardiogram artifacts ~Woods & Clayworth, 1985!. ERPs were averaged separately for each site and stimulus type. Only the analysis of ERPs to auditory stimuli are presented in this report. The analysis epoch included 200 ms of prestimulus activity and 800 ms of poststimulus activity. Trials contaminated by excessive eye movements, excessively large peak-to-peak deflection ~6 150 mv!, or amplifier saturation were automatically rejected before averaging. The MMNs for each deviant type were identified in the difference wave between the standard and the deviant stimuli. MMN peak latency was defined as the maximum negativity between 100 and 300 ms following deviant onset. For each deviant type, the MMN peak amplitude was computed for the 40-ms interval centered at the peak and measured relative to the mean amplitude of the 200-ms prestimulus activity. The MMN peak latencies and amplitudes were then submitted to an analysis of variance ~ANOVA! with repeated measures with deviant type, ear, and electrode as factors. Scalp topographies were first analyzed using ANOVA for repeated measure after voltage normalization ~McCarthy & Wood, 1985! and included 23 electrode sites. The electrodes lateral and below the left eye, the two mastoid electrodes, the two preauricular electrodes, and the one placed on the tip of the nose were omitted from the analysis. Type I errors associated with inhomogeneity of variance were controlled by decreasing the degrees of freedom using the Greenhouse Geisser epsilon. For the alternating sequence, the MMNs elicited by the repetitions ~i.e., break in the alternation! of low or high frequency tones did not differ in amplitude and0or topographic composition. Therefore, the MMNs to low and high frequency tones were pooled together. Significant differences in scalp topography revealed by the ANOVA were examined further using the procedure developed by Achim and Plourde ~1997! to determine whether the different MMN distributions reflected different generators or a different weighting of the same generators. Differences in amplitude distribution over time were quantified using principal component analysis ~PCA! and a signal identification method. The signal identification test, called PC1 ~Achim, 1995!, is applied to the wave shapes at a given channel. The first principal component is the single wave shape that best reproduces the individual wave shapes, irrespective of polarity, and the factor scores are the corresponding required amplitudes. A signal is detected when the mean factor score differs significantly from zero, showing that the various contributing wave shapes have the same signal embedded in the noise. For noise only, the shape of the first principal component should be strictly opportunistic with random polarities of its factor scores. In the current experiment, PC1 was applied to the residuals in each condition after removing the signal accounted for by a given number of principal components ~Achim, Richer, & Saint-Hilaire, 1988!. All statistical signal identification tests have a negative bias when applied to residuals of a model that is least-squares fitted on data, which includes principal component models. Simulations have shown that separate signal detection tests at several channels on the residuals of a least squared fitted model do not lead to excessive false positives ~Achim & Plourde, 1997!. For the present purpose of deciding whether differences in topography reflect different sources or different combinations of the same sources, the first step was to objectively determine the minimum number of active generators for the MMN to frequencyor pattern-deviant sounds. We used a procedure described by Achim, Richer, Alain and Saint-Hilaire ~1988! except for using PC1 as the signal detection test. This procedure requires that the signal be accounted for at all channels, that is, that filtering out the first N principal components from the data of each participant at a given channel leaves only noise in the residuals. Successive spatial principal components extracted from the sum of cross products across latencies and cases are included in the model until that criterion is reached. The number of components required to account for the signal constitutes the estimated dimensionality. In multiple dipole modeling, this estimated dimensionality represents the minimum number of dipoles that could account for all the signals in the data. In the current paper, a model-free approach is taken that consists of verifying whether the same minimal dimensionality is sufficient to account for different types of MMN. This method would indicate that they likely emanate from the same generators, despite apparent differences in topography. Results Behavioral Performance Overall, the participants had little difficulty performing the visual task. They were fast ~mean RT 426 ms! and accurate, detecting 88% of targets and making less than 1% FAs. The type of auditory sequence presented in the background did not influence the participants performance on the visual task. ERPs Figure 2 shows the group average ERPs for standard and deviant stimuli presented to the left ear. In all cases, ERPs to standard stimuli were characterized by a small negative ~N1! and positive ~P2! deflections peaking around 110 and 180 ms from stimulus onset. Both frequency- and pattern-deviant tones elicited a broad

4 740 C. Alain, A. Achim, and D.L. Woods Figure 3. Mismatch negativity elicited by deviant sounds presented to the left ear. Solid line mismatch negativity ~MMN! to small frequencydeviant; dashed line MMN to large frequency-deviant; dotted line MMN to pattern-deviant. Figure 2. Group means of 22 subjects event-related potentials ~ERPs! to tones presented to the left ear. ~A! ERP to repetitive tones, that is, standard ~solid!, small ~dashed!, and large ~dotted! frequency-deviant sounds. ~B! ERPs to tones that alternated in frequency, that is, standard ~solid! and pattern-deviant sounds ~dashed!. In this figure and in Figure 3, the stimulus onset is shown by the vertical bar and negativity is plotted upward. negative deflection that began around 80 ms poststimulus and superimposed the N1 wave. Figure 3 shows the corresponding MMNs isolated by the difference between the ERP to standard and deviant tones. The MMN to large frequency deviant sounds began about 20 ms before the MMN elicited by small frequency deviants or by changes in the alternating pattern. Table 1 shows the group MMN mean peak latencies and amplitudes at the midline frontal site ~Fz! for the deviant sounds presented in the left and right ears. ANOVAs on the mean peak latencies with deviant-type and ear of presentation as factors yielded a main effect of deviant type, F~2,42! 14.42, p.001. The MMN peak latency was earliest for large frequency deviants, intermediate for pattern deviants, and latest for small frequencydeviant sounds ~all pairwise comparisons significant at p.02!. There was no difference in MMN peak latencies as a function of the ear of stimulation, nor was there an interaction between ear and deviant-type. A similar pattern of results was observed for the mean peak MMN amplitude. There was a main effect of deviant type, F~2,42! 20.74, p.001, with large frequency deviants producing more prominent MMNs than small frequency or pattern deviants ~ p.01 in both cases!. MMN peak amplitudes were similar for small frequency- and pattern-deviant stimuli. Moreover, MMN amplitudes did not differ as a function of ear of presentation nor was there an interaction between the ear and deviant type. Table 1. Group Mean Peak Latencies and Amplitude (6SD) for the MMN Measured at the Midline Frontal Site (i.e., Fz) for Both Ears of Stimulation Ear Deviant type Peak latencies Peak amplitudes Left Small frequency-deviant 167 ~40! 4.0 ~1.9! Large frequency-deviant 130 ~25! 5.8 ~2.6! Pattern-deviant 145 ~17! 3.3 ~1.5! Right Small frequency-deviant 172 ~42! 3.9 ~2.6! Large frequency-deviant 144 ~44! 5.2 ~2.6! Pattern-deviant 159 ~34! 3.6 ~1.9! Note: MMN mismatch negativity.

5 Mismatch negativity to frequency- and pattern-deviant stimuli 741 Figure 4 shows the MMN amplitude distribution as a function of deviant type presented to the left and right ears. Because the MMN latency varied with deviant type, the MMN amplitude distributions were examined over a 40-ms interval during which the MMN was observed to be maximal at Fz for each deviant type. An ANOVA with electrode location, deviant type, and ear of presentation as factors yielded a significant Electrode Deviant Type interaction, F~44,924! 4.59, p.001, E Pairwise comparison revealed that the MMN distribution to pattern-deviant sounds differed from the ones generated by either small or large frequency changes. When all three deviant types were considered together, the Electrode Ear interaction was not significant, F~22,462! 1.76, nor was the Electrode Ear Deviant-Type interaction, F~44,924! Similar results were obtained when all measurements were taken in the ms interval. An ANOVA on MMN amplitudes from the midline electrodes yielded a main effect of electrode, F~4,84! 57.91, p.001, E 0.573, revealing a larger MMN at Fz than FPz, Cz, Pz, or Oz ~all pairwise comparisons significant at p.02!. The interaction between Electrode Deviant Type was also significant, F~8,168! 3.76, p.02, E The MMN peak amplitude was more anterior for small frequency-deviant stimuli than for large frequencyor pattern-deviant stimuli ~ p.05 in both cases!. Both frequencyand pattern-deviant stimuli generated an MMN that was maximum over right frontal regions. This asymmetry in MMN scalp distribution was examined by comparing the mean MMN amplitude recorded over the left ~F3, F7! and right ~F4, F8! frontal regions. The ANOVA revealed a significant difference in amplitude with the MMN being larger over the right ~ 3.5 mv! than the left frontal cortex ~ 2.9 mv!, F~1,21! 20.41, p.001. The interaction between hemisphere and deviant type was not significant. At central sites, the MMN tended to be larger over the left ~T3, C3! than the right ~C4, T4! hemisphere, but the difference failed to reach significance, F~1,21! 4.08, p.054. Again the interaction between hemisphere and deviant type was not significant, nor were the effect of ear or the interaction between ear, deviant type, and hemisphere significant. Differences in amplitude distributions as a function of the deviant type were examined further using scalp current density ~SCD! analysis ~Figure 5!. In all conditions, the MMN was associated with marked current fields at frontal and temporal regions. The SCD maps varied with the deviant type. The MMN to small and large frequency-deviant stimuli were associated with stronger current sources over the temporal region compared with the MMN elicited by pattern-deviant stimuli. This finding suggests that MMN to frequency- and pattern-deviant stimuli may have involved different neural generators or at least differential contributions from common generators. The amplitude distribution of the frequency- and patterndeviant MMNs was also compared using the PCA and Achim s signal identification method. First, the number of dimensions required to explain the signal for the MMN to frequency- and patterndeviant stimuli was examined separately for left and right ears. The MMN to frequency-deviant ~small and large! and pattern-deviant Figure 4. Isopotential gray-scale maps of the normalized distribution of the mismatch negativity ~MMN! as a function of the type of stimulus deviation. A view from the top and the right side of the head are shown respectively for the deviant tones presented to the left or right ear. The maps show MMN amplitude distribution for a mean voltage taken over a 40-ms interval centered around the peak obtained at Fz for each MMN. The original data ~27 scalp sites! were interpolated with a spherical spline algorithm ~cf. Perrin, Pernier, Bertrand, & Echallier, 1987!.

6 742 C. Alain, A. Achim, and D.L. Woods Figure 5. Scalp current density of the mismatch negativity ~MMN! elicited by deviant tones presented in the left or right ear. Views from the left side and the right side of the head are shown respectively for the deviant tones presented to the left or right ear. The maps show MMN amplitude distribution for a mean voltage taken over a 40-ms interval centered around the peak obtained at Fz for each MMN. stimuli presented in either the left or right ear required a single component to account for the signal at all channels. The second step in the dimensional analysis consisted of testing the null hypothesis that the same topographies could account for a pair of MMNs ~e.g., MMN to small-frequency deviant tones vs. MMN to pattern-deviant tones!. This implies that a common set of spatial principal components obtained from a matrix of cross products cumulated over the ms interval of both conditions could account simultaneously for the signal in both MMNs. In the present case, the first common principal component should eliminate all signals from each data set if one spatial dimension was represented overall. The first common spatial principal component was thus extracted and filtered out from each participant. The signal identification test was then applied to the residuals to determine whether the retained first common spatial principal component accounted for the entire signal. For the MMNs to small frequency- and pattern-deviant sounds presented to the left ear, the first common principal component did not remove all of the signal from the pattern MMN; significant signals were left at F7, FP1, FPz, FP2, F8, Fz, F4, C3, Cz, C4, and O1 ~ p.05, in all cases!. This residual signal indicates that the MMNs to small frequency- and pattern-deviant sounds originate in distinct generators. The observed differences over the frontal region probably reflect a change in dipole orientation located in the posterior cortices rather than the contribution of a generator in the frontal lobe because for both MMNs only one dimension was needed to account for the signal. A similar pattern of results was observed for the right ear stimulation. The first common spatial filter did not remove the entire signal from both conditions, with signals left at C3, Cz, and F8 ~ p.05 in all cases!. For the MMNs to large frequency- and pattern-deviant sounds presented to the left ear, application of the filter based on the first common dimension could not remove all signal from the MMN to pattern-deviant stimuli. There was a significant signal detected at F7, T1, and F8 ~ p.05, in all cases!, suggesting that MMN to large frequency- and pattern-deviant sounds also originate in distinct neural systems. The remaining signal over the lateral frontal regions may reflect differences in dipole orientations located in posterior auditory cortices. However, for the same deviant sounds presented in the right ear, application of the filter based on the first common dimension did remove all the signal from the MMN to pattern-deviant stimuli. Significant differences were also found between the MMN elicited by small and large frequency-deviant sounds presented to the left ear. When the first common dimension was removed from the MMN to small frequency-deviant sounds, there was a significant signal at FP1, FPz, FP2, F3, F8, C3, Cz, C4, and T4 ~ p.05 in all cases!. Lastly, there was also a difference in topographic structure between the MMN elicited by small and large frequency deviants presented to the right ear. When the first common principal component was removed from the MMN to small frequency deviants, there was a significant signal detected at FP2, C3, and Cz ~ p.05 in all cases!. Discussion For all participants, deviant tones differing in frequency from the standard stimuli elicited an MMN wave. MMN amplitude and latency varied with the magnitude of deviation. This finding is consistent with many studies showing a decrease in the amplitude

7 Mismatch negativity to frequency- and pattern-deviant stimuli 743 and increase in the latency of the MMN with reduced discriminability between standard and deviant stimuli ~Näätänen, 1992; Tiitinen, May, Reinikainen, & Näätänen, 1994!. Also in agreement with previous findings ~Alain et al., 1994; Tervaniemi, Saarinen, et al., 1994!, occasional repetitions of a tone in a sequence of tones that otherwise alternated regularly in frequency generated an MMN wave. This MMN wave to pattern-deviant stimuli may reflect a neural mismatch between the incoming stimulus and the expected stimulus based on the organization of the previous stimuli ~Alain, Cortese, & Picton, 1998; Alain et al., 1994; Tervaniemi, Maury, et al., 1994!. In other words, the pattern of stimuli leads the auditory system to anticipate one signal and not another, and the MMN occurs when an incoming sound violates that expectancy. The MMN to a change in auditory pattern also implies that the neural model underlying the MMN generation includes complex features such as the spectral and possibly temporal relationships between elements composing those patterns. The novel finding of the current study is that the MMN amplitude distribution varied as a function of deviant type ~frequency vs. pattern!. The MMN elicited by pattern-deviant stimuli was more central than the one elicited by small frequency-deviant tones. This reliable finding was replicated for small and large frequency deviants presented to the left ears as well as for small frequency deviants presented to the right ear. The observed topographic differences are unlikely to reflect changes in arousal because the performance in the visual task was not affected by the sequence type presented in the background. Our results replicate and extend the study of Alho et al. ~1996!, indicating that the MMF to changes in frequency and sound pattern have different sources in the auditory cortex. In Alho s study, the MMF to changes in sound pattern was generated by a more medial source than frequency-deviant MMFs. A more medial generator for the MMN to pattern-deviant sounds could account for the lower amplitude at the temporal sites observed in the SCD maps in the present study. Because the MMN generation depends on representation of previously presented stimuli, difference in MMN amplitude distribution for frequency- and pattern-deviant stimuli indicate separate memory-related processing for auditory frequency and patterns. One possibility is that information about sound frequency and the interstimulus relationship are stored in distinct cortical areas. There is evidence that MMNs to deviant complex patterns, such as vowels, can be selectively impaired in patients with aphasia whereas the MMN elicited by pure tone deviant sounds is little affected by such lesions ~Aaltonen, Tuomainen, Laine, & Niemi, 1993!. Coupled with our findings, these results indicate that the processing and encoding of more abstract features such as the interstimulus relationship among auditory stimuli activate separate auditory areas from that activated when detecting changes along a simple physical dimension such as frequency. The MMN amplitude was larger over the right frontal regions than over the left frontal lobe, irrespective of the ear of stimulation. This finding is consistent with numerous studies showing right hemisphere dominance for the MMN elicited by changes along a physical dimension ~Giard, Perrin, Pernier, & Peronnet, 1990; Paavilainen et al., 1991! as well as for changes in auditory pattern ~Alain, Cortese et al., 1998; Alain et al., 1994!. This right hemisphere dominance of the MMN may be related to an asymmetry in the MMN generators. Levanen et al. ~1996! proposed two distinct sources in the right hemisphere and one in the left hemisphere to account for the MMF amplitude distribution: a bilateral source in the supratemporal cortex that would reflect featurespecific processing and an additional right hemisphere source in the inferior parietal cortex that would be associated with more global auditory change detection. Such a model could accounted for the hemispheric difference in MMN amplitude observed in the current study. A generator in the right frontal lobe has also been proposed to account for the asymmetry in MMN amplitudes ~Giard et al., 1990!. In the present study, the stronger current field observed over the right frontal lobe is consistent with such a hypothesis. Evidence from lesion studies in humans provide further support for a generator located in the prefrontal cortex. Unilateral brain lesion centered around the dorsolateral prefrontal cortex markedly reduced the amplitude of the MMN ~Alain, Woods, & Knight, 1998; Alho, Woods, Algazi, Knight, & Näätänen, 1994!. The dorsolateral prefrontal cortex may play a critical role in maintaining auditory traces for comparison with incoming sensory stimuli or may reflect an involuntary attention switching to changes in the auditory environment ~Giard et al., 1990; Näätänen, 1992!. The MMN topographies also varied as a function of the magnitude of deviation in frequency. This finding may reflect contributions from a tonotopically organized generator ~Tiitinen et al., 1993!. However, this might not be a general feature of all MMNs, but a specific characteristic of the MMN elicited by infrequent changes in tone frequency, as we did not observe topographic differences in the pattern MMN elicited by the repetition of either the low or the high frequency tones. Because the frequency- MMN was generated by rare changes in frequency in a sequence of identical tones, differences in topographies between frequencyand pattern-mmn may also be associated with contributions from N1 generators not strictly involved in the MMN proper. Evidence from brain electrical source analysis ~Scherg, Vajsar, & Picton, 1989! suggests that the difference wave used to isolate the MMN contains at least two sources: ~1! a component that represents the activation of nonrefractory auditory neurons; and ~2! a true mismatch process. Thus, MMNs from physically deviant stimuli may include other components that may account for the observed difference in topography for the MMNs to small and large deviant tones. One possibility to consider for the generation of the MMN would be a two-step evaluation of background auditory stimuli. Large frequency deviation would cause an immediate and large MMN, whereas tones neither classified as standard nor as clearly deviant would be subjected to a second evaluator, eventually leading to a later and smaller MMN. Such a system could account for the later onset of the MMN for small deviant tones observed in the current study. Another possibility would be that large frequency-deviant sounds recruit additional activity in the supratemporal plane or frontal lobes ~Alain, Woods, et al., 1998; Alho et al., 1994; Giard et al., 1990!. The frontal activity would occur primarily with large frequencydeviant stimuli and could be related to an involuntary attention switching to rare salient deviant sounds ~Giard et al., 1990; Näätänen, 1992!. In summary, both infrequent frequency- and pattern-deviant stimuli generate an MMN. The topographies of the MMN were consistent with generators in posterior auditory areas with possible contributions from additional sources located in the prefrontal cortices. The amplitude distribution varied as a function of the deviant type, suggesting that the memory for auditory frequency and pattern are represented in distinct cortical maps. The results show that the neural circuit underlying the deviance detection system associated with the MMN generation varies as a function of the type of stimulus deviation.

8 744 C. Alain, A. Achim, and D.L. Woods REFERENCES Aaltonen, O., Tuomainen, J., Laine, M., & Niemi, P. ~1993!. Cortical differences in tonal versus vowel processing as revealed by an ERP component called mismatch negativity. Brain and Language, 44, Achim, A. ~1995!. Signal detection in averaged evoked potentials: Monte Carlo comparison of the sensitivity of different methods. Electroencephalography and Clinical Neurophysiology, 96, Achim, A., & Plourde, G. ~1997!. Differences between ERP conditions: Different sources or different activity of the same sources. Psychophysiology, 34, S17. Achim, A., Richer, F., Alain, C., & Saint-Hilaire, J.-M. ~1988!. A test of model adequacy applied to the dimensionality of multi-channel average auditory evoked potentials. In D. Samson-Dollfus, J. D. Guieu, J. Gotman, & P. Etevenon ~Eds.!, Statistic and topography in quantitative EEG ~pp !. Paris: Elsevier. Achim, A., Richer, F., & Saint-Hilaire, J.-M. ~1988!. Methods for separating temporally overlapping sources of neuroelectric data. Brain Topography, 1, Alain, C., Cortese, F., & Picton, T. W. ~1998!. Event-related brain activity associated with auditory pattern Process Citation#. NeuroReport, 9, Alain, C., & Woods, D. L. ~1997!. Attention modulates auditory pattern memory as indexed by event-related brain potentials. Psychophysiology, 34, Alain, C., Woods, D. L., & Knight, R. T. ~1998!. A distributed cortical network for auditory sensory memory in humans. Brain Research, 812, Alain, C., Woods, D. L., & Ogawa, K. H. ~1994!. Brain indices of automatic pattern processing. NeuroReport, 6, Alho, K., Tervaniemi, M., Huotilainen, M., Lavikainen, J., Tiitinen, H., Ilmoniemi, R. J., Knuutila, J., & Näätänen, R. ~1996!. Processing of complex sounds in the human auditory cortex as revealed by magnetic brain responses. Psychophysiology, 33, Alho, K., Woods, D. L., Algazi, A., Knight, R. T., & Näätänen, R. ~1994!. Lesions of frontal cortex diminish the auditory mismatch negativity. Electroencephalography and Clinical Neurophysiology, 91, Cowan, N. ~1984!. On short and long auditory stores. Psychological Bulletin, 96, Cowan, N. ~1995!. Attention and memory: An integrated framework. New York: Oxford University Press. Giard, M. H., Lavikainen, J., Reinikainen, K., Perrin, F., Bertrand, O., Pernier, J., & Näätänen, R. ~1995!. Separate representation of stimulus frequency, intensity, and duration in auditory sensory memory: An event-related potentials and dipole-model analysis. Journal of Cognitive Neuroscience, 7, Giard, M.-H., Perrin, F., Pernier, J., & Peronnet, F. ~1990!. Brain generators implicated in processing of auditory stimulus deviance: A topographic event-related potential study. Psychophysiology, 27, Gomes, H., Bermstein, R., Ritter, W., Vaughan H. G., Jr., & Miller, J. ~1997!. Storage of feature conjunctions in transient auditory memory. Psychophysiology, 34, Kraus, N., McGee, T. J., & Koch, D. B. ~1998!. Speech sound representation, perception, and plasticity: A neurophysiologic perceptive. Audiology Neurootology, 3, Levanen, S., Ahonen, A., Hari, R., McEvoy, L., & Sams, M. ~1996!. Deviant auditory stimuli activate human and right auditory cortex differentially. Cerebral Cortex, 6, Massaro, D. W. ~1972!. Preperceptual images, processing time, and perceptual units in auditory perception. Psychological Review, 79, McCarthy, G., & Wood, C. C. ~1985!. Scalp distributions of event-related potentials: An ambiguity associated with analysis of variance models. Electroencephalography and Clinical Neurophysiology, 62, Näätänen, R. ~1992!. Attention and brain function. Hillsdale, NJ: Erlbaum. Näätänen, R., Lehtokoski, A., Lennes, M., Cheour, M., Huotilainen, M., Livonen, A., Vainio, M., Alku, P., Ilmoniemi, R. J., Luuk, A., Allik, J., Sinkkonen, J., & Alho, K. ~1997!. Language-specific phoneme representations revealed by electric and magnetic brain responses. Nature, 385, Näätänen, R., Paavilainen, P., Alho, K., Reinikainen, K., & Sams, M. ~1989!. Do event-related potentials reveal the mechanisms of the auditory sensory memory in the human brain? Neuroscience Letters, 98, Nordby, H., Roth, W. T., & Pfefferbaum, A. ~1988!. Event-related potentials to time-deviant and pitch-deviant tones. Psychophysiology, 25, Paavilainen, P., Alho, K., Reinikainen, K., Sams, M., & Näätänen, R. ~1991!. Right hemisphere dominance of different mismatch negativities. Electroencephalography and Clinical Neurophysiology, 78, Perrin, F., Pernier, F., Bertrand, O., & Echallier, J. F. ~1987!. Spherical splines for scalp potential and current density mapping. Electroencephalography and Clinical Neurophysiology, 66, Ritter, W., Deacon, D., Comes, H., Javitt, D. C., & Vaughan, H. G. ~1995!. The mismatch negativity of event-related potentials as a probe to transient auditory memory: A review. Ear and Hearing, 16, Scherg, M., Vajsar, J., & Picton, T. W. ~1989!. A source analysis of the late human auditory evoked potentials. Journal of Cognitive Neuroscience, 1, Tervaniemi, M., Maury, S., & Näätänen, R. ~1994!. Neural representations of abstract stimulus features in the human brain as reflected by the mismatch negativity. NeuroReport, 5, Tervaniemi, M., Saarinen, J., Paavilainen, P., Danilova, N., & Näätänen, R. ~1994!. Temporal integration of auditory information in sensory memory as reflected by the mismatch negativity. Biological Psychology, 38, Tiitinen, H., Alho, K., Huotilainen, M., Ilmoniemi, R., Simola, J., & Näätänen, R. ~1993!. Tonotopic auditory cortex and the magnetoencephalographic ~MEG! equivalent of the mismatch negativity. Psychophysiology, 30, Tiitinen, H., May, P., Reinikainen, K., & Näätänen, R. ~1994!. Attentive novelty detection in humans is governed by pre-attentive sensory memory. Nature, 372, Winkler, I., & Näätänen, R. ~1992!. Event-related potentials in auditory backward recognition masking: A new way to study the neurophysiological basis of sensory memory in humans. Neuroscience Letters, 140, Winkler, I., & Näätänen, R. ~1995!. The effects of auditory backward masking on event-related brain potentials. Electroencephalography and Clinical Neurophysiology, 44, ~Suppl.!, Woods, D. L., & Clayworth, C. C. ~1985!. Click spatial position influences middle latency auditory evoked potentials ~MAEPs! in human. Electroencephalography and Clinical Neurophysiology, 60, ~Received May 13, 1998; Accepted January 26, 1999!