Temporo-basal sulcal connections: a manual annotation protocol and an investigation of sexual dimorphism and heritability

The temporo-basal region of the human brain is composed of the collateral, the occipito-temporal, and the rhinal sulci. We manually rated (using a novel protocol) the connections between rhinal/collateral (RS-CS), collateral/occipito-temporal (CS-OTS) and rhinal/occipito-temporal (RS-OTS) sulci, using the MRI of nearly 3400 individuals including around 1000 twins. We reported both the associations between sulcal polymorphisms as well with a wide range of demographics (e.g. age, sex, handedness). Finally, we also estimated the heritability, and the genetic correlation between sulcal connections. We reported the frequency of the sulcal connections in the general population, which were hemisphere dependent. We found a sexual dimorphism of the connections, especially marked in the right hemisphere, with a CS-OTS connection more frequent in females (approximately 35–40% versus 20–25% in males) and an RS-CS connection more common in males (approximately 40–45% versus 25–30% in females). We confirmed associations between sulcal connections and characteristics of incomplete hippocampal inversion (IHI). We estimated the broad sense heritability to be 0.28–0.45 for RS-CS and CS-OTS connections, with hints of dominant contribution for the RS-CS connection. The connections appeared to share some of their genetic causing factors as indicated by strong genetic correlations. Heritability appeared much smaller for the (rarer) RS-OTS connection.


Introduction
During fetal development of the human cerebral cortex, in the second trimester of pregnancy, the initially smooth cortex starts growing into multiple folds. This process, called gyrification, produces multiple sulci on the surface of the brain (Chi et al. 1977;Zilles et al. 1988). Individuals largely share the same set of sulci that present high interindividual variability in terms of shape, length and sulcal connections, whose origin is not well-known (Ronan and Fletcher 2015;Borrell 2018;Kroenke and Bayly 2018).
From a genetic point of view, many studies have highlighted the possible genetic influences on the sulcal structure. The study by Pizzagalli et al. (2020) concluded that sulcal width was the most heritable measure (comparing variations in the length, depth, width, and surface area of several sulci associated with certain diseases). The authors also showed earlier forming sulci exhibited higher heritability (on average, for all four measurements) and that each hemisphere had partly specific genetic influences. The study by Yang et al. (2020), based on young Chinese Extended author information available on the last page of the article 1 3 adults, identified significant sex differences in sulcal morphology in several brain regions. In addition, monozygotic twins have been shown to have more similar sulcal patterns (sulcal depth, overall brain shape, study of geometric features by graph matching approach) than unrelated individuals (Lohmann 1999;Mohr et al. 2004;Im et al. 2011), which suggests a genetic contribution. Nevertheless, the study by Troiani et al. (2022) targeting the variability of the orbital sulci and using monozygotic and dizygotic twins showed that their variability was not of genetic origin, suggesting the complexity of being able to generalize findings across the whole brain.
From a clinical point of view, numerous studies have highlighted associations between sulcal morphology and various pathologies. The study by Kippenhan (2005) showed a reduction in the depth of certain sulci (particularly the collateral sulcus) in patients with Williams syndrome. Other pathologies such as Down syndrome (Yun et al. 2021) autism (Nordahl et al. 2007;Auzias et al. 2014;Ecker et al. 2016;Libero et al. 2019), dyslexia (Im et al. 2016), bipolar disorder and unipolar depression (Penttilä et al. 2009), schizophrenia Penttilä et al. 2008), Parkinson's disease (Pereira et al. 2012) or Alzheimer's disease (Im et al. 2008) have shown associations with certain sulcal morphologies (sulcal depth, surface-based morphometry, local gyrification index, etc.).
Connection-based sulcal variations are mainly studied by neurosurgeons, especially in the medio-basal part of the temporal lobe, which is a path of surgical approach used to treat pathologies such as some tumors and arteriovenous malformations, or pharmaco-resistant temporal lobe epilepsy (Cikla et al. 2016;Ovalioglu et al. 2018). Several post mortem studies have focused on describing the sulci of the inferior surface of the temporal lobe (collateral, occipito-temporal, and rhinal), by measuring their length, depth and mutual connections using dissected brains (Cikla et al. 2016;Ovalioglu et al. 2018). Another study , using MRI images, showed a possible link between temporal lobe epilepsy and the mutual connections of these sulci. In these three studies, each time a limited number of individuals was studied (N < 100) and the classifications used were not strictly similar (the meaning of what a connection is varies), producing results difficult to generalise and compare between them.
Here, we focused on the connections between sulci of the medio-basal part of the temporal lobe, namely the rhinal (RS), collateral (CS) and occipito-temporal (OTS) sulci. We selected the medio-basal temporal lobe because these connections have been understudied (small studies, inconsistent protocols and results) and they may relate to incomplete hippocampal inversion (IHI), a phenotype of interest due to its possible association with epilepsy (Bajic et al. 2009). The association between the abnormal positioning of the hippocampus and the depth of the collateral sulcus has been described in several studies (Baulac et al. 1998;Bernasconi et al. 2005;Cury et al. 2015) but the link between the sulcal depths and connections between these three sulci has not yet been established.
We aimed to study sulci connections using MRI images from large population samples, in order to produce robust results about sulcal connections frequencies, and associations with demographics and IHI. In addition, we used twins to estimate the extent to which individual differences can be explained by genetics (heritability) and environmental contributions. Finally, we also reported the co-occurrence of sulcal connections, which indicates whether similar genetic or environmental factors contribute to local or global sulcal polymorphism. In absence of an established manual protocol for large datasets and due to the complexity of automating this task, we relied on manual rating of sulcal connections and we created a modular rating protocol, which may be easily extended to other sulci. Indeed, sulcal connections are not easily extracted using automated image processing algorithms unlike sulcal depths. Our protocol and approach ensure that the data generated in this study will be compatible with future extended descriptions of sulcal connections that may focus on complementary brain regions.
Our starting hypothesis is that there is a stable frequency of sulcal connections that can be observed across independent samples of healthy subjects. This precise frequency of sulcal connections is yet unclear as previous studies relied on a limited number of individuals inducing very wide confidence intervals (Table S4). The second hypothesis is that of a heritable component to these sulcal connections, in line with the genetic influence that have been reported for sulcal characteristics such as length, depth or width (Pizzagalli et al. 2020). Lastly, our investigation of associations between sulcal connection and demographics or IHI is largely exploratory, and we do not have specific hypotheses about the possible associations.

Participants
We studied three databases: IMAGEN, QTIM, and QTAB. We kept only the MRIs whose quality made it possible to clearly see the sulci of interest (visual exclusion during subjects rating). The multi-centric European database IMAGEN (The IMAGEN consortium et al. 2010) contains data collected in 2,089 young individuals from four European countries (France, Germany, United Kingdom, and Ireland). We used the MRI images and demographic information acquired at baseline, when participants were 14 years old. The MRIs of 2005 individuals were judged of sufficient quality for sulci assessment. The Queensland Twin IMaging (QTIM) database (de Zubicaray et al. 2008;Strike et al. 2022a) contains data collected in over 1000 Australian individuals between 18 and 30 years of age. The MRIs of 979 individuals were judged of sufficient quality, which consisted of 144 complete monozygotic (MZ) pairs, 180 complete dizygotic (DZ) pairs, and 331 single members of a twin pair and siblings of twins. Finally, The Queensland Twin Adolescent Brain (QTAB) sample (Strike et al. 2022b; contains 422 Australian twins aged from 9 to 14 years. MRIs of 405 individuals were judged of sufficient quality, which comprised 101 complete monozygotic twin pairs, 94 complete dizygotic pairs and 15 unrelated individuals. We summarised the sample demographics in Table 1.

Patient consent statement and ethics approval
IMAGEN was approved by the local ethic committees and a detailed description of recruitment, assessment procedures, and exclusion/inclusion criteria have been published in (The IMAGEN consortium et al. 2010), QTIM was approved by the Human Research Ethics Committees (HREC) at the University of Queensland (de Zubicaray et al. 2008), and QTAB was approved by the Children's Health Queensland HREC and the University of Queensland HREC (Strike et al. 2022b).

MRI acquisition
We relied on 3-Dimensional T1-weighted anatomical MRI, to determine the connections between sulci.
For IMAGEN, the MRIs were acquired on a range of 3 Tesla scanners (Siemens Verio and TimTrio, Philips Achieva, General Electric Signa Excite, and Signa HDx), which depended on the acquisition site. All sites used the same acquisition parameters of an MPRAGE (Magnetization Prepared Rapid Acquisition Gradient Echo) sequence (TR = 2300 ms; TE = 2.8 ms; flip angle = 9°; resolution = 1.1 × 1.1 × 1.1 mm). In order to assess the sulcal connections with a standardized MRI orientation, the images were registered toward the MNI152 atlas using the automated affine transformation method FLIRT from FSL software (Jenkinson and Smith 2001;Jenkinson et al. 2002), as in Cury et al. In QTIM, the MRIs were acquired on a 4 Tesla Bruker Medspec scanner using an inversion recovery rapid gradient echo protocol (TI = 700 ms; TR = 1500 ms; TE = 3.35 ms; flip angle = 8°; resolution = 0.94 × 0.98 × 0.98 mm). We registered the MRIs in the MNI152 space using the t1-linear pipeline in Clinica (Wen et al. 2020;Routier et al. 2021).
The differences in acquisition parameters between the three databases did not impact the manual morphological classification. For example, Figure S1 shows a coronal view from the three databases. Overall, the image resolution can be a limiting factor for manual rating, but the voxel sizes ranged from 0.8 × 0.8 × 0.8 mm (QTAB) to 1.1 × 1.1 × 1.1 mm (IMAGEN), which were not visually disturbing to classify the sulcal connections. Note that the image registration in the MNI152 space simplifies the classification as it homogenizes the orientation and position of the brain as well as the resolution of the databases (1 × 1 × 1 mm). More advanced image processing and harmonization remains warranted (and an active field of 1 3 research) when extracting automatic structural measurements (Gebre et al. 2023).

A connection-based classification
We propose a classification based on the physical connections between the collateral (CS), rhinal (RS) and occipito-temporal (OTS) sulci. To facilitate manual rating, we have only considered connections in the anterior part of the medio-temporal lobes (the posterior part is more variable and more complex to classify) (Fig. 1). For the same reason, we have not considered the possible morphological sulcal variations (subcomponents, side branches, number of segments, etc.). For example, we have considered the anterior transverse collateral sulcus to be part of the collateral sulcus. We considered each connection as a binary variable, which makes our classification modular and easy to extend to other sulci. Each variable denotes the presence (coded 1) or absence (coded 0) of connection between the rhinal and collateral sulci (noted RS-CS connection), between the collateral and occipito-temporal sulci (CS-OTS connection) and between the rhinal and occipito-temporal sulci (RS-OTS connection). Our rating considers the connections individually (connection-based classification). It allows studying specific associations with variables of interest (e.g. demographics, or clinical), but also allows studying the co-occurrence of connections as well as the total number of connections. Previous works relied on pattern-based classification, i.e. patterns comprising several sulci rather than individual connections were rated Cikla et al. 2016;Ovalioglu et al. 2018), which does not facilitate the analyses and is more difficult to update when considering additional sulci. Importantly, our connection-based classification may be easily converted into pattern-based classification if needed (see supplementary materials for the correspondence with pattern-based classification).

Collateral sulcus variations and anterior/posterior boundary of the medio-temporal lobe
Although the collateral sulcus (CS) has a very constant morphology in its anterior part, it can be present as a singlebranch, separated into two branches (an occipital branch and a temporal branch), or can be composed of two nonconnected sulci in its posterior part (Fig. 2). This sulcus delimits the medial occipito-temporal gyrus (formed by the para/hippocampal gyrus in the anterior part and the lingual gyrus in the posterior part) from the lateral occipitotemporal gyrus, also called the fusiform gyrus (Fig. 2). We considered the anterior/posterior boundary of the medio-temporal lobe to be located at the split of the CS into two branches or, in absence of a split, around the center of the CS, approximately at the separation of the hippocampal gyrus and the lingual gyrus (Fig. 2).

Occipito-temporal and Rhinal sulci variations
The occipito-temporal sulcus (OTS) is the sulcus with the most frequent morphological variations among the three sulci considered, with up to six distinct sections recorded (Cikla et al. 2016). It is located laterally to the collateral sulcus. It delimits, medially, the lateral occipitotemporal (fusiform) gyrus and, laterally, the inferior temporal gyrus (Fig. 2).
The rhinal sulcus (RS) is the least described sulcus of the three in the scientific literature because it has long been considered part of the collateral sulcus (so called anterior collateral sulcus) rather than a separate sulcus. It is located in the antero-medial position of the collateral sulcus.

Assessing connections between sulci and visual rating guidelines
We assessed the T1w images with the medInria visualization software (https:// med. inria. fr/). For each hemisphere, we scrolled through the axial view to distinguish each sulcus with their mutual connections (Fig. 3). Connection is observed when there is a meeting point at any depth between two sulci (Fig. 3, III axial, III coronal and IV axial). For a good appreciation of the connections, we recommend considering both the coronal view and the axial view, although we often found the coronal view to be the most conclusive (Fig. 3). We recommend starting by locating the collateral sulcus (which has the least morphological variations) and then the other two sulci. Another figure is also available in supplementary materials ( Figure S2).

Ambiguous connection
A difficulty arises when the CS and OTS sulci merge and connect with RS, as it becomes difficult to know which of the CS or OTS connects with the RS (Fig. 4). In this particular case, we rated that all sulci connect (RS-CS = 1, CS-OTS = 1 and RS-OTS = 1).

Reproducibility of the classification (intra/inter-observer)
Kevin De Matos (KDM) performed manual assessment on all the MRI images. We estimated the intra-observer reproducibility on the first 100 IMAGEN individuals rated a second time by KDM. In addition, we estimated the interobserver reproducibility thanks to a second rater (LC = Lydia Chougar, neuroradiologist) who evaluated the same 100 individuals. We reported the reproducibility using Cohen's kappa without weighting. For a better understanding, we have also reported the proportions of true positives/true negatives/false positives/false negatives compared with KDM.

Descriptive analysis of sulcal connections
First, we reported the frequency of each sulcal connection across the three databases. Next, we tested the association between sulcal polymorphisms and demographics or general variables (sex, age, weight, height, intracranial volume (ICV), handedness, BMI and site). Then, we evaluated possible associations between sulcal polymorphisms and incomplete hippocampal inversions (IHI). For IMAGEN, the IHI data came from the manual rating used in the study by Cury et al. (2015) and for QTIM/QTAB the IHI data were manually rated by KDM using the protocol described in Cury et al. Temporal lobe anatomy and connection example. 0 = hippocampus; 1 = collateral sulcus; 2 = occipito-temporal sulcus; 3 = inferior temporal sulcus; 4 = superior temporal sulcus; 5 = lateral sulcus; 6 = rhinal sulcus; A = hippocampal gyrus; B = fusiform gyrus; C = inferior temporal gyrus; D = middle temporal gyrus; E = superior temporal gyrus. Images II to IV (axial and coronal views) are from the same subject. We can see a connection between the collateral sulcus and the occipito-temporal sulcus in III axial, III coronal and IV axial (red circle) We used a generalized linear model (GLM) for the statistical analysis and significance testing using the "statsmodels" python package. Specifically, we used a logistic regression as the sulcal connections were binary variables, and we reported the OR (odd-ratios), and p-value from the Wald/ log-likelihood ratio test. We evaluated the correlations between sulcal connections and all general and demographic variables, using the joint model: We used IMAGEN as a discovery sample and evaluated if the significant associations replicated in QTIM and QTAB. We used a significance threshold of 7.81e−04 = 0.05/64 (64 = (3 connections + nbr of connection) × 2 hemispheres × 8 demographic variables) tested in the discovery sample, which accounts for the number of tests performed (Bonferroni correction). As twins from the same family may not yield independent observations (a hypothesis of GLM modeling), we selected a single individual per family in the replication analyses. Similarly, when testing associations between sulcal polymorphisms and IHI, we used the model: we used a significance threshold of 3.57e−04 = 0.05/140 (140 = 5 IHI criteria × 2 hemispheres × 14 variables).

Heritability
We used the twin samples QTIM and QTAB to estimate the heritability of sulcal connections. The heritability quantifies how much of the individual differences (variance) in sulcal connections may be attributable to genetic differences. Heritability ranges between 0 and 1, with 0 corresponding to a trait with no genetic influence, and 1 indicating that the genetic differences account for all of the trait variability in the population.
We first reported the intra-pair tetrachoric correlations estimated with the R programming packages umx (Bates et al. 2019) and psych (Revelle 2022). Tetrachoric correlation is best suited to quantify correlations between dichotomous variables of interest. Intra-pair correlations that are larger in MZ pairs than in DZ pairs suggest that a trait is heritable. Next, we estimated the heritability using ACE and ADE models (Neale and Cardon 2011;Verweij et al. 2012) with the R programming package umx and openMx (Neale et al. 2016). The ACE model decomposes the trait variance into additive genetic contributions (A), shared or familial environmental contributions (C) and a residual term (E), which includes individual specific environmental sources of variance as well as measurement error. The ADE model allows estimating the contribution of genetic dominant effects (D). In the case of an ADE model, the broad sense heritability corresponds to the A + D contributions, by opposition to the narrow sense heritability which consists of the sole additive genetic contribution. For both models, we used age, sex, weight, height and ICV as covariates.

Co-occurrence of sulcal connections
We also reported the co-occurrence between the different sulcal connections, either intra-hemispheric or inter-hemispheric. We estimated these co-occurrences, controlling for covariates, using a GLM with the following formula: We used a significance threshold of 1.85e−04 = 0.05/270 (270 = 30 possible co-occurrence combinations × 9 variables) tested in the discovery sample, which accounts for the number of tests performed (Bonferroni correction).

Genetic correlations
Lastly, we reported genetic correlations (rG), estimated using twin models. Genetic correlations indicate how much of the genetic sources of variance may be common between the sulcal connections. In other words, how much are sulcal connections influenced by the same genetic variants. Note that we have calculated genetic correlations [between the traits liability, i.e. assuming a liability threshold model (Neale 2014)] in this section whereas the phenotypic co-occurrences are reported as odds ratio.

Descriptive analysis: frequency
We observed that the frequencies of the sulcal connections were comparable across the three different samples (based on overlapping confidence intervals in most cases, Fig. 5).
In the left hemisphere, for IMAGEN, 18.2% of individuals had zero connections, 57.3% had a single connection, 20.3% had two while 4.2% had three connections (which includes the ambiguous case). For QTIM + QTAB, these figures were respectively 23.4%, 54.3%, 21.0%, and 1.4%. In the right hemisphere, for IMAGEN, 36.2% of individuals had zero connections, 52.4% had a single connection, 9.0% had two while 2.3% had three connections. For QTIM + QTAB, these figures were respectively 42.3%, 47.5%, 10.0%, and 0.2%. See Table S2 for separate frequencies of QTIM and QTAB, and see Table S3 and Figure S3 for frequencies by pattern in supplementary materials.

Descriptive analysis: demographic GLM model with replication
In the GLM model, after Bonferroni correction (significance threshold of 7.81e−04), we found in IMAGEN that the sulcal connection frequency was strongly associated with sex, in particular for the right RS-CS connection (28% in females, 44% in males, OR = 2.0 p-value = 1.41e−08, controlling for all other covariates) and CS-OTS connection (39% in females, 23% in males, OR = 0.5 and p-value = 1.15e−07). In the left hemisphere, the RS-CS connection was also more common in males (35% in females, 46% in males, OR = 1.55 and p-value = 2.19e−04) (Tables 3, 4 and Fig. 5).
We replicated the sexual dimorphism of the right CS-OTS in QTIM + QTAB (35% in females, 21% in males, OR = 0.51   Table 5). The contralateral side showed partly the same association but with lower odds ratios. In Table 6, we found that replication with QTIM and QTAB was satisfactory for the left hemisphere with odds ratios going in the same direction and also statistically significant. The right hemisphere gives less consistent replication results with four of the seven odds ratios close to one and becoming non-significant.

Twin pair tetrachoric correlation
In Table 7, showing the within-pair tetrachoric correlations, we observed a seemingly larger correlation between monozygotic (MZ) twins than between dizygotic (DZ) twins for the RS-CS and CS-OTS connections on both sides. Note that MZ > 2*DZ (e.g. for left RS-CS) suggests a non-additive (e.g. dominant) genetic contribution. The negative tetrachoric correlation for opposite sex pairs, might suggest a sex-limitation model, even if the confidence intervals remained wide, considering the current sample size. On the other hand, we found no evidence of twin pair correlation for the RS-OTS suggesting little heritability or at an undetectable level considering the current sample size.

Heritability: ACE and ADE models
We fitted both ACE and ADE models and used the model with the lowest AIC value (best fitting) to further test the significance of A and C/D (  Table 9 presents the co-occurrence of sulcal connections. Overall, after Bonferroni correction (significance threshold of 1.85e−04), we found the strongest correlations between the same type of connection in the left and right hemispheres (e.g. left RS-CS: OR = 12.12 and p-value = 2.38e−96 with the right RS-CS). The third line represents the co-occurrence in percentage (e.g. when left RS-CS was present, left CS-OTS was also present at 49.2% but when left CS-OTS was present, left RS-CS was also present at only 33%). In Table 10, the replication with QTIM + QTAB confirmed to us that the same type of connection in the left and right hemispheres remained significant (e.g. left RS-CS: OR = 16.22 and p-value = 6.74e−43 with the right RS-CS).

Genetic correlations
We estimated the genetic correlation (A) for each pair of two connections. In detail, we detected a high and significant correlations between left/right CS-OTS connections (A = 1.00, 95% CI = 0.80-1.00 and p-value = 7.00e−12) and between left/right RS-CS connections (A = 0.94, 95% CI = 0.78-1.00 and p-value = 4.80e−11) (Table 11). It was the same connection pairs that obtained the highest odds ratios for the co-occurrence. The RS-OTS connections on both sides have not been tested due to lack of heritability. We also estimated the environmental correlation (E) giving several significant results but weak correlations.

Discussion
In this study, we introduced a simple manual classification of the morphological variations of the rhinal, collateral and occipito-temporal sulci by focusing only on their mutual connections. We have obtained many novel results: precise characterization of sulcal connection frequencies with evidence of important hemispheric and sexual differences as well as evidence of moderate broad-sense heritability for two of the connections with also the sharing of genetic and environmental causal factors. Our classification is modular and may be extended to other sulci and lobes. We showed using multiple raters that our classification was reproducible, in particular for the RS-CS (kappa = 0.77) and CS-OTS (kappa = 0.77) connections (Table 2). Reproducibility was lower for the RS-OTS connection (kappa = 0.48), possibly due to ambiguous situations and a rarer phenotype.
We performed a radiological evaluation of nearly 3400 healthy young individuals from three databases. We reported precise frequency of each connection between sulci (Table 3). Thanks to our approach, we have for the first time quantified this on a large-scale. This is in itself a novel contribution to neuroanatomical knowledge, and it could serve as a basis for future research (for example by comparing our results with databases of patients with various neurological diseases, by using our data to create automated classification methods, by increasing the number of sulcus taken into account, etc.). Moreover, the hemispheric differences found would be in support of different sulcal development patterns between hemispheres. Indeed, there was, for the left hemisphere, a frequency of 55-60% for the CS-OTS connection and 35-40% for the RS-CS connection while, for the right hemisphere, there was a frequency of 30-35% for RS-CS connection and 25-30% for CS-OTS connection. We therefore found that sulcal connections were more frequent in the left hemisphere. On this subject, the ENIGMA-Laterality Working Group has published several studies of brain asymmetries (Guadalupe et al. 2017;Kong et al. 2022). In particular, the study by Kong et al. (2018) focused on cortical thickness and surface area in 17,141 healthy subjects. It revealed global hemispheric asymmetries, with thicker cortex in the left hemisphere but smaller cortical surface area. At the regional level, many asymmetries were reported, including some in the anterior temporal lobe. The entorhinal cortex and the temporal pole tended to be thicker in the right hemisphere, but the cortical surface area of all gyri (inferior temporal, fusiform, entorhinal, and parahippocampal) appeared larger in the left hemisphere. More work is needed   to evaluate whether these asymmetries could be associated with the asymmetries in sulcal connections we reported here. Unexpectedly, when individuals were grouped by sex, we found large frequency differences for CS-OTS and RS-CS connections (Tables 3 and 4). Specifically, the left and right RS-CS connections were significantly more frequent in males (left RS-CS: 35% in females, 46% in males, OR = 1.55 and p-value = 2.19e−04; right RS-CS: 28% in females, 44% in males, OR = 2.0 and p-value = 1.41e−08) and the right CS-OTS connection was more common in females (39% in females, 23% in males, OR = 0.5 and p-value = 1.15e−07). At this level, it is difficult to understand the underlying mechanisms responsible for these differences. However, we ruled out that this sexual dimorphism could be explained by intracranial volume or other demographic data. However, we only replicated the sexual dimorphism (using QTIM + QTAB) of the right CS-OTS (35% in females, 21% in males, OR = 0.51 and p-value = 5.06e−03). To our knowledge, our study is the largest to date to study sulci connection and the first one to report precise frequency in the general population as well as a sex difference. The study of Novak et al. (2002), with a partially compatible classification, used 50 individuals MRIs, and highlighted a sexual dimorphism for the RS-CS connection which went in the same direction as our study (more frequent in men than in women) and concluded that the sex of individuals was a significant factor in sulcal patterns. Other studies done in the meantime have not confirmed this tendency. This sexual dimorphism may also relate to previous reports of sexual dimorphism in the brain. It is well established that males have on average a 10-12% larger total brain volume (Ruigrok et al. 2014;Kruggel and Solodkin 2020) as well as larger cortical surface area, while females tend to exhibit thicker cortical thickness (Ritchie et al. 2018). At a regional level, those sexual differences are observed throughout the anterior temporal pole, but are greatly reduced when controlling for intracranial volume differences (Ritchie et al. 2018). In addition, sex was also associated with brain asymmetry, with males showing more asymmetry in cortical thickness of the parahippocampal and entorhinal gyri, as well as less asymmetry in surface area (Kong et al. 2018). Finally, several psychiatric and behavioral dimensions are associated with sex, and more studies are required to understand their relationships with brain sexual dimorphism. Only significant values of Table 5   - In addition, we did not observe an association between sulcal connection and handedness, although the association may be small and undetectable with current statistical power. The ENIGMA consortium came to a similar conclusion when looking at cortical thickness, cortical surface area (Kong et al. 2018) or the volume of subcortical structures (Guadalupe et al. 2017).
Our work follows and extends that of four other published studies, that studied sulci connection and used a compatible classification system Huntgeburth and Petrides 2012;Cikla et al. 2016;Ovalioglu et al. 2018) ( Table S4). The number of participants in those studies were small, varying between 18 and 51, leading to wide confidence intervals of frequency estimate (Table S4). Overall, the studies of Cikla et al. and Ovalioglu et al. reported connection frequencies consistent with our results. For the study of Huntgeburth et Petrides, there was a higher frequency of the B pattern (only RS-CS connection) from the left hemisphere and a lower frequency of the E pattern (more than one connection) on both sides. The study of Kim et al., on healthy individuals, had the same frequency differences as Huntgeburth et Petrides but with a lower frequency of the Table 7 Tetrachoric correlation   The Table 7 presents the tetrachoric correlations in the pairs of monozygotic (MZ) and dizygotic (DZ) twins resulting from the addition of the QTIM and QTAB databases with their confidence intervals as well as the separation of the male/female data     A pattern (no connection) from the right hemisphere. The origin of these differences have not been identified but could be caused by the manual ratings which are not strictly similar, the proportion of males/females not controlled for and the material used (MRI images versus post mortem studies). For the study of Kim et al., the frequencies obtained in individuals with temporal lobe epilepsy (TLE) would require a large-scale study to be confirmed. Regarding the association between sulcal connections and IHIs, an anatomical variation of the hippocampus, we were able to confirm our initial hypothesis that a link existed between the two. More specifically, RS-CS and CS-OTS connections on both sides are associated with a deeper collateral sulcus and more medial position of the hippocampus ( Table 5). Most of the associations replicated in QTIM + QTAB (twelve out of sixteen and especially all those in the left hemisphere) ( Table 6). To better understand this link, it will be necessary to consider more morphometric measurements of the sulci (surface, maximum depth, average depth, length, sulcal opening, etc.) in future studies. In a next step, we could also imagine targeting patients with IHI-related brain disorders (epilepsy, possibly autism, schizophrenia, etc.) to try to highlight a phenotypic association between these pathologies and some sulcal variations.
The heritability studies we performed have shown broad-sense heritability coefficients h2 (Additive + Dominance) of 0.45 on the left and 0.30 on the right for the RS-CS connection as well as 0.30 on the left and 0.28 on the right for CS-OTS connection ( Table 8). The sexual dimorphism for RS-CS (Table 4) as well as the negative tetrachoric correlation for opposite sex pairs (Table 7), could warrant to study whether the connection may be more heritable for one sex, and/or if different genetic loci contribute to the phenotype in each sex (sex-limitation models) but a larger number of e twin pairs would be required to yield conclusive results. Reporting heritability is a first for these phenotypes, and opens the way to genome-wide association studies (GWAS), which could identify the genetic loci that cause these sulcal connections. Several studies have investigated the heritability of sulcal metrics (length, average depth, maximum depth, width, surface area), which vary greatly between metrics and brain regions (overall range 0-0.72) (Le Guen et al. 2018;Pizzagalli et al. 2020). Pizzagalli et al. meta-ana-lyzed several twin studies (incl. QTIM) and showed that the collateral sulcus structure was under moderate genetic influence (e.g. h2 = 0.51 for mean depth, h2 = 0.42 sulcal width or 0.32 for surface area). In contrast, heritability of the anterior occipito-temporal sulcus appeared smaller (h2 between 0.26 and 0.32) while it was around 0.19-0.24 for the rhinal sulcus structure. In addition, thickness and surface area of the neighboring gyri also display a moderate heritability (Strike et al. 2019). Future work should clarify which genetic variants cause the structural variation in the population, and whether sulci and gyri structure are   influenced by variants that also contribute to the sulcal connections we reported here. Sulcal connections tended to be largely symmetric as indicated by the large correlations between hemispheres that replicated in QTIM and QTAB (Tables 9 and 10). Other research based on patterns has come to the same conclusion Ovalioglu et al. 2018). We also found that, in general, there was a negative correlation between RS-CS and CS-OTS connections. Finally, we found large genetic correlations, particularly for the RS-CS and CS-OTS connections, which indicate that the co-occurrence of the sulcal connection is likely influenced by shared genetic factors (Table 11).
Our study has some limitations. First, we only considered three sulci and cannot generalise our findings throughout the brain. Our modular rating protocol allows to build on our results and progressively study other sulci. We will share our manual ratings for future reuse by interested researchers. Secondly, the number of twin pairs limited our investigation of the heritability, in particular to detect non-additive genetics or sex specific contributions. Thirdly, our use of a replication sample is a strength of the study and ensures replicated results are robust, however some sex dysmorphisms, co-occurrences, and association with IHIs did not replicate. The lack of replication is, to some extent, expected due to the winner's curse and a smaller replication sample, but it warrants future investigations in larger samples. Finally, the confidence interval of the frequencies for each connection does not systematically overlap between each database (IMAGEN versus QTIM versus QTAB). We can only hypothesise that this might be caused by a slight change in how the rater classified individuals after thousands of classifications, unknown differences between the twin databases, or differences in image quality between studies.
Having demonstrated that the connections between the rhinal, collateral and occipito-temporal sulci have a stable frequency determined by genetic and environmental factors, our work calls to establish more precise classifications that include more sulci that would pave the way towards a global understanding of what create these anatomical variations. It remains unclear whether sulci connections could have a clinical significance. Using our classification principle and data, the development of an automatic sulci-connection classifier for future large-scale studies could accelerate this type of research.

Conclusion
In conclusion, our manual classifications over 3,400 individuals provide accurate evaluations of sulcal connection frequencies between the rhinal, collateral and occipitotemporal sulci. Our protocol may be extended to include connections with other sulci or the different sulcus patterns (side branches, number of segments, …). In addition, we reported hemispheric asymmetries and sexual dimorphisms in sulci connections and showed that the connections exhibit broad sense heritability. Our study is the largest to date on sulcal connections, and suggests they are worth further Funding The research leading to these results has received funding from the French government under management of Agence Nationale de la Recherche as part of the Investissements d'avenir program, reference ANR-19-P3IA-0001 (PRAIRIE 3IA Institute) and reference ANR-10-IAIHU-06 (Agence Nationale de la Recherche-10-IA Institut Hospitalo-Universitaire-6). BCD is supported by the NHMRC (CJ Martin Fellowship, APP1161356). The Imagen study is supported by the following sources. This work received support from the following sources: the European Union-funded FP6 Integrated Project IMAGEN (Reinforcement-related behaviour in normal brain function and psychopathology) (LSHM-CT-2007-037286), the Horizon 2020 funded ERC Advanced Grant 'STRATIFY' (Brain network based stratification of reinforcement-related disorders) (695313), Human Brain Project (HBP SGA 2, 785907, and HBP SGA 3, 945539), the Medical Research Council Grant 'c-VEDA' (Consortium on Vulnerability to Externalizing Disorders and Addictions) (MR/N000390/1), the National Institute of Health (NIH) (R01DA049238, A decentralized macro and micro gene-by-environment interaction analysis of substance use behavior and its brain biomarkers), the National Code availability Code used to process the data and perform the analyses will be available upon publication as https:// github. com/ Kevin DMR.

Declarations
Conflict of interest Disclosure of interests related to the present article: none to disclose. Disclosure of interests unrelated to the present article: OC reports having received consulting fees from AskBio and Therapanacea, having received fees for writing a lay audience short paper from Expression Santé, and that his laboratory has received grants (paid to the institution) from Qynapse. Members from his laboratory have co-supervised a PhD. thesis with myBrainTechnologies and with Qynapse. OC's spouse is an employee of myBrainTechnologies. OC holds a patent registered at the International Bureau of the World Intellectual Property Organization (PCT/IB2016/0526993, Schiratti J-B, Allassonniere S, Colliot O, Durrleman S, A method for determining the temporal progression of a biological phenomenon and associated methods and devices). Dr. Banaschewski served in an advisory or consultancy role for eye level, Infectopharm, Lundbeck, Medice, Neurim Pharmaceuticals, Oberberg GmbH, Roche, and Takeda. He received conference support or speaker's fee by Janssen, Medice and Takeda. He received royalties from Hogrefe, Kohlhammer, CIP Medien, Oxford University Press. The present work is unrelated to the above grants and relationships. Dr. Barker has received honoraria from General Electric Healthcare for teaching on scanner programming courses. Dr. Poustka served in an advisory or consultancy role for Roche and Viforpharm and received speaker's fee by Shire. She received royalties from Hogrefe, Kohlhammer and Schattauer.