Functional anatomy and topographical organization of the frontotemporal arcuate fasciculus

doi:10.21203/rs.3.rs-4614103/v1

Download PDF

Article

Functional anatomy and topographical organization of the frontotemporal arcuate fasciculus

https://doi.org/10.21203/rs.3.rs-4614103/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Traditionally, the frontotemporal arcuate fasciculus (AF) is viewed as a single entity in anatomo-clinical models. However, it is unclear if distinct cortical origin and termination patterns within this bundle correspond to specific language functions. We used track-weighted dynamic functional connectivity, a hybrid imaging technique, to study the AF structure and function in a large cohort of healthy participants. Our results suggest the AF can be subdivided based on dynamic changes in functional connectivity at the streamline endpoints. An unsupervised parcellation algorithm revealed spatially segregated subunits, which were then functionally quantified through meta-analysis. This approach identified three distinct clusters within the AF - ventral, middle, and dorsal frontotemporal AF - each linked to different frontal and temporal termination regions and likely involved in various language production and comprehension aspects.

Biological sciences/Neuroscience/Cognitive neuroscience/Language

Health sciences/Anatomy/Nervous system/Central nervous system/Brain

Connectivity

resting-state functional MRI

language

phonology

tractography

The arcuate fasciculus is a prominent frontotemporal association pathway in the human brain ^1–6. Since its first description in the 19th century, the arcuate has been associated with language processes and, when damaged, deficits in language functions. With the advent of in vivo diffusion-weighted tractography, the brain’s white matter was studied in many healthy participants and clinical populations. These studies identified that the anatomy of the perisylvian white matter extends beyond the original frontotemporal arcuate and includes direct connections to the parietal lobes ^1,7. Accordingly, a novel nomenclature has been proposed subdividing the arcuate into long (frontotemporal or arcuate proper), anterior (frontoparietal, largely equivalent to the third branch of the superior longitudinal fasciculus, SLF3), and posterior (temporoparietal) segments ¹. In addition to the extended cortical reach, these studies also identified a significant left lateralization of the frontotemporal arcuate fasciculus ^8–10. This asymmetry has been considered in light of the left hemisphere dominance for language ^11–13.

Functionally, the left AF is considered crucial for language processing ^1,8,14–19, albeit not exclusively ¹⁸. Several studies combined tractography and functional MRI, or lesion mapping studies in clinical populations, to map the functions of the arcuate segments ^17,20. In particular, the frontoparietal segment of the arcuate fasciculus has been suggested to mediate phonology-to-movement mapping, phonological working memory, and phonology-based word retrieval ^21–24 as well as speech fluency, informativeness and rate ^25–27; the parietotemporal segment would contribute to reading, word comprehension, and vocabulary knowledge (Turken and Dronkers 2011; Thiebaut de Schotten et al. 2014; Teubner-Rhodes et al. 2016).However, the precise role of the direct, frontotemporal segment of the AF is still a matter of lively debate. Early investigations in patients with conduction aphasia consistently linked damage to the left AF to sensory-to-motor speech production impairment, such as phonemic paraphasia and repetition deficits ^31–33. Accordingly, further investigations have suggested a role of the AF proper in low-level phonemic and phonological processing in the context of language production, including word and non-words (sublexical) repetition ^25,34–36. However, there is evidence suggesting that the direct AF may take part in higher-order language production processes that may involve the integration of lexical and phonological information, such as naming and phonologic fluency ^14,37,38. This dissociation between phonological and semantic processing in the frontotemporal AF may be grounded in the underlying bundle anatomy, as suggested by several ex vivo and in vivo investigations which provided evidence of “ventral” and “dorsal” sub-segments within the frontotemporal AF, with distinct course, origin, and termination ^5,6.

Further, preliminary evidence comparing structural connectivity findings with activation maps obtained from previous task-based fMRI studies suggested that the ventral AF mediates mostly phonemical-phonological processing, while the dorsal AF is involved in lexical-semantical processes ^19,39. However, while these dissociations were anatomically driven, potential functional dissociations within the frontotemporal arcuate have poorly been explored because of the lack of methods that conveniently combine tractography and functional MRI data.

Recently, novel methods advanced this framework by mapping the functional white matter networks as identified by statistically combining task or task-free functional and structural imaging ^40–42. In addition, track-weighted dynamic functional connectivity (tw-dFC) is a recently developed technique that allows for a joint analysis of structural and functional connectivity by mapping time-windowed functional connectivity, sampled from resting-state functional MRI, back onto the underlying white matter anatomy, reconstructed by tractography ⁴³. In previous work, independent component analysis (ICA) applied to tw-dFC time series was suitable for identifying highly reliable and biologically meaningful functional units within the human white matter ⁴⁴. This feature makes it a useful tool to parcellate white matter structures in an entirely data-driven fashion based on the fluctuations of functional connectivity at their cortical endpoints.

In the present work, we adapt these methods to test the hypothesis of functionally independent clusters within the human AF. We obtained bundle-specific tw-dFC time series of the human AF by combining high-quality resting-state and diffusion data from the Human Connectome Project ⁴⁵. We could identify anatomically and functionally dissociable AF clusters by applying an ICA-based parcellation protocol to these time series. We demonstrated in an unsupervised fashion that the AF may be subdivided into two independent components according to dynamic changes in functional connectivity at the streamline endpoints. This solution maximized the reproducibility of the results over split-half samples of the same dataset, test-retest functional and diffusion data, and an independent validation dataset ⁴⁶. Moreover, these independent components were spatially segregated into three anatomical clusters with distinct courses and cortical termination. Finally, the functional meaning of such a peculiar anatomical organization was investigated through a meta-analytic decoding approach based on the NeuroQuery predictive model and database ⁴⁷.

1. Functional activity in the arcuate fasciculus is best decomposed into two independent components

Preprocessed diffusion-weighted imaging (DWI) data of the primary, test-retest and validation datasets underwent an automatic AF reconstruction pipeline through TractSeg, an algorithm that directly segments white matter bundles from the Fiber Orientation Distribution (FOD) peaks ⁴⁸. For each voxel traversed by streamlines, tw-dFC time series at a given time window (~ 40-second length) were computed as the average functional connectivity (FC) value at the endpoints of the streamlines traversing that voxel (Fig. 1).

Bundle-specific tw-dFC volumes underwent a spatial group ICA framework implemented in the Group ICA of FMRI Toolbox (GIFT) ^49,50. Group analysis was performed separately for the primary, test-retest (Human Connectome Project, HCP) and validation (Leipzig Study for Mind-Body-Emotion Interactions, LEMON) datasets, and for the left and right AF. For each cardinality of components ranging from k = 2 to k = 5, group ICA was successfully performed in all datasets for both the left and right AF. Measures of between-subjects, within-subject, and between-cohorts spatial similarity and similarity to static functional connectivity (Functionnectome) were considered to select an optimal number of components for left and right AF parcellations (Fig. 2).

The similarity of group ICA results over split-half resamples of the main dataset was higher for k = 2 (both left and right r = 0.99), with slightly lower similarity for other k values. (Fig. 2A).

The within-subject similarity was found to be maximal for k = 2 both for left (r = 0.98) and right AF (r = 0.98); for left AF, lower values were obtained with k = 3, while for right AF, a drop in similarity values was observed for k = 4 and k = 5 (Fig. 2B).

The between-cohorts spatial similarity was substantially higher for k = 2 ICA solution both for left (r = 0.90) and right (r = 0.89) AF, with markedly lower values for higher values of k (Fig. 2C).

Finally, the similarity to functionnectome-based ICA was found higher for the k = 2 ICA solution both for left (r = 0.79) and right (r = 0.85) AF; the correlation was found to decrease gradually with increasing values of k, with a slight increase for k = 5 (Fig. 2D).

Taking these results together, k = 2 was selected as the optimal k value for ICA analysis, and the resulting components were considered for AF parcellation.

2. Independent components map on three anatomically distinct segments of the AF

The results of group ICA suggest a topographical organization of the left and right AF (Fig. 3A).

For the left and right AF, component maps included voxels of all the AF but with different weights, corresponding to distinct patterns of correlated and anti-correlated activity within the AF: for each voxel, positive weights indicate that the activity is positively correlated to the overall component time series. In contrast, negative weights indicate that the voxel activity negatively correlates with the component time series. This differentiation in voxel weights across the AF allows us to delineate the network's functional connectivity, highlighting the importance of understanding both synchronized and distinct patterns of neural activity. At the selected number of components of k = 2, the time series of these components, while not fully independent of each other, showed a relatively weak temporal correlation (left: r = 0.17, right: r = 0.18) (Fig. 3B).

Given the overlapping spatial distribution of these functional components, we opted for a hard parcellation of AF sub-units based on k-means clustering. Based on the silhouette plot (Fig. 4A), an optimal number of clusters of c = 3 was identified both for left and right AF.

Hard parcellation revealed a tripartite topographical organization of AF clusters following a ventral-dorsal topographical organization: a ventral cluster that extends from the superior temporal gyrus, superior temporal sulcus, and anterior part of the middle temporal gyrus to the most ventral portion of the inferior frontal gyrus; a middle cluster which connects the posterior middle temporal gyrus and anterior inferior temporal gyrus to the dorsal parts of the inferior frontal gyrus; and a dorsal cluster reaching the inferior temporal sulcus and middle frontal gyrus (Fig. 4B).

To clarify the relation between functional independent components, we plotted the average component weights from the group ICA for each of the three clusters (Fig. 4C). As highlighted from the plots, component 1 weights progress from extreme values in the ventral AF cluster to extreme opposite values in the middle cluster, while component 2 weights span from extreme values in the middle cluster to opposite extreme values in the dorsal cluster.

3. Distinct AF clusters correlation patterns with meta-analytic maps

To functionally characterize white matter clusters from AF parcellation, a custom meta-analytic approach using the Neuroquery database was employed. Voxel-wise inverse distance maps were generated for each cluster centroid to create continuous distribution maps, indicating voxel proximity to centroids. These maps were masked with a mean GM mask and used in Neuroquery's image search to find top correlated neuroscience terms ⁴⁷. After removing duplicates and anatomical terms, unthresholded statistical maps for the remaining terms were converted to track-weighted predictive term maps using MRtrix3 ^51,52. Pairwise Pearson’s correlation quantified the similarity between cluster maps and term maps. Statistical significance was assessed with a permutational approach ⁵³ and multiple comparisons were corrected using the Benjamini-Hochberg method. Correlation effect sizes were evaluated using the R² determination coefficient.

The analysis of correlation to track-weighted predictive term maps revealed dissociable correlation patterns for each AF cluster. Cluster inverse distance maps were correlated to meta-analytic terms, such that positive correlation values indicate that a meta-analytic map correlates to the proximity to cluster centroid (i.e. positively associated with a given cluster). Negative correlation values mean that a meta-analytic map is correlated to the distance from the cluster centroid (i.e., it is negatively associated with a given cluster). The initial screening of meta-analytic terms resulted in 33 neuroscience terms being selected for the following analysis (Tables 1–2). Supplementary file 1 provides the links to the publications related to the meta-analytic terms.

In brief, the ventral AF cluster positively correlated with auditory-related terms (“pitch”: left r = 0.39, R² = 0.15, p < 0.001; right r = 0.41, R² = 0.17, p < 0.001; “sound”: left r = 0.47, R² = 0.22, p < 0.001; right r = 0.51, R² = 0.26, p < 0.001) and terms specific to voice processing (“vocal”: left r = 0.47, R² = 0.22, p < 0.001; right r = 0.57, R² = 0.33, p < 0.001; “voice”: left r = 0.29, R² = 0.08, p < 0.001; right r = 0.50, R² = 0.26, p < 0.001); it also yielded positive correlation to phonology-related terms (“pseudo”, which includes studies referring mostly to words-pseudowords discrimination: left r = 0.40, R² = 0.16, p < 0.001; right r = 0.57, R² = 0.33, p < 0.001; “phonological”: left r = 0.33, R² = 0.11, p < 0.001; right r = 0.55, R² = 0.30, p < 0.001). It negatively correlated to terms related to semantic processing (“semantic”: left r = -0.19, R² = 0.03, p < 0.001; right r = 0.55, R² = 0.30, p < 0.001; “semantic memory”: left r = 0.33, R² = 0.11, p < 0.001; right r = 0.55, R² = 0.30, p < 0.001). Correlation values were generally higher for the right AF, and the right ventral clusters was also correlated to terms related to reading and text processing (“read”: right r = 0.56, R² = 0.31, p < 0.001; left r = 0.04, p > 0.05).

The middle cluster of the AF positively correlated with most of the selected terms, with higher values for language-related terms, especially terms related to semantic processing (“semantic”: left r = 0.81, R² = 0.66, p < 0.001; right r = -0.24, R² = 0.06, p < 0.001; “semantic memory”: left r = 0.77, R² = 0.59, p < 0.001; right r = -0.15, R² = 0.02, p < 0.001, “semantic processing”: left r = 0.81, R² = 0.66, p < 0.001; right r = -0.03, p > 0.05, “meaning”: left r = 0.79, R² = 0.63, p < 0.001; right r = 0.21, R² = 0.04, p < 0.001, “noun”: left r = 0.75, R² = 0.56, p < 0.001; right r = 0.35, R² = 0.12, p < 0.001) followed by syntactic (“syntactic”: left r = 0.56, R² = 0.32, p < 0.001; right r = 0.24, R² = 0.06, p < 0.001, “syntax”: left r = 0.66, R² = 0.44, p < 0.001; right r = 0.44, R² = 0.19, p < 0.001, “violations”: left r = 0.74, R² = 0.55, p < 0.001; right r = 0.67, R² = 0.44, p < 0.001) and phonological processing (“phonological”: left r = 0.28, R² = 0.08, p < 0.001; right r = 0.14, R² = 0.02, p < 0.001). It also yielded correlations to non-linguistic terms such as those related to higher-order cognitive functions (“consideration”: left r = 0.46, R² = 0.21, p < 0.001; right r = 0.41, R² = 0.17, p < 0.001, “campus” which mostly refers to works on autobiographical memory: left r = 0.40, p < 0.001; right r = 0.39, p < 0.001) and social function (“social cognition”: left r = 0.68, p < 0.001; right r = 0.62, p < 0.001). Overall, correlations with language-related terms were weaker or negative in the right hemisphere, while correlations to social-related terms were slightly stronger. The middle cluster distance map also negatively correlated with purely auditory-related terms.

Finally, the dorsal cluster of the AF was anti-correlated with terms related to acoustic or language processing. It shared with the middle cluster positive correlation to terms related to higher-order cognitive functions (“campus”, which includes studies related to autobiographical memory processes: left r = 0.33, R² = 0.11, p < 0.001; right r = 0.42, R² = 0.18, p < 0.001; “consideration”: left r = 0.39, R² = 0.15, p < 0.001; right r = 0.41, R² = 0.16, p < 0.001 “locked”: left r = 0.50, R² = 0.25, p < 0.001; right r = 0.55, R² = 0.18, p < 0.001) and to semantic memory in the right hemisphere (right r = 0.39, R² = 0.16, p < 0.001; left r = -0.02, p > 0.05). The most relevant terms for each cluster are summarized in word clouds (Fig. 4D).

Table 1

**Meta-analytic decoding of left AF clusters.** For each of the 33 neuroscience terms derived from the meta-analytic screening procedure, Pearson’s correlation coefficients (r values) to track-weighted meta-analytic maps as well as their effect size (R² coefficient of determination) and spatial autocorrelation-corrected p-values are displayed. All p-values are corrected for multiple comparisons using the Benjamini-Hochberg method.
		Ventral AF				Middle AF			Dorsal AF
	Neuroscience term	r	R²		p	r	R²	p	r	R²	p
Non-linguistic (cognitive)	ambiguous	-0.14	0.02	0.0469		0.79	0.63	< 0.001	-0.28	0.08	< 0.001
	campus	-0.33	0.11	< 0.001		0.4	0.16	< 0.001	0.33	0.11	< 0.001
	consideration	-0.45	0.21	< 0.001		0.46	0.21	< 0.001	0.39	0.15	< 0.001
	gaze	-0.03	0.00	1		0.51	0.26	< 0.001	-0.32	0.10	< 0.001
	locked	-0.48	0.23	< 0.001		0.22	0.05	< 0.001	0.5	0.25	< 0.001
	order	0.18	0.03	0.015		0.13	0.02	0.035	-0.31	0.09	< 0.001
	time	-0.1	0.01	0.2051		0.25	0.06	< 0.001	0.01	0.00	1
	violations	-0.1	0.01	0.2491		0.74	0.55	< 0.001	-0.3	0.09	< 0.001
Non-linguistic (social)	social	0.38	0.14	< 0.001		0.12	0.01	0.018	-0.59	0.35	< 0.001
	social_cognition	-0.18	0.03	< 0.001		0.68	0.46	< 0.001	-0.19	0.03	< 0.001
	social_interaction	0.29	0.08	< 0.001		0.34	0.11	< 0.001	-0.58	0.34	< 0.001
Linguistic (general)	english	0.16	0.02	0.0562		0.57	0.32	< 0.001	-0.55	0.31	< 0.001
	german	-0.06	0.00	1		0.73	0.53	< 0.001	-0.35	0.12	< 0.001
	language	0.09	0.01	0.3998		0.63	0.40	< 0.001	-0.52	0.27	< 0.001
	language_processing	0.23	0.05	< 0.001		0.49	0.24	< 0.001	-0.61	0.37	< 0.001
	linguistic	0.16	0.03	0.0469		0.57	0.33	< 0.001	-0.55	0.31	< 0.001
Linguistic (semantic-comprehension)	comprehension	0.03	0.00	1		0.71	0.50	< 0.001	-0.46	0.21	< 0.001
	meaning	-0.15	0.02	0.0257		0.79	0.63	< 0.001	-0.25	0.06	< 0.001
	noun	-0.22	0.05	< 0.001		0.75	0.56	< 0.001	-0.14	0.02	< 0.001
	semantic	-0.19	0.03	< 0.001		0.81	0.66	< 0.001	-0.18	0.03	< 0.001
	semantic_memory	-0.29	0.08	< 0.001		0.77	0.59	< 0.001	-0.02	0.00	1
	semantic_processing	-0.14	0.02	0.0257		0.81	0.66	< 0.001	-0.24	0.06	< 0.001
Linguistic (phonological-syntactic)	phonological	0.33	0.11	< 0.001		0.28	0.08	< 0.001	-0.61	0.37	< 0.001
	pseudo	0.4	0.16	< 0.001		0.02	0.00	1	-0.57	0.33	< 0.001
	syntactic	0.08	0.01	0.6361		0.56	0.32	< 0.001	-0.43	0.19	< 0.001
	syntax	0.02	0.00	1		0.66	0.44	< 0.001	-0.42	0.18	< 0.001
	sentence	-0.15	0.02	0.0257		0.8	0.64	< 0.001	-0.29	0.09	< 0.001
Linguistic (Reading)	read	0.04	0.00	1		0.65	0.42	< 0.001	-0.46	0.21	< 0.001
Linguistic (Reading)	text	-0.19	0.04	< 0.001		0.78	0.61	< 0.001	-0.21	0.05	< 0.001
Acoustic/vocal	pitch	0.39	0.15	< 0.001		-0.15	0.02	< 0.001	-0.4	0.16	< 0.001
	sound	0.47	0.22	< 0.001		-0.24	0.06	< 0.001	-0.53	0.28	< 0.001
	vocal	0.47	0.22	< 0.001		-0.02	0.00	1	-0.57	0.33	< 0.001
	voice	0.29	0.08	< 0.001		0.37	0.14	< 0.001	-0.5	0.25	< 0.001

Table 2

**Meta-analytic decoding of right AF clusters.** For each of the 33 neuroscience terms derived from the meta-analytic screening procedure, Pearson’s correlation coefficients (r values) to track-weighted meta-analytic maps as well as their effect size (R² coefficient of determination) and spatial autocorrelation-corrected p-values are displayed. All p-values are corrected for multiple comparisons using the Benjamini-Hochberg method.
		Ventral AF				Middle AF			Dorsal AF
	Neuroscience term	r	R²		p	r	R²	p	r	R²	p
Non-linguistic (cognitive)	ambiguous	-0.03	0.00	1		0.58	0.34	< 0.001	-0.14	0.02	< 0.001
	campus	-0.53	0.28	< 0.001		0.39	0.15	< 0.001	0.42	0.18	< 0.001
	consideration	-0.52	0.27	< 0.001		0.41	0.17	< 0.001	0.41	0.16	< 0.001
	gaze	-0.14	0.02	< 0.001		0.66	0.43	< 0.001	-0.1	0.01	0.0093
	locked	-0.57	0.32	< 0.001		0.43	0.19	< 0.001	0.43	0.18	0
	order	-0.1	0.01	< 0.001		0.35	0.12	< 0.001	0.16	0.02	0
	time	0.15	0.02	< 0.001		0.24	0.06	< 0.001	-0.1	0.01	0.0348
	violations	0.07	0.01	0.1393		0.67	0.44	< 0.001	-0.3	0.09	< 0.001
Non-linguistic (social)	social	-0.07	0.00	0.1349		0.66	0.44	< 0.001	-0.19	0.04	< 0.001
	social_cognition	-0.41	0.17	< 0.001		0.62	0.38	< 0.001	0.17	0.03	< 0.001
	social_interaction	0.16	0.03	< 0.001		0.59	0.34	< 0.001	-0.35	0.12	< 0.001
Linguistic (general)	english	0.43	0.19	< 0.001		0.21	0.04	< 0.001	-0.45	0.20	< 0.001
	german	0.42	0.18	< 0.001		0.43	0.19	< 0.001	-0.59	0.35	< 0.001
	language	0.28	0.08	< 0.001		0.1	0.01	< 0.001	-0.18	0.03	< 0.001
	language_processing	0.53	0.29	< 0.001		0.02	0.00	1	-0.44	0.19	< 0.001
	linguistic	0.45	0.20	< 0.001		0.23	0.05	< 0.001	-0.46	0.21	< 0.001
Linguistic (semantic-comprehension)	comprehension	0.37	0.14	< 0.001		0.32	0.10	< 0.001	-0.43	0.19	< 0.001
	meaning	0.49	0.24	< 0.001		0.21	0.04	< 0.001	-0.52	0.27	< 0.001
	noun	0.41	0.17	< 0.001		0.35	0.12	< 0.001	-0.52	0.27	< 0.001
	semantic	0.28	0.08	< 0.001		-0.24	0.06	< 0.001	-0.06	0.00	0.0348
	semantic_memory	-0.18	0.03	< 0.001		-0.15	0.02	< 0.001	0.39	0.16	< 0.001
	semantic_processing	0.5	0.25	< 0.001		-0.03	0.00	1	-0.4	0.16	< 0.001
Linguistic (phonological-syntactic)	phonological	0.55	0.30	< 0.001		0.14	0.02	< 0.001	-0.57	0.32	< 0.001
	pseudo	0.57	0.33	< 0.001		-0.19	0.04	< 0.001	-0.47	0.22	< 0.001
	syntactic	0.17	0.03	< 0.001		0.24	0.06	< 0.001	-0.09	0.01	0.0422
	syntax	0.33	0.11	< 0.001		0.44	0.19	< 0.001	-0.41	0.17	< 0.001
	sentence	0.13	0.02	0.0186		0.4	0.16	< 0.001	-0.21	0.04	< 0.001
Linguistic (Reading)	read	0.56	0.31	< 0.001		0.07	0.01	0.1393	-0.53	0.28	< 0.001
Linguistic (Reading)	text	0.01	0.00	1		0.47	0.22	< 0.001	-0.1	0.01	0.0572
Acoustic/vocal	pitch	0.41	0.17	< 0.001		0.21	0.04	< 0.001	-0.48	0.23	< 0.001
	sound	0.51	0.26	< 0.001		-0.33	0.11	< 0.001	-0.33	0.11	< 0.001
	vocal	0.57	0.33	< 0.001		0.14	0.02	< 0.001	-0.59	0.35	< 0.001
	voice	0.5	0.26	< 0.001		0.29	0.08	< 0.001	-0.59	0.34	< 0.001

We fractionated the anatomy of the frontotemporal AF in the healthy adult human brain in the left and right hemispheres based on resting-state fMRI differences and characterized functionally these subcomponents by means of comparison with metanalytic maps. Three main findings emerged from our work. First, distinct patterns of correlated and anti-correlated activity exist within the AF. Second, the frontotemporal arcuate can be subdivided into three anatomically distinct clusters. Third, each of these divisions connects areas classically activated with different fMRI paradigms.

1. Structure and function in the AF: independent components of dynamic brain activity along the AF

The present work provides an account of the functional anatomy of the AF in the human brain, by investigating tract-specific, time-dependent fluctuations in spontaneous functional activity. This paradigm has been recently employed to identify spatially independent functional units in the whole-brain white matter, representing fiber bundles sharing coordinated oscillations of connectivity at their endpoints ^43,44. Herein, we leveraged this method to elucidate the structure-function relationship in the AF.

By applying independent component analysis to the track-weighted, time-varying functional connectivity data, we identified distinct patterns of correlated and anti-correlated activity within the frontotemporal AF. In particular, we described that the patterns of functional activity within the AF are best characterized by the least possible number of independent components (k = 2), as confirmed by between-subject and within-subject (test-retest) reliability measures. Additionally, the two components identified by group ICA showed very high out-of-sample reproducibility (above 0.90), suggesting that they may capture functional features of AF that are robust to experimental differences in data acquisition and processing. It is worth noting that the two datasets employed in the analysis showed several demographical (larger age range, different gender proportion) and technical differences both in DWI and rs-fMRI acquisition and processing ^46,54,55 further highlighting the robustness of these results.

While maintaining a well-recognizable polarity between different clusters of the AF, the activity patterns highlighted by group-ICA components are not entirely segregated to previously anatomically defined segments of the arcuate ^1,7, but span across the whole tract. In other words, the spatial patterns highlighted by ICA may be interpreted as spatially distinct and temporally coherent “modes” of dynamic connectivity along the AF, in which distinct portions of the tract show anti-correlated dynamic connectivity (i.e. while functional connectivity increases at the endpoints of a certain cluster of the tract, it decreases at the endpoints of another cluster and vice-versa).

In recent years, a growing interest in functional connectivity's dynamic, time-varying properties has arisen ^56–59. While the exact biological meaning of such fluctuations is still far from being fully understood ⁶⁰, there is substantial consensus that they may reflect, to a certain extent, neuronal sources of activity at different frequency bands ^61–63. In addition, emerging evidence suggests that time-varying functional connectivity is partly constrained by anatomical connectivity ^62,64 and that it may be related to spontaneous mental processes, such as arousal, perceptual fluctuations, mind-wandering, or daydreaming ^65–68. In keeping with this view, projecting the spatial patterns of fluctuations in functional connectivity at the endpoints of the AF onto the tract itself may help understand how different portions of the tracts constrain intrinsic activity and whether they show dissociable functional profiles during spontaneous mentation.

To gain additional insight into the functional meaning of these independent components, we benchmarked our results against those obtained with a similar method, the Functionnectome ⁴¹. While not entirely identical to tw-dFC, this method also involves the mapping of functional data on white matter priors to combine structural and functional information, and can be applied to resting-state data ⁴²; on the other hand, as it is based on direct mapping of BOLD signal onto the white matter, the resulting components can be seen as an estimate of “static” functional connectivity units in the AF, compared to the time-windowed dynamic functional connectivity of tw-dFC.

The high correlation observed between the components identified with these methods (Fig. 2) is in line with previous accounts describing relatively similar results between “static” and dynamic connectivity-based ICA of resting state data ^44,69. This suggests that fluctuations in functional connectivity follow the same spatial patterns of functional activity and are organized along the same white matter pathways.

2. Three new divisions of the frontotemporal arcuate fasciculus

We applied a clustering algorithm to the component weights to explore the inherent anatomy underlying the functional patterns revealed by spatial ICA. We observed that the spatial distribution of components points out a tripartite subdivision of the AF with a defined ventro-dorsal and latero-medial topography.

A similar subdivision of fiber bundles in the frontotemporal AF has been described by various studies, both in vivo using diffusion tractography. While the most widely accepted model of AF anatomy substantially regarded the frontotemporal segment as a whole ¹, a substantial body of anatomical evidence suggests that it may be instead composed of multiple, potentially independent units. Anatomical findings based on in vivo tractography and ex vivo fiber dissection subdivided the frontotemporal AF into an inner or ventral pathway, interconnecting the pars opercularis and the most ventral portion of precentral gyrus with the STG and rostral MTG, and an outer or dorsal pathway that connects ventral precentral gyrus, caudal middle frontal gyrus, and dorsal pars triangularis/dorsal prefrontal cortex with MTG and ITG ^5,6. However, it is worth noting that anatomical methods alone are inherently blind both to the exact origin and termination of fiber bundles and that the functional segregation of these distinct segments can only be assumed as a hypothesis. Our data-driven subdivision is in line with the overall ventral-dorsal topography identified by these investigations by showing a ventral cluster running between the posterior STG and anterior MTG and the ventral pars orbitalis and triangularis, as in Yagmurlu et al. (2016), and two outer-most clusters (middle and dorsal) corresponding substantially to the anatomically-defined dorsal AF: the middle cluster connects the posterior MTG and ventral frontal lobe, while the dorsal cluster connects the dorsal pars triangularis, middle frontal gyrus, and ventral premotor cortex to the posterior MTG and ITG.

Lastly, our results suggest a slightly asymmetrical pattern between the left and right AF sub-units. However, in contrast to the available literature ⁶, we did not observe marked left-right differences in the terminations of the ventral cluster; rather, different termination patterns were demonstrated in the middle cluster, which in left hemisphere terminates mostly on the dorsal pars triangularis, while in the right hemisphere expands more to the middle frontal gyrus (that in the left hemisphere is covered mostly by the dorsal cluster).

3. Functional characterization of AF clusters: relevance to language processing and cognitive function

Along with providing a multimodal, functional connectivity-informed, and data-driven anatomical segmentation of the AF, we also sought to elucidate the cognitive relevance of our model by applying a custom-designed meta-analytic decoding paradigm to the AF clusters derived from ICA-based parcellation.

Since earlier investigations, anatomical models of AF and its structural subunits have been tightly linked to functional language processing models ¹. There has been a considerable effort towards clarifying the role of AF in the context of the so-called “dual stream model” of information flow between language-related areas, which is anchored to specific white matter tracts ^70,71.

This model has been formulated in terms of two diverging, parallel but mutually interacting cortical streams related to speech production: a ventral stream involved in mapping sound to meaning (semantic processing) and a dorsal stream involved in mapping sound to motion (phonological processing) (Hickok and Poeppel 2007; Friederici and Gierhan 2013). Within this model, there is general agreement that the AF is mostly involved in dorsal stream functions. At the same time, other white matter structures (e.g. uncinate fasciculus, inferior fronto-occipital fasciculus (IFOF)) have been proposed to subserve ventral stream functions ^36,72–74. However, whether the AF may be involved in ventral stream functions, such as lexical or semantic retrieval, is still a matter of lively debate ^75–78.

Our meta-analytic decoding found partially overlapping yet dissociable correlation profiles between our AF clusters and language-related terms. The ventral cluster of AF, covering the superior and anterior middle temporal gyrus, was correlated with the results of functional imaging studies concerning pitch and voice recognition. Evidence from direct electrical stimulation (DES) and microelectrode recording in awake surgical patients suggests that the STG may be involved in syllable and word recognition ^79–82 and pitch ⁸³. In keeping with these findings, DES in the middle and superior temporal gyrus and peri-insular white matter results in word deafness and phonemic paraphasia ^37,76,84,85, strengthening the hypothesis that the ventral AF may be involved in both sensory and motor phonological processes.

Conversely, the middle cluster of the AF, connecting the dorsal inferior frontal gyrus and the posterior middle temporal gyrus and sulcus, correlated with semantic processing. This finding is in line with the involvement of these regions in semantic tasks, as suggested by non-invasive stimulation and lesion studies ^86–88. Semantic access and retrieval have been described to be mediated by other white matter structures, such as the IFOF ^78,88,89. However, a recent observational study on patients with unilateral left hemisphere stroke suggests that structural connectivity estimates in the dorsal arcuate fasciculus may be related to semantic functions ⁹⁰. Similarly, a recent tractography and task-based fMRI investigation reported that structural integrity of the left AF segment connecting the middle temporal gyrus with the dorsal pars opercularis predicts performance in a lexical-semantic verb-generation task ¹⁹, suggesting that this portion of AF may be partially involved in semantic processing, especially during speech production. Our middle AF cluster is closely similar to the dorsal, “semantic” AF subtract identified by Janssen et al. (2023). This AF cluster was mostly, but not exclusively, correlated to semantic processing-related terms. At the same time, it showed a significant correlation with almost all the language-related terms, suggesting that this part of the AF may have a role in the integration between ventral and dorsal stream processing.

Lastly, the most dorsal cluster of the AF, connecting ITG to the middle frontal gyrus, strongly anticorrelated with phonological and semantic language terms, in partial contrast with studies suggesting a putative linguistic role of the inferior temporal terminations of the AF ^77,91–93.

Along with enforcing the notion of different functional implications for different segments of the left AF, our findings also shed light on the functional significance of the right AF, which is far less understood than its left hemisphere homologue ^94,95. Our results suggest that the right and the left AF share a substantially similar morpho-functional organization, although with some relevant differences. The ventral cluster of right AF, similarly to its left homologue, shows a high correlation with phonological functions, voice, and pitch recognition; for some terms, the correlation was even higher than for the left AF (e.g. “voice”, “phonological”, “read”). Traditionally, “core” functions of language processing such as phonological, lexical, and semantic processing are considered as left-lateralized. In contrast, the right hemisphere is thought to be involved in processing “secondary” language functions such as prosody, pitch, and intonation ^96–99. It is then possible that the correlation pattern found for the right hemisphere may reflect non-specific language-related activity in the contralateral hemisphere in task-based studies focused on phonological and semantic features of language processing in general. However, recent evidence from post-stroke patients suggests that the right AF may play a role in the long-term recovery of language functions, likely compensating for the loss of function of the contralateral AF ^100,101: this may suggest that, at least partially, the right AF may be involved in primary language functions such as phonological or semantic processing, even in the healthy brain. In keeping with its role in mediating the emotional components of language processing, the right AF has also been suggested to be relevant for social cognition. Some studies described a relation between social cognitive functions, such as emotional intelligence, mentalization, or mind-reading abilities, and AF integrity and microstructure ^102–104. In our study, we observed a higher correlation between social cognition and social communications for the middle cluster of the AF. Of note, this component shows also the most marked asymmetry in terms of frontal termination sites between the left and right hemispheres, which may concur to explain why correlation to social cognition terms is higher in the right hemisphere. In this purely explorative context, the identified finding enables us to hypothesize a functional dissociation between social processing and linguistic processing within the right AF. This parallels the phonological-semantical dissociation previously proposed for the left AF. Collectively, these results provide valuable insights into the anatomical substrates of various linguistic and non-linguistic functions dependent on AF integrity and connectivity. Additionally, they shed light on the spatial topographical arrangement of these functions.

4. Technical issues and limitations

Some limitations of the current approach need to be acknowledged. First, we adopt an automatic tract reconstruction algorithm based on machine learning ⁴⁸ paired with deterministic tractography, which is known to provide more conservative estimates of white matter trajectories ¹⁰⁵. As such, we adopt the definition of the AF as characterised in the TractSeg software, which in turn adopts a tract definition system based on cortical termination patterns ¹⁰⁶. Equally valid, automatic segmentation algorithms for AF reconstruction ^107,108 may provide slightly different anatomical reconstructions, potentially affecting the results.

Second, choosing a pre-defined anatomical scaffold of the AF to guide tw-dFC generation forces the assumption that functional connectivity fluctuations sampled at the bundle termination are entirely driven by activity along the AF. This excludes the contributions of other fiber tracts that may share the same cortical projections, such as the uncinate fasciculus ², superior longitudinal fasciculus ¹⁰⁹, inferior fronto-occipital fasciculus ¹¹⁰, or short-range U-fibers. This extends to non-neuronal sources of functional connectivity fluctuations ¹¹¹.

Third, the quality of the tw-dFC maps strongly depends on the correct alignment of tractography and BOLD-fMRI data, making non-linear registration of tractograms, and distortion correction of both DWI and BOLD data critical steps for accurate and reliable results.

Lastly, meta-analytic decoding provides a quantitative estimate of similarity between a given brain map and meta-analytic maps derived by collating results from multiple imaging studies ¹¹². The NeuroQuery approach synthesizes meta-analytic maps for each term from studies in which that term, or other semantically related terms, are frequently mentioned ⁴⁷. This approach was preferred over other meta-analytic approaches as it permits obtaining meaningful statistical activation maps even for terms with few studies available, unlike other methods that are based on a reduced set of cognitive latent variables (“topics”) ¹¹³. However, the resulting statistical maps do not correspond to any specific fMRI task or mental state and do not necessarily represent a single task-positive functional network ¹¹⁴.

The arcuate fasciculus may be subdivided into two independent components according to dynamic changes in functional connectivity at its streamline endpoints. These independent components spatially subdivide AF into three clusters with distinct courses and cortical termination. Using a custom meta-analytic approach, we hypothesize that each cluster may be related to distinct functional processes: in particular, we suggest a partial dissociation between auditory/phonemic and lexical/semantic functions in ventral vs middle left AF, possibly paralleled by a partial dissociation between auditory-phonemic and social communication processing in ventral vs middle right AF, while the dorsal cluster would be involved in non-linguistic processing in both hemispheres. Overall, our findings provide data-driven evidence for the morpho-functional segregation of the arcuate fasciculus in both hemispheres.

1. Participants and data acquisition

1.1. Primary and test-retest datasets (HCP)

Structural, diffusion, and resting-state functional MRI data were obtained from the HCP repository (https://humanconnectome.org). Two datasets have been employed for the present work: the first dataset (primary dataset) consisted of 210 healthy participants (males = 92, females = 118, age range 22–36 years), and the second dataset (test-retest dataset) included 44 participants with available test-retest MRI scans (males = 13; females = 31; age range: 22–36 years). Data have been acquired by the Washington University, University of Minnesota and Oxford University (WU-Minn) HCP consortium. Participants recruitment procedures, informed consent, and sharing of de-identified data were approved by the Washington University in St. Louis Institutional Review Board (IRB) ¹¹⁵

For MRI data acquisition, a custom-made Siemens 3T “Connectome Skyra” (Siemens, Erlangen, Germany), provided with a Siemens SC72 gradient coil and maximum gradient amplitude (Gmax) of 100 mT/m (initially 70 mT/m and 84 mT/m in the pilot phase ¹¹⁶.

For high-resolution T1-weighted MPRAGE scan acquisition, the following parameters were used: voxel size = 0.7 mm, TR = 2400 ms, TE = 2.14 ms ¹¹⁵.

Multi-shell diffusion-weighted imaging (DWI) data (b-values: 1000, 2000, 3000 mm/s²) were acquired using a single-shot 2D spin-echo multiband Echo Planar Imaging (EPI) sequence. DWI volumes were acquired with 90 directions per shell in addition to 18 non-diffusion-weighted (b = 0 mm/s²) volumes, and a spatial isotropic resolution of 1.25 mm ⁵⁴

Resting-state functional MRI data (rs-fMRI) were acquired with a gradient-echo EPI sequence, using the following parameters: voxel size = 2mm isotropic, TR = 720 ms, TE = 33.1 ms, 1200 frames, ~ 15 min/run. While data were acquired separately on different days along two different sessions, each session consisting of a left-to-right (LR) and a right-to-left (RL) phase encoding acquisition ^55,115,116, the present work features LR and RL acquisitions of the first session only.

1.2. Validation dataset (LEMON)

High-quality structural, diffusion, and rs-fMRI data of 213 healthy subjects (males = 138, females = 75, age range 20–70 years) were retrieved from the Leipzig Study for Mind-Body-Emotion Interactions (LEMON) dataset (http://fcon_1000.projects.nitrc.org/indi/retro/MPI_LEMON.html). The study was carried out in accordance with the Declaration of Helsinki and the study protocol was approved by the ethics committee at the medical faculty of the University of Leipzig. A 3T scanner (MAGNETOM Verio, Siemens Healthcare GmbH, Erlangen, Germany) equipped with a 32-channel head coil was employed for MRI data acquisition.

The parameters of the MP2RAGE sequence used for structural T1w data acquisition were: voxel size = 1 mm, TR = 5000 ms, TE = 2.92 ms. DWI data (single shell, b = 1000 s/mm²) were acquired using a multi-band accelerated sequence with spatial isotropic resolution = 1.7 mm, and 60 diffusion-encoding directions plus 7 non-diffusion-weighted (b = 0 s/mm²) volumes. For rs-fMRI data, a gradient-echo EPI was acquired with the following parameters: phase encoding = AP, voxel size = 2.3 mm isotropic, TR = 1400 ms, TE = 30 ms, 15.30 min/run ⁴⁶.

2. Data preprocessing

2.1. Structural preprocessing

Skull-stripped T1 weighted images provided by the HCP were segmented into cortical and subcortical gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) using FAST and FIRST FSL’s tools ^117,118. A 5-tissue-type (5TT) image, which was required later for diffusion signal modeling, was obtained from structural segmented images. For the HCP dataset, the already available MNI-space transformations included in the minimal preprocessing pipeline were employed (FLIRT 12 degrees of freedom affine; FNIRT nonlinear registration) ¹¹⁹. For the LEMON dataset, T1-weighted volumes were also non-linearly registered to the 1-mm resolution MNI 152 asymmetric template using a FLIRT 12 degrees of freedom affine transform and FNIRT non-linear registration ^120–122 and direct and inverse transformations were saved; a visual quality check as in Benhajali et al. 2020 was performed to ensure proper alignment of major sulcal and gyral structures.

2.2. DWI preprocessing

The DWI scans of the two datasets underwent different preprocessing pipelines: namely, the HCP datasets were available in a minimally preprocessed form, while the LEMON DWI scans were available only in raw form and were preprocessed entirely with a dedicated pipeline included in the MRtrix3 software ¹²⁴. We decided to keep the preprocessing pipelines different, as in our previous work ⁴⁴, to further highlight the reproducibility of our findings.

The minimal preprocessing pipeline of the HCP data includes eddy currents, EPI distortion and motion correction, and cross-modal linear registration of structural and DWI images ¹²⁵.

The LEMON DWI scans were preprocessed following the subsequent steps: 1) denoising using Marchenko-Pastur principal component analysis (MP-PCA) ¹²⁶, 2) removal of Gibbs ringing artifacts ¹²⁷, 3) eddy currents, distortion (by exploiting the available reverse-phase encoding scans) and motion correction using EDDY and TOPUP FSL’s tools ^118,128,129 and 4) bias field correction using the N4 algorithm ¹³⁰.

2.3. Resting-state fMRI preprocessing

Both HCP and LEMON rs-fMRI data were obtained in preprocessed and denoised form, though the featured preprocessing steps are different between the two datasets. As mentioned above, we decided to use the already preprocessed functional volumes, considering the different acquisition features of the two datasets and further highlighting the robustness of the derived findings.

The HCP data minimal preprocessing pipeline included the following steps: 1) artifact and motion correction; 2) registration to 2-mm resolution MNI 152 standard space, 3) high pass temporal filtering (> 2000 s full width at half maximum) ¹²⁵, 4) denoising, which features ICA-based artifact identification (ICA-FIX) ¹³¹ as well as regression of artifacts and motion-related parameters ⁵⁵. In addition to the minimal preprocessing, data were band-pass filtered (0.01–0.09 Hz), and the global WM and CSF signal was regressed out to improve ICA-based denoising further ¹³².

The LEMON dataset processing pipeline included the following steps: 1) removal of the first 5 volumes to allow for signal equilibration, 2) motion and distortion correction, 3) outlier and artifact detection (rapidart) and denoising using component-based noise correction (aCompCor), 4) mean-centering and variance normalization of the time series and 5) spatial normalization to 2-mm resolution MNI 152 standard space ^46,133.

To minimize BOLD partial volume sampling from the white matter, both HCP and LEMON rs-fMRI time series were additionally smoothed through convolution with a relatively large Gaussian kernel (6mm full width at half maximum) in line with the reference tw-dFC work ⁴³. All the additional preprocessing was carried out using CONN toolbox ¹³⁴.

2.4. Bundle-specific tractography and tw-dFC

Diffusion signal modeling was performed on the preprocessed DWI data using the constrained spherical deconvolution (CSD) framework, which estimates white matter Fiber Orientation Distribution (FOD) function from the diffusion-weighted deconvolution signal using a single fiber response function as reference ¹³⁵. For HCP DWI data (multi-shell), a multi-shell multi-tissue (MSMT) CSD signal modeling algorithm was applied to estimate separate response functions in WM, GM, and CSF ¹³⁶. For LEMON DWI data (single-shell), a single-shell 3-tissue (SS3T) CSD signal modeling was applied, despite the very low b-value, to keep the processing as consistent as possible between the two datasets, since it is necessary for the following automatic tract extraction step. In addition, it has been suggested to outperform the tensor model for tract reconstruction even at very low b values ¹³⁷. SS3T-CSD is a variant of the MSMT model optimized for RF estimation in single-shell datasets and was performed using MRtrix3Tissue ¹³⁸, a fork of MRtrix3 software (https://3tissue.github.io).

A robust and unbiased reconstruction of the AF is of key importance for the good quality and generalizability of results. For bundle-specific tractography of the AF, TractSeg (https://github.com/MIC-DKFZ/TractSeg/), a convolutional neural network-based tract segmentation approach, was employed. The TractSeg algorithm directly segments white matter bundles for each subject from the FOD peaks; importantly, this method has been demonstrated to be less affected by the original data quality compared to other automatic tract segmentation algorithms ⁴⁸, and was then preferred to grant high comparability of the results between the different data-quality HCP and LEMON datasets.

Participant-level binary tract masks and ending masks obtained from TractSeg were employed to guide tractography of the AF, which was performed on FODs using deterministic tractography (SD-STREAM algorithm) ¹³⁹ with default tracking parameters up to a fixed number of 2000 streamlines.

For each participant of both HCP and LEMON datasets, tractograms of left and right AF were registered to the MNI 152 standard space by applying the aforementioned non-linear transformations (paragraph 2.1), to bring individual tractography and rsfMRI data in the same space for track-weighted dynamic functional connectivity (tw-dFC) analysis. Then, tractograms were combined with preprocessed rs-fMRI time series using MRtrx3’s tckdfc command to generate a 4-dimensional tw-dFC time series ⁴³. Separate tw-dFC time series were obtained for the left and right AF, and the following parameters were used: spatial resolution: 2 mm2, sliding window shape: rectangular, sliding window length: ~40 s (55 time points for the HCP data, TR = 0.72 s; 29 time points for the LEMON data, TR = 1.4 s). The length of the sliding window was chosen in line with previous works to maximize the stability of the time-varying connectivity profiles ^{56,69,111,140}. For the HCP data, tw-dFC derived from LR and RL phase encoding volumes were temporally concatenated for each participant. Estimation of tw-dFC of the AF is summarized in Fig. 1.

3. Group-level analysis

3.1. Group ICA-based parcellation

Bundle-specific tw-dFC volumes underwent a spatial group ICA framework implemented in the Group ICA of FMRI Toolbox (GIFT) ^49,50. Group analysis was performed separately for the primary dataset (HCP) and the validation dataset (LEMON) and for the left and right AF. For each AF tract, a binary mask for group ICA was built after transforming all the subject bundle masks to template space by summing up all individual masks and applying a probability threshold of 25% (i.e., voxels that were part of the AF in at least 75% of participants were considered for group analysis). The pipeline for group ICA analysis involved 1) a first dimensionality reduction step in which a subject-level principal component analysis (PCA) was applied to tw-dFC data to obtain 50 principal components per subject; 2) a second step in which dimensionality-reduced data of all subjects were temporally concatenated and a secondary PCA dimensionality reduction was applied along directions of maximal group variability; 3) the proper group ICA step, in which a given number (k) of independent components (ICs) is obtained from low-dimensional data using the Infomax algorithm ¹⁴¹; 4) back reconstruction, to obtain subject-specific spatial maps and time courses for each components using the group information guided ICA (GIG-ICA) algorithm ¹⁴²; 5) back-reconstructed individual components were then averaged and normalized to obtain group-level z-maps of each component.

3.2. Data-driven dimensionality selection

To select the most appropriate number of components (k) for AF parcellation given our data, the ICA pipeline was iterated for the left and right AF separately at different values of k, ranging from 2 to 5, and for each ICA solution and the following measures were calculated:

1) Between-subjects spatial similarity: the primary dataset (HCP, 210 subjects) was split into symmetrical, random halves (105 subjects each) and the group ICA pipeline was run on each split for all the k values; the average Pearson’s correlation coefficient between group spatial ICA maps of the first and second split was employed as a measure of split-half similarity;

2) Within-subject spatial similarity: the test-retest dataset (HCP, 44 subjects with test-retest diffusion and rsfMRI data) was employed. The group ICA pipeline was run separately on test and retest tw-dFC data for all the k values; the average Pearson’s correlation coefficient between group spatial ICA maps of the test and retest data was employed as a measure of test-retest similarity;

3) Within-cohorts spatial similarity: the primary dataset (HCP, 210 participant) and the validation dataset (LEMON, 213 participant) were considered. Group ICA was performed separately on each dataset for all the k values; the average Pearson’s correlation coefficient between group spatial ICA maps of the primary dataset and the validation dataset was employed as a measure of external validation.

4) Similarity to “static” functional connectivity-based ICA: we benchmarked our results against a recently developed algorithm that involves mapping of the function signal from fMRI to tractography-derived priors of white matter anatomy, the “Functionnectome” ⁴¹. In a recent implementation, this method has been extended to resting-state functional connectivity ⁴² Here, we employed the participant-level tractograms of left and right AF to derive group-level AF tractography priors. Then, we applied the Functionnectome algorithm to map the functional signal from individual BOLD volumes on these priors. The resulting 4-dimensional volumes underwent group ICA with the same parameters as for tw-dFC data. The group-ICA results for each value of k were compared using the average Pearson’s correlation coefficient.

5) Functional correlation between component time series: to investigate the temporal independence of time series at each value of k, we employed the back-reconstructed time series for all subjects of the main dataset. Pearson’s correlation was calculated pairwise for each pair of components on each individual and then averaged across the entire dataset. For values of k > 2, the average between pairwise correlation values was considered.

The optimal number of components was decided based on a consensus approach between all these metrics: the value k for which most metrics showed the highest value.

3.3. Component-driven AF parcellation

To derive a parcellation of the AF from the component maps, we applied a k-means clustering procedure in the component space. Briefly, after selecting an optimal number of components according to the measures described above, each voxel in the AF was clustered according to its similarity in weight in each component z-map. The ideal number of clusters (c) was determined using the silhouette coefficient ¹⁴³.

3.4. Meta-analytic functional decoding

To provide a functional characterization for the white matter clusters derived from AF parcellation, we employed a custom meta-analytic approach, specifically designed to account both for the discrete nature of binary clusters and the white matter nature of the underlying spatial maps. Meta-analytic decoding was based on the Neuroquery database, which contains predictive activation maps estimated from over 7547 neuroscience terms ⁴⁷.

Initially, we generated voxel-wise inverse distance maps from each cluster centroid to transform binary cluster maps into a continuous distribution of values reflecting the extent of each voxel belonging to each cluster. This process was carried out separately for the left and right AF clusters. Within voxel-wise distance maps, the value of each voxel indicates its proximity to cluster centroids, with higher values signifying closer distances. In an initial screening of terms from the Neuroquery database, these inverse distance maps underwent a masking procedure using a mean GM mask. Subsequently, the masked maps were utilized as inputs for the Neuroquery image search tool (https://github.com/neuroquery/neuroquery_image_search), retrieving the top 20 most correlated terms to each cluster distance map, resulting in a total of 120 neuroscience terms (2 hemispheres x 3 clusters x 20 terms). To this initial selection, a first screening procedure was applied by discarding duplicate terms and terms referring to anatomy or localization (e.g. “left”, “right”, “cortex”, “hemisphere” “pfc”). A template tractogram was obtained for left and right AF by summing all the subject-specific AF tractograms in standard space. Subsequently, the unthresholded statistical maps corresponding to each of the remaining terms were retrieved and converted to track-weighted predictive term maps using the left and right template tractograms ¹⁴⁴ via MRtrix3’s tckmap command employing the option -scalar_map to provide the statistical maps as input ^51,52.

Finally, pairwise Pearson’s correlation was employed to quantify the similarity between each cluster distance map and the resulting track-weighted term maps. To address the spatial autocorrelation (SA) properties of arcuate maps, statistical significance was assessed using a permutational approach described in Burt et al. (2020). This approach involved the generation of SA-preserving surrogated maps through 1000 permutations. The resulting p-values underwent correction for multiple comparisons using the Benjamini-Hochberg method. The effect sizes of correlations were assessed by computing the R² determination coefficient.

Acknowledgments: The primary dataset (HCP) was provided by the Human Connectome Project, WU‐Minn Consortium (Principal Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091657), founded by the 16 NIH institutes and centers that support the NIH Blueprint for Neuroscience Research, and by the McDonnell Center for Systems Neuroscience at Washington University. We also gratefully acknowledge the Mind-Body-Emotion group at the Max Planck Institute for Human Cognitive and Brain Sciences, for the data of the “Leipzig Study for Mind-Body-Emotion Interactions” (LEMON) that have been used as validation dataset.

This research was funded by the Italian Ministry of Health, Current Research Funds 2023. This work was also supported by DOD N° W81XWH-19-1-0810 (AQ and AlC). This project received funding from the Donders Mohrmann Fellowship No. 2401512 (SJF, NEUROVARIABILITY). M.T.d.S is supported by the European Union’s Horizon 2020 research and innovation programme under the European Research Council (ERC) Consolidator grant agreement No. 818521 (DISCONNECTOME) and by the University of Bordeaux’s IdEx ‘Investments for the Future’ program RRI ‘IMPACT’ and the IHU ‘Precision & Global Vascular Brain Health Institute – VBHI’ funded by the France 2030 initiative.

Competing interests: The authors have nothing to declare.

Data and code availability: The primary dataset (HCP) was provided by the Human Connectome Project, WU‐Minn Consortium (Principal Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091657). The data are openly available from https://www.humanconnectome.org/

The “Leipzig Study for Mind-Body-Emotion Interactions” (LEMON) data used as a validation dataset was provided by the Mind-Body-Emotion group at the Max Planck Institute for Human Cognitive and Brain Sciences. The data are openly available from http://fcon_1000.projects.nitrc.org/indi/retro/MPI_LEMON.html.

The code and maps obtained in the present work are available at https://github.com/BrainMappingLab.

Authors Contribution: G.A.B. conceived the study, implemented the methods, performed the analyses, and wrote the manuscript. V.N. implemented part of the methods and revised the manuscript. A.Q. revised the manuscript and provided funding. A.G. wrote and revised the manuscript. A.I. implemented part of the methods and performed part of the analyses. Ant.C. wrote and revised the manuscript. D.M. provided fundamental intellectual comments and revised the manuscript. M.A. provided fundamental intellectual and methodological comments and extensively revised the manuscript. S.J.F provided fundamental intellectual and methodological comments and extensively revised the manuscript. M.T.S. provided fundamental intellectual and methodological comments and extensively revised the manuscript. Alb.C. conceived and coordinated the study, implemented the methods, performed the analyses, wrote and revised the manuscript, and provided funding.

Catani, M., Jones, D. K. & ffytche, D. H. Perisylvian language networks of the human brain. Ann Neurol 57, 8–16 (2005).
Catani, M. & Thiebaut de Schotten, M. A diffusion tensor imaging tractography atlas for virtual in vivo dissections. Cortex 44, 1105–1132 (2008).
Rilling, J. K. et al. The evolution of the arcuate fasciculus revealed with comparative DTI. Nat Neurosci 11, 426–428 (2008).
Martino, J. et al. Fiber dissection and diffusion tensor imaging tractography study of the temporoparietal fiber intersection area. Neurosurgery 72, (2013).
Yagmurlu, K., Middlebrooks, E. H., Tanriover, N. & Rhoton, A. L. Fiber tracts of the dorsal language stream in the human brain. J Neurosurg 124, 1396–1405 (2016).
Fernández-Miranda, J. C. et al. Asymmetry, connectivity, and segmentation of the arcuate fascicle in the human brain. Brain Struct Funct 220, 1665–1680 (2015).
Frey, S., Campbell, J. S. W., Pike, G. B. & Petrides, M. Dissociating the human language pathways with high angular resolution diffusion fiber tractography. J Neurosci 28, 11435–11444 (2008).
Catani, M. et al. Symmetries in human brain language pathways correlate with verbal recall. Proc Natl Acad Sci U S A 104, 17163–17168 (2007).
Parker, G. J. M. et al. Lateralization of ventral and dorsal auditory-language pathways in the human brain. Neuroimage 24, 656–666 (2005).
Berthier, M. L., Lambon Ralph, M. A., Pujol, J. & Green, C. Arcuate fasciculus variability and repetition: the left sometimes can be right. Cortex 48, 133–143 (2012).
Powell, H. W. R. et al. Hemispheric asymmetries in language-related pathways: A combined functional MRI and tractography study. Neuroimage 32, 388–399 (2006).
Vernooij, M. W. et al. Fiber density asymmetry of the arcuate fasciculus in relation to functional hemispheric language lateralization in both right- and left-handed healthy subjects: A combined fMRI and DTI study. Neuroimage 35, 1064–1076 (2007).
Silva, G. & Citterio, A. Hemispheric asymmetries in dorsal language pathway white-matter tracts: A magnetic resonance imaging tractography and functional magnetic resonance imaging study. Neuroradiol J 30, 470–476 (2017).
Ivanova, M. V., Zhong, A., Turken, A., Baldo, J. V. & Dronkers, N. F. Functional Contributions of the Arcuate Fasciculus to Language Processing. Front Hum Neurosci 15, (2021).
Negwer, C. et al. Loss of Subcortical Language Pathways Correlates with Surgery-Related Aphasia in Patients with Brain Tumor: An Investigation via Repetitive Navigated Transcranial Magnetic Stimulation–Based Diffusion Tensor Imaging Fiber Tracking. World Neurosurg 111, e806–e818 (2018).
Fridriksson, J. et al. Anatomy of aphasia revisited. Brain 141, 848–862 (2018).
Price, C. J. The anatomy of language : contributions from functional neuroimaging. J. Anat 197, 335–359 (2000).
Forkel, S. J. et al. Anatomical evidence of an indirect pathway for word repetition. Neurology 94, e594–e606 (2020).
Janssen, N. et al. Dissociating the functional roles of arcuate fasciculus subtracts in speech production. Cerebral Cortex 33, 2539–2547 (2023).
López-Barroso, D. et al. Word learning is mediated by the left arcuate fasciculus. Proc Natl Acad Sci U S A 110, 13168–13173 (2013).
Rizio, A. A. & Diaz, M. T. Language, aging, and cognition. Neuroreport 27, 689–693 (2016).
Kljajevic, V. & Erramuzpe, A. Dorsal White Matter Integrity and Name Retrieval in Midlife. Curr Aging Sci 12, 55–61 (2019).
Schwartz, M. F., Faseyitan, O., Kim, J. & Coslett, H. B. The dorsal stream contribution to phonological retrieval in object naming. Brain 135, 3799–3814 (2012).
Bohland, J. W., Bullock, D. & Guenther, F. H. Neural Representations and Mechanisms for the Performance of Simple Speech Sequences. J Cogn Neurosci 22, 1504–1529 (2010).
Breier, J. I., Hasan, K. M., Zhang, W., Men, D. & Papanicolaou, A. C. Language Dysfunction After Stroke and Damage to White Matter Tracts Evaluated Using Diffusion Tensor Imaging. American Journal of Neuroradiology 29, 483–487 (2008).
Fridriksson, J., Guo, D., Fillmore, P., Holland, A. & Rorden, C. Damage to the anterior arcuate fasciculus predicts non-fluent speech production in aphasia. Brain 136, 3451–3460 (2013).
Halai, A. D., Woollams, A. M. & Lambon Ralph, M. A. Using principal component analysis to capture individual differences within a unified neuropsychological model of chronic post-stroke aphasia: Revealing the unique neural correlates of speech fluency, phonology and semantics. Cortex 86, 275–289 (2017).
Thiebaut de Schotten, M., Cohen, L., Amemiya, E., Braga, L. W. & Dehaene, S. Learning to Read Improves the Structure of the Arcuate Fasciculus. Cerebral Cortex 24, 989–995 (2014).
Teubner-Rhodes, S. et al. Aging-Resilient Associations between the Arcuate Fasciculus and Vocabulary Knowledge: Microstructure or Morphology? Journal of Neuroscience 36, 7210–7222 (2016).
Turken, A. U. & Dronkers, N. F. The Neural Architecture of the Language Comprehension Network: Converging Evidence from Lesion and Connectivity Analyses. Frontiers in System Neuroscience 5, (2011).
Benson, D. F. Conduction Aphasia. Arch Neurol 28, 339 (1973).
Tanabe, H. et al. Conduction aphasia and arcuate fasciculus. Acta Neurol Scand 76, 422–427 (1987).
Bernal, B. & Ardila, A. The role of the arcuate fasciculus in conduction aphasia. Brain 132, 2309–2316 (2009).
Kim, S. H. & Jang, S. H. Prediction of Aphasia Outcome Using Diffusion Tensor Tractography for Arcuate Fasciculus in Stroke. American Journal of Neuroradiology 34, 785–790 (2013).
Shinoura, N. et al. Damage to the left ventral, arcuate fasciculus and superior longitudinal fasciculus -related pathways induces deficits in object naming, phonological language function and writing, respectively. International Journal of Neuroscience 123, 494–502 (2013).
Saur, D. et al. Ventral and dorsal pathways for language. Proceedings of the National Academy of Sciences 105, 18035–18040 (2008).
Duffau, H. et al. Intraoperative mapping of the subcortical language pathways using direct stimulations. Brain 125, 199–214 (2002).
Marchina, S. et al. Impairment of Speech Production Predicted by Lesion Load of the Left Arcuate Fasciculus. Stroke 42, 2251–2256 (2011).
Glasser, M. F. & Rilling, J. K. DTI Tractography of the Human Brain’s Language Pathways. Cerebral Cortex 18, 2471–2482 (2008).
Nozais, V., Theaud, G., Descoteaux, M., Thiebaut de Schotten, M. & Petit, L. Improved Functionnectome by dissociating the contributions of white matter fiber classes to functional activation. Brain Struct Funct 228, 2165–2177 (2023).
Nozais, V., Forkel, S. J., Foulon, C., Petit, L. & Thiebaut de Schotten, M. Functionnectome as a framework to analyse the contribution of brain circuits to fMRI. Commun Biol 4, 1–12 (2021).
Nozais, V. et al. Atlasing white matter and grey matter joint contributions to resting-state networks in the human brain. Commun Biol 6, 726 (2023).
Calamante, F., Smith, R. E., Liang, X., Zalesky, A. & Connelly, A. Track-weighted dynamic functional connectivity (TW-dFC): a new method to study time-resolved functional connectivity. Brain Struct Funct 222, 3761–3774 (2017).
Basile, G. A. et al. White matter substrates of functional connectivity dynamics in the human brain. Neuroimage 258, 119391 (2022).
Van Essen, D. C. et al. The WU-Minn Human Connectome Project: An overview. Neuroimage (2013) doi:10.1016/j.neuroimage.2013.05.041.
Babayan, A. et al. A mind-brain-body dataset of MRI, EEG, cognition, emotion, and peripheral physiology in young and old adults. Sci Data 6, 180308 (2019).
Dockès, J. et al. NeuroQuery, comprehensive meta-analysis of human brain mapping. Elife 9, (2020).
Wasserthal, J., Neher, P. & Maier-Hein, K. H. TractSeg - Fast and accurate white matter tract segmentation. Neuroimage 183, 239–253 (2018).
Calhoun, V. D., Adali, T., Pearlson, G. D. & Pekar, J. J. A method for making group inferences from functional MRI data using independent component analysis. Hum Brain Mapp (2001) doi:10.1002/hbm.1048.
Erhardt, E. B. et al. Comparison of multi-subject ICA methods for analysis of fMRI data. Hum Brain Mapp 32, 2075–2095 (2011).
Basile, G. A. et al. In Vivo Super-Resolution Track-Density Imaging for Thalamic Nuclei Identification. Cerebral Cortex (2021) doi:10.1093/cercor/bhab184.
Basile, G. A. et al. In vivo probabilistic atlas of white matter tracts of the human subthalamic area combining track density imaging and optimized diffusion tractography. Brain Struct Funct 227, 2647–2665 (2022).
Burt, J. B., Helmer, M., Shinn, M., Anticevic, A. & Murray, J. D. Generative modeling of brain maps with spatial autocorrelation. Neuroimage 220, 117038 (2020).
Sotiropoulos, S. N. et al. Advances in diffusion MRI acquisition and processing in the Human Connectome Project. Neuroimage 80, 125–43 (2013).
Smith, S. M. et al. Resting-state fMRI in the Human Connectome Project. Neuroimage 80, 144–168 (2013).
Leonardi, N. & Van De Ville, D. On spurious and real fluctuations of dynamic functional connectivity during rest. Neuroimage 104, 430–436 (2015).
Heitmann, S. & Breakspear, M. Putting the “dynamic” back into dynamic functional connectivity. Network Neuroscience 2, 150–174 (2018).
Liégeois, R., Laumann, T. O., Snyder, A. Z., Zhou, J. & Yeo, B. T. T. Interpreting temporal fluctuations in resting-state functional connectivity MRI. Neuroimage 163, 437–455 (2017).
Laumann, T. O. et al. On the Stability of BOLD fMRI Correlations. Cerebral Cortex (2016) doi:10.1093/cercor/bhw265.
Lurie, D. J. et al. Questions and controversies in the study of time-varying functional connectivity in resting fMRI. Network Neuroscience 4, 30–69 (2020).
Allen, E. A., Damaraju, E., Eichele, T., Wu, L. & Calhoun, V. D. EEG Signatures of Dynamic Functional Network Connectivity States. Brain Topogr 31, 101–116 (2018).
Kucyi, A. et al. Intracranial Electrophysiology Reveals Reproducible Intrinsic Functional Connectivity within Human Brain Networks. The Journal of Neuroscience 38, 4230–4242 (2018).
Matsui, T., Murakami, T. & Ohki, K. Neuronal Origin of the Temporal Dynamics of Spontaneous BOLD Activity Correlation. Cerebral Cortex 29, 1496–1508 (2019).
Liégeois, R. et al. Cerebral functional connectivity periodically (de)synchronizes with anatomical constraints. Brain Struct Funct 221, 2985–2997 (2016).
Sadaghiani, S., Poline, J.-B., Kleinschmidt, A. & D’Esposito, M. Ongoing dynamics in large-scale functional connectivity predict perception. Proceedings of the National Academy of Sciences 112, 8463–8468 (2015).
Shine, J. M. et al. The Dynamics of Functional Brain Networks: Integrated Network States during Cognitive Task Performance. Neuron 92, 544–554 (2016).
Kucyi, A. & Davis, K. D. Dynamic functional connectivity of the default mode network tracks daydreaming. Neuroimage 100, 471–480 (2014).
Kucyi, A. Just a thought: How mind-wandering is represented in dynamic brain connectivity. Neuroimage 180, 505–514 (2018).
Fan, L. et al. Brain parcellation driven by dynamic functional connectivity better capture intrinsic network dynamics. Hum Brain Mapp 42, 1416–1433 (2021).
Axer, H., Klingner, C. M. & Prescher, A. Fiber anatomy of dorsal and ventral language streams. Brain Lang 127, 192–204 (2013).
Dick, A. S., Bernal, B. & Tremblay, P. The Language Connectome. The Neuroscientist 20, 453–467 (2014).
Dick, A. S. & Tremblay, P. Beyond the arcuate fasciculus: consensus and controversy in the connectional anatomy of language. Brain 135, 3529–3550 (2012).
Vavassori, L., Sarubbo, S. & Petit, L. Hodology of the superior longitudinal system of the human brain: a historical perspective, the current controversies, and a proposal. Brain Struct Funct 226, 1363–1384 (2021).
Cocquyt, E.-M. et al. The white matter architecture underlying semantic processing: A systematic review. Neuropsychologia 136, 107182 (2020).
Sarubbo, S. et al. Structural and functional integration between dorsal and ventral language streams as revealed by blunt dissection and direct electrical stimulation. Hum Brain Mapp 37, 3858–3872 (2016).
Sarubbo, S. et al. Mapping critical cortical hubs and white matter pathways by direct electrical stimulation: an original functional atlas of the human brain. Neuroimage 205, 116237 (2020).
Giampiccolo, D. & Duffau, H. Controversy over the temporal cortical terminations of the left arcuate fasciculus: a reappraisal. Brain 145, 1242–1256 (2022).
Duffau, H. et al. New insights into the anatomo-functional connectivity of the semantic system: a study using cortico-subcortical electrostimulations. Brain 128, 797–810 (2005).
Hamilton, L. S., Oganian, Y., Hall, J. & Chang, E. F. Parallel and distributed encoding of speech across human auditory cortex. Cell 184, 4626-4639.e13 (2021).
Mesgarani, N., Cheung, C., Johnson, K. & Chang, E. F. Phonetic Feature Encoding in Human Superior Temporal Gyrus. Science (1979) 343, 1006–1010 (2014).
Oganian, Y. & Chang, E. F. A speech envelope landmark for syllable encoding in human superior temporal gyrus. Sci Adv 5, (2019).
Yi, H. G., Leonard, M. K. & Chang, E. F. The Encoding of Speech Sounds in the Superior Temporal Gyrus. Neuron 102, 1096–1110 (2019).
Ozker, M. et al. Speech-induced suppression and vocal feedback sensitivity in human cortex. bioRxiv 2023.12.08.570736 (2024) doi:10.1101/2023.12.08.570736.
Tate, M. C., Herbet, G., Moritz-Gasser, S., Tate, J. E. & Duffau, H. Probabilistic map of critical functional regions of the human cerebral cortex: Broca’s area revisited. Brain 137, 2773–2782 (2014).
Duffau, H., Gatignol, P., Mandonnet, E., Capelle, L. & Taillandier, L. Intraoperative subcortical stimulation mapping of language pathways in a consecutive series of 115 patients with Grade II glioma in the left dominant hemisphere. J Neurosurg 109, 461–471 (2008).
Hart, J. & Gordon, B. Delineation of single‐word semantic comprehension deficits in aphasia, with anatomical correlation. Ann Neurol 27, 226–231 (1990).
Hoffman, P., Pobric, G., Drakesmith, M. & Lambon Ralph, M. A. Posterior middle temporal gyrus is involved in verbal and non-verbal semantic cognition: Evidence from rTMS. Aphasiology 26, 1119–1130 (2012).
Python, G., Glize, B. & Laganaro, M. The involvement of left inferior frontal and middle temporal cortices in word production unveiled by greater facilitation effects following brain damage. Neuropsychologia 121, 122–134 (2018).
Janssen, N. et al. How the speed of word finding depends on ventral tract integrity in primary progressive aphasia. Neuroimage Clin 28, 102450 (2020).
Hula, W. D. et al. Structural white matter connectometry of word production in aphasia: an observational study. Brain 143, 2532–2544 (2020).
Cohen, L., Jobert, A., Le Bihan, D. & Dehaene, S. Distinct unimodal and multimodal regions for word processing in the left temporal cortex. Neuroimage 23, 1256–1270 (2004).
Nobre, A. C., Allison, T. & McCarthy, G. Word recognition in the human inferior temporal lobe. Nature 372, 260–263 (1994).
Matsumoto, R. et al. Functional connectivity in the human language system: a cortico-cortical evoked potential study. Brain 127, 2316–2330 (2004).
Talozzi, L. et al. Latent disconnectome prediction of long-term cognitive-behavioural symptoms in stroke. Brain 146, 1963–1978 (2023).
Pacella, V., Nozais, V., Talozzi, L., Forkel, S. J. & de Schotten, M. T. Unravelling the fabric of the human mind: the brain-cognition space. Res Sq (2022) doi:https://doi.org/10.21203/rs.3.rs-2260331/v1.
Ross, E. D. & Monnot, M. Neurology of affective prosody and its functional–anatomic organization in right hemisphere. Brain Lang 104, 51–74 (2008).
Witteman, J., van IJzendoorn, M. H., van de Velde, D., van Heuven, V. J. J. P. & Schiller, N. O. The nature of hemispheric specialization for linguistic and emotional prosodic perception: A meta-analysis of the lesion literature. Neuropsychologia 49, 3722–3738 (2011).
Davis, C. L. et al. White matter tracts critical for recognition of sarcasm. Neurocase 22, 22–29 (2016).
Gajardo-Vidal, A. et al. How right hemisphere damage after stroke can impair speech comprehension. Brain 141, 3389–3404 (2018).
Lin, B., Hon, F., Lin, M., Tsai, P. & Lu, C. Right arcuate fasciculus as outcome predictor after low‐frequency repetitive transcranial magnetic stimulation in nonfluent aphasic stroke. Eur J Neurol 30, 2031–2041 (2023).
Forkel, S. J. et al. Anatomical predictors of aphasia recovery: a tractography study of bilateral perisylvian language networks. Brain 137, 2027–2039 (2014).
Parkinson, C. & Wheatley, T. Relating Anatomical and Social Connectivity: White Matter Microstructure Predicts Emotional Empathy. Cerebral Cortex 24, 614–625 (2014).
Barbey, A. K., Colom, R. & Grafman, J. Distributed neural system for emotional intelligence revealed by lesion mapping. Soc Cogn Affect Neurosci 9, 265–272 (2014).
Cabinio, M. et al. Mind-Reading Ability and Structural Connectivity Changes in Aging. Front Psychol 6, (2015).
Sarwar, T., Ramamohanarao, K. & Zalesky, A. Mapping connectomes with diffusion MRI: deterministic or probabilistic tractography? Magn Reson Med 81, 1368–1384 (2019).
Wassermann, D. et al. The white matter query language: a novel approach for describing human white matter anatomy. Brain Struct Funct 221, 4705–4721 (2016).
Yendiki, A. Automated probabilistic reconstruction of white-matter pathways in health and disease using an atlas of the underlying anatomy. Front Neuroinform 5, (2011).
O’Donnell, L. J. et al. Automated white matter fiber tract identification in patients with brain tumors. Neuroimage Clin 13, 138–153 (2017).
Thiebaut De Schotten, M. et al. A lateralized brain network for visuospatial attention. Nat Neurosci 14, 1245–1246 (2011).
Forkel, S. J. et al. The anatomy of fronto-occipital connections from early blunt dissections to contemporary tractography. Cortex 56, 73–84 (2014).
Preti, M. G., Bolton, T. A. & Van De Ville, D. The dynamic functional connectome: State-of-the-art and perspectives. Neuroimage 160, 41–54 (2017).
Laird, A. R. et al. Investigating the Functional Heterogeneity of the Default Mode Network Using Coordinate-Based Meta-Analytic Modeling. The Journal of Neuroscience 29, 14496–14505 (2009).
Yarkoni, T., Poldrack, R. A., Nichols, T. E., Van Essen, D. C. & Wager, T. D. Large-scale automated synthesis of human functional neuroimaging data. Nat Methods 8, 665–670 (2011).
Poldrack, R. Can cognitive processes be inferred from neuroimaging data? Trends Cogn Sci 10, 59–63 (2006).
Van Essen, D. C. et al. The Human Connectome Project: A data acquisition perspective. Neuroimage 62, 2222–2231 (2012).
Uǧurbil, K. et al. Pushing spatial and temporal resolution for functional and diffusion MRI in the Human Connectome Project. Neuroimage 80, 80–104 (2013).
Patenaude, B., Smith, S. M., Kennedy, D. N. & Jenkinson, M. A Bayesian model of shape and appearance for subcortical brain segmentation. Neuroimage 56, 907–922 (2011).
Smith, S. M. et al. Advances in functional and structural MR image analysis and implementation as FSL. in NeuroImage (2004). doi:10.1016/j.neuroimage.2004.07.051.
Glasser, M. F. et al. The minimal preprocessing pipelines for the Human Connectome Project. Neuroimage 80, 105–124 (2013).
Andersson, J. L. R., Jenkinson, M., Smith, S. & Andersson, J. FNIRT — FMRIB’ Non-Linear Image Registration Tool. Oxford Centre for Functional Magnetic Resonance imaging of the Brain, Department of Clinical Neurology, Oxford University, Oxford, UK (2007).
Jenkinson, M. & Smith, S. A global optimisation method for robust affine registration of brain images. Med Image Anal (2001) doi:10.1016/S1361-8415(01)00036-6.
Jenkinson, M., Bannister, P., Brady, M. & Smith, S. Improved optimization for the robust and accurate linear registration and motion correction of brain images. Neuroimage (2002).
Benhajali, Y. et al. A Standardized Protocol for Efficient and Reliable Quality Control of Brain Registration in Functional MRI Studies. Front Neuroinform 14, 7 (2020).
Tournier, J. D. et al. MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation. NeuroImage Preprint at https://doi.org/10.1016/j.neuroimage.2019.116137 (2019).
Glasser, M. F. et al. The minimal preprocessing pipelines for the Human Connectome Project. Neuroimage 80, 105–124 (2013).
Veraart, J. et al. Denoising of diffusion MRI using random matrix theory. Neuroimage (2016) doi:10.1016/j.neuroimage.2016.08.016.
Kellner, E., Dhital, B., Kiselev, V. G. & Reisert, M. Gibbs-ringing artifact removal based on local subvoxel-shifts. Magn Reson Med 76, 1574–1581 (2016).
Andersson, J. L. R. & Sotiropoulos, S. N. An integrated approach to correction for off-resonance effects and subject movement in diffusion MR imaging. Neuroimage 125, 1063–1078 (2016).
Andersson, J. L. R., Skare, S. & Ashburner, J. How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging. Neuroimage 20, 870–888 (2003).
Tustison, N. J. et al. N4ITK: Improved N3 Bias Correction. IEEE Trans Med Imaging 29, 1310–1320 (2010).
Salimi-Khorshidi, G. et al. Automatic denoising of functional MRI data: Combining independent component analysis and hierarchical fusion of classifiers. Neuroimage 90, 449–468 (2014).
Plachti, A. et al. Multimodal Parcellations and Extensive Behavioral Profiling Tackling the Hippocampus Gradient. Cerebral Cortex 29, 4595–4612 (2019).
Mendes, N. et al. A functional connectome phenotyping dataset including cognitive state and personality measures. Sci Data 6, 180307 (2019).
Whitfield-Gabrieli, S. & Nieto-Castanon, A. Conn : A Functional Connectivity Toolbox for Correlated and Anticorrelated Brain Networks. Brain Connect 2, 125–141 (2012).
Tournier, J. D. et al. Resolving crossing fibres using constrained spherical deconvolution: Validation using diffusion-weighted imaging phantom data. Neuroimage 42, 617–625 (2008).
Jeurissen, B., Tournier, J. D., Dhollander, T., Connelly, A. & Sijbers, J. Multi-tissue constrained spherical deconvolution for improved analysis of multi-shell diffusion MRI data. Neuroimage (2014) doi:10.1016/j.neuroimage.2014.07.061.
Calamuneri, A. et al. White Matter Tissue Quantification at Low b-Values Within Constrained Spherical Deconvolution Framework. Front Neurol 9, 716 (2018).
Dhollander, T., Raffelt, D. & Connelly, A. Unsupervised 3-tissue response function estimation from single-shell or multi-shell diffusion MR data without a co-registered T1 image. ISMRM Workshop on Breaking the Barriers of Diffusion MRI (2016).
Descoteaux, M., Deriche, R., Knösche, T. R. & Anwander, A. Deterministic and probabilistic tractography based on complex fibre orientation distributions. IEEE Trans Med Imaging (2009) doi:10.1109/TMI.2008.2004424.
Zalesky, A. & Breakspear, M. Towards a statistical test for functional connectivity dynamics. Neuroimage 114, 466–470 (2015).
Bell, A. J. & Sejnowski, T. J. An Information-Maximization Approach to Blind Separation and Blind Deconvolution. Neural Comput 7, 1129–1159 (1995).
Du, Y. & Fan, Y. Group information guided ICA for fMRI data analysis. Neuroimage 69, 157–197 (2013).
Rousseeuw, P. J. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. J Comput Appl Math 20, 53–65 (1987).
Calamante, F. et al. Track-weighted functional connectivity (TW-FC): A tool for characterizing the structural–functional connections in the brain. Neuroimage 70, 199–210 (2013).

There is NO Competing Interest.

SupplementaryFile1.xlsx

Download PDF

Version 1

posted

You are reading this latest preprint version

Functional anatomy and topographical organization of the frontotemporal arcuate fasciculus

Status:

Version 1

Abstract

Figures

Introduction

Results

1. Functional activity in the arcuate fasciculus is best decomposed into two independent components

2. Independent components map on three anatomically distinct segments of the AF

3. Distinct AF clusters correlation patterns with meta-analytic maps

Discussion

1. Structure and function in the AF: independent components of dynamic brain activity along the AF

2. Three new divisions of the frontotemporal arcuate fasciculus

3. Functional characterization of AF clusters: relevance to language processing and cognitive function

4. Technical issues and limitations

Conclusion

Methods

1. Participants and data acquisition

1.1. Primary and test-retest datasets (HCP)

1.2. Validation dataset (LEMON)

2. Data preprocessing

2.1. Structural preprocessing

2.2. DWI preprocessing

2.3. Resting-state fMRI preprocessing

2.4. Bundle-specific tractography and tw-dFC

3. Group-level analysis

3.1. Group ICA-based parcellation

3.2. Data-driven dimensionality selection

3.3. Component-driven AF parcellation

3.4. Meta-analytic functional decoding

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1