Electrophysiological Correlates of Absolute Pitch in a Passive Auditory Oddball Paradigm: a Direct Replication Attempt

Discussion

In the present study, we attempted to replicate Rogenmoser et al. (2015)’s finding of electrophysiological group differences between AP and non-AP musicians during passive listening. Rogenmoser et al. (2015) investigated the automatic nature of AP by recording EEG during a passive auditory oddball paradigm. By analyzing MMN and P3a, they intended to assess the contribution of both pre-attentive (as reflected by the MMN) and more cognitive processes (as reflected by the P3a) in AP. To compare the tone processing between AP and non-AP musicians under different deviation conditions, they applied a paradigm with multiple tuned and mistuned deviants. In line with previous research (Tervaniemi et al., 1993; Matsuda et al., 2013, condition with tuned tones), they did not find any significant group differences in the MMN. In contrast, Rogenmoser et al. (2015) observed smaller P3a amplitudes in AP musicians. This group difference was only found in conditions in which the deviation magnitude was larger than one semitone (264-Hz deviant in block A and all deviants in block C), suggesting that AP musicians process between-pitch but not within-pitch categories differentially than non-AP musicians. Because the P3a has been associated with an early reallocation of attention (Escera et al., 1998; Friedman et al., 2001; Kujala et al., 2007; Polich, 2007), the smaller amplitudes in AP musicians were interpreted as an indication for more efficient cognitive tone processing in AP. The authors concluded that the “P3a component turned out to be a specific marker for AP” (Rogenmoser et al., 2015).

In the current direct replication study, we found no significant group differences in the MMN, confirming the results of the original study. However, and most critically, there were also no significant group differences in the P3a. Additional Bayes factor analyses revealed that the data are more likely under the null hypothesis, implying that AP and non-AP musicians’ tone processing, as indicated by MMN and P3a peak amplitudes and latencies, does not differ during passive listening. Thus, our results challenge the view of cognitive facilitation in AP during passive listening.

In passive auditory oddball paradigms, the MMN typically occurs in response to a change (deviation) in auditory stimulation within a sequence of repeated stimuli (standard tone). The main generator of the MMN is located in the auditory cortex (for review, see Näätänen et al., 2007), where the repeated presentation of a stimulus potentially causes the formation of a short-term memory trace (Näätänen and Winkler, 1999). The MMN is generated when a new auditory input differs from the representation in this sensory memory trace. Because this mismatch detection process does not require that the stimuli are attended, it is thought to be automatic (Sussman et al., 2003; Paavilainen et al., 2007). Accordingly, the MMN is considered an objective measure of auditory discrimination accuracy (Näätänen et al., 2007). Consistent with this view, it has been shown that the amplitude of the MMN increases when discrimination performance improves through training (Näätänen et al., 1993; Menning et al., 2000; Atienza et al., 2002). The MMN amplitude also correlates more generally with behavioral discrimination accuracy (Novak et al., 1990; Näätänen et al., 1993). Similarly, the MMN is also influenced by the deviation magnitude, with larger, and therefore more salient, deviations evoking larger amplitudes and shorter latencies (Sams et al., 1985; Berti et al., 2004; Novitski et al., 2004).

The original study reported an effect of deviation magnitude for block A but not for block C. The authors provided a possible explanation that in block C, all deviants were clustered around an extreme deviation level, with a distance between eight and nine semitones from the standard tone. Consequently, all deviants were probably equally easy to detect. In accordance with the original study, our results showed larger MMN amplitudes and shorter MMN latencies for larger deviations in block A. In block C, the effect also reached significance, but like in the original study, amplitudes did not unambiguously increase with deviation magnitude (compare Fig. 3), suggesting a context effect in this specific block.

More importantly, we also replicated the result of non-significant group differences between the AP and non-AP musicians in MMN measures. The Bayes factor analysis additionally provided support for the null hypothesis. Thus, our data were more likely under the hypothesis that there were no differences in the MMN amplitudes and latencies between the two groups than under the H₁. Our results are not only consistent with the original study but also with other previous research. Using tuned and mistuned pure tones and piano tones, Tervaniemi et al. (1993) did not find group differences between AP and non-AP musicians in MMN amplitudes and latencies. In Matsuda et al. (2013)’s study, MMN amplitudes of AP and non-AP musicians did also not differ for tuned tones, but AP musicians showed larger MMN amplitudes for mistuned tones. However, this effect might have been influenced by the fact that their AP musicians were musically more experienced than the non-AP musicians. Previous research has shown that musical experience can increase MMN amplitudes (Koelsch et al., 1999; Putkinen et al., 2014), specifically in response to mistuned tones (Tervaniemi et al., 2014).

Because the MMN is associated with a passive discrimination process, Tervaniemi et al. (1993) concluded from their results that “pitch naming and discrimination are based on different brain mechanisms.” This coincides with results from behavioral studies showing that pitch-labeling accuracy is not correlated with behavioral pitch-discrimination accuracy (Sergeant, 1969; Fujisaki and Kashino, 2002). Thus, evidence from both behavioral and electrophysiological data suggests that AP does not simply rely on refined pitch discrimination.

In passive auditory oddball paradigms, the MMN is often followed by the P3a, a subcomponent of the P300. Both components have been proposed to play a role in the reallocation of attention to unattended stimuli (Näätänen, 1990; Escera et al., 2000; Kujala et al., 2007), with the processes underlying MMN probably initiating the attention switching and the P3a directly reflecting it. The P3a is affected by the magnitude of deviation in similar ways as the MMN (Berti et al., 2004). As for the MMN, the original study found such a deviation modulation only in block A, probably again due to the more extreme deviation levels in block C. The present study successfully replicated these results. In block A, P3a amplitudes increased and P3a latencies decreased with increasing deviation, and as in the original study, no similar effect was observed in block C. Future studies should more systematically investigate this dependence on specific contexts.

Although the modulation of the MMN and P3a as a function of deviation magnitude is an interesting aspect of general pitch processing, the main finding of the original study was the reduced P3a amplitudes in AP musicians. This result was compared to findings from the parietal P3b, another subcomponent of the P300, which is elicited in active oddball paradigms and often called P300 in these studies. The P3b has been linked to working memory updating (for review, see Kok, 2001; Polich, 2007) and has been investigated more thoroughly in AP research than the P3a. The first study to detect differences in ERPs during pitch processing reported the absence of a P3b in individuals with AP (Klein et al., 1984). This was regarded as an indication that individuals with AP did not need to update their auditory working memory during the task because their pitch representations are permanent. Subsequently, some studies replicated the absence or diminution of P3b amplitudes in AP (Hantz et al., 1992; Wayman et al., 1992; Crummer et al., 1994), but others did not (Hantz et al., 1995; Hirose et al., 2002). . This inconsistency was shown to be caused by differential pitch-processing strategies (RP or AP) employed by the participants based on the specific task instructions, the task difficulty, and the individual level of AP (Bischoff Renninger et al., 2003).

Individual differences in listening strategies could explain why we did not replicate the effect of AP on the P3a. However, this seems rather unlikely as the use of top-down strategies was controlled with the help of a distractor task (watching a silent film) in both the original and the replication study. Given how unreliable the effect of AP on ERPs is even in active tasks, we believe it is more plausible that the differences in passive pitch processing are too subtle to be reliably detectable with ERP peak measures. Alternatively, it could also be speculated that the pitch labeling is only initiated when actively attending the auditory stimuli or when performing a labeling-related task (e.g., bimodal Stroop task; Akiva-Kabiri and Henik, 2012). Compelling evidence for an automatic pitch-labeling process comes from behavioral studies, in which the auditory stimuli had to be attended to solve the task. For instance, individuals with AP performed poorer in auditory Stroop tasks when they heard sung tone names and were instructed to repeat the syllable while ignoring the pitch it was sung in (Miyazaki, 2004; Itoh et al., 2005; Schulze et al., 2013). AP also hindered performance in a RP task, in which participants had to compare a visual notation with the auditory presentation of a melody (Miyazaki and Rakowski, 2002). Further evidence for the automaticity of pitch labeling was provided by neuroscientific studies that observed differential electrophysiological or hemodynamic responses in AP musicians during attentive listening (Zatorre et al., 1998; Itoh et al., 2005). Contrary to these studies, in the present study, participants were instructed to focus their attention on a silent film and to ignore the auditory stimuli altogether. AP musicians can label tones fast and effortlessly, but they may not necessarily do so under all circumstances. Apart from the specific task, also other situational factors like stress and fatigue might influence pitch-labeling performance and pitch-labeling automaticity. Additionally, it is also possible that there are considerable interindividual differences in the level of automaticity of AP per se. Future studies will hopefully uncover the role of such influences on this extraordinary ability and its neural underpinnings in more detail.

Although this study could not demonstrate a cognitive facilitation in AP during passive listening, we believe our results do not challenge existing cognitive theories of AP, like the two-component model (Levitin, 1994). The two-component model focuses on the use of long-term pitch memory representations and their association with labels in AP. This mechanism in turn poses less demands on working memory in some tasks than using RP (Klein et al., 1984; Itoh et al., 2005; Schulze et al., 2009). In contrast to these mnemonic processes, the P3a in passive auditory oddball paradigms is mostly associated with attentional processes, which are not explicitly postulated as part of AP by the two-component model. Further research should be undertaken to determine the influence of attention on pitch processing in AP.

We attempted a direct replication of the original study, still there are some mentionable differences between the original and the replication study that might have influenced the results. While questionnaires on musical experience and the pitch-labeling test were assessed with paper-pencil in the original study, we used online questionnaires and an online pitch-labeling test in the present study. Because our participants underwent an extensive test protocol in the context of the larger AP project spanning several days during which they participated in various (f)MRI and EEG experiments, we tried to keep the travel burden for them as low as possible by providing the opportunity to work on several tests at home. For our statistical analyses, we used the software R instead of SPSS, and we performed Welch’s t tests instead of Student’s t test because they are more robust for groups with unequal sample sizes (Ruxton, 2006; Delacre et al., 2017). For ANOVAs, we reported generalized eta-squared instead of partial eta-squared as recommended by Bakeman (2005). Like in the original study, groups were defined based on self-report. Contrary to the original study, in our replication study, the non-AP musicians performed above chance in the pitch-labeling test. Accordingly, it could be argued that the groups were less homogenous than in the original study and that this is the reason for the unsuccessful replication. However, because trials in the pitch-labeling test lasted 15 s instead of 5 s, participants probably had enough time to employ RP strategies in our test. It can be expected that highly-trained musicians perform above chance levels when given the opportunity to use RP strategies. For the same reason, it is possible that the pitch-labeling performance of AP musicians was also overestimated. The longer maximal trial duration was due to the online implementation of the pitch-labeling test. In a pilot study, we tested a version with the original trial duration of 5 s, which turned out to be very demanding and difficult to solve even for AP musicians because of the multiple-choice format with 36 answer options. We would recommend future studies to measure reaction times in pitch-labeling tests to be able to better disentangle the effortless and fast AP strategy from the slower RP strategy, or to apply a pitch-labeling test that impedes the usage of RP strategies (e.g., as suggested in Wengenroth et al., 2014). Yet, it still remains unclear which is the best way to objectively identify AP ability and if it is even possible to do so, a question that has been asked frequently and was also discussed in an early influential review on AP (Takeuchi and Hulse, 1993). The authors addressed several methods to quantify AP, ranging from producing tones to different variants of pitch-labeling tests. Up to date, the pitch-labeling tests applied in AP research differ considerably in procedure (e.g., trial duration, answer registration, sine tones/instrumental tones), the number of used tones, and the presentation technique (e.g., online vs lab). Most importantly, no specific cutoff has been established to distinguish AP from non-AP possessors. Thus, in the present study, the pitch-labeling test only served as a validation tool. For group assignment, we relied on self-report since only the participants themselves can judge whether they possess the ability to employ AP strategies. In addition, as demonstrated in the exploratory subgroup analyses, the conclusions of the results remained the same even when just considering participants with the lowest and highest pitch-labeling scores, suggesting that this sample difference between studies did not cause the absence of the AP effect. Similarly, conclusions about the P3a amplitude did not change when just looking at the female participants. Thus, although the original study was less balanced in terms of gender than the present study, the absence of an effect of AP on the P3a amplitude in the present study does not seem to be caused by gender distribution differences between studies. Also, according to current scientific understanding gender differences in neuroscientific cognitive studies are most often due to small sample sizes and should only be interpreted when the influence of hormonal levels was controlled for (Jäncke, 2018). It should also be mentioned that in the present study, the AP and non-AP musicians showed a statistically significant, albeit small in absolute terms (less than three points out of 80 possible points), difference in musical aptitude (AMMA). However, scores are comparable to those reported in the original study, and additional covariance analyses with the AMMA score as covariate showed the same results as the replication analyses.

Finally, it is important to note that a single replication study can never conclusively confirm or disconfirm previous findings. Nevertheless, our results cast reasonable doubt that there is cognitive facilitation in AP during passive tone processing as indicated by the P3a. The more so since our sample was four times the size of the original study, and Bayes factors analyses provided evidence that the proposed effect does not exist. Although it is possible that additional factors we did not control for moderated the effect, we reduced such moderators to a minimum by doing a direct replication. Thus, if an effect of AP on the P3a really exists, its true effect size is probably much smaller than reported in the original study as it is not reliably detectable in a large sample, and its generalizability might be limited.

Considering the large effect size obtained in the original study, the results of the current study demonstrate that only through replications a better estimate of the true effect can be obtained. We believe replications are desirable in science in general and particularly in research fields that are prone to false-positive results and to overestimations of effect sizes due to small samples. Neuroscientific studies often use small samples because of the high financial costs and time-consuming data acquisition and analysis. Collaborative efforts between multiple research groups are suggested as a means to recruit larger sample sizes.

In summary, our direct replication of Rogenmoser et al. (2015) successfully replicated the non-significant results for group differences in the MMN. In contrast, we did not replicate the finding of smaller P3a amplitudes in AP musicians. Taken together, our study does not support electrophysiological differences between AP and non-AP musicians during passive listening. It is conceivable that the different pitch-processing modes of AP and RP can only be reliably distinguished either with more sensitive measures or in more attention-engaging tasks. In more general terms, the results of the present study underline both the importance of replications and of larger sample sizes in neuroscientific research.