Data for functional MRI connectivity in transgender people with gender incongruence and cisgender individuals

We provide T2*-weighted and T1-weighted images acquired on a 3T MRI scanner obtained from 17 transwomen and 29 transmen with gender incongruence; and 22 ciswomen and 19 cismen that identified themselves to the sex assigned at birth. Data from three different techniques that describe global and regional connectivity differences within functional resting-state networks in transwomen and transmen with early-in-life onset gender incongruence are provided: (1) we obtained spatial maps from data-driven independent component analysis using the melodic tool from FSL software; (2) we provide the functional networks interactions of two functional atlases’ seeds from a seed-to-seed approach; (3) and global graph-theoretical metrics such as the smallworld organization, and the segregation and integration properties of the networks. Interpretations of the present dataset can be found in the original article, doi:10.1016/j.neuroimage.2020.116613[1]. The original and processed nifti images are available in Mendeley datasets. In addition, correlation matrices for the seed-to-seed and graph-theory analyses as well as the graph-theoretical measures were made available in Matlab files. Finally, we present supplementary information for the original article.

Functional MRI gender incongruence gender identity graph theory independent component analysis (ICA) resting-state connectivity seed-based analysis transmen transwomen a b s t r a c t We provide T2 * -weighted and T1-weighted images acquired on a 3T MRI scanner obtained from 17 transwomen and 29 transmen with gender incongruence; and 22 ciswomen and 19 cismen that identified themselves to the sex assigned at birth. Data from three different techniques that describe global and regional connectivity differences within functional resting-state networks in transwomen and transmen with early-in-life onset gender incongruence are provided: (1) we obtained spatial maps from data-driven independent component analysis using the melodic tool from FSL software; (2) we provide the functional networks interactions of two functional atlases' seeds from a seedto-seed approach; (3) and global graph-theoretical metrics such as the smallworld organization, and the segregation and integration properties of the networks. Interpretations of the present dataset can be found in the original article, doi: 10.1016/j.neuroimage.2020.116613 [1] . The original and pro-cessed nifti images are available in Mendeley datasets. In addition, correlation matrices for the seed-to-seed and graphtheory analyses as well as the graph-theoretical measures were made available in Matlab files. Finally, we present supplementary information for the original article.
© 2020 The Author(s Value of the data • These data provide structural and resting-state functional MRI scans of transgender individuals with gender incongruence. This condition is rare with prevalence rates around 1% in the population [8] . Therefore, study samples tend to be small making this dataset valuable to potentially increase other studies samples.
• These data can be valuable for researchers who aim to explore the underlying neurobiology of different gender variants from cisgender identities to transgenderism by means of MRI. • These data can be used as an independent sample for studies that aim to characterize and decompose spatial and temporal components into functional group-networks using restingstate imaging in young adults. The use of an independent sample in MRI studies can also be helpful to deal with the replication crisis .

Data
Data include the original T1-weighted and T2 * -weighted images of the 87 participants in the original article [1] that can be found in https://doi.org/10.17632/hjmfrv6vmg.2 [2] . In addition, this dataset includes the 200 functional seeds from the Craddock's atlas [9] , and the 56 seeds (from the default mode, salience, executive control and sensorimotor networks) extracted from the Stanford findlab atlas [10] (see Fig. 1 ). Seeds from these two functional atlases were used in the seed-to-seed and the graph theoretical approches.
The general linear model (GLM) comparing groups of transgender people with cisgender groups can be found in https://doi.org/10.17632/ts8c7fm8dj.1 [7] and in Figs. 2 -5 . GLM regressing out the effects of age and education can be found in the original article [1] .

Group comparisons from ICA spatial maps analyses without confounding variables
For the seed-to-seed and graph-theory approaches, unsmoothed images (see https://doi.org/ 10.17632/rw2yhtpj96.3 [5] and https://doi.org/10.17632/bgyzz94mz9.3 [6] ) were used to obtain the correlation connectivity matrices. These matrices can be found in https://doi.org/10.17632/ ts8c7fm8dj.1 [7] . Table 2 summarizes the significant test stats and P-values of the edges that differed between cismen > transmen.    Significant test stats and P-values without any covariates can be found in the cis-men_transmen_transwomen_noncovs_Stanford_atlas_results.mat file [7] and in Fig. 6 .
Additionally, data for the seed-to-seed group comparisons with the 200 functional ROIs of the Craddock atlas is provided (see the cismen_transmen_Craddock_atlas results.mat file [7] and Fig. 7 ).
Finally, global graph theoretical measures were calculate setting up relative thresholds of sparsity and taking the 100% of the connections (absolute) after deleting connections with negative values. Graph theory measures can be found in the Mendeley repository (doi: https: //doi.org/10.17632/ts8c7fm8dj.1 [7] ) both for absolute and relative thresholds for the two atlases employed [ 9 , 10 ]. Table 3 , Table 4 , Table 5 and Table 6 summarizes group means for each graphtheory parameter.

Participants
Twenty-nine transmen (TM) and 17 transwomen (TW) with gender incongruence according to the ICD-11 with an identification with the other gender; and 22 ciswomen (CW) and 19 cismen (CM) underwent MRI evaluation.
We selected 56 ROI from a functional template [10] . The downloaded ROI had a resolution of 2 mm in standard MNI space and they were transformed to a slice thickness of 3 mm with the flirt tool from FSL software.

Head motion parameters and noise correction
To control for head motion, an exclusion cut-off was established for mean interframe head motion at ≥ 0.3 mm translation or 0.3 °rotation; and for maximum interframe head motion at ≥ 1 mm translation or 1 °rotation.
To remove the effects of head motion and other non-neural sources of signal variation from the functional data, we used an ICA-based strategy for Automatic Removal of Motion Artifacts (ICA-AROMA) [11] . ICA-AROMA breaks data down via ICA and automatically identifies, which, if any, of these components are related to head motion by using four robust and standardized features ( https://github.com/maartenmennes/ICA-AROMA ).
As a quality control measure to assess the efficacy of ICA-AROMA in reducing relationship between signal variation and motion, we performed correlations between framewise head displacement and overall signal variation after regressing the ICA-AROMA components.

ICA spatial maps and dual regression
Melodic from FSL v5.0.10 ( https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/ ) was used to obtain temporalconcatenated spatial maps based on an ICA approach. Temporally and spatially coherent patterns of signal variation were extracted from functional images with a predetermined dimensionality of 25. The SN, the DMN and the bilateral ECN were considered to be the networks of interest.
The set of spatial maps from the group-average analysis was used to generate subject-specific versions of the spatial maps, and associated timeseries, using FSL's dual regression. First, for each subject, the group-average set of spatial maps is regressed (as spatial regressors in a multiple regression) into the subject's 4D space-time dataset. This ends in a set of subject-specific timeseries, one per group-level spatial map. Next, those timeseries are regressed (as temporal regressors, again in a multiple regression) into the same 4D dataset, resulting in a set of subject-specific spatial maps, one per group-level spatial map. We then tested for group differences using the FSL permutation-testing tool (5,0 0 0 permutations) with threshold-free cluster enhancement (TFCE). A binarized mask for each network was applied.

Intra and internetwork functional connectivity differences
The first eigenvariate of the BOLD signal temporal series was extracted for the ROI form the two atlases with the fslmeants command ( https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Fslutils ). The connectivity between two ROI was estimated using Pearson's correlation between their time series. Therefore, a 56 × 56 matrix and a second 200 × 200 matrix were obtained for each of the 87 subjects. Seed-to-seed intergroup differences in the strength of the edges were obtained with the threshold-free network-based statistics, TFNBS [12] . This method performs statistical inference on brain graphs and combines network-based statistics [13] , frequently used for statistical analysis of brain graphs, and TFCE, a method commonly used in voxel-wise statistical inference [14] . Matlab R2017a (The MathWorks Inc., Natick, MA, USA) was used to perform t test and Montecarlo permutation testing with 1,0 0 0 iterations between each of the four groups with and without considering age or education as confounding variables. Reported information survived Bonferroni connectome-wise correction for multiple comparisons at P < 0.05.

Graph-theory measures
A graph-theory approach was applied using the same matrices (56 × 56 × 87 for the Stanford atlas and 200 × 200 × 87 for the Craddock's atlas). Networks were constructed using only positive r values, i.e., setting negative values to 0. We used a sparsity threshold to create a set of undirected graphs (existing number of edges in a graph divided by the number of all possible edges) using the r correlation values as edge weights for each pair of seeds for each subject. Sparsity is a measure of network density that ensures that all subjects' networks would have the same number of edges to facilitate group comparisons. The range of sparsities was 5 to 25% with incremental steps of 2.5%. These percentages mean that, in the first threshold, 95% of the weakest connections will be deleted, 92.5% for the second and so on. The global graph theory measurements computed were: The clustering coefficient, which quantifies the number of connections that exist between the neighbors of a node as a proportion of the maximum number of possible connections. As a global measurement, it is the average of the clustering coefficient of all nodes. This measurement was normalized by 1,0 0 0 random networks (generated by random rewiring of the original network, maintaining degree distribution).
The characteristic path length of a node, which is the average of the minimum number of edges that must be traversed to go from this node to any other network node. As a global measurement, it is the average of the characteristic path length of all nodes. This measurement was normalized by 1,0 0 0 random networks.
Modularity, indicates the degree to which a network can be subdivided into well-delineated modules, each containing several densely interconnected nodes with relatively few connections between nodes in different modules.
Small world coefficient , defined as the ratio of the average clustering coefficient to the characteristic path length divided by the ratio of the same measurements obtained from equivalent random networks. Networks that have this small-world property usually have coefficients > 1.