Music Perception in Dementia

INTRODUCTION

Despite much recent interest [1–3], the impact on music processing of Alzheimer’s disease (AD) and other dementias has not been fully defined. Music is first and foremost a complex acoustic phenomenon, and the perception of music requires the parsing of a musical stimulus of interest against the acoustic background (musical scene analysis: [4]), representation of the musical source (instrumental or vocal timbre), and tracking of pitch (melody) and temporal (rhythm, meter) information to create a coherent musical ‘object’ [5]. This formulation suggests that music presents the brain with a difficult problem of auditory information processing, entailing the decoding of a number of perceptual and cognitive modules [6, 7]. On both computational and neuroanatomical grounds, these processes are likely to be vulnerable to the effects of neurodegenerative diseases, most notably AD and primary progressive aphasia syndromes that target peri-Sylvian cortex (progressive nonfluent aphasia (PNFA) and logopenic aphasia (LPA): [8–11]). A substantial body of structural and functional neuroimaging work in the healthy brain and in patients with focal brain lesions has delineated distributed cortico-subcortical networks that analyze the dimensions of music [6, 13]: these networks closely overlap the networks targeted in canonical dementia syndromes [14, 15]. However, to date most studies of music in dementia have focused on the interaction of music and memory [16–18], preserved abilities in trained musicians developing dementia [16, 20], and potential benefits of music more widely in dementia [21–25].

Aside from its intrinsic interest, music is an attractive candidate paradigm for assessing the processing of complex information streams or patterns in both the healthy and the diseased brain. In the domain of musical pitch, patterns of pitch change can be analyzed at two levels: pitch interval (the magnitude of change between consecutive notes) and pitch change direction (the overall pattern of ‘ups’ and ‘downs’ comprising the contour of the melody: [7, 26]).By analogy with the visual domain, pitch interval and melody contour entail the processing of ‘local’ and ‘global’ pitch pattern information, respectively; according to this formulation, pitch intervals can be considered fine-grained musical features while combining these intervals to create a melody contour can be considered an overall (global) ‘gestalt’ of the musical piece. The distinction between these levels is evident in everyday music listening; changing individual pitch intervals is often perceived as a jarring distortion to the musical line, whereas simultaneously changing all pitch intervals but maintaining the relations between them (as in transposition of a melody to another key) retains the same musical gestalt (the tune is still recognizably the same). The concept of local versus global processing levels is fundamental for understanding how percepts are organized and relevant to many sensory domains. Music can be considered a non-visual test case for assessing the generality of effects on sensory information streams and the relative impact on featural (local) versus gestalt (global) perception of clinical disorders such as the dementias.

The local (pitch interval) and global (melody contour) levels of music perception can be differentially affected by focal brain lesions distributed between the cerebral hemispheres [26–30]. Functional neuroimaging studies in the healthy brain have demonstrated separable mechanisms in posterior superior temporal lobe and parietal and prefrontal projection pathways for the processing of pitch interval and melody contour [30, 31]. Available evidence suggests that the decoding of musical patterns may be affected by common dementias and may help to stratify dementia syndromes and pathologies without relying on more specialized (and potentially confounding) verbal mechanisms. Elementary pitch discrimination may be retained in AD and impaired in PNFA, consistent with relatively greater involvement of early auditory areas in neurodegenerative processes that target peri-Sylvian cortex [18, 32–36]. However, the effects of these diseases on more complex pitch pattern processing have not been resolved. Studies in the visual domain suggest that patients with AD may have disproportionate difficulty in the analysis of global structure with relatively intact analysis of local features: this profile is likely to reflect dysfunction of integrative mechanisms in parietal cortex that are particularly targeted by AD pathology but may be more difficult to interpret in the context of associated executive, verbal or spatial deficits [37–43]. In the musical domain, it follows that AD should produce more severe impairment for processing global (melody contour) than local (pitch interval) patterns; whereas in PNFA, a more pervasive impairment of local and global pitch pattern processing would be anticipated. However, currently available neuropsychological instruments for assessing pitch pattern processing often rely on comparisons between paired musical sequences [44]. Such comparisons are vulnerable to concurrent auditory working memory deficits that accompany AD and the progressive aphasias [17, 45–47]; moreover, the explicit serial comparison of sequential melodies is seldom required in everyday music listening. Whereas specific musical working memory systems are likely to be integrally linked to the perception of pitch and temporal patterns in music, these are separable from verbal and other working memory systems that might be generically involved in any auditorytask [48–50].

Temporal patterns in music can similarly be represented at interval (rhythmic, local) and longer duration stress or accent (metrical, global) levels of analysis [51]. Deficits in these dimensions of musical temporal perception occur with focal lesions involving temporal and parietal cortices [52–55] but frequently dissociate from pitch impairment [26, 56] and further dissociate from each other [52, 57]. In the healthy brain perceptual analysis of rhythm and meter engages cortico-subcortical circuitry jointly involved in preparing motor output [58–61]. While evidence in AD is not conclusive [19, 63], impairments of temporal pattern processing have been described in PNFA associated with involvement of peri-Sylvian cortex [64]: this may be attributable both to loss of dynamic precision mediated by the dominant hemisphere and the high temporal resolution required for accurate processing of speech signals[65, 66].

Under most circumstances, the listener must simultaneously decode more than one stream of musical information (whether produced by an ensemble of instruments or a single instrument played polyphonically). Such an analysis is fundamental to the initial parsing of a musical ‘scene’, before more detailed analysis can occur [7]; it is likely to entail an interaction of bottom-up mechanisms for coding perceptual structure with top-down mechanisms for resolving perceptual ambiguities based on stored templates or schemas derived from past experience of music [5, 67].Musical scene analysis has not been widely studied neuropsychologically in clinical populations but is likely to engage posterior superior temporal and parietal lobe regions and their dorsal projections [68–72]. AD has been shown to produce a generic impairment of auditory scene analysis under diverse listening tasks and conditions, including the streaming of sound sequences that bear some similarities to musical melodies; this has been linked to dysfunction of posterior temporo-parietal areas overlapping those involved in music perception [33, 73–75]. On both neuroanatomical and neuropsychological grounds, patients with AD might therefore be anticipated to have difficulties with musical scene analysis; however, this has not been addressed directly in previous work.

In this study we assessed the perceptual components of music processing systematically in a cohort of patients representing major dementia syndromes. Based on the above synthesis of the available literature in both the auditory and visual domains, we anticipated that global versus local levels of musical pitch and temporal information processing and the effect of presenting a melody against a musical background (i.e., processing of musical ‘scenes’) would be the most informative components of music perception to target in the principal neurodegenerative dementias. The framework we addressed in designing the experimental music perception battery is outlined in Fig. 1, adapting the modular model of music cognition proposed by Peretz and Coltheart [7]. We studied patients with typical AD in relation to patients with a syndromic diagnosis of primary PNFA and patients with the LPA clinical variant presentation of AD pathology. Inclusion of these syndromic groups allowed us to assess the effects of disease topography in dominant peri-Sylvian cortex in relation to the predicted underlying molecular pathology (PNFA in relation to LPA). We designed novel neuropsychological tests requiring continuous tracking of musical patterns and detection of deviants from the established pattern in the domains of pitch (interval, melody) and time (rhythm, meter). Our rationale was that detection of a deviant or ‘wrong’ note played during a performance more closely approximates natural music listening than does sequential comparison of melodies or related neuropsychological procedures and also reduces working memory and associated, extraneous executive demands. In addition, we created a test to assess detection of melody patterns within a musical ‘scene’. These dimensions of perceptual pattern processing were assessed in relation to detection of timbral deviants (a measure of sustained auditory attention and executive processing of sound sequences) and recognition of familiar tunes (a widely used index of musical semantic processing). To allow musical perceptual effects to be interpreted without potentially confounding effects from auditory working memory impairment, we controlled for this factor in analyzing the musical performance profiles of our patient groups: our concern here was to adjust for generic, task-related auditory working memory capacity rather than any more specifically musical working memory subsystem.

In line with previous evidence including studies of the healthy brain and focal brain damage, we hypothesized that musical deficits would be produced by all three target dementia syndromes, with distinctive profiles of impairment in each syndrome. More specifically, we hypothesized that typical AD would be associated with relatively greater impairment of global than local levels of musical pattern analysis and impaired musical scene analysis, with a similar profile of deficits in LPA; while PNFA would be associated with deficient analysis of both local and global pitch patterns but with more severely impaired analysis of temporal patterns in music.

METHODS

RESULTS

DISCUSSION

Here we have shown that canonical dementia syndromes of typical AD, LPA, and PNFA may be associated with profiles of impaired music perception relative to healthy older individuals. Deficits exhibited by the present syndromic groups affected the analysis of pitch pattern and were not simply attributable to prior musical expertise, general cognitive, elementary perceptual, or task factors. After taking general auditory working memory performance into account, detection of acoustic deviants (indexed by varying note timbre) was comparable to healthy controls in all syndromic groups. Patients’ performance on pitch pattern analysis tasks deteriorated with increasing perceptual difficulty (as indexed by the more versus less difficult versions of the pitch pattern tests), consistent with a true deficit of pitch pattern processing. Patients with typical AD had impaired processing of global pitch (melody contour) information but (after accounting for general auditory working memory capacity) intact processing of local pitch (interval) and temporal pattern, as well as intact tune recognition whether in isolation or within a polyphonic ‘musical scene’. Patients with LPA (a syndrome generally underpinned by AD pathology) showed a similar profile with predominant impairment of global pitch processing, albeit the evidence of impairment was most apparent in the more demanding processing of direction-only contour variation (produced by deviance in the ordering of the same note sequence). In contrast, patients with PNFA exhibited deficits affecting local (interval) as well as global (melody) information in pitch patterns but (again, after controlling for general auditory working memory capacity) performance that did not reach statistical significance when compared to control processing of temporal pattern, tune recognition and musical scene analysis.

These findings are broadly consistent with a modular organization of music cognition, as previously proposed [7]. More specifically, the relatively greater impairment of global than local pitch pattern analysis in typical AD and the similar profile in LPA corroborate our experimental predictions. Impaired global processing of pitch information in music is in line with other evidence for defective formation of coherent global stimulus representations in AD: this deficit might reflect increased demand for coordinated integrative computations between temporo-parietal association cortices vulnerable to AD pathology [81–85], though any disadvantage with respect to the coding of local stimulus features is likely to be relative rather than absolute [37–40, 86]. While the processing of global stimulus characteristics unfolding over longer time windows requires attentional resources [40, 86], it is unlikely that the profile of pitch deficits here was entirely underpinned by attentional compromise: as our paradigm required a single response to consecutively presented stimuli, it is unlikely to have taxed divided attention, while demands on sustained attention are likely to have been similar in the timbral deviance detection task, on which the present AD group performed normally. Moreover, pitch pattern deficits in our patient groups were documented after taking auditory working memory capacity into account. On the other hand, the present data suggest any claim that dementia syndromes differentially affect particular components of music cognition must be qualified. Syndromic profiles were documented in the context of wide individual variability (Fig. 3). Moreover, across the patient cohort, correlated performance was observed for processing local and global information and pitch and temporal patterns. As the neural mechanisms mediating different components of music perception are likely to be affected together by the spreading neurodegenerative process, the finding of correlation (or absence of differential impairment) in this setting cannot be used to draw inferences about the underlying cognitive architecture.

Our findings provide further evidence that LPA and PNFA have associated phenotypes of nonverbal auditory impairment [32, 87–89]. The musical phenotype was more severe in the PNFA group here; the involvement of pitch pattern analysis in this syndrome is in line with previous work [32] and suggests a putative mechanism linking generic mechanisms of dynamic auditory encoding with speech production via the dorsal auditory cortical pathway, extending over a range of timescales relevant to processing of individual and sequential speech sounds [9, 90–92], Marked involvement of musical perceptual mechanisms might be anticipated from the severe and focal involvement of auditory association areas in the progressive aphasias [9, 11]. Although we did not demonstrate a correlation of musical measures with standard measures of verbal encoding, pitch processing mechanisms are likely to be more relevant to prosody (a crucial non-linguistic attribute of speech signals) than phonemic sequencing, at least for non-tonal languages. Both perception and production of prosody are abnormal in PNFA [89], raising the possibility of a common mechanism linking musical pitch encoding with the programming of pitch variations in speech.

Allowing for the relatively small cohorts here, the present data offer relatively little support for specific musical signatures of particular dementia pathologies: when syndromic groups were directly compared, no measures indicated robust differences. Our findings suggest that certain musical perceptual attributes such as melody (pitch contour) tax neural computational resources across dementia syndromes; the data do not suggest any simple dichotomization of dementias according to whether they degrade or spare the perception of music. Though the overall profile of pitch pattern deficits suggested some selectivity for particular syndromes (predominantly affecting global pitch characteristics in AD and LPA and more widespread in PNFA), any syndromic effects were relative rather than syndrome-specific. An important theme emerging from this study is that auditory working memory deficits are likely to amplify any purely musical deficits (compare Fig. 3 prior to adjustment for this factor and the unadjusted analysis summarized in Supplementary Table 4 with the adjusted significance attributions in Table 2): patients presenting with impairments of music processing may be comparably impaired on processing of other extended auditory information streams. The extent to which musical deficits reflect music-specific processes might then depend on the nature of the interaction between auditory working memory and the relevant musical characteristic, as suggested by previous work [48–50]. This factor may partly explain the lack of evidence here for specific deficits of temporal pattern processing from music, which we anticipated particularly in the PNFA group [64]. While in principle this could also reflect the small study cohort or failure to sample relevant temporal windows (as temporal characteristics of music are less constrained than pitch variations), temporal analysis of musical sequences may be more intimately reliant on auditory working memory capacity than pitch analysis; moreover, the linkage between temporal analysis and working memory mechanisms may have a neuroanatomical substrate (including insular cortex) that is targeted in PNFA [93]. This is a difficult issue to resolve, as particular subsystems of working memory are likely to be music-specific [48–50]. On the other hand, it has been shown that musical listening tasks also engage domain-general working memory circuits [94]. In this study, we set out to adjust for a general index of (verbal) auditory working memory capacity that might affect performance on auditory tracking tasks; however, the relative effects of music-specific and music-independent buffer systems on the perception of musical structure will only be resolved by assessing indices of musical pitch and temporal short-term memory directly alongside standard working memory measures. Even if underpinned by separable neural substrates, music-specific and music-independent working memory systems may be affected together in neurodegenerative disease.

Perhaps more surprisingly, we found no evidence for a specific deficit of musical scene analysis (as indexed by the processing of polyphonic melodies) in any patient group. This contrasts with previous work characterizing a generic impairment of auditory scene analysis in AD [33, 72–75] and may in part reflect the relatively wide variation in healthy control performance on our tune streaming test (Fig. 3). However, it is also possible that the analysis of musical scenes benefits to a greater degree than other kinds of auditory scenes from the availability of stored templates, here familiar tunes. If (as the present data also suggest) recognition of familiar tunes is relatively preserved in these dementia syndromes, patients may be able to engage ‘top-down’ mechanisms for parsing the musical scene even despite degraded mechanisms of early scene analysis [95, 96].

The present findings have certain practical and clinical implications. Deficits of pitch pattern analysis here were demonstrated using stimuli that required tracking of musical information over time. Conventional neuropsychological (including music psychology) procedures that assess discrete stimulus tokens presented in isolation may not fully capture information processing deficits in dementia, particularly earlier in the disease course. Novel neuropsychological instruments that require on-line tracking of information streams could be relevant for assessing the encoding of verbal as well as musical sequences in these diseases. Though conclusions must be qualified pending further detailed investigation, our findings suggest that particular musical attributes (such as rhythm) might be used as a vehicle for designing musical interventions in at least some patients with dementia. At present, formal trials of music therapy in dementia often yield disappointing results despite anecdotal reports of benefit [97]. Targeting of those musical components where the prospect of benefit is greatest would provide a rational basis for music therapy in patients with dementia; moreover, rhythm-based interventions might be more straightforward to deliver and outcomes (for example, patient motor responses) may be easier to code than more complex musical interventions [98].

Considered together, our findings suggest that music perception may be a useful paradigm for assessing neural computational processes that support the analysis of information streams over different time windows and levels of complexity and the impact of dementias on those processes. Impaired encoding of pitch contour may have potential utility as a novel nonverbal and nonvisual biomarker across dementia syndromes while the overall profile of pitch pattern processing may have relative selectivity for particular pathologies such as AD. In addition, potential linkages between musical pitch processing and the processing of speech prosody warrant further investigation particularly in patients with progressive aphasia. This study has several limitations that should direct further work. The numbers of patients recruited here to particular syndromic groups were small and additionally, particular tests were not completed by all patients (Table 2), further underlining the need to study larger patient cohorts to corroborate these findings. Future study cohorts should ideally encompass a wider range of neurodegenerative syndromes and diseases with longitudinal assessments to determine the sensitivity and specificity of particular musical perceptual indices and patterns of evolution over time, ultimately with histopathological and molecular correlation. Combining multi-center patient cohorts might improve power to detect effects and potentially, to stratify neurodegenerative syndromes and pathologies. Even within the AD spectrum, factors such as age and disease stage (severity) may importantly modify phenotype [39]; moreover the present data underline the need to take into account individual variability, which may be amplified by prior musical competence. Besides the analysis of local and global information per se, dementia syndromes might degrade associated cognitive operations, such as perceptual learning, executive shifting between processing levels, or top-down attentional modulation of perceptual mechanisms [39, 43]: future musical paradigms should address these possibilities. Structural and functional neuroanatomical studies comparing patient and healthy older cohorts will be required to delineate the alterations in brain mechanisms of music processing produced by these diseases and to more fully understand the musical phenotypes demonstrated here. More broadly, our findings may provide a prima facie case for tackling theoretical and practical issues of sensory information processing in the dementias that go beyond the domains of language and vision.