The Development of Socially Directed Attention: A Functional Magnetic Resonance Imaging Study in Infant Monkeys

Abstract

Socially guided visual attention, such as gaze following and joint attention, represents the building block of higher-level social cognition in primates, although their neurodevelopmental processes are still poorly understood. Atypical development of these social skills has served as early marker of autism spectrum disorder and Williams syndrome. In this study, we trace the developmental trajectories of four neural networks underlying visual and attentional social engagement in the translational rhesus monkey model. Resting-state fMRI (rs-fMRI) data and gaze following skills were collected in infant rhesus macaques from birth through 6 months of age. Developmental trajectories from subjects with both resting-state fMRI and eye-tracking data were used to explore brain–behavior relationships. Our findings indicate robust increases in functional connectivity (FC) between primary visual areas (primary visual cortex [V1] – extrastriate area 3 [V3] and V3 – middle temporal area [MT], MT and anterior superior temporal sulcus area [AST], as well as between anterior temporal area [TE]) and amygdala (AMY) as infants mature. Significant FC decreases were found in more rostral areas of the pathways, such as between temporal area occipital part – TE in the ventral object pathway, V3 – lateral intraparietal (LIP) of the dorsal visual attention pathway and V3 – temporo-parietal area of the ventral attention pathway. No changes in FC were found between cortical areas LIP-FEF and temporo-parietal area – Area 12 of the dorsal and ventral attention pathways or between Anterior Superior Temporal sulcus area (AST)-AMY and AMY-insula. Developmental trajectory of gaze following revealed a period of dynamic changes with gradual increases from 1 to 2 months, followed by slight decreases from 3 to 6 months. Exploratory association findings across the 6-month period showed that infants with higher gaze following had lower FC between primary visual areas V1–V3, but higher FC in the dorsal attention areas V3-LIP, both in the right hemisphere. Together, the first 6 months of life in rhesus macaques represent a critical period for the emergence of gaze following skills associated with maturational changes in FC of socially guided attention pathways.

INTRODUCTION

Attention to social cues emerges in the first 2 years of life in humans, beginning with preferred attention to faces right after birth, followed around 3 months of age by engagement in mutual eye gaze and then by 1 year of age by gaze following that is the attention to changes in gaze direction of others (Tremblay & Rovira, 2007; Hood, Willen, & Driver, 1998). Gaze following is a necessary precursor to achieving joint attention between 10 and 12 months of age (Striano & Rochat, 2000). These early developing socially guided attention skills provide information about other individuals’ current focus of attention and clues to their future intentions. The importance of this information is underscored by evidence that, as humans, great apes (Tomasello, Hare, & Fogelman, 2001) and monkeys (Mosher, Zimmerman, & Gothard, 2011; Shepherd & Platt, 2008; Emery, Lorincz, Perrett, Oram, & Baker, 1997; Maestripieri & Wallen, 1997; Redican, 1975) automatically orient their attention to face-like stimuli, engage in mutual eye gaze, and follow the direction of another’s gaze. In addition, like humans, these social skills emerge in infancy in macaques (Wang et al., 2020; Muschinski et al., 2016; Parr et al., 2016; Rosati, Arre, Platt, & Santos, 2016; Teufel, Gutmann, Pirow, & Fischer, 2010; Farroni, Massaccesi, Pividori, & Johnson, 2004; Kuwahata, Adachi, Fujita, Tomonaga, & Matsuzawa, 2004; Tomasello et al., 2001). It is believed that these early developing skills act to accelerate the processing of social information and provide an innate mechanism for which gaze following and elaborated forms of social cognition are built (Baron-Cohen, 1992). Interestingly, disruption in these early developing social skills in humans has been described in several pervasive neurodevelopmental disorders, such as autism spectrum disorder and Williams syndrome (Mundy, 2018; D’Souza et al., 2015; Jones & Klin, 2013; Riby, Hancock, Jones, & Hanley, 2013; Lord, Luyster, Guthrie, & Pickles, 2012). Atypical appearance of these skills are often used as an early diagnostic marker of the social disorders in children at risk (American Psychiatric Association, 2013).

From birth, infant rhesus macaques need to learn to navigate the rich and complex dynamic of social groups (Machado & Bachevalier, 2003) and have provided an excellent animal model to study the development of social skills and their underlying neural substrates. They allow the collection of longitudinal behavioral and neuroimaging data at a higher sampling density, about a 4 times faster maturation rate, and with a greater image resolution for neuroimaging studies not possible in human infants. They also allow to control for difficult confounding variables in human studies (e.g., diet, resource access, maternal care, prenatal exposure to drugs).

The critical similarity in brain organization and development in rhesus macaques and humans offers a translational model to understand the critical developmental periods of specific neural networks associated with covert and overt behavioral changes in social-visual attention in humans. Longitudinal structural neuroimaging studies in both species show that the first months of life mark the most rapid period of postnatal structural growth, especially for the occipital cortex (Kovacs-Balint et al., 2019; Li, Patel, Auerbach, & Zhang, 2013; Gilmore et al., 2012), with primary sensory networks and structures maturing before integration and association networks (Kovacs-Balint et al., 2019; Gao et al., 2015; Deoni et al., 2011). Furthermore, recent studies in infant monkeys have shown the presence of a retinotopic proto-organization of the visual temporal cortex at birth, which serves as scaffolding for subsequent category-selective organization, such as faces (Arcaro & Livingstone, 2017), of high relevance for understanding the development of social visual systems in humans. Indeed, studies in infant monkeys deprived of face-related stimuli since birth indicate that face experience is necessary for the formation of face domains (Arcaro, Schade, Vincent, Ponce, & Livingstone, 2017; Sugita, 2008) and that early visual experience is crucial for normal visual cortical development during infancy. Face stimuli are processed along at least four visual cortical pathways: the object detection pathway, the motion pathway, and the dorsal and ventral visual attention pathways. Maturation of functional connectivity (FC) along these pathways (Pitcher & Ungerleider, 2021; Kovacs-Balint et al., 2019; Wang, 2019) during the first 3 months of life indicates that primary visual areas (primary visual cortex [V1] – extrastriate area 3 [V3]) show strong FC earlier than the temporal visual association areas mediating face perception (temporal area occipital part [TEO] – inferior temporal area [TE]; Mosher, Zimmerman, & Gothard, 2010; Gothard, Battaglia, Erickson, Spitler, & Amaral, 2007; Hoffman, Gothard, Schmid, & Logothetis, 2007), the motion detection areas within the superior temporal sulcus associated with perception of facial expressions, gaze following, and face and body motion (middle temporal area [MT] – AST; Roy, Shepherd, & Platt, 2014; Kamphuis, Dicke, & Thier, 2009; Pelphrey, Morris, & McCarthy, 2004; Pelphrey, Singerman, Allison, & McCarthy, 2003; Puce & Perrett, 2003; Puce et al., 2003) and the dorsal visual stream areas within the parietal cortex and the FEF, supporting covert visuospatial attention to social cues (lateral intraparietal [LIP] – FEF; Breu, Ramezanpour, Dicke, & Their, 2023; Kovacs-Balint et al., 2019; Liu, Yttri, & Snyder, 2010; Dickinson et al., 2003; Ungerleider & Mishkin, 1982). This pattern of developmental trajectory suggests that the primary visual temporal areas might support the gradual increase of looking at faces early on—before 2 months of age—whereas the maturational shifts to higher association areas in the ventral, motion, and attention pathways might relate to the strengthening of early social skills as the infants mature (Wang, Payne, Moss, Jones, & Bachevalier, 2020; Rosati et al., 2016; Teufel et al., 2010; Farroni et al., 2004; Tomasello et al., 2001). Indeed, exploratory analyses by our group showed that rhesus infants with greater increases in V1–V3 FC in early infancy have an earlier and stronger increase in attention to the eye region of faces before 2 months (Ford et al., 2023). Furthermore, the delayed maturation of FC within the more rostral, integrative, and associative socio-visual areas seen during the first 3 months of life may be associated with the delayed maturation of more complex social skills, such as gaze following becoming stronger around 5–6 months in infant monkeys (Wang, 2019; Rosati et al., 2016; Teufel et al., 2010; Tomasello et al., 2001). To test this later possibility in this study, we extended FC analyses from birth to 6 (instead of just 3) months of age and addressed individual variability by increasing the sample size of male infants that had been part of our earlier studies (Kovacs-Balint et al., 2019). We also included FC analyses in additional social visual attention networks (Vossel, Geng, Fink, 2014; Vossel, Weidner, Driver, Friston, & Fink, 2012), mainly (a) the dorsal visual attention network, including the LIP area and FEF, responsible for orienting the subjects’ attention to salient social visual stimuli (Bogadhi, Bollimunta, Leopold, & Krauzlis, 2018; Sapountzis, Paneri, & Gregoriou, 2018; Bisley, Mirpour, Arcizet, & Ong, 2011) and (b) the ventral attention network, including the TPJ and superior temporal sulcal areas and ventrolateral prefrontal cortex (vlPFC), responsible for covert (involuntary, subconscious), bottom–up (stimulus-driven) attention use to reorient attention toward salient stimuli (Ramezapour & Thier, 2020; Bogadhi et al., 2018; Herrington, Nymberg, & Schultz, 2011; Levy & Wagner, 2011; Corbetta, Patel, & Shulman, 2008; Allison, Puce, & McCarthy, 2000; Corbetta, Kincade, Ollinger, McAvoy, & Shulman, 2000; Grossman et al., 2000; Hoffman & Haxby, 2000). The selection of the ventral and dorsal visual attentional pathways for this developmental study was dictated by (a) the well-established description of these two partially segregated attention networks in the human brain (Corbetta et al., 2000, 2008) and (b) by developmental studies in humans indicating differential maturation of these two attentional networks in early infancy (3 months to 1 year; see Richards & Hunter, 2002). Finally, given that a subset of the infant monkeys in our neuroimaging study were also studied in eye-tracking experiments while they viewed social videos of conspecifics, we added an exploratory regression analysis between cortical areas with significant maturational changes in FC within the visual, motion, and attention pathways and changes in frequency of mutual eye gaze and gaze following abilities obtained at the same time points (Wang, 2019).

METHODS

Subjects

Twenty-one newborn male monkeys (Macaca mulatta), born and raised by their mothers in large, socially complex groups at Emory National Primate Research Center (ENPRC) Field Station (Lawrenceville, GA) were included in the neuroimaging study and had quality scans for resting-state FC analysis (see below). A subset of these infants, n = 12, had quality eye-tracking data for behavioral analyses of mutual gaze and gaze following skills and were also included in the regression analyses.

All infants were estimated full-term (>450 g) offspring of mid-ranking multiparous mothers, and the mother–infant pairs remained socially housed for the duration of the study. Although the goal of the study was to begin experimental procedures as soon after birth as possible, the first testing session was not done until at least 3 days postpartum to assure that mother–infant attachment was not impacted by the procedures. Mother–infant pairs were also immediately returned to their social groups after experimental procedures to limit any social impact of our studies. Exclusion criteria included: (1) infants’ or mothers’ health complications that necessitated regular veterinary care and removal from their social group; (2) infants requiring separation from their mother because of issues with lactation; or (3) infants’ rejection by their mother shortly after birth, or any situations that deprived the infant of proper maternal care. Animals were fed standard monkey chow (Purina Mills Int., Lab Diets), and water was freely available. Animals were also provided with fruits and vegetables twice daily, and additional enrichment. All procedures were approved by the Institutional Animal Care and Use Committee of Emory University, in accordance with the Animal Welfare Act and the U.S. Department of Health and Human Services “Guide for Care and Use of Laboratory Animals.” The ENPRC is fully accredited by AAALAC, International.

Resting-state fMRI

Methodological details were previously published in Kovacs-Balint and colleagues (2019). Here, we added 11 infants to that study and extended the data analyses from 12 weeks to 24 weeks. The methods are summarized below.

MRI Image Acquisition

Structural and resting-state functional MRI (rs-fMRI) data were collected longitudinally through the first 6 months of life, at 2, 4, 8, 12, 16, 20, and 24 weeks of age (the equivalent of 2 months to 2 years in human infants; see Figure 1). On the day of each scan, infants were removed from their social groups and transported to the ENPRC Imaging Center with their mothers. They were returned the next day after a 24-hr veterinary monitoring. All infants were scanned on a 3 T Siemens Magnetom Trio Tim scanner, using an 8-channel phase array head coil. To minimize motion artifacts, infants were intubated and anesthetized using telazol (2.56 ± 0.05 mg/kg body weight) and isoflurane (0.8–1% via inhalation), as low as possible to limit the effect of anesthesia on the BOLD signal used to index brain activity (Miranda-Dominguez et al., 2014; Hutchison, Womelsdorf, Gati, Everling, & Menon, 2013; Li et al., 2013; Vincent et al., 2007). During each imaging session, we collected a high-resolution, T1-weighted structural scan using a 3-D magnetization prepared rapid gradient echo parallel image sequence: inversion time = 900 msec, repetition time/echo time = 2600/3.46 msec, voxel size = 0.5 mm³ isotropic; and two 15-min rs-fMRI BOLD images using a T2*-weighted gradient-EPI sequences (EPI parameters = 400 volumes, repetition time/echo time = 2060/25 msec, voxel size = 1.5 mm³ isotropic), with a reverse phase encoding rs-fMRI scan collected for distortion correction in the EPI scans (Andersson, Skare, & Ashburner, 2003).

Figure 1.

Study design indicating the ages for collection of MRI scans and eye tracking (ET) data. Rhesus macaque developmental charts, milestones, and ages (brown) compared with humans (blue).

Rs-fMRI Data Processing

All processing steps were completed using the FMRIB Software Library (FSL, RRID: SCR_002 823; Woolrich et al., 2009; Smith et al., 2004), 4dfp tools, and a Nipype-based pipeline built in-house (Gorgolewski et al., 2011) and used in our previous infant rhesus studies (Ford et al., 2023; Kovacs-Balint et al., 2019). To improve signal-to-noise ratio and reduce signal artifacts, functional images underwent (1) unwarping using TOP-UP correction (Andersson et al., 2003), (2) slice-time correction, (3) motion correction, and (4) signal normalization. Motion correction included (1) within-run rigid body correction, (2) linear registration of EPI to T1, and (3) nonlinear registration of T1 to template. Subjects’ EPI functional images were registered to their T1-weighted structural images nonlinearly registered to age-specific rhesus macaque structural MRI atlases developed by our group (Shi et al., 2017; available at https://www.nitrc.org/projects/macaque_atlas). On the basis of best match of neuroanatomical characteristics, infant scans completed at 2 and 4 weeks were registered to the 2-weeks atlas, scans completed at ages 8, 12, and 16 weeks were registered to the 12-weeks atlas, and scans acquired at 20 and 24 weeks were registered to the 24-weeks atlas. Additional steps consisted of (1) detrending the functional signal, (2) removal of nuisance regressors using global signal regression, and (3) application of a second-order Butterworth filter (Miranda-Dominguez et al., 2014; Fair et al., 2007, 2009, 2012). Use of global signal regression is explained and justified in further detail in Kovacs-Balint and colleagues (2019).

FC Analysis

FC values were generated between (1) five ROIs along the visual object pathway (V1–V3, TEO–TE, posterior part [TEp] – amygdala [AMY], and the insula [INS]; see Figure 3 for schematic of the full pathway); (2) the ventral motion pathway (V3 – MT – superior temporal sulcus area AST – AMY; see Figure 4); (3) the dorsal visual attention pathway (V3–LIP–FEF; see Figure 5); and (4) the ventral attention pathway (V3 – temporo-parietal area [TPt] – vlPFC [Area 12]; see Figure 6). These ROIs were defined using three parcellation schemes. Label maps for the left and right AMY were manually delineated (Styner et al., 2007) and propagated to the University of North Carolina-Emory rhesus infant atlases via Advanced Normalization Tools (ANTS) (Shi et al., 2017). Remaining ROIs were mapped from established parcellations (Markov et al., 2014; Lewis & Van Essen, 2000) to the University of North Carolina-Emory atlas space. All ROIs were then evaluated and, if necessary, manually edited by experimenters in a quality control step to ensure neuroanatomical accuracy and prevent ROI overlap and signal drop out. For each subjects’ scans, FC values were calculated for every ROI. BOLD timeseries across all the voxels within each distinct ROI were averaged to determine the overall ROI activity value. These values were then correlated with the activity values of the other ROIs alongside the given pathway, creating a pathway-specific FC correlation matrix. The resulting R-coefficients were then transformed into Fisher z scores, which were used to build longitudinal trajectories of change in FC. A z score of 0 corresponds to undetected FC between two regions, meaning the two regions are not temporally correlated, that is, functionally coupled.

Figure 3.

FC within the cortical areas of the ventral visual pathway shown in the lateral view of the brain. FC between the analyzed ROIs is shown for the two hemispheres separately. FC matrices showing every possible ROI–ROI connectivity used for the network analysis within the ventral visual pathway at each age point. 1 = V1; 2 = V3; 3 = TEO; 4 = TEp; 5 = Amy; 6 = Ins; L = Left; R = Right hemisphere.

Figure 4.

FC within the cortical areas of the ventral motion pathway shown on the lateral view of the brain. FC between the analyzed ROIs is shown for the two hemispheres separately. FC matrices showing every possible ROI–ROI connectivity used for the network analysis within the ventral motion pathway at each age point. 1 = V1; 2 = V3; 7 = MT; 8 = AST; 5 = Amy; L = Left; R = Right hemisphere.

Figure 5.

FC within cortical areas of the dorsal visual attention pathway shown in the lateral view of the brain. FC between the analyzed ROIs is shown for the two hemispheres separately. FC matrices showing every possible ROI–ROI connectivity used for the network analysis within the dorsal visual attention pathway at each age point. 1 = V1; 2 = V3; 9 = LIP; 10 = FEF; L = Left; R = Right hemisphere.

Figure 6.

FC within cortical areas of the ventral attention pathway. FC between the analyzed ROIs is shown for the two hemispheres separately. FC matrices showing every possible ROI–ROI connectivity used for the network analysis within the ventral attention pathway at each age point. 1 = V1; 2 = V3; 11 = TPt; 12 = Area12; L = Left; R = Right hemisphere.

Eye-tracking Procedures

For this exploratory study, we used eye-tracking data on 12 infants that also participated in the FC neuroimaging study reported above. These infants received 14 total eye-tracking sessions at ages 1, 2, 3, 4, 5, 7, 9, 11, 13, 15, 17, 19, 21, and 23 weeks (see Figure 1) during which they spontaneously looked at short videoclips of unfamiliar conspecifics. This spontaneous viewing task has already convincingly demonstrated critical shifts in viewing patterns to faces in infant monkeys during the first 6 months of life (Wang et al., 2020; Muschinski et al., 2016; Parr et al., 2016). It also has the advantages to maintain the infant and its dam in their familiar social environment while allowing us to access them frequently for short behavioral testing periods and avoid the use of highly constrained operant tasks lacking ecological validity, leading to overestimated gaze responses (Mosher et al., 2011; Høgh-Olsen, 2006) and not possible to administer in few week-old infants. In this study, we used the same eye-tracking data set but measured gaze following skills using procedures that have demonstrated gaze following abilities on adult rhesus monkeys using a similar spontaneous viewing task and eye-tracking measures (Leonard, Blumenthal, Gothard, & Hoffman, 2012; Mosher et al., 2011). Full details on the eye-tracking procedures are available in earlier reports (Muschinski et al., 2016; Parr et al., 2016) and are briefly summarized below.

Eye-tracking Data Acquisition

Mother–infant pairs were removed from their social groups and brought into our onsite testing facility where the mother was anesthetized (3- to 5-mg/kg-1 telazol, im) and placed comfortably in a reclining seat with her infant placed on her ventrum, facing a computer monitor onto which stimuli were presented (see Figure 2, left). An infrared eye-tracking camera (https://www.iscan.com/; 60 Hz) was mounted underneath the computer monitor on a motorized gimbal, controlled by an experimenter, to track the location of the infant’s eye. After successful calibration, testing sessions were restricted to 30 min to limit the mother’s time under anesthesia. After the mother had recovered from anesthesia, the mother–infant pair was returned to their social group.

Figure 2.

On the left, the testing chamber built for eye-tracking of rhesus macaque infants at the ENPRC, with adjustable reclining chair for the dam. The chamber was enclosed at the time of data collection to create a quiet, dark environment, and the infant was free to continue nursing while voluntarily viewing stimuli on the monitor. On the right is an example of two consecutive video frames from one of the stimuli. In all stimuli, a single close-up monkey was shown of various ages and of both sexes. Gaze following occurred when (A) the video monkey first made direct eye contact to the camera. A ROI encompassing the eyes was used to detect if the subject infant made a fixation establishing mutual eye contact. Following, (B) the direction of the video monkey’s attention was coded, and saccades were considered gaze following if the saccade path followed was between ±70° within the video monkey’s attention direction.

Stimuli

Stimuli were short 10-sec videos of unfamiliar rhesus monkeys from the breeding colony maintained by the Caribbean Primate Research Center in Cayo Santiago, Puerto Rico, played with accompanying sound. Videos showed close-up images of a single monkey, with approximately equal representation of female and male, and juvenile and adult monkeys (see Figure 2, right). Only stimuli with neutral emotional facial expressions were used to avoid emotional reactions from the infant viewer.

Thirty-three unique videos were created for use in the testing sessions. A specific pseudorandom combination of 12 videos was presented with an approximate 4:1 ratio of repeated to novel videos at each time point. Intertrial interval of 2 sec separated each video during which a centering stimulus (circular, chiming target on an otherwise blank screen) was presented to maintain infant attention. All videos were examined frame by frame to identify episodes during which the stimulus monkey looked directly into the camera to engage mutual eye contact with an observer then shifted its eye/head. Each video contained at least one to two potential episodes that could engage the viewer monkeys to make eye contact with the stimulus monkey (direct gaze) and then follow its gaze (averted gaze in the same direction as the on-screen monkey; see Figure 2).

Eye-tracking Data Processing

Fixation and eye movements data were analyzed and coded with custom-written software run in MATLAB (MathWorks; see Wang, 2019, for details). Briefly, data analysis consisted of (1) an automated identification of nonfixation data, comprising blinks, saccades, and fixations directed away from the presented screen; (2) saccades identification by eye velocity using threshold of 30°/sec; and finally (3) fixation locations within ROIs, specifically the eye region for the detection of mutual eye engagement, were identified and hand traced on each frame of the video and stored as binary bitmaps (through software written in MATLAB). A line was hand drawn during a period of averted eye direction, and then the direction of attention was calculated as radial angle of the line. Gaze following was then operationalized as comprising of a fixation to the eyes of the stimulus monkey—a moment of “mutual eye contact” when the onscreen monkey was looking directly at the viewer (Figure 2, right), followed by a gaze shift of the viewer in the direction in which the stimulus animal subsequently shifted its gaze (Figure 2, right). Mathematically, this was defined as saccades that fall within −10° to +70° of the line drawn representing the video monkey’s direction of attention.

Statistical Analyses

Statistical analyses were conducted with IBM SPSS Statistical package (IBM Corporation; Version 29.0.0.0).

Resting-state FC Data Analysis

First, FC measures were tested for normality using the Shapiro–Wilk test. When data were not normally distributed, measures were tested for extreme outliers using the third interquartile range and ran the analyses with and without those outliers to review the robustness of the findings. Because the statistical findings were not affected by the outliers, results are presented including outliers to illustrate individual variability. Then, developmental changes in ROI–ROI FC (Fisher Z-transformed data) along each pathway were statistically analyzed using repeated measures ANOVA with Age (2, 4, 8, 12, 16, 20, and 24 weeks) and Brain Hemisphere (left vs. right) as repeated factors. When sphericity was violated, Greenhouse–Geisser corrected results are provided. Whenever a main Age effect was significant (p < .05) for measures with > 2 timepoints, post hoc pairwise comparisons were conducted using Bonferroni correction to adjust for multiple pairwise comparisons.

Data are reported as mean Fisher Z-transformed FC data ± SEM, and effect sizes and significance level was set at p < .05.

Eye-tracking Data Analysis

Because of the high degree of variation in available data across testing sessions and between monkeys, mutual eye contact opportunities realized and gaze following initiated were analyzed and weighted proportionally to the amount of available data per monkey each week. Because only a few animals elicited mutual eye gaze and gaze following at weeks 2 and 17, these two weeks were not included in the analyses because of zero inflation. In addition, to address missing data at each time point, data obtained on two adjacent testing ages were averaged (e.g., weeks 3–4, 5–7, 9–11, 13–15, and 21–23). Repeated measures ANOVA was conducted using Age as the repeated factor, and significance level was set at p < .05.

Association Analyses

Fixed effects regression analysis was performed to examine whether changes in ROI–ROI FC were predicted as a function of specific eye-tracking data (i.e., mutual eye-contact and gaze following) over time, by way of the Least Squares Dummy Variable approach. Reversed models—assuming that brain FC predicts mutual eye contact or gaze following behavioral changes over time—were also tested, using the same approach. In each regression model, dummy variables (Model 1) and time-variant behavioral (eye tracking [ET]) or neural (FC) data (Model 2) were entered stepwise, to examine the between-subjects variability and the combined effect of the between-subjects predictor and the time-variant predictors (R² and R² change). Data from a subset of 12 subjects with both ET and FC data were analyzed at four age points: 4–8 – 12–24 weeks. Ages when less than 50% of the subjects showed mutual eye contact or gaze following behaviors (2, 3, 15, 17, 19, and 21 weeks) were excluded from the analysis. FC data were collected at 2 weeks, and every 4 weeks after 1 month of age, whereas eye tracking data were collected biweekly after 5 weeks of age, on alternating weeks. Thus, for the association analysis, we used ET data collected at an age closest to the scan age (e.g., FC data obtained at 8 weeks were compared with the averaged eye tracking data collected at 7 and 9 weeks, and FC data collected at 12 weeks were compared with the averaged eye tracking data collected at 11 and 13 weeks).

RESULTS

Note that for clarity purposes, only significant main effects or interactions will be reported below.

Neuroimaging Data: FC

Ventral Object Pathway (Figure 3)

ROI–ROI FC analysis along the ventral object pathway revealed significant FC changes between visual areas V1 and V3 during the first 6 months of life, Age effect: F(6, 96) = 5.395, p = 7.7 × 10⁻⁵, η² = .252). Post hoc Bonferroni test showed strong V1–V3 FC early after birth (at 2 weeks), which remained stable until week 8 and increased significantly afterward to reach a plateau at week 20. FC was higher in the left compared with the right hemisphere, Laterality effect: F(1, 16) = 24.299, p = 1.51 × 10⁻⁴, η² = .603, especially between 12 and 24 weeks of age, Age × Laterality interaction: F(6, 96) = 4.852, p = 2.29 × 10⁻⁴, η² = .233.

FC between visual areas V3 and TEO was weak at 2 weeks of age and remained unchanged throughout the first 6 months of life (no significant age or laterality effect, Age × Laterality interaction was observed).

Temporal areas TEO and TEp became uncoupled by 12 weeks of age, Age effect: F(6, 96) = 4.016, p = .006, η² = .201; Bonferroni post hoc test revealed significant FC decrease from 2 to 12 weeks, and from 4 to 12 weeks of age. FC was higher in the right hemisphere compared with the left throughout the first 6 months of life, F(1, 16) = 7.686, p = .014, η² = .324.

FC between area TEp and the AMY showed significant shifts across the 6 months, Age effect: F(6, 96) = 4.258, p = .007, η² = .210 (post hoc Bonferroni test revealed significant increases from 4 to 8 and from 4 to 12 weeks of age). Connectivity between area TEp and AMY was stronger in the left compared with the right hemisphere, Laterality effect: F(1, 16) = 5.532, p = .032, η² = .257.

FC between AMY and INS was also uncoupled at the earliest ages (2 and 4 weeks), but the FC slightly increased from 2 to 8 weeks of age and then plateaued, Age effect: F(6, 96) = 2.244, p = .045, η² = .123, post hoc Bonferroni test: p = .053.

Ventral Motion Pathway (Figure 4)

FC between visual areas V3 and MT increased from 2 weeks to 20 weeks, age effect: F(6, 96)=2.523, p=.026, η² = .136, although post hoc Bonferroni test revealed no significant age-specific changes. The increase in FC between V3 and MT was significantly higher in the left compared with the right hemisphere, Laterality effect: F(1, 16) = 43.499, p = 6 × 10⁻⁶, η² = .731.

Connectivity between cortical areas MT and AST slightly, but significantly, increased during the first 6 months of life, age effect: F(6, 96) = 2.354, p = .036, η² = .128, although post hoc Bonferroni test revealed no age-specific FC changes. FC was significantly higher in the left compared with the right hemisphere, laterality effect: F(1, 16) = 6.284, p = .023, η² = .282.

The rostral part of the ventral motion pathway, cortical area AST, was uncoupled with the AMY at 2 weeks after birth, and FC between these areas remained weak during the first 6 months of life (no significant age effect). FC was similar in the left compared with the right hemisphere (no laterality effect observed).

Dorsal Visual Attention Pathway (Figure 5)

Visual area V3 and parietal area LIP showed moderate FC during the first 8 weeks of life and became uncoupled by week 12, Age effect: F(6, 96) = 3.719, p = .002, η² = .189. Post hoc Bonferroni test revealed significant FC decrease specifically from 2 to 16 weeks of age. FC decrease was more robust in the left compared with the right hemisphere, Laterality × Age interaction: F(6, 96) = 2.409, p = .033, η² = .131.

FC between LIP and the FEF remained weak during the first 6 months of life (no significant Age effect), with no hemispheric differences observed (no Laterality effect).

Ventral Attention Pathway (Figure 6)

Visual area V3 and cortical area TPt showed moderate FC at 2 weeks of age, but became uncoupled later, Age Effect: F(6, 96)=3.120, p = .008, η² = .163, although post hoc Bonferroni test revealed no significant age-specific FC changes. FC was stronger in the right compared with the left hemisphere, Laterality effect: F(1, 16) = 9.407, p = .007, η² = .370. Areas V3 and TPt FC became negatively correlated in the left hemisphere after 12 weeks of age, Laterality × Age interaction: F(6, 96) = 2.755, p = .016, η² = .147. Cortical areas TPt and vlPFC (Area12) were weakly connected during the first 6 months of life; post hoc Bonferroni test revealed significant FC increase—switching from negative to positive correlation—from 4 to 8, and from 4 to 12 weeks of age, Age effect: F(6, 96) = 2.937, p = .011, η² = .155. No Laterality effects were observed (p = .056).

Eye-tracking data.

The data showed high individual variability in the number of mutual eye gaze and gaze following initiated by each monkey at each age. As shown in Table 1, the number of animals initiating successful mutual eye gaze varied from 2 to 12, with more than half of the animals initiating this behavior between 4 and 13 weeks and then between 21 and 23 weeks. Following mutual eye gaze, the number of animals initiating successful gaze follows varied from 0 to 9, with more than half of the animals initiating this behavior between 5 to 15 weeks and then between 21 and 23 weeks.

Table 1.

Summary of Eye-tracking Data

Age (Weeks)	Animals with Successful Mutual Gaze	Animals with Successful Gaze Following	Mean % Weighted Mutual Gaze	Mean % Weighted Gaze Following
2	3	3	20.00 ± 8.16	83.33 ± 8.31
3	5	3	18.18 ± 7.04	52.00 ± 22.45
4	8	4	48.00 ± 12.72	36.67 ± 12.65
5	9	6	38.63 ± 8.87	47.53 ± 10.28
7	6	5	28.57 ± 7.14	65.28 ± 15.28
9	12	9	50.00 ± 10.84	59.99 ± 8.75
11	7	7	27.02 ± 5.40	50.17 ± 5.99
13	7	7	35.16 ± 9.60	67.44 ± 10.17
15	5	5	51.11 ± 11.50	58.33 ± 16.67
17	2	0	13.33 ± 13.33	0.00 ± 0.00
19	3	2	27.45 ± 4.49	45.55 ± 27.24
21	6	6	57.05 ± 10.46	60.61 ± 12.66
23	9	8	52.92 ± 11.42	42.10 ± 12.11

Number of animals at each age with successful calibration and mutual gaze initiated, and gaze following initiated. The ages between 4 and 13 weeks (1–3 months) and 23 weeks (~5–6 months) were those with the higher number of animals initiating mutual gaze and gaze follow (gray highlights). Percent weighted scores (mean ± SEM) are provided for mutual gaze and gaze following at each age. Shaded areas indicate ages at which over 50% of mutual gaze and gaze follow were initiated.

Mutual eye contact engagement was calculated as the percentage of times mutual eye contact (fixations to the eyes) was established between the onscreen monkey and subject monkey compared with the total number of events in which mutual eye contact was possible for each video clip (i.e., the number of times the onscreen monkey stared directly at the camera). As shown in Table 1 and Figure 7, the percent of successful mutual eye gazes was low and fluctuated during the first 6 months of age with peak values at weeks 4, 9, and 15 (48.0%. 50%, and 51.1%, respectively) and then at weeks 21 and 23 (58.5% and 52.9%, respectively; see Table 1). The Age effect was not significant, F(4, 44) = 0.408, p > .05.

Figure 7.

(A) The percentage of mutual gaze cues during which infants engaged in mutual eye contact (mean ± SEM). (B) From the instances in which mutual eye contact is established between the movie monkey and infant monkey, the percentage of gaze following events during which the infant would shift attention in the same direction as the movie monkey (mean ± SEM). Each monkey’s data are weighted proportional to how much total data are available at each week.

Gaze following was then defined as a saccade within the coded direction of attention of the video monkey following successful mutual eye contact event. Gaze following engagement was calculated as the percentage of time the subject monkey followed the gaze of the onscreen monkey after mutual eye contact was established (weighted percent; see Table 1). The high percentage of gaze follows was really seen only in week 2 (83.3%) and could have been inflated because of the very few monkeys (only 3 out of the 12 infants) and the low percentage of mutual gaze these infants initiated, that is, 20% of the total possible eye gazes contained in all social videos played at that age. After a drop to 36.7% in week 4, percent gaze following increased progressively until weeks 7 (65.28%) and plateaued thereafter. The repeated measures ANOVA showed a significant effect of Age, F(4, 44) = 3.001, p < .028. Post hoc comparisons revealed significant decrease between weeks 3–4 (42.5%) and 5–7 (62.5%), between weeks 5–7 (62.5%) and 19–23 (49.7%), and between weeks 9–11 (55.0%) and 19–23 (49.7%; all Bonferroni corrected p < .05).

Association between FC and eye-tracking data.

Stepwise regression models were used to explore whether between-subjects variability and/or age-related changes in specific ET behaviors (mutual gaze and gaze following) were associated with age-related changes in ROI–ROI FC values over time (see Table 2).

Table 2.

Regression Analysis Summary

Brain ROI–ROI FC	Model	R Square	F Value	p Value
V1–V3 right	1	0.408	1.318	.282
	2	0.591	2.109	.068
	Δ	0.182	4.23	.03
TEO–TEp left	1	0.558	2.409	.04
	2	0.641	2.609	.028
	Δ	0.083	2.197	.139
AMY–INS right	1	0.584	2.685	.025
	2	0.64	2.6	.029
	Δ	0.056	2.6	.255
V3-LIP right	1	0.157	0.355	.96
	2	0.418	1.05	.45
	Δ	0.261	4.269	.029

Significant stepwise linear regression models. Bold font indicates significant predictive models of brain FC. Model 1 = predictive effect of the within-subject variability, measured by the way of dummy variables; Model 2 = predictive effect of within-subject variability and the specific ET behaviors (percent mutual gaze and gaze following) together; Δ = predictive effect of the specific ET behaviors (percent mutual gaze and gaze following) alone.

Regression models showed no significant association with FC between visual areas V1–V3 in the right hemisphere, Model 1: R² = .408, F(11, 21) = 1.318, p = .282; Model 2: R² = .591, F(13, 19) = 2.109, p = .068. However, specific ET behaviors alone predicted FC changes between visual areas V1–V3 in the right hemisphere, R² change = .182, F(2, 19) = 4.230, p = .030. Specifically, gaze following behavior showed significant negative slope of regression, which suggests that lower gaze following percentage was associated with higher FC between V1–V3 in the right hemisphere (unstandardized B = −0.002, p = .013).

Between-subjects variability was a significant predictor of TEO-TEp FC in the left hemisphere, Model 1: R² = .558, F(11, 21) = 2.409, p = .040, and adding the specific ET behaviors to the model increased the strength of the association, Model 2: R² = .641, F(13, 19) = 2.609, p = .028. Adding the specific ET behaviors to the model increased the effect size (R² change=.083); however, specific ET behaviors alone did not predict FC between the areas mentioned above, F(2, 19) = 2.197, p = .139.

FC between AMY-INS in the right hemisphere was predicted by the between-subjects variability, Model 1: R² = .584, F(11, 21) = 2.685, p = .025, and adding the specific ET behaviors to the model did not increase the strength of the association, Model 2: R² = .640, F(13, 19)=2.600, p=.029. Adding the specific ET behaviors to the model slightly increased the effect size (R² change = .056); specific ET behaviors alone did not predict FC between the areas mentioned above, F(2, 19) = 1.470, p = .255.

Between-subjects variability did not predict FC between cortical areas V3-LIP in the right hemisphere, Model 1: R² = .157, F(11, 21) = 0.355, p = . 960, and although adding the specific ET behaviors to the model increased the strength of the association, the model was still not significant, Model 2: R² = .418, F(13, 19) = 1.050, p = .450. However, specific ET behaviors alone predicted FC between V3 and LIP in the right hemisphere, R² change = .261, (2, 19) = 4.269, p = .029. Significant positive slope of regression observed with gaze following suggests that higher gaze following percentage was associated with higher FC between V3 and LIP in the right hemisphere (unstandardized B = 0.002, p = .021).

When using mutual gaze and gaze following as dependent variables (separately), and ROI–ROI FCs as time-variant predictors, no significant associations were observed.

DISCUSSION

Our findings indicate that during the first 6 months of life, FC increased significantly in primary visual areas (V1–V3 and V3-MT), ventral motion areas MT–AST, as well as between area TE and the AMY. Significant FC decreases were found, instead, between areas TEO-TE in the ventral object pathway, areas V3-LIP of the dorsal visual-attention pathway, and areas V3-TPt of the ventral attention pathway. Weak FC was detected between the more rostral cortical areas (between LIP-FEF and TPt-vlPFC Area 12) of the dorsal and ventral attention pathways as well as between AST-AMY and AMY-INS. Developmental trajectory of gaze following skills revealed a period of slow increases from 1 to 2 months that plateaued until 6 months. Regression analyses between FC and eye-tracking behavioral measures indicated that infants with higher gaze following percentage had lower FC between primary visual areas V1–V3 of the ventral visual pathway. By contrast, infants with higher gaze following percentage had higher FC between areas V3-LIP of the dorsal visual-attention pathway. Gaze following percentage was not associated with FC between any other regions along the visual and attention pathways. Thus, the gaze following frequency during this 6-month period was driving the changes in FC of the primary visual areas V1–V3 and of the parietal attention areas V3-LIP but in an opposite direction. Together, the first 6 months of life in rhesus macaques represent a critical period for the emergence of gaze following skills that seem to drive the development of socially guided visual and attention pathways.

FC along the Visual and Attention Pathways

From birth to 6 months, FC along the visual and attentional pathways changes rapidly and with region-to-region specific developmental trajectories.

The trajectories of FC between cortical areas in the ventral object pathway replicate the changes reported earlier by Kovacs-Balint and colleagues (2019) from 1 to 3 months and Ford and colleagues (2023) with a small subset of the infants used in the present study. Thus, from 2 weeks to 2 months of age FC between V1 and V3 was strong, sharply increased to peak at 4 months, and then plateaued until 6 months. Furthermore, the FC between V3 and TEO was strong and remained stable from 2 weeks to 6 months, whereas FC between TEO and TE was weak from 2 weeks to 2 months and became uncoupled thereafter. After a robust increase in FC between TEp and AMY from 2 weeks to 3 months, FC slightly decreased from 3 to 6 months. Finally, the FC between AMY and INS, although uncoupled in the first month, significantly increased to reach a peak at 2 months and then plateaued until 6 months. Our data indicate that local feature detection mediated by the primary visual areas (V1–V3) of the ventral object pathway develops early. By contrast, the rostral visual temporal areas (V3-TEO and TEO-TE) containing neurons specifically devoted to perception of faces (face patches) do not fully specialize for face recognition until 5–6 months of age (Livingstone et al., 2017) and their maturation is experience-dependent (Arcaro et al., 2017). Thus, socio-visual experience that capitalizes on the architecture and physiology of the face patch system may be required to promote processing of face information and social visual cues after 6 months of age. The delayed maturation of the patch may also be why we did not observe associations between gaze following and FC in the more anterior regions of the visual object pathway before 6 months. The significant FC changes between the most rostral region of the inferior temporal cortex (TE) and the AMY between 2 and 4 months could support the increase in gaze following abilities during this period. TE–AMY interactions have been shown to be critical for the detection of face identity and facial expressions (Schwiedrzik, Zarco, Everling, & Freiwald, 2015; Mosher et al., 2010; Gothard et al., 2007; Hoffman et al., 2007), and their strong coupling from 2 to 5 months could provide a subcortical route to enable greater precision in the assessment of faces and the emergence of mutual eye gaze and gaze following skills in the first months after birth in monkeys (Wang, 2019; Muschinski et al., 2016; Parr et al., 2016; Sugita, 2008; Kuwahata et al., 2004; Lutz, Lockard, Gunderson, & Grant, 1998).

FC between cortical areas V3–MT and MT–AST along the ventral motion pathway, known to be critical for facial and eye motion from expressions as well as gaze monitoring and gaze following, was moderate but stable during the first 3 months. FC between these areas then slightly increased thereafter, suggesting further maturation after 3 months. This developmental pattern is in line with the emergence of detection of motion direction around 2 weeks (Kiorpes, Price, Hall-Haro, & Movshon, 2012; Kiorpes & Movshon, 2004), which improves during the first 3 months, although not reaching adult-like levels until 3 years (Mikami & Fujita, 1992). Furthermore, the ability to rapidly evaluate direct gaze faces and gaze following emerges by 3 months (Wang, 2019; Muschinski et al., 2016; Mendelson, Haith, & Goldman-Rakic, 1982) but drastically increases after 5–6 months to reach a peak in the juvenile period (Rosati et al., 2016; Teufel et al., 2010; Tomasello et al., 2001). Thus, the FC between visual areas along the superior temporal sulcus may begin to support the early reflex-like gaze following skills emerging around 3 months of age.

Areas within the dorsal visual attention pathway, which are intimately involved in the control of oculomotor and spatially directed attention (Milner & Goodale, 1995), showed a decrease in FC between V3 and LIP during the first 3 months of age and then plateaued until 6 months. By contrast, FC between LIP and FEF remained weak during the first 6 months of age and may not strengthen until older ages because spatially directed attention emerges only in the juvenile and adolescent periods (Condé, Lund, & Lewis, 1996). In addition, the late maturation of FC in more rostral prefrontal areas within the dorsal visual attention pathway are consistent with the progressive development of gaze following that begins in infancy (current data; Wang, 2019) but becomes stronger after 6 months of age and peaks in the juvenile period (Rosati et al., 2016; Teufel et al., 2010).

Cortical areas within the ventral attention pathway (V3-TPt and TPt-pFC Area 12) support face-emotion processing, attention, and theory of mind operations in humans (Gazzaniga, Ivry, & Mangun, 2018; Krall et al., 2015; Corbetta et al., 2008). Recent studies have shown some anatomical and functional homologies of the middle part of the macaque superior temporal sulcus (mid-superior temporal sulcal) with human TPJ—a crucial node in social cognitive processes associated with mentalizing (Ninomiya, Noritake, & Isoda, 2021; Roumazeilles et al., 2021; Mars et al., 2013). Here, we showed that FC between V3 and TPt was weak and decreased from 2 weeks to 3 months after which it switched to become negative from 3 to 6 months. In addition, FC between TPt and Prefrontal Area 12 was uncoupled during the 6-month period, indicating that these areas were not fully functionally coupled at these early ages. These weak FC patterns during the first 6 months of life are also in line with the late development of social cognitive processes, such as theory of mind in monkeys (Arre, Clark, & Santos, 2020).

Mutual Eye Gaze and Gaze Following

The ability to derive from another individual’s eye gaze the target of their interest and to follow their gaze is hypothesized to be a basic building block for the development of “theory of mind,” the ability to infer the mental state of another individual (Emery, 2000; Langton, 2000; Langton, Watt, & Bruce, 2000). Our results indicate that mutual eye gaze and gaze following saccades were present, but relatively infrequent, during the first 6 months of life and parallel the earlier findings using a larger sample of infant monkeys (Wang, 2019).

The relative infrequent and unstable percentage of mutual gazes as measured by fixations to the eye ROI may be related to the stringent procedures we used to assess gaze following skills; that is, we required the infant monkeys to first fixate at the eyes of the on-screen monkey when looking directly at the camera (and thus at the infant). Yet, eye contact between monkeys can signal affiliative but, in many instances, agonistic intentions from a conspecific (Maestripieri & Wallen, 1997; Redican, 1975). Therefore, gaze aversion by the viewer is the most common response to threatening or dominant individuals, especially for infants when faced with adult unfamiliar individuals (Masataka & Fujii, 1980; Redican, 1975). In addition, Muschinski and colleagues (2016) demonstrated using the same spontaneous task that infant monkeys made shorter fixations when looking at the direct gaze compared with averted gaze faces. Thus, the overall low levels of mutual gaze reported could have resulted from many instances when the infant viewer was not directly looking at the monkey’s eye on the screen but would occasionally make a saccade and/or a single fixation to the face of the on-screen monkey to check–look on what was happening. In this instance, the score for the infant monkeys would have been “zero” mutual gaze. In addition, these instances could have occurred more frequently on certain testing sessions than others depending on the monkey stimuli presented in the movie clips (juveniles vs. adults; males vs. females) resulting in the irregular pattern of percent mutual gaze across ages. These low levels of percent mutual gaze would have in turn affected the percentage of gaze follows the infant will initiate. Despite this shortcoming, the present data showed clear evidence of gaze follows, especially between 2–4 months and 5–6 months, which extends earlier reports indicating that gaze following skills become stronger after 5–6 months of age, reaching adult level at 1 year of age (Rosati et al., 2016; Teufel et al., 2010; Tomasello et al., 2001). Teufel and colleagues (2010), combining longitudinal and cross-sectional observational field data on Barbary macaques with live conspecifics emitting directional cues, showed relatively few gaze follows before 5 months with a rapid increase between 5 and 6 months to reach adult-like levels by 1 year of age. Rosati and colleagues (2016) found that human experimenter-initiated gaze following in infant monkeys living in semi-free-ranging social groups does not emerge until 6 months. Using densely sample data, controlled laboratory eye-tracking procedures that provide precise automated identification of gaze follow behaviors while maintaining species-relevant dynamic social stimuli (Mosher et al., 2011), the current study show the presence of immature gaze following skills earlier than 5–6 months. Interestingly, the emergence of these skills in monkeys around 7 weeks (~7–8 months in humans) is more protracted than in human infants (3 months; Gredebäck et al., 2010; D’Entremont, 2000; D’Entremont, Hains, & Muir, 1997), suggesting that nonhuman primates may require more experience with relevant social interactions during infancy (Teufel et al., 2010; Ferrari, Kohler, Fogassi, & Gallese, 2000). Like human infants (Gredebäck et al., 2010), significant individual differences in gaze following skills were noted in the infant monkeys as well.

Associations between Gaze Following and FC within Visual and Attention Pathways

The results demonstrated association between early developing gaze following skills and functional brain development in infant rhesus macaques. Across the 6-month period, we showed that higher frequency of gaze following predicted lower FC between V1 and V3 but higher FC between V3 and LIP, both in the right hemisphere. The direction of the association between gaze following and FC V1–V3 was unexpected as our earlier findings showed that greater increases in early FC between left V1 and V3 predicted earlier increases in visual attention to the eyes of conspecifics in the first 2 months of life (Ford et al., 2023). Although differences between the two sets of data could be because of different statistical regression models employed, they could also indicate that V1 to V3 FC could be suppressed during gaze following by the increased FC between area TE and the AMY. The AMY is known to project back to all visual areas, including V1 and could modulate the early visual perceptual processes (Amaral, Price, Pitkanen, & Carmichael, 1992). These later developing social skills may require additional socio-visual experience that capitalizes on the architecture and physiology of the face patch system to promote processing of face information and social visual cues. The delayed maturation of the patch may also be why we did not observe associations between gaze following and FC in the more anterior regions of the visual object pathway, including TEp to AMY, before 6 months.

No significant associations were noted between the behavioral data and FC in cortical areas within the visual motion pathway, although metabolic activity within these areas reaches adult-like levels between 2 and 3 months of age (Distler, Bachevalier, Kennedy, Mishkin, & Ungerleider, 1996). The results were surprising given that these cortical areas are involved in the perception of motion, including facial movements during production of facial expressions and the perception of eye gaze shifts to initiate gaze follows. One possibility is that the motion pathway may be required to support social stimuli related to eye gaze only later after 6 months of age.

Gaze following requires not only the perception of the face and a preference looking to the eyes but also initiation of mutual eye gaze followed by attention toward another individual shift in looking direction. The positive association we found between gaze following and FC between V3 and LIP is consistent with the role of the parietal areas in orienting the subjects’ attention to salient social visual stimuli (Bogadhi et al., 2018; Sapountzis et al., 2018; Bisley et al., 2011). Specifically, the higher frequency of gaze following during this 6-month period drove the strengthening of FC in the dorsal visual attention networks, which could then set up stronger attention to social visual face cues later in life. It also could serve as a reflex-like process emerging around 3 months that will be later modulated by volitional attentional mechanisms supported by connectivity between frontoparietal areas (LIP-FEF). The delayed maturation of the prefrontal areas may be a reason why we did not observe associations between gaze following and FC connectivity in the more anterior regions of the dorsal visual-attention pathway (LIP-FEF).

Finally, the lack of associations between gaze following and FC between areas of the ventral attention pathway may have resulted from the uncoupling of FC between areas V3-TPt and TPt – Area 12 during the first 6 months of age because of the immaturity of these cortical areas at these early ages. The weak FC patterns during the first 6 months of life may indicate that these cortical areas developed later to support the protracted development of more complex visual social cognitive processes that involve less reflexive but more conscious/deliberate processes, such as theory of mind in humans (Luo & Johnson, 2009; Onishi & Baillargeon, 2005; Wellman, Cross, & Watson, 2001; Wimmer & Perner, 1983) and possibly more social experience in monkeys (Arre et al., 2020).

Because of the exploratory nature of the association analyses, our sample size was too small to detect statistically significant brain–behavior associations after correction for multiple comparisons. Thus, future studies are necessary to replicate those results and to map trajectories of social neurobehavioral development in a larger sample out to 9 or 12 months.

Neural Mechanisms Underlying FC Changes

The specific mechanisms that underlie FC changes in visual and attention cortical areas in early infancy are still poorly understood and remain speculative. Nevertheless, as already discussed in our earlier report (Ford et al., 2023; Kovacs-Balint et al., 2019), the progressive coupling in activity patterns between the most caudal cortical areas V1–V3 between 8 and 12 weeks is consistent with the synaptic proliferation and rapid dendritic spine increase (Rakic, Bourgeois, Eckenhoff, Zecevic, & Goldman-Rakic, 1986), the rise of metabolic activity (Distler et al., 1996; Bachevalier, Hagger, & Mishkin, 1991), and the presence of adult-like selective responsivity of neurons activity (Rodman, Scalaidhe, & Gross, 1993) reported in many cortical areas during the first 3 months of life, including the most caudal visual areas. Concurrent with these remodeling of gray matter, white matter microstructure matures in parallel until about 9 months of age (Kim et al., 2020). By contrast, FC between higher order association areas of all visual pathways (TEO–TE, MT–AST, LIP–FEF, TPt – Area 12) were weaker even at 3 months of age. Again, these findings are in line with those of previous studies showing a protracted development of metabolic and neuronal activity of these higher-order cortical areas until 3 months of age (Kiorpes, Hawken, & Movshon, 2007; Movshon, Rust, Kohn, Kiorpes, & Hawken, 2004; Distler et al., 1996; Rodman et al., 1993). These anatomical and physiological changes may contribute to the less pronounced change measured in FC between V3-TEO and TEO-TEp, relative to V1–V3, and ultimately, the specific trajectories of FC changes along the regions of inferior temporal cortex—that is, TEO to TEp. Furthermore, the changes in TE–AMY FC parallel the progressive retraction of exuberant projections from TE to AMY (Webster, Ungerleider, & Bachevalier, 1991; Webster, Bachevalier, & Ungerleider, personal communication). This anatomical refinement of connections, presumably resulting from additional experience interacting with other conspecifics, could increase the functional strength of remaining TE inputs to AMY neurons. Together with significant volumetric increase and morphological changes—including myelination—within the AMY during the first 3 months postpartum in monkeys (Chareyron, Lavenex, Amaral, & Lavenex, 2012; Payne, Machado, Bliwise, & Bachevalier, 2010) and the fine-tuning of TE–AMY connections may provide a subcortical route to support reflexive perception of social stimuli at these early ages. Lastly, overproduction of synaptic contacts and wiring, and synaptic density peak relatively late (1–2 years in monkeys) for the prefrontal cortical areas of the ventral and dorsal visual attention pathways (Goldman-Rakic, 1982, 1987; Goldman-Rakic, Isseroff, Schwartz, & Bugbee, 1983; Goldman & Alexander, 1977). This protracted refinement of morphological changes in the prefrontal cortical areas may parallel the weak FC between the most rostral areas of the dorsal and ventral attention pathway (LIP-FEF and TPt – Area 12, respectively). This later development of prefrontal areas could provide a more volitional control of reflex-like behavior given the social context (Breu et al., 2023).

Relevance to Human Gaze Following Neural Development

Abundant knowledge has been gathered on the development of human visual and socially guided attention function (e.g., Braddick & Atkinson, 2011; Klaver et al., 2008), yet little is known on the brain changes that underlie such maturational processes especially in the first months of life. Developmental neuroimaging studies in humans suggest that visual and attentional neural networks are just beginning to develop during infancy as they do in monkeys. For example, both PET and fMRI studies showed (1) maturation of primary visual areas (V1/V2/V3) early in life (Gao et al., 2015; Chugani, Phelps, & Mazziotta, 1987; Chugani & Phelps, 1986); (2) presence of unique resting state networks, including primary visual cortex, sensorimotor areas, and lateral parietal cortex in infant brain (Fransson et al., 2007); and (3) emergence of FC between primary visual cortex and higher-order visual areas during infancy—neonates to 2 years (Lin et al., 2008; Deen et al., 2017). Given the difficulty in scanning typically developing human infants, our findings may shed light on the normative development of social visual networks in humans and their derailment in those at risk for developing autism spectrum disorder and Williams syndrome.

Limitations

MRI scans were collected under low levels of anesthesia to minimize the potential effects of isoflurane and telazol on functional activity and subsequent BOLD signal (Hutchison et al., 2013; Li et al., 2013; Vincent et al., 2007). Although previous studies reported similar changes in BOLD signal between awake versus anesthetized macaques, particularly in signal from the visual system (Hutchison et al., 2013; Li et al., 2013; Vincent et al., 2007), FC differences with wakeful states may still exist (Jovellar & Doudet, 2019). However, in our previous study (Kovacs-Balint et al., 2019), we reported that controlling for levels of telazol and isoflurane in the same protocol and partial sample used did not change the FC results in the visual object pathway.

As noted above, the low levels of percent mutual gazes and gaze follows during the 6-month period could have resulted from our stringent analyses of the infants’ looking patterns, requiring them to fixate at the direct eye gaze of the on-screen monkeys, known to be a threatening gesture for rhesus monkeys. Infant monkeys may have thus averted their gaze and/or occasionally made a saccade and/or a single fixation to the face of the on-screen monkey to check–look on what was happening but reducing the percentage of direct mutual gazes. Future studies could ease the procedures by counting not only the fixations to the eyes but also to the face of the on-screen monkeys, especially given that in almost all videoclips shown, the on-screen monkeys shifted not only its eyes (averted eyes) but also its face (averted face) when switching gaze direction.

Finally, only male subjects were used in this study to avoid potential sex differences in brain development and increase overall statistical power. Future studies using a larger sample size, including female subjects, and extending our assessment of neurodevelopmental changes of social visual networks further into the juvenile and adolescent period are clearly warranted.

Diversity in Citation Practices

Retrospective analysis of the citations in every article published in this journal from 2010 to 2021 reveals a persistent pattern of gender imbalance: Although the proportions of authorship teams (categorized by estimated gender identification of first author/last author) publishing in the Journal of Cognitive Neuroscience (JoCN) during this period were M(an)/M = .407, W(oman)/M = .32, M/W = .115, and W/W = .159, the comparable proportions for the articles that these authorship teams cited were M/M = .549, W/M = .257, M/W = .109, and W/W = .085 (Postle and Fulvio, JoCN, 34:1, pp. 1–3). Consequently, JoCN encourages all authors to consider gender balance explicitly when selecting which articles to cite and gives them the opportunity to report their article’s gender citation balance.

Data Availability Statement

Data will be made available on request.