Degeneracy in the neurological model of auditory speech repetition

1051 

Abstract Submission 

Noor Sajid¹, Andrea Gajardo-Vidal², Justyna Ekert¹, Diego Lorca-Puls³, Thomas Hope¹, David Green⁴, Karl Friston¹, Cathy Price¹

¹Wellcome Centre for Human Neuroimaging, University College London, London, London, ²CICS Centro de Investigación en Complejidad Social, Universidad del Desarrollo, Concepción, N/A, ³Sección de Neurología, Departamento de Especialidades, Universidad de Concepción, Concepción, N/A, ⁴Experimental Psychology, University College London, London, London

Noor Sajid  
Wellcome Centre for Human Neuroimaging, University College London
London, London

Andrea Gajardo-Vidal  
CICS Centro de Investigación en Complejidad Social, Universidad del Desarrollo
Concepción, N/A

Justyna Ekert  
Wellcome Centre for Human Neuroimaging, University College London
London, London

Diego Lorca-Puls  
Sección de Neurología, Departamento de Especialidades, Universidad de Concepción
Concepción, N/A

Thomas Hope  
Wellcome Centre for Human Neuroimaging, University College London
London, London

David Green  
Experimental Psychology, University College London
London, London

Karl Friston  
Wellcome Centre for Human Neuroimaging, University College London
London, London

Cathy Price  
Wellcome Centre for Human Neuroimaging, University College London
London, London

The neurological language model (1) posits that auditory speech repetition engages four left hemisphere brain regions sequentially: primary auditory cortex (A1), Wernicke's area (WA), Broca's area (BA), and primary motor cortex (M1), with the arcuate fasciculus mediating information relay. Recent studies challenge this, emphasising the importance of areas near WA and BA (2). Here, we investigate the bilateral interaction amongst these and their involvement with A1 and M1 during auditory speech repetition. Using previously identified activations (2) we estimate effective connectivity (i.e., directed interactions) across these areas using Dynamic Causal Modelling (DCM)3. Our findings reveal variable effective connectivity across word or pseudoword repetition, indicative of functional degeneracy.

We studied 59 right-handed native English speakers during auditory speech repetition using 3T fMRI scanning runs with 40 stimuli in 4 blocks. The stimuli had an average syllable count of 1.68 (words) and 1.5 (pseudowords) and pre-scan training ensured accuracy. Data processing included spatial realignment, unwarping, normalisation, and smoothing. Activation timeseries were extracted from peak responses (p<0.05 FWE-corrected) in anatomical constrained regions (Figure 1A): Te1.0 & Te1.2 for A1, rostro-posterior superior temporal sulcus (pSTS) for WA, dorsal and ventral pars opercularis for BA, and face and tongue & larynx subregion for M1 (M1-f & M1-tl). Effective connectivity amongst these was estimated using participant (3) and group-level (6) DCM hypothesising input from A1, A1's connections to all regions barring M1, connection from pSTS to pOp, and M1 as the output region (Figure 1B). The winning group-level model quantified estimated connection strengths (posterior probability>0.75) after Bayesian model comparison over 256 models.

The winning models revealed excitatory connectivity from A1 to pSTS, pSTS to M1, A1 to pOp, and pOp to M1 (Figure 1C-D); positive extrinsic connections were excitatory & inhibitory if negative). There were inhibitory connections from M1 to A1, pSTS to A1, and surprisingly, between pSTS and pOp. The results were consistent across task and subregional configurations. To investigate the unanticipated effective connectivity from pSTS to M1, we evaluated individual models (Figure 2A). Each model was assigned given estimated connectivity from pOp or pSTS to M1: A had excitatory connections from both to M1; B only from pSTS to M1; C only from pOp to M1; and D had an absence of significant connections. More than half the models were assigned to Group A, 20% to Group B and D and <5% in Group C (Figure 2C). Importantly, only 2% of models were consistent with the neurological model (pSTS->pOp->M1). We measured degeneracy via high intra-subject variability in how M1 was influenced by pSTS and/or pOp (Figure 2B). Connectivity variability was quantified by assigning participants to groups A-D. Entropy measurement, reflecting functional degeneracy (7), showed an average entropy in individual group membership of 0.49 (word) and 0.55 (pseudoword). 73% were assigned to ≥2 groups, indicating the ability to execute auditory speech repetition in diverse ways, exemplifying degeneracy.

We used fMRI DCM to evaluate the effective connectivity between pSTS and pOp, and with A1 and M1. Contrary to the neurological model, we show that pSTS drives M1 independently of pOp, there is bilateral inhibitory connectivity between pOp and pSTS, and participants vary in the degree to which M1 activity is driven by pSTS or pOp. This demonstrates a distributed, functional architecture between pSTS and pOp, implying alternative pathways for auditory speech repetition (degeneracy), and serve to generate hypotheses about how auditory speech repetition can be maintained or recovered after brain damage.

Speech Perception

Speech Production ¹

Activation (eg. BOLD task-fMRI)

Bayesian Modeling

fMRI Connectivity and Network Modeling ²

FUNCTIONAL MRI

Language