A Spatiotemporal Map of Reading Aloud

TL;DR: This article found that lexicality is encoded by early activity in mid-fusiform (mFus) cortex and precentral sulcus, followed by later engagement of the inferior frontal gyrus (IFG) and inferior parietal sulcus (IPS), and orthographic neighborhood is encoded solely in the IPS.

Abstract: Reading words aloud is a foundational aspect of the acquisition of literacy. The rapid rate at which multiple distributed neural substrates are engaged in this process can only be probed via techniques with high spatiotemporal resolution. We used direct intracranial recordings in a large cohort to create a holistic yet fine-grained map of word processing, enabling us to derive the spatiotemporal neural codes of multiple word attributes critical to reading: lexicality, word frequency and orthographic neighborhood. We found that lexicality is encoded by early activity in mid-fusiform (mFus) cortex and precentral sulcus. Word frequency is also first represented in mFus followed by later engagement of the inferior frontal gyrus (IFG) and inferior parietal sulcus (IPS), and orthographic neighborhood is encoded solely in the IPS. A lexicality decoder revealed high weightings for electrodes in the mFus, IPS, anterior IFG and the pre-central sulcus. These results elaborate the neural codes underpinning extant dual-route models of reading, with parallel processing via the lexical route, progressing from mFus to IFG, and the sub-lexical route, progressing from IPS to anterior IFG.

Summary

Introduction

• Reading a word aloud requires multiple complex transformations in the brain -mapping the visual input of a letter string into an internal sequence of sound representations that are then expressed through orofacial motor articulations.
• Ventral temporal cortex, particularly mid-fusiform cortex (mFus), is strongly associated with the lexical route.
• The two routes are proposed to converge in the inferior frontal gyrus (IFG) (Taylor et al., 2013) .
• The authors utilized intracranial recordings in a large cohort of patients (44 patients, 3,642 electrodes), with medically intractable epilepsy, while they read aloud known and novel words.

Results

• Participants were visually presented with phonologically regular words, exception words and novel pseudowords that they read aloud .
• Electrophysiological recordings were performed from a total of 3,642 separate intracranial electrodes placed for the localization of intractable epilepsy.

Spatiotemporal Mapping of Single Word Reading

• The authors used a mixed-effects, multilevel analysis (MEMA) of broadband gamma activity (BGA; 70-150 Hz) in group surface normalized space to create a population level map of cortical activation across the population.
• All correctly articulated trials across all word classes, were used.
• To create a more focused visualization of the spatiotemporal progression across reading-sensitive cortex, the authors selected 12 regions of interest (ROIs) in areas thought to be important to written word processing, speech production and speech monitoring .
• CC-BY-NC-ND 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

Spatiotemporal Representation of Lexical Factors

• To distinguish activity patterns across word classes the authors contrasted grouped gamma power activations between exception vs. pseudowords and exception vs. regular words using MEMA.
• With some patients showing >80% decoding accuracy .
• The authors observed lexicality distinctions between known words (regular and exception) and novel pseudowords broadly across the previously defined ROIs .
• Sensitivity to orthographic neighborhood of pseudowords was only seen in IPS (500-700 ms).
• Sensitivity to word frequency was observed earliest in mFus (200 ms) followed by IPS and aIFG (425 ms) .

Discussion

• This large population intracranial study comprehensively maps the spatiotemporal spread of cortical activation across the left hemisphere during word reading to derive the dynamics of cortical networks underlying literacy.
• It is commonly assumed that sensitivity to statistical properties of language such as word frequency seen in ventral temporal cortex are as a result of top-down modulation from IFG (Heim et al., 2013; Price and Devlin, 2011; Woodhead et al., 2014) .
• This consolidates mFus's role as a specialized orthographic lexicon, organized based on statistical regularities of individual words in natural language.
• The IPS was the only region with sensitivity to orthographic neighborhood.
• Given the association of pCS with articulation phonology and phonological dyslexia, this may represent part of the process of constructing novel phonologies.

Materials and Methods

• All participants were semi-chronically implanted with intracranial electrodes for seizure localization of pharmaco-resistant epilepsy.
• Electrode Implantation and Data Recording: Data were acquired from either subdural grid electrodes (SDEs; 4 patients) or stereotactically placed depth electrodes (sEEGs; 40 patients).
• Each stimulus was displayed for 1,500 ms with an inter-stimulus interval of 2,000 ms.
• Analyses were performed by first bandpass filtering raw data of each electrode into broadband gamma activity (BGA; 70-150Hz) following removal of line noise (zero-phase 2nd order Butterworth bandstop filters), also known as Signal Analysis.
• The authors quantified word frequency as the base-10 log of the SUBTLEXus frequency (Brysbaert and New, 2009) .

Full text

1
A Spatiotemporal Map of Reading Aloud
Oscar Woolnough
1,2
, Cristian Donos
1,3
, Aidan Curtis
1
, Patrick S. Rollo
1,2
, Zachary J. Roccaforte
1,2
,
Stanislas Dehaene
4,5
, Simon Fischer-Baum
6
, Nitin Tandon
1,2,7,*
1
Vivian L. Smith Department of Neurosurgery, McGovern Medical School at UT Health Houston,
Houston, TX, 77030, United States of America
2
Texas Institute for Restorative Neurotechnologies, University of Texas Health Science Center at
Houston, Houston, TX, 77030, United States of America
3
Faculty of Physics, University of Bucharest, 050663, Bucharest, Romania
4
Cognitive Neuroimaging Unit CEA, INSERM, NeuroSpin Center, Université Paris-Sud and
Université Paris-Saclay, 91191, Gif-sur-Yvette, France
5
Collège de France, 11 Place Marcelin Berthelot, 75005, Paris, France
6
Department of Psychological Sciences, Rice University, Houston, TX, 77005, USA
7
Memorial Hermann Hospital, Texas Medical Center, Houston, TX, 77030, United States of
America
* Correspondence: nitin.tandon@uth.tmc.edu
Abstract
1
Reading words aloud is a foundational aspect of the acquisition of literacy. The rapid rate at which
2
multiple distributed neural substrates are engaged in this process can only be probed via
3
techniques with high spatiotemporal resolution. We used direct intracranial recordings in a large
4
cohort to create a holistic yet fine-grained map of word processing, enabling us to derive the
5
spatiotemporal neural codes of multiple word attributes critical to reading: lexicality, word frequency
6
and orthographic neighborhood. We found that lexicality is encoded by early activity in mid-fusiform
7
(mFus) cortex and precentral sulcus. Word frequency is also first represented in mFus followed by
8
later engagement of the inferior frontal gyrus (IFG) and inferior parietal sulcus (IPS), and
9
orthographic neighborhood is encoded solely in the IPS. A lexicality decoder revealed high
10
weightings for electrodes in the mFus, IPS, anterior IFG and the pre-central sulcus. These results
11
elaborate the neural codes underpinning extant dual-route models of reading, with parallel
12
processing via the lexical route, progressing from mFus to IFG, and the sub-lexical route,
13
progressing from IPS to anterior IFG.
14
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted May 23, 2021. ; https://doi.org/10.1101/2021.05.23.445307doi: bioRxiv preprint

2
Introduction
15
Reading a word aloud requires multiple complex transformations in the brain - mapping the visual
16
input of a letter string into an internal sequence of sound representations that are then expressed
17
through orofacial motor articulations. Models of how this mapping occurs during reading invoke a
18
dual-route architecture (Coltheart et al., 2001; Perry et al., 2007, 2010, 2019; Taylor et al., 2013),
19
with a lexico-semantic route for rapidly reading known words and a sub-lexical route for constructing
20
the phonology of novel words. A common method of targeting these two routes is to look at
21
contrasts between phonological exception words and pseudowords (Fiebach et al., 2002; Sebastian
22
et al., 2014; Shim et al., 2012; Taylor et al., 2013). Exception words contain irregular grapheme-
23
phoneme associations (e.g. yacht, sew) and their phonologies must be retrieved from internal
24
lexical representations as they cannot be accurately constructed de novo. In contrast, pseudowords
25
have no stored representation and their phonology must be constructed rather than retrieved.
26
Ventral temporal cortex, particularly mid-fusiform cortex (mFus), is strongly associated with the
27
lexical route. mFus is heavily implicated as the site of the orthographic lexicon, the long-term
28
memory storage of which letter strings map onto known words (Glezer et al., 2015; Hirshorn et al.,
29
2016; Kronbichler et al., 2004; Lochy et al., 2018; White et al., 2019; Woolnough et al., 2021). This
30
region is sensitive to lexicality and word frequency (Kronbichler et al., 2004; White et al., 2019;
31
Woolnough et al., 2021), and shows selective changes during visual word learning (Glezer et al.,
32
2015; Taylor et al., 2019). The sub-lexical route, essential for articulating novel words, is thought to
33
engage the inferior parietal lobe (IPL), dysfunction of which is associated with dyslexia (Raschle et
34
al., 2011; Temple et al., 2003; Tomasino et al., 2020), dysgraphia (Rapp et al., 2016), in addition to
35
phonological and semantic deficits (Binder et al., 2009; Hula et al., 2020; Numssen et al., 2021).
36
The two routes are proposed to converge in the inferior frontal gyrus (IFG) (Taylor et al., 2013).
37
The majority of our knowledge regarding the neural architecture underlying reading aloud is derived
38
from lesion data and functional MRI which provide accurate spatial localizations of function but lack
39
crucial temporal information. We utilized intracranial recordings in a large cohort of patients (44
40
patients, 3,642 electrodes), with medically intractable epilepsy, while they read aloud known and
41
novel words. This allowed us to comprehensively map the flow of information through these cortical
42
networks and track the spatiotemporal dynamics of the cortical representation of behaviorally
43
relevant lexical and sub-lexical factors.
44
45
Results
46
Participants were visually presented with phonologically regular words, exception words and novel
47
pseudowords that they read aloud (Figure 1A). Electrophysiological recordings were performed from
48
a total of 3,642 separate intracranial electrodes placed for the localization of intractable epilepsy
49
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted May 23, 2021. ; https://doi.org/10.1101/2021.05.23.445307doi: bioRxiv preprint

3
(Figure 1B,C) - 4 participants had subdural grid electrodes (SDEs) and 40 had depth recordings
50
using stereotactic EEG electrodes (sEEGs).
51
52
Figure 1: Experimental Design and Electrode Coverage. (A) Schematic representation of the
53
reading task. (B) Representative coverage map (44 patients) and (C) individual electrode locations
54
(3,642 electrodes) for the left hemisphere, highlighting responsive electrodes (1,158 electrodes;
55
>20% activation above baseline).
56
57
Behavioral Analysis
58
Mean (± SD) response times (RTs) were: regular words (743 ± 122 ms), exception words (747 ±
59
125 ms) and pseudowords (923 ± 193 ms) (Figure 2A). Regular and exception words showed no
60
difference in RT (Wilcoxon sign rank, p = 0.75; ln(Bayes Factor (BF
10
)) = -1.5) though pseudoword
61
RT was slower than for exception words (p < 10
-8
, ln(BF
10
) = 28).
62
63
Figure 2: Population Word Response Times. (A) Response time distribution for each of the three
64
word classes, averaged within participant, (B) Mean (± SE) response times for each item within the
65
three word classes, averaged across participants.
66
67
To determine the underlying properties of the words that modulate RT within this cohort, we
68
performed linear mixed effects (LME) and Bayes factor (BF) analyses on each word class with fixed
69
kaize
+
group
+
2000ms
1500ms
2000ms
1500ms
A
B
C
1 8+
# Patients
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted May 23, 2021. ; https://doi.org/10.1101/2021.05.23.445307doi: bioRxiv preprint

4
effects modelling linguistic factors commonly linked to word identification and articulation (Table 1).
70
Regular words and exception word RTs showed the greatest modulation by word frequency.
71
Pseudoword RT was most strongly associated with orthographic neighborhood.
72
Regular
df = 3170, r
2
= 0.36
Exception
df = 3098, r
2
= 0.35
Pseudowords
df = 3185, r
2
= 0.40
β (SE)
p
ln(BF
10
)
β (SE)
p
ln(BF
10
)
β (SE)
p
ln(BF
10
)
Length
49
(17)
0.004
0.9
48
(22)
0.03
-0.7
23
(26)
0.38
-3
Word Frequency
-186
(17)
<10
-27
59
-154
(16)
<10
-21
43
-
-
-
Orthographic
Neighborhood
52
(27)
0.05
-0.9
-97
(35)
0.005
1.4
227
(33)
<10
-11
21
Phonological
Neighborhood
20
(18)
0.26
-2.5
-7
(16)
0.63
-3.2
58
(20)
0.004
0.5
Positional Letter
Frequency
13
(14)
0.89
-2.9
-16
(16)
0.29
-2.9
-50
(19)
0.009
-0.3
73
Table 1: Statistical Modelling of Response Time. As predictors were normalized, β values
74
approximate change in RT between extreme values within the entire stimulus set (Supplementary
75
Table 1). Factors with strong evidence of an effect (ln(BF
10
) > 2.3) are highlighted.
76
77
Spatiotemporal Mapping of Single Word Reading
78
We used a mixed-effects, multilevel analysis (MEMA) of broadband gamma activity (BGA; 70-150
79
Hz) in group surface normalized space to create a population level map of cortical activation across
80
the population. This analysis is specifically designed to account for sampling variations and to
81
minimize effects of outliers (Argall et al., 2006; Conner et al., 2014; Esposito et al., 2013; Fischl et
82
al., 1999; Kadipasaoglu et al., 2014; Saad and Reynolds, 2012). All correctly articulated trials
83
across all word classes, were used. 4D representations of the spread of activation across the
84
cortical surface were generated by performing MEMA on short, overlapping time windows (150 ms
85
width, 10 ms spacing) to generate successive images of cortical activity, time locked to stimulus
86
onset (Video 1) or the onset of articulation (Video 2). The spatial distribution of activations was
87
highly comparable across word classes (Supplementary Figure 1).
88
By collapsing across these frames, we visualized peak activations at each point on the cortical
89
surface (Figure 3A). To create a more focused visualization of the spatiotemporal progression
90
across reading-sensitive cortex, we selected 12 regions of interest (ROIs) in areas thought to be
91
important to written word processing, speech production and speech monitoring (Figure 3B,C). This
92
analysis highlights regions displaying primarily pre-articulatory processes, in ventral
93
occipitotemporal cortex, inferior parietal lobe and the inferior frontal gyrus.
94
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted May 23, 2021. ; https://doi.org/10.1101/2021.05.23.445307doi: bioRxiv preprint

5
95
Figure 3: Spatiotemporal Profile of Cortical Activations. (A) Collapsed articulation-locked
96
activation movie (Video 2) highlighting the amplitude of peak activation. (B) Representative ROIs in
97
12 anatomically and functionally distinct regions, showing all responsive electrodes. (C) Mean
98
activation during word reading of each ROI, averaged within patient, time locked to stimulus onset
99
(left) and articulation onset (right). Standard errors omitted for visual clarity. LOT, Lateral
100
OccipitoTemporal cortex; mFus, mid-Fusiform Cortex; IPS, Inferior Parietal Sulcus; pCS, pre-
101
Central Sulcus; pIFG, posterior Inferior Frontal Gyrus; aIFG, anterior Inferior Frontal Gyrus; FO,
102
Frontal Operculum; iMC, inferior Motor Cortex; SMG, Supra Marginal Gyrus; SMA, Supplementary
103
Motor Area; PI, Posterior Insula; STG, Superior Temporal Gyrus.
104
105
Spatiotemporal Representation of Lexical Factors
106
To distinguish activity patterns across word classes we contrasted grouped gamma power
107
activations between exception vs. pseudowords (lexicality) and exception vs. regular words
108
(regularity) using MEMA. The lexicality contrasts demonstrated clusters in mFus, precentral sulcus
109
(pCS), inferior parietal sulcus (IPS) and anterior inferior frontal gyrus (aIFG).
110
-500 0 500 1000
Time (ms)
0
25
50
75
100
125
% BGA
LOT
mFus
IPS
pCS
pIFG
aIFG
FO
iMC
SMG
SMA
PI
STG
-1000 -500 0 500
0
25
50
75
100
125
Peak %BGA
10 50
A
C
B
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted May 23, 2021. ; https://doi.org/10.1101/2021.05.23.445307doi: bioRxiv preprint

Frequently asked questions

Q1. What contributions have the authors mentioned in the paper "A spatiotemporal map of reading aloud" ?

The authors used direct intracranial recordings in a large 4 cohort to create a holistic yet fine-grained map of word processing, enabling us to derive the 5 spatiotemporal neural codes of multiple word attributes critical to reading: lexicality, word frequency 6 and orthographic neighborhood. 4. 0 International license made available under a ( which was not certified by peer review ) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. 

Q2. How many stimuli were presented in a pseudorandom order?

Stimuli were presented in two 250 recording sessions, each containing presentation of 120 stimuli in a pseudorandom order with no 251repeats. 

Q3. What is the timing of the cessation of sub-lexical processes?

Given the word frequency 191 dependence of lexical identification, the timing of the cessation of sub-lexical processes should also 192 be frequency dependent. 

Q4. What is the role of pCS in phonological processing?

Their data demonstrate that pCS activation begins early, preceding the IFG, suggesting 198 a role in early linguistic or phonological processing, potentially as part of the sub-lexical route. 

Q5. What was the amplitude of the sigmoid flanks?

A frequency domain bandpass Hilbert transform (paired sigmoid flanks 257 with half-width 1.5 Hz) was applied and the analytic amplitude was smoothed (Savitzky - Golay finite 258 impulse response, 3rd order, frame length of 201 ms). 

Q6. What is the effect of fMRI on lexicality?

These data 174 minimize the impact of response time variations, which confounds modalities with lower temporal 175 resolution (e.g. fMRI) and may artificially inflate lexicality effects in regions such as IFG (Taylor et 176 al., 2014). 

Q7. What is the mean response time for the three word classes?

72Regulardf = 3170, r 2 = 0.36Exceptiondf = 3098, r 2 = 0.35Pseudowordsdf = 3185, r 2 = 0.40β (SE) p ln(BF10) β (SE) p ln(BF10) β (SE) p ln(BF10)Length 49(17) 0.004 0.948 (22)0.03 -0.7 23(26) 0.38 -3Word Frequency -186 (17)<10 -2759 -154 (16)<10 -2143 - - -Orthographic Neighborhood52 (27)0.05 -0.9 -97 (35)0.005 1.4 227 (33)<10 -1121Phonological Neighborhood20 (18)0.26 -2.5 -7(16) 0.63 -3.258 (20)0.004 0.5Positional Letter Frequency13 (14)0.89 -2.9 -16 (16)0.29 -2.9 -50 (19)0.009 -0.373Table 1: Statistical Modelling of Response Time. 

Q8. What was the underlying property of the word class that modulates RT?

6667To determine the underlying properties of the words that modulate RT within this cohort, the authors 68 performed linear mixed effects (LME) and Bayes factor (BF) analyses on each word class with fixed 69kaize+group+2000ms1500ms2000ms1500msA 

Q9. What was the procedure for implantation of sEEGs?

Following surgical implantation, electrodes were localized by co-registration of 232 pre-operative anatomical 3T MRI and post-operative CT scans in AFNI (Cox, 1996). 

Q10. What is the lexicality effect of mFus?

This consolidates mFus’s role as a specialized orthographic lexicon, organized based on 184 statistical regularities of individual words in natural language.