Dissociable developmental trajectories of Orbitofrontal subregion grey matter volume

Adolescence is a period of development which is characterised by distinct differences in decision-making strategies relative to adults. While it is broadly established that there are relative differences in the structural maturation of the prefrontal cortex (PFC) and subcortical reward nuclei, such as the amygdala and ventral striatum, heterogeneity within the PFC is often neglected. In particular very little is known about the fine-scale gray matter (GM) development of the Orbitofrontal Cortex (OFC), itself critical to a number of learning and decision-making mechanisms which show delayed development trajectories. Here we applied voxel-based morphometry to examine subregional differences in OFC grey matter in high-quality structural MRI scans of 125 subjects aged 11-35yrs from the Human Connectome Project. First, we examined fine-scale GM maturation in 5 anatomically dissociable OFC subregions and identified the best-fitting polynomial model. Next, we directly compared developmental trajectories across 3 functionally dissociable subregions, revealing a complex topological developmental profile from medial to lateral subregions. Collectively, the two complementary analyses suggest that while unequivocally the phylogenetically younger lateral OFCs showed the greatest shift in GM volume across adolescence, with maturation continuing well into young adulthood, the differences between the medial and central OFC subregions suggested a more complex pattern of maturation than a simple graded medial to lateral topological development. We argue that knowledge of these fine-scale anatomical differences in maturation could explain precise mechanistic differences in goal-directed behaviours.


Introduction
Adolescence is a late, sensitive period of development, during which young people gain independence and develop mature social goals (Fuhrmann et al., 2015;van Duijvenvoorde et al., 2016). Yet it is also a dangerous time of life, notoriously characterised by irrational decisions, that can often lead to accidental death (Viner et al., 2012;Wolfe et al., 2014), and coincides with the onset of many serious mental health illnesses (Paus et al., 2008). Neuroanatomically, the teenage brain is also undergoing significant structural reorganisation (Paus et al., 1999;Blakemore & Choudhury, 2006;Colby et al., 2011;Whitaker et al., 2016). The relative maturity of subcortical brain regions appears to overshadow the underdeveloped and poorly connected prefrontal cortex (PFC) (Mills et al., 2014;Casey, 2015), potentially resulting in teenagers' risky reward-seeking behaviour (Burnett et al., 2010;Blakemore & Robbins, 2012;Galvan & McGlennen, 2013). While the importance of the well-established finding of the late maturation of the PFC cannot be understated, the PFC is composed of a number of fundamentally distinct sub-regions (Passingham & Wise, 2012;Sallet et al., 2013;Neubert et al., 2014;Neubert et al., 2015b) and this heterogeneity is neglected in current whole-brain developmental maps of cortical maturation (Sowell et al., 2003;Gogtay et al., 2004). As we move to a circuit-based understanding of learning and decision-making (Hunt & Hayden, 2017) and the altered network states during adolescence (Casey, 2015), a critical fundamental advancement then becomes the fine-scale quantification of the developmental of anatomically dissociable PFC subregions.
One particularly developmentally delayed subregion of the PFC is the orbitofrontal cortex (OFC), a region critically involved in learning and decision-making, which broadly does not reach gross maturity until the early 20s (Gogtay et al., 2004). The OFC is itself comprised of numerous cytoarchitectonically and connectionally distinct subregions (Carmichael & Price, 1996;Neubert et al., 2015b). Using diffusion weighted imaging, Neubert and colleagues created detailed parcellation maps based on the unique connectivity profiles of each subregion and delineated 21 key human orbital and medial frontal brain regions, of which five anatomically dissociable regions are consistently considered part of the orbitofrontal cortex itself. The first corresponds to medial area 14 which extends between the medial orbitofrontal sulcus and the rostral sulcus in ventromedial prefrontal cortex, as such it is often referred to as mOFC/vmPFC in humans (Ongur & Price, 2000). Laying ventral to area 14, is area 11m. Area 11m encompasses the gyrus rectus and cortex beneath, and medial to the olfactory tract.
Despite area 11m laying medial to area 13 and extending rostrally onto the medial surface, its projections are more interlinked with the more central and lateral orbital networks (Carmichael & Price, 1996) and as such it is not considered part of mOFC/vmPFC. Area 11 (excluding area 11m), which lies immediately posterior to the frontopolar region of area 10, and the more posterior area 13 are located laterally from the medial orbitofrontal sulcus and used to be referred to as the lateral orbitofrontal sulcus are now referred to as central OFC (Rudebeck et al., 2017;Sallet et al., 2020).
Central OFC is now also distinguished from cortex beyond the lateral orbitofrontal sulcus which transitions into ventrolateral prefrontal cortex and aligns mostly with the gyral region of the orbital part of inferior frontal gyrus. This corresponds to the orbital part of area 12 (12o) in macaques and Brodmann's area 47o in humans (referred to from here as area 47/12o).
The anatomically-defined dissociations between OFC subregions are complemented by evidence of fine-scale functional dissociations. For example, lesion studies support in adult humans and macaques show distinct functional contributions of medial area 14, central OFC (areas 11 and 13, typically lesioned together) and lateral OFC area 47/12o in learning and decision-making tasks. These studies demonstrate that mOFC/vmPFC in both species is necessary for decisions that involve multiple valuable options to choose between and for focusing a decision on the most relevant information (Noonan et al., 2010;Vaidya & Fellows, 2016;Noonan et al., 2017). Central OFC regions are involved in the updating of object valuations according to current motivational states (Rudebeck et al., 2013b).
Finally, the lateral regions of the OFC are critical for behavioural flexibility and learning the consequences of choices (Noonan et al., 2010;Noonan et al., 2011;Rudebeck et al., 2013b;Noonan et al., 2017;Sallet et al., 2020). To date, fine-scale developmental differences in subregion maturation have not been investigated but quantifying these differences is likely to be key in interpreting the behavioural developments in contingent learning (Javadi et al., 2014;Hauser et al., 2015) and valueguided decision-making (Kwak et al., 2015), as well as more broadly advancing our understanding of the maturation of reward circuits in adolescents (Galvan et al., 2005). Therefore the current study combined recently released high-quality HCP developmental data (Somerville et al., 2018) with the HCP young adult dataset , and used voxel based morphometry methods (Good et al., 2001) to identify and quantify relative grey matter (GM) differences in fine-scale OFC subregions across an age range of 11-35yrs.

Methods
Participants 125 individuals were sampled from the HCPYA (n = 56) and HCPD (n = 69) databases. Participants were aged 11-35 years and categorised into three age groups, 11-17, 18-23 and 24-35 years old. These categories were informed by societal and neuroanatomical developmental changes (Viner et al., 2012;van Duijvenvoorde et al., 2016) with particular acknowledgement of the late maturation of the OFC (Gogtay et al., 2004). Approximately equally-sized populations were sampled within these age ranges in order to create a non-biased study specific GM template. 43 subjects, aged 11-17 years from HCPD had pre-processed T1w images available at the time of analysis. We matched this sample size for the 24-35 age group, randomly selecting 43 subjects with T1w images and necessary demographic data from HCPYA. Finally, bridging upper and lower ranges of HCPD and HCPYA, we populated the 18-23 age group by randomly sampling 26 HCPD subjects between 18-21 years and 13 HCPYA subjects between 22-23 years. The non-biased study-specific template was generated by randomly selecting the N of the smallest group (n = 39) from each of the larger groups (see Figure 1 for a histogram of participant ages). Demographic data, sex and handedness (overall score of number of responses of right hand minus those of left hand, divided by right plus left, using items of the Edinburgh handedness inventory (Oldfield, 1971)) for all participants can be found in Table 1. For the purposes of template creation, participants were categorised a priori into three age groups: 11-17 (light grey); 18-23 (mid grey); 24-35 (dark grey).

Image Acquisition and Minimal Pre-Processing
T1w structural images were acquired with HCP acquisition parameters (Glasser et al., 2013): 256 sagittal slices were acquired with TR = 2400ms, TE = 2.14ms, flip angle = 8°, FOV = 224mm, matrix = 320. The primary aim of these selected parameters was to achieve high spatial resolution: T1w images had a voxel size of 0.7mm isotropic. All structural images underwent minimal pre-processing via the three HCP structural pipelines PreFreeSurfer, FreeSurfer, and PostFreeSurfer. The PreFreeSurfer pipeline generates a Native Volume Space brain-extracted, undistorted T1w image which is linearly aligned to the axes of MNI space. Full details of structural image acquisition and minimal preprocessing pipelines are reported by Glasser et al (2013).

Voxel Based Morphometry
Based on the voxel-based morphometry (VBM) (Douaud et al., 2007) analysis pipeline, we conducted analysis to identify fine-scale differences in GM between OFC subregions across adolescence and early adulthood. The HCP-provided skull stripped T1w images were individually segmented into GM, white matter (WM) and cerebral spinal fluid (CSF) before being affine-registered to the GM ICBM-152 template using FLIRT (Jenkinson & Smith, 2001) followed by nonlinear registration using FMRIB's Nonlinear Image Registration Tool (FNIRT) (Andersson et al., 2008). Of the resulting images, those randomly selected to compose the non-bias study-specific template (equal number of subjects from each of the age ranges 11-17, 18-23 and 24-35 years) were averaged, the native grey matter images of all subjects were then non-linearly re-registered to this template, and concatenated into a 4D image. The resulting unsmoothed 4D image is modulated for the contraction or enlargement of each subject's image to the template with each voxel of each image being multiplied by the Jacobian of the warp field (Good et al., 2001). The final step of our VBM analysis was to extract all subject's values from the 4D GM modulation image -from here on these values will be referred to as "GMm" -for each of our ROIs, providing a between-subjects comparison of volume which could be regressed against age to map developmental trajectory. ROI average Jacobian values were extracted using fslmeants in combination with the corresponding mask. Higher GMm values correspond to more GM, reflecting regional expansion of the native GM images relative to the template, whereas lower values correspond to less GM and reflect regional GM contraction of native images.

Statistical Analysis
Partialing out confounding covariates To control for the effects of subjects' sex, handedness and total brain volume, these variables were regressed out of the GMm values for each of the five parcels across each hemisphere using MATLAB (2014a, The Mathworks, Inc., Natick, MA, United States). The resulting residual GMm were used in analyses.

Correlations and Model Fits
We conducted bivariate correlations to investigate the relationship between age and GMm of each of the five anatomically defined OFC subregions in each hemisphere. After demonstrating that such a correlation exists for the specific subregion, we then fitted linear, quadratic and cubic functions, using lowest Akaike Information Criterion (AIC) value to identify the best-fitting model, and adjusted R 2 to metricise goodness-of-fit. In addition, we ran a likelihood-ratio test to confirm whether the best fitting model explained significantly more variance than a null model. The null model was constructed by assuming no relationship with age and instead fitting the mean of the GMm.
Linear model: = + 1 ( ) Cubic model: Finally, we used the best-fitting model estimate of the residual GMm values at ages 11yrs and 35yrs to identify the percentage change in grey matter volume (GMv) for each functional subregion in the age range sampled using the following function: (model estimate of GMm age 11yrs -model estimate of GMm age 35yrs)*100. These follow-up analyses allowed us to characterise differences in shape and steepness between subregions.

Analysis of Covariance
To investigate and compare GM trajectories across functionally dissociable subregions of the OFC we subjected medial area 14, central OFC (cOFC average GMm of area 11 and area 13) and lateral area 47/12o to a repeated measures ANCOVAs. All key assumptions necessary to subject the data to a repeated measures ANCOVA were tested and met. With both the dependent variable, GMm, and the covariate, age, measured on continuous scales, and the independent variables being subregions of OFC and hemispheres. No significant outliers were identified (all data were less than 3 standard deviations from the mean, except two subjects with 2 data points <4SD above the mean, and two subjects with 1 data point <4xSD above the mean Residuals were subjected to ANCOVA methodology outlined by Schneider and colleagues (2015). The covariate measure of age was manually centered across all of the participants in the experiment before entered into 2-way repeated measures ANCOVAs. We included OFC subregion and brain hemisphere as independent variables, with age as a covariate (3(ROI: cOFC [11 & 13], area 14 and area 47/12o) x 2(Hemisphere) x age ANCOVA), and examined evidence in favour of the hypothesis of a differential effect of age on OFC subregion GMm, i.e. an interaction between OFC subregion GMm and age. Schneider et al (2015) then advise employing a standard ANOVA (3(OFC subregion) x 2(hemisphere) to evaluate any effects not involving the age covariate. However, due to the studyspecific GM template being an age-wise representative average of the brain structure of the subjects studied, and GMm values being relative to this average template (zero), the mean of all values will always be approximately zero for each of the OFC subregions. We therefore did not anticipate any independent main effects of OFC subregion or hemisphere. In this analysis, the primary focus examines relative differences in GMv trajectories and percentage changes in GMv, not differences in absolute GMv between brain regions. Accordingly, we do not report the results of follow up ANOVA.
We next examined the three functionally dissociable subregions in three separate 2(ROI: cOFC vs area 14 | cOFC vs area 47/12o | area 14 vs area 47/12o) x 2(Hemisphere) x age ANCOVAs. Again, we do not report results of follow uppost-hoc ANOVAs.

GM developmental model fits for all anatomical OFC subregions
To investigate the relationship between age and GM of each of the five anatomically defined OFC subregions we performed bivariate correlation analyses with complementary model fit comparisons.
While GMm in left and right hemispheres of each OFC subregion were significantly positively correlated (all rs > 0.532, ps < 0.001) we fit models to the GMm separately for each hemisphere of all five OFC subregions. Figure 3 illustrates the GMm values for each OFC subregions correlated against age and, if this correlation is significant, fit with their respective best fitting polynomial model.

Quantifying and comparing GMm across a priori functional OFC subregions
Next, we sought to test the key hypothesis that three functionally distinguishable OFC subregions mature at different rates. First, we quantified differences in grey matter between the youngest and oldest sampled participants in medial area 14, central OFC (average GMm of area 11 and area 13), and lateral area 47/12o. Averaging GMm across hemispheres and using the same model fit procedure describe above, we showed that medial, central and lateral regions were best fit, according to the AIC, each by a cubic model (area 14: adjusted R 2 = 0.097, LR p < 0.001, cOFC: adjusted R 2 = 0.107, LR p < 0.001, 47/12o: adjusted R 2 = 0.310, LR p < 0.001). Taking these model fits we determined the difference in GM across the age range sampled and report that area14 has a 4.69% reduction while cOFC shows a 7.86% reduction in GMv from age 11yrs to 35yrs. Note, when the relationship between the left cOFC is assessed in isolation this subregion was also best fit by a cubic model (adjusted R 2 = 0.279 LR p < 0.001) and showed a 10.91% reduction in GMv. By far the greatest reduction in GMv is in lateral area 47/12o which showed a 14.44% loss across the age range tested.
To examine whether these differences in GMv between subregions were meaningful we conducted ANCOVA analyses, comparing differences in GMm as a function of age among the three a priori functional OFC subregions of interest. GMm values from medial area 14, cOFC (area 11 and area 13) and lateral area 47/12o from both hemispheres were subjected to a 3 (OFC subregion) x 2 We performed follow up comparisons of each functional subregion with each other and report the key effects below. confidence intervals (dashed lines) are shown for each OFC subregion against each other. Model fits were used to estimate the percentage difference in GM volume between the youngest and oldest participants, reported on the right of each subplot. Lines to the right represent the proportionate reductions in GM volume from the youngest to the oldest age sampled. The pattern of results suggests a loose graded maturation of GM from medial to lateral areas as indexed by the differences between subregion developmental trajectories, with the clearest difference between the most medial OFC subregion (area 14) and the most laterally considered region (area 47/12o).

Medial area 14 v Lateral area 47/12o
We compared area 14 and 47/12o subregions directly using ANCOVA analysis. The results reveal differential developmental trajectories of GMm between these subregions as indexed by a significant age x OFC subregion interaction (F1,123 = 9.896, p = 0.002, Ƞ 2 = 0.074, Power = 0.877). Age was also a significant independent predictor of GMm (F1,123 = 26.437, p < 0.001, Ƞ 2 = 0.177, Power = 0.999). There were no further interactions involving age ( GMv in the right hemisphere of the cOFC has already matured by approximately 11 years (as there is no a relationship with age), driving the significant differences with right hemisphere of area 47/12o, but that the left hemisphere cOFC by contrast was still maturing. In the left hemisphere, GM maturation of the cOFC and area 47/12o do not differ in terms of the overall timing.

Discussion
The aim of this study was to quantify the relative differential patterns of structural maturation between anatomically and functionally dissociable OFC subregions. Analyses of high-quality structural MRI scans of 125 Human Connectome Project subjects of adolescent and young adult brains shows unique developmental profiles for each anatomically dissociable subregion. While there are fine-scale regional and hemispheric differences in the maturation profiles within the orbitofrontal cortex, the pattern of results provide evidence of a loose graded maturation with the phylogenetically younger lateral area 47/12o showing a longer developmental maturation profile compared medial OFC and showing a significant decrease in GMv across adolescence that continues well into young adulthood.
At the more fine-scale level, we observed that medial OFC showed a relatively small percentage reduction in GMv across age compared to other functional regions and the relationship with age was best characterised by quadratic functions. Central OFC subregions 11 and 13, though phylogenetically younger than their medial neighbour, showed hemispherically heterogeneous relationship between GMv and age. Their relationship with age was best fit by cubic and quadratic functions respectively, when it showed a significant relationship with age at all. The more lateral region 47/12o, however, unequivocally showed the greatest percentage reduction in GMv, with GM  (Rudebeck et al., 2013b). By contrast, area 14, plays a critical role in expected value coding  and biasing choice on the basis of value (Noonan et al., 2010). We recently translated and adapted the multi-option probabilistic decision-making paradigm used in these previous studies and tested a large online sample of adolescents and young adults between 11-35yrs (Wittmann et al., 2021). As the OFC is notably one of the last prefrontal regions to mature across adolescence, we predicted developmental delays in these two behavioural mechanisms. This prediction was borne out for learning strategies supported by the lateral region but not for decision-making strategies supported by medial OFC. Specifically, we found evidence that learning the causal relationships between choice and reward, mediated by the lateral area 47/12o, correlates with age. By contrast, we found no evidence that the children's choices were influenced by high values of irrelevant alternatives, as mediated by medial OFC.
Understanding OFC grey matter changes across adolescence at this fine-scale is critical to understanding the development of reward guided behaviours more generally, as well as predicting problematic behavioural and mental health outcomes. Broadly, the adolescent OFC shows heightened activity to large rewards and reward predicting cues, compared to adults (Galvan et al., 2006). This enhanced activity to large rewards, particularly in socially rewarding contexts, is associated with reduced cognitive control (Breiner et al., 2018). While too much OFC activity can translate to poor performance, so too can too little OFC grey matter. Diminished cortical thickness in this region (particularly lateral OFC) is associated with adolescent impulsive choice as measured by greater delay discounting (Pehlivanova et al., 2018) and Behavioural Activation and Inhibition questionnaire score (Ide et al., 2020). Collectively, these findings and others have shifted the emphasis of developmental models from broad imbalances between the PFC and subcortical structures to theories suggesting regional neurochemical, structural and functional brain changes within development lead to imbalances within brain circuitry. It is therefore necessary to underlie these theories with detailed examination of brain development with a precise anatomical resolution and to Altered OFC structure and function in adolescents is associated with a range of mental health disorders. For example, accelerated OFC thinning (particularly more lateral areas) is observed in adolescents who go on to develop depressive symptoms (Bos et al., 2018) and in individuals with bipolar with a history of suicide attempts compared to controls (Huber et al., 2019). While OFC surface morphology differentiates individuals with, or at risk of schizophrenia (Nakamura et al., 2007;Nakamura et al., 2019). Disordered OFC activity is also a common element in mental health conditions, with enhanced activity in tasks in which adolescents make social  or reward-based decisions. For example, Tegelbeckers and colleagues (2018) observed enhanced signal to large vs small expected rewards in the OFC of ADHD individuals relative to typically developing children/adolescents, with stronger reward-related activity correlated with individual differences in hyperactive/impulsive symptoms in the ADHD group. While the breadth of mental health disorders that the OFC is associated with could indicate a shared underlying dysfunction across all conditions the current apparent overlap and commonalities could also stem from a failure in the precision of which the anatomical foci are reported. If the development of the OFC is not uniform, this heterogeneity has important implications for theories and potential treatment strategies born out of this body of literature. Improving our localisation of these structural and functional differences, and linking them to their unique developmental trajectories, would then facilitate direct comparison to animal models which are already identifying fundamental dissociations in depression and anxiety in the neighbouring ACC's subregions (Wallis et al., 2017). This is also a crucial step in investigating pharmacological interventions, which is particularly important when there are such few viable treatment options for adolescents (Cipriani et al., 2016).