Background. It is a commonplace notion that melodies are tone sequences which are neither too random nor too predictable in their structure. Little is known, however, about patterns of brain response as a function of the structure of tone sequences.

Aims. This study sought to determine if differences in the statistical structure of tone sequences are reflected in measurable, dynamic neural responses, and if sequences that are melody-like in their statistical properties have a distinct neural signature.

Methods. Subjects listened to 1-minute long diatonic tone sequences while neural signals were recorded using 148-channel whole-head magnetoencephalography (MEG). Sequences were random, deterministic (scalar), or one of two categories of 'fractal' sequences differing in their balance of predictability and unpredictability. (One of the fractal categories had melody-like statistics). Amplitude-modulation of the tone sequences was used to generate an ongoing, identifiable neural response whose amplitude and timing (phase) could be studied as a function of sequence structure.

Results. Ongoing timing patterns in the neural signal showed a strong dependency on the structure of the tone sequence. At certain sensor locations, timing patterns covaried with the pitch contour of the tone sequences, with increasingly accurate tracking as sequences became more predictable. In contrast, interactions between brain regions (as measured by temporal synchronization), particularly between left posterior regions and the rest of the brain, were greatest for the tone sequences with melody-like statistics. This may reflect the perceptual integration of local and global pitch patterns in melody-like sequences.

Conclusions. Dynamic neural responses reveal a neural correlate of pitch contour in the human brain, and show that interactions between brain regions are greatest when tone sequences have melody-like statistical properties.