Background
Autism spectrum disorders or conditions (henceforth “autism”) are more common in people with mathematical [
1], visuospatial [
2], musical [
3], or “savant” abilities [
4], e.g., rapid mental mathematical calculation [
5,
6], calendar calculation [
7], or extreme memory [
8,
9]. Autism, a set of neurodevelopmental condition, is characterized by social and communication difficulties, alongside unusually repetitive behaviors and unusually narrow interests [
10], sensory hypersensitivity, and difficulties in adjusting to unexpected change (DSM-5, APA 2013).
Absolute pitch (AP), the ability to name or produce a musical tone without the use of a reference tone [
11], is a common special ability in professional musicians with a proportion of up to 7–25% [
12‐
14] but less than 1% [
15] in the general population. AP is an excellent model for the investigation of a joint influence of genetic and environmental factors on the brain and on human cognitive abilities [
16]. Even if the ability is relatively rare, various studies suggest that the ability might be distributed more gradually than expected [
17‐
19]. Partial AP ability seems to be common in professional musicians, who might in conjunction with good relative pitch strategies (interval judgements) yield moderate to good results in absolute pitch tests [
18]. An influence of age of onset of musical training [
20‐
22], ethnicity [
12,
14,
22], and type of musical education (label to fixed pitch vs. label to interval, unfixed to pitch) techniques [
12]) suggest environmental aspects in the acquisition of AP. In contrast, AP often clusters in families, genetically overlaps with other familial aggregated abilities (e.g., synesthesia [
23]), and has a higher proportion in autistic people [
3,
7,
24‐
29] and in Williams syndrome [
30,
31], both strongly genetic conditions [
32‐
39]. Remarkably, some studies on musically untrained children with and without ASC could even show increased long-term memory for pitch [
3,
27,
28] in ASC children. Finally, a sensitive or critical period before the age of seven is considered due to the importance of the early onset of musical training [
14,
16,
20,
40‐
43]. Relative pitch abilities, i.e., the ability to perceive equal intervals between musical tones as similar and to be able to judge the pitch height of tones relative to each other, is very common in the general population. RP abilities also show high variability in absolute pitch possessors [
18,
44‐
46]. Musically trained people, however, often exhibit more explicitly developed RP abilities (e.g., verbal labeling of musical intervals to a similar proficiency as tone labeling of absolute pitch possessors) than less musically trained or musically untrained people [
18].
Recently, two studies have given evidence for heightened autistic traits in musicians with AP [
47,
48]. Both AP and autism are associated with similarly altered brain connectivity in terms of the relation between hyper- and hypoconnectivity [
41,
49‐
58]. The theory of veridical mapping [
7] tries to explain absolute pitch, synesthesia, and other abilities like hyperlexia, frequently seen in autistic people or in savant syndrome, with the neurocognitive mechanism of associating homolog patterns of two perceptual or cognitive structures (veridical mapping). According to this framework, an enhanced low-level perception [
59,
60] and an increased ability to detect patterns (“systemizing” [
61]) are associated with regional hyper- as well as global hypoconnectivity in absolute pitch [
49,
51,
62‐
67] and autism [
50,
52,
54,
68]. It is also noteworthy that autism and abilities like absolute pitch share excellent attention to detail [
40,
69] and a shift in the direction of higher segregation with reduced integration in the brain [
69]. Investigating disconnection syndromes or integration deficit disorders, as well as phenomena with similar brain network characteristics, may therefore provide insights into the variability of brain network structure and function and its relation to perception, cognition, and behavior.
The present study tests if and to what extent AP and autistic traits share the same neurophysiological network connectivity. To our knowledge, this study is the first to investigate (1) the relation of pitch adjustment ability (active absolute pitch; in contrast to (passive) pitch identification) and brain as well as behavioral correlates; (2) the relation of AP ability, autistic traits, and functional brain connectivity within one study; and (3) graph theoretical network parameters in AP during resting-state electroencephalography. We used graph theoretical analysis [
70,
71] of resting-state EEG data to estimate differences in global network structure of the brain. We analyzed three graph theoretical network parameters reflecting segregation (average clustering coefficient) and integration (average shortest path length) and the so called small-worldness (a combination of clustering and path length) [
70,
71]. To our knowledge, this is also the first study investigating the global average connectivity parameters over the whole brain between AP and RP (relative pitch) musicians, while prior studies [
49,
51] have focused on parameters for single regions (e.g., degree, single node clustering, and single node characteristic path length). We expected higher autistic traits, higher path length (reduced integration), and lower clustering (underconnectivity) for AP and an interrelation among those variables. Further, we expected these differences to specifically occur in low- (delta, theta) vs. high-frequency (beta) ranges for integration vs. segregation, respectively.
Discussion
The results of the present study underline a possible interrelation between autistic traits, brain connectivity, and absolute pitch ability. We investigated the EEG-resting state connectivity using a graph theory approach in professional musicians with and without absolute pitch, the Autism Spectrum Quotient [
80], and a test of pitch naming and pitch adjustment ability. The analyses revealed higher autistic traits, higher average path length (delta 2–4 Hz), lower average clustering (beta 13–20 Hz), lower small-worldness (gamma 30–60 Hz), and a tendency for an earlier start of musical training in absolute pitch musicians. Furthermore, pitch naming was well predicted by autistic traits, path length, and clustering values, explaining a total of 44% of the variance. Pitch adjustment (i.e., active absolute pitch) was explained by the same predictors plus the age of the beginning of musical training summing up to an
R2 = 0.38. However, in the latter case, the starting age of musical training and path length remained marginally significant.
It is noteworthy that the start of playing a musical instrument in our models did not significantly improve the prediction of AP performance but only in pitch adjustment. Furthermore, the total amount of musical training during life neither was predictive of any AP performance in the general linear model, nor did show a group difference. The typical human brain exhibits a small-world-like structure with a much higher clustering compared to a random network, while maintaining an efficient information transfer and low wiring cost through an equally low path length [
70,
103,
107]. In this context, the results of the present study indicate a less efficient and less small-world structured functional network in AP compared to RP, in line with the structural results of Jäncke et al. [
49] and the results from the autism research [
52,
53,
56,
100,
113] but extends the results to EEG functional connectivity networks.
It is further interesting that both correlations and regressions between autistic traits and the two AP tests show higher correlations and better prediction of pitch naming than pitch adjustment by AQ. This can be explained by the aforementioned theory of veridical mapping [
7,
69]. This framework explains savant abilities and other unusual abilities in autism by their common characteristic of one-to-one mappings between elements of two conceptual or perceptual structures (e.g., letters-musical tones, letters-colors). According to this theory, all of these abilities share further commonalities including hyper-systemizing [
61]and enhanced perceptual functioning [
59,
60], and depend on the exposure to material, and—if they occur as autistic savant ability—the related elements can also be recalled without a strategy [
7,
69]). This explicit recall in the absolute pitch, i.e., the naming of the pitch, therefore might be a more savant-like ability, leading to a higher correlation with autistic traits.
Furthermore, we observed a reduced connectivity for AP compared to RP in interhemispheric connections when compared to the participants own distribution of connectivities (z-standardized calculation)—especially between the left auditory-located electrodes and various right temporal, parietal, and frontal electrodes.
While higher path length in low frequency bands (delta, therefore reduced integration) and lower clustering in higher frequencies (beta, reduced segregation on sensor level) are in line with our a priori hypotheses, we did not expect the reduced small-worldness within the gamma band for AP compared to RP (found during eyes closed). Nevertheless, this result can be explained by previous research findings: Cantero et al. [
114] reported an increased gamma band measured by intracranial electrodes between hippocampal areas and neocortex in humans during wakefulness but not during sleep, pointing to a relation of gamma band couplings and awareness states in humans. This also suggests that gamma band activity, probably useful for the storage and retrieval of memory [
115‐
117] and binding of perceptual features [
116,
117], might even play a role during resting (awake more than asleep) states. AP ability, similarly, is often described as the ability to associate tones and verbal labels in a stable, hyper-memorized way, pointing to the importance of long-term memory processes [
118‐
122]. Furthermore, Bhattacharya et al. [
123,
124] found increased long-range gamma synchronization between distributed cortical areas during music listening in musicians compared to non-musicians, which might reflect musical memory and binding of musical features. In contrast, Sun et al. [
125] found reductions in gamma-band phase locking and power in participants with autism associated with perceptual organization tasks (visual), while Brown et al. [
126] found higher gamma peaks in response to illusory figures in autism. Generally, abnormal gamma activity is found in a range of neuropsychiatric disorders, with reduced gamma in negative schizophrenic symptoms, Alzheimer’s disease, and task-specific gamma decrease in autism, but an increase in gamma in ADHD, positive schizophrenic symptoms, and epilepsy (for a review see [
127,
128]). Thus, the results of reduced small-worldness in AP are in line with an integration-deficit hypothesis of AP, both in perceptual organization and binding of musical stimuli and in brain connectivity, which is again similar to autism (see [
50,
52,
129‐
132]). However, the findings in gamma band did not show correlations with autistic symptoms.
Our results replicate the results of Dohn et al. [
47] showing higher autistic traits, which reached significance in the subscales “imagination” (similar to [
47]), “attention to detail” (marginally), and “social skills” (marginally). Furthermore, autistic traits were also correlated not only to pitch naming as already shown by Dohn et al. [
47], but also to pitch adjustment accuracy (MAD, mean absolute deviation to target tone in cent; 100 cent = 1 semitone) and adjustment consistency (SDfoM, pitch template tuning). However, similar to [
47], the group mean autistic traits did not reach the cutoff for diagnostic relevance, indicating a high variability regarding autistic traits even in the AP group (with seven AP compared to one RP scoring above cutoff or borderline). This fits with the analyses of the broader autism phenotype [
133] and might implicate joint as well as divergent phenotypic and endotypic characteristics of AP and autism.
In contrast to our study, various previous studies have shown an influence of the start of musical training in AP, making the onset of training before the age of 7 necessary, but not sufficient to acquire absolute pitch [
12,
16,
20‐
22,
41]. For example, Loui et al. [
41] recently found that early onset of musical training was associated with an enlarged tract between pSTG and pMTG in the left hemisphere, but the degree of AP proficiency still correlated with the size of the tract after partialling out the age of onset. Gregersen et al. [
12] further analyzed the familiar aggregation of AP in different samples of musicians and non-musicians with early and late onset of musical training comparing different types of musical education and found no general differences of AP between early or late starting siblings of AP. Their results further indicated a higher influence of genetic disposition and the type of education used, which both had a more pronounced influence than the age of onset per se [
12].
Higher average path length (delta 2–4 Hz)), lower average clustering (beta 13–20 Hz), and lower small-worldness (gamma 30–60 Hz) for AP compared to RP are also in line with previous studies showing structural local hyper- vs. global hypoconnectivity in AP [
49] and reduced clustering and higher path length in participants with autism [
113,
134]. In contrast, Loui et al. [
51] reported overall increased degrees, clustering, and local efficiency coefficients of functional networks in AP using fMRI during music listening and rest. The authors further speculate that there might be a “dichotomy” between the structural and functional hyperconnectivity in AP, where the structure is locally hyperconnected but the function is globally hyperconnected [
51]. The present study, however, provided more evidence for an also functionally underconnected brain in AP musicians compared to relative pitch musicians. Diverging results compared to Loui et al. [
51] might be due to differences in methods (EEG vs. fMRI) or different definition of nodes (electrode positions vs. brain regions) and edges (wPLI vs. functional correlations).
Differences seen in single connection analysis might reflect the connections that lead to differences in clustering values described above. Similarly to the prediction of clustering by AP and autistic traits, single connection differences in the beta range are in line with the findings from the autism literature: First, various others have reported reduced interhemispheric connectivity in autism [
56,
112,
113,
135,
136]. Second, hypoconnectivity between the left FT7 (BA:22) and right frontal-temporal-occipital electrodes (F8, T8, TP8, P8, P4; BA:45/47, 4, 21/22/20/37, 19/37, 39/7/40/19; see Table
4 for the anatomical interpretation of electrode positions) might reflect a specific underconnectivity between the left STG and right IFOF, of which alterations have already been described in both AP [
137] and autism [
138]. Especially reduced interhemispheric connectivity between the left auditory-related cortex and right IFOF might reflect autism-like personality traits and perception of (some) absolute pitch possessors. The IFOF, especially the right IFOF, has been shown to play an important role in music perception and the integration of musical features, as it connects various brain regions from the frontal over temporal to posterior parts of the brain [
139]. A reduced white matter integrity of IFOF was found in amusics [
139,
140], whereas people with synesthesia and absolute pitch were shown to have a higher IFOF integrity [
67,
137]. More importantly, however, increased interhemispheric connectivity in musicians was found by several studies [
141‐
145] showing the importance of interhemispheric integration in music perception. A reduced interhemispheric functional connectivity, especially between bilateral auditory regions as found in the present study, perhaps might result in less perceptual integration of musical features (i.e., auditory weak central coherence) and hence a more detail-oriented processing of music and musical pitches (i.e., absolute vs. relative) in those participants. An exaggeration of those features might also lead to symptoms of amusia, which has also been associated with alterations in the left and right STG and right IFOF [
139,
140,
146] and with autism [
147]. However, it must be clearly said that we cannot explicitly conclude the anatomical differences from connectivity differences on the sensor levels. Further structural or functional studies using methods with high anatomical precision have to be conducted to evaluate this hypothesis.
Limitations
Some caveats of the present approach are warranted. First, we did not use a source-based approach of functional connectivity, making conclusions with respect to anatomical associations of the obtained differences very speculative. Second, various different configurations of local and global hyper- vs. hypoconnectivity can be assumed to result into the same averaged network measures; therefore, no conclusions can be made about the exact relative structure within the brain and among different regions. Nevertheless, higher path length (EC, delta 2–4 Hz) can be interpreted as weaker integration in the network and higher clustering (EO,13–20 Hz) as higher local segregation of functions [
95] and therefore might again reflect a local hyper- over global (integrative) hypoconnectivity in the brain of AP musicians. This interpretation is further encouraged by studies showing that long-range connectivity (integration) is more reflected in low frequency bands, whereas short-range connectivity is more in high frequency bands [
110,
148]. This again fits to the results of our study, as higher clustering, indicative for local segregation, was found in the beta range and path length—indicative for global integration in the network and therefore long-range associations—in the delta range.
Furthermore, evidence for a higher proportion of AP among people with autism or Williams syndrome, as mentioned in the introduction, is currently mainly based on case studies and case reports, as systematic epidemiologic studies have not been conducted yet and studies with respect to Williams syndrome are rare. Therefore, the actual co-occurrence between the phenomena remains to be evaluated. On top of that, the R2 values for predicting brain connectivity by AP and autism seem comparably weak. Mediation analysis did not reveal a mediating influence of autistic traits on the relation between absolute pitch and brain connectivity (both directions). However, the nature of altered brain connectivity is a common phenotype for numerous neuropsychiatric disorders and phenomena, not just autism and absolute pitch. Therefore, we have to admit that a range of other factors influencing brain connectivity would have been necessary to get a more detailed insight into what influences brain connectivity and vice versa. Most likely a relation between autistic traits and absolute pitch is only true for a subgroup of absolute pitch possessors. Bigger sample sizes are necessary to investigate this hypothesis.
In addition, significant group differences were highly selective for certain frequency bands, states (EO vs. EC), and thresholds. Nevertheless, we can rule out the possibility that we obtained those differences by chance. First, there were significant differences for at least one threshold in a frequency band, and effect sizes of the other thresholds in the same frequency band never (exclusive: Crand EO alpha) indicated reverse effects (see color code in Fig.
1). Second, we did only consider differences relevant if at least two neighboring thresholds exhibited a significant group difference. Third, the three network parameters selected via group differences always could also predict AP performance with a reasonable high
R2 and/or showed bivariate correlations with AP performance in both tests of AP.
Conclusion
For the first time, we included a pitch adjustment test of active absolute pitch [
79] into a study on brain connectivity in AP, so we are not only referring to pitch naming as were previous studies [
41,
47,
49,
51]. Also, whereas Jäncke et al. [
49] were using structural cortical thickness covariations and Loui et al. [
51] functional correlations of fMRI activity (during rest and music listening) as weights for connections in graph analysis, we for the first time applied graph theory on the resting-state EEG connectivity of AP musicians, both in eyes closed and eyes open conditions. This is similar to methods used in analyzing brain connectivity in autism [
57,
113]. Finally, while Elmer et al. [
118] used phase synchronization as an estimate for functional EEG connectivity, we used wPLI (weighted phase lag index, [
85]), which is less contaminated by volume conduction [
85‐
88,
91], thus contributing to a higher validity and reliability with respect to true brain connectivity and graph theoretical parameters [
89,
93,
94].
In summary, differences in network and connectivity analysis in the beta band seem to be specifically associated with the relation of autistic traits and absolute pitch, whereas path length in delta range and small-worldness in gamma range might reflect other influences on the acquisition of the ability (e.g., environmental factors, genetic factors not attributable to autistic traits, musical education method, instrument, learning, sensitive periods). To our knowledge, this is the first study to combine measures on autistic traits and brain networks on musicians with and without absolute pitch. We conclude that this is further evidence showing that both AP and autism have shared and distinct neuronal and phenotypic characteristics. This might also be reflected in subgroups of AP with different genesis, providing new arguments for the discussion about a dichotomous or continuous view on AP. However, the causal relationship between AP, autistic traits, and brain connectivity remains to be evaluated.