Our present results confirmed the hypothesis that under focused auditory attention and relative to random stimulus sequencing, constant stimulus sequencing sharpens population-level frequency tuning in human auditory cortex in the tonotopic region of the constant frequency. N1m responses were significantly larger when the test stimulus (TS) had a constant frequency than with random TS frequencies, particularly when band-eliminated noises (BENs) with narrow stop-bands were simultaneously presented. Because the total amount of stimulation received at each frequency was identical between the constant and random sequencing conditions, it is the difference in patterning of the stimuli that must be responsible for the findings, with one pattern allowing processing resources to be attracted or allocated to a specific frequency, and the other pattern not.
In order to investigate the mechanism underlying neural population-level frequency tuning, we utilized overlays of TS and BEN and measured auditory evoked fields by means of MEG. Neural activity, which was evoked by TS-BEN overlays, could be schematically divided into three categories: (1) neural activity evoked solely by the TS (dark gray areas in Figure ), (2) neural activity triggered merely by the BEN (light gray areas), and (3) neural activity elicitable by both the TS as well as the BEN (black areas). The N1m responses analyzed in this experiment represent neural groups solely activated by TS onset (dark gray area), since distinct neural groups (black and light gray areas) had already been activated and masked by preceding BENs when TS appeared. We found that the smaller notch-width of BEN caused smaller N1m source strength, as shown in Figure . The presentation of narrow BENs might result in comparably large overlap between neural populations representing BEN versus TS, and therefore comparably little neural activity was elicited by the late TS onset. Constant stimulus sequencing under focused auditory attention may cause sharper and larger neural activity at the attended (constantly presented) frequency, and broader and smaller neural responses at the other frequencies, compared to the random sequencing condition (as schematized in Figure ). This results in little neural activity overlap (black area in Figure ) and large neural activity elicited by the TS onset (dark gray area), especially in case of narrow BEN conditions. We confirmed this hypothesis by demonstrating large N1m source strength differences between the constant sequencing and random sequencing conditions in case of narrow BENs, but similar N1m responses between these two sequencing conditions in case of wide BENs (Figure ).
Our findings cannot be easily explained by invoking attentional gain alone [7
]. It is possible that attentional gain may have been higher for the constant sequencing compared to the random sequencing condition, because subjects could allocate their processing resources to a specific frequency in the constant sequencing condition, but had to divide them across frequencies in the random sequencing condition. However, the differential dependence of N1m enhancement on BEN type, with N1m enhancement declining with the bandwidth of the notch more in the random sequencing relative to the constant sequencing condition, implies that the sharpness of tuning was an important additional factor. Inhibitory neural interactions in the auditory system are known to contribute to sharpening frequency tuning [30
]. Recent animal studies recording single neural activity demonstrated that afferent auditory neurons project broadly tuned inhibitory inputs, in addition to focally tuned excitatory inputs. This results in relatively stronger inhibition of the auditory neurons corresponding to frequencies that neighbour the test frequency [34
]. Such balanced (excitatory and inhibitory inputs) neural activity contributes to sharpening the frequency tuning and to improving spectral contrasts. In the model of Figure , enhanced inhibitory effects on the task-irrelevant neural activity is depicted as reduced activity evoked by the BEN sound in the constant sequencing compared to the random sequencing condition.
In the present design, the subjects rapidly appreciated when a constant sequencing block was presented. Under these conditions, they could focus their attention on a particular stimulus frequency for the duration of the block (30 trials). Similarly, in a random sequencing block, the subjects understood that attention had to be divided across several stimulus frequencies. Because of this evident task knowledge, it is possible that frequency tuning was differentially modulated by "top-down" attentional mechanisms between these two conditions [37
]. These top-down neural inputs targeting at one specific region within the tonotopic map may have enhanced and sharpened the neural activity corresponding to the constant test stimulus as compared to the random sequencing condition, where the subjects would have distributed the top-down processing resources across the task-relevant tonotopic area, which was defined by the wide range of presented frequencies.
Alternatively, the cumulative bottom-up inputs within a constant-sequencing block may have driven a dual tuning process. The constant stimulus sequencing could have configured a regular auditory stream, which was perceivable for the listeners as an auditory object [38
], whereas the random sequencing could not configure such an auditory object. The encoding of an auditory object in noisy environments might enhance the corresponding neural activity [39
], and might have resulted in better auditory performance in the present study. Either of these mechanisms ("top-down" or "bottom-up") is compatible with evidence for a "winner take all" strategy of auditory tuning reported by Schulze et al. [40
] and Kurt et al. [41
]. Their findings indicated that slightly higher neural activity elicited by one specific sound object ('winner') inhibited neural activity corresponding to other sounds ('losers'). In the present study, the repetition of constant TS within a block might have unconsciously formed a neural representation of an auditory object corresponding to the constant TS sequence in the auditory cortex by means of a bottom-up process. Alternatively, top-down auditory focused attention during constant stimulus sequencing could have defined the neural activity corresponding to the constant TS as 'winner' in advance of the TS onset, dynamically sharpening frequency tuning for the relevant sound in constant sequencing blocks. These neural processes might have lead to sharper population-level frequency tuning and better auditory performance, as evident in the constant sequencing condition during auditory focused attention.
In the present study, we observed larger N1m source strengths in the left compared to the right hemisphere. Noteworthy, it is known that the N1m response elicited by a pure tone in a silent environment has similar or even larger amplitudes [42
] and shorter latencies [43
] in the right hemisphere than in the left hemisphere. Therefore, the results of the present study support the hypothesis that the left hemisphere plays a dominant role in monitoring and processing auditory signals in noisy environments [44
Previous studies demonstrated that the repetition of auditory stimuli with an identical or a similar frequency reduces corresponding neural activity ('stimulus-specific adaptation') [45
]. In the present study, TS were identical in the constant sequencing condition, which theoretically could have lead to larger stimulus-specific adaptation effects and smaller N1m responses than in random sequencing. However, the N1m responses were significantly larger in the constant sequencing condition. The important difference between our study and previous studies is that whereas we presented the BENs between as well as during the presentation of the test sounds, silent intervals between test stimuli were used in previous studies. In our study, all BENs in a constant sequencing block contained a spectral notch around the constant TS frequency, whereas in the random sequencing block most of the preceding BENs (not the simultaneously presented BENs) had a spectrum overlapping with the subsequent TS. The BENs had a power that was 15 dB larger compared to the TS. Therefore, the spectral overlap between a preceding BEN and the subsequent TS in the random sequencing condition might have caused larger N1m decrements as compared to the constant sequencing condition. However, considering the long (2200 msec) time interval between a preceding BEN and the subsequent TS, the adaptation effect on the N1m response should be quite small [49
]. Thus, adaptation alone cannot explain the relatively small N1m source strength difference between the constant and the random sequencing conditions in the wide BEN compared to the narrow BEN conditions.