Neurogenetics and brain connectomics: insights from healthy individuals, monogenic and polygenic disorders

Neurogenetics and brain connectomics: insights from healthy individuals, monogenic and polygenic disorders Antonio Jim´enez Mar´ın PhD Supervisor Jesus M. Cortes Diaz 2023 (cc)2024 ANTONIO JIMENEZ MARIN (cc by-nc 4.0) A la familia que tengo desde que nac´ı y a la que he ido incluyendo en estos 30 a˜nos Gracias por estar ah´ı Resumen La Resonancia Magn´etica (RM) es una t´ecnica no invasiva que se puede utilizar para adquirir im´agenes de alta resoluci´on de los ´organos de un animal vivo. Existen una amplia variedad de modalidades para adquirir im´agenes que involucran diferentes aspectos de los tejidos que forman los ´organos, por ejemplo, propiedades de difusi´on del agua, oxigenaci´on, perfusi´on, susceptibilidad, etc. Para estudiar el cerebro in vivo, la resonancia magn´etica se ha utilizado ampliamente en la pr´actica cl´ınica y en investigaci´on, proporcionando informaci´on sobre la integridad del tejido cerebral, sus malformaciones, la organizaci´on de la microestructura y la din´amica funcional. A partir de im´agenes de RM, el cerebro humano se puede modelar como una red de redes, que se denomina conectoma y puede ser tanto estructural como funcional. Estas redes, representadas matem´aticamente como grafos, se componen por nodos, que se corresponden con las diferentes ´areas del cerebro que se pretende estudiar, y enlaces o links que definen c´omo se conectan unos nodos con otros. El conectoma cerebral se puede utilizar para estudiar poblaciones con trastornos neurol´ogicos o psiqui´atricos, pero tambi´en para comprender la organizaci´on a gran escala del cerebro sano. Dado que cada conectoma humano tiene su propia huella [1], es posible anotar sus caracter´ısticas en funci´on de sus funciones cognitivas de alto orden, variables cl´ınicas, datos gen´eticos u otras caracter´ısticas fenot´ıpicas [2]. Adem´as, las caracter´ısticas del conectoma no s´olo difieren entre participantes sanos y patol´ogicos, sino que tambi´en cambian durante el envejecimiento, lo que permite encontrar biomarcadores basados en conectomas que podr´ıan ser ´utiles para el diagn´ostico y pron´ostico de patolog´ıas [3, 4, 5, 6, 7, 8, 9, 10] La multimodalidad de los datos cerebrales adquiridos con resonancia magn´etica, junto con otras t´ecnicas de adquisici´on, permite buscar una amplia gama de biomarcadores mediante enfoques computacionales que aprovechan las diferentes fuentes de datos. En los ´ultimos a˜nos, la publicaci´on de datos de transcripci´on a nivel celular y de todo el cerebro ha permitido a los investigadores vincular las iv caracter´ısticas del conectoma con datos gen´eticos. Esta estrategia permite identificar genes y tipos de c´elulas espec´ıficos implicados en el desarrollo y progreso de trastornos relacionados con el cerebro [11]. T´ecnicas como los estudios de asociaci´on de todo el genoma (GWAS) o los estudios de asociaci´on de todo el transcriptoma (TWAS) se han utilizado ampliamente para vincular fenotipos derivados de im´agenes (IDP) con polimorfismos de un solo nucle´otido (SNP), no solo en variantes neurodegenerativas y neuropsiqui´atricas sino tambi´en en poblaciones sanas [12, 13, 14, 15, 16, 17, 18]. Ambos enfoques requieren un gran n´umero de participantes e hip´otesis s´olidas sobre los rasgos gen´eticos que se van a estudiar. Adem´as, la informaci´on gen´etica se extrae de la sangre o la saliva, por lo que no puede dar informaci´on sobre c´omo la variante gen´etica se distribuye espacialmente y est´a afectada a lo largo del tejido cerebral [11]. Un enfoque alternativo que resuelve los problemas antes mencionados de GWAS o TWAS es el uso de atlas transcript´omicos cerebrales [19, 20, 21, 22, 23, 24], que proporcionan la expresi´on de miles de genes que cubren diferentes ubicaciones del cerebro. Entre las bases de datos disponibles, el m´as utilizado es el Allen Human Brain Atlas (AHBA), que incluye datos de seis donantes sin ninguna patolog´ıa conocida – por lo tanto, permite estudiar mecanismos fundamentales organizativos – y cubre todo el cerebro [19]. Para utilizar este atlas en estudios de resonancia magn´etica, lo que hacemos es considerar como datos normativos los perfiles transcript´omicos extra´ıdos de los cerebros, extrayendo todas las muestras de cada donante en un atlas ´unico. Despu´es de varios pasos de preprocesamiento [25], necesarios para limpiar y estandarizar los datos, es posible correlacionar la expresi´on transcript´omica de los genes disponibles con los IDPs que se desean estudiar. Hay algunas cuestiones a tener en cuenta al utilizar esta base de datos: •Varios genes en la base de datos tienen diferentes patrones de expresi´on entre los donantes. En [26], los mismos autores que publicaron la base de datos original definieron una medida que tiene en cuenta esa cuesti´on, a saber, la Estabilidad Diferen- v cial (DS), que indica la reproducibilidad de la expresi´on en los seis donantes de cada gen. El par´ametro DS se puede utilizar para seleccionar la sonda m´as reproducible anotada en un gen espec´ıfico o para filtrar genes de baja reproducibilidad para los an´alisis. •Para la construcci´on de este atlas, s´olo se utilizaron muestras de dos de los seis cerebros en ambos hemisferios. Una opci´on es utilizar ´unicamente muestras del hemisferio izquierdo (muestreadas en los seis cerebros), que reduce considerablemente los grados de libertad (DOF) y, por lo tanto, la significancia estad´ıstica en los an´alisis de asociaci´on entre los IDPs y la neurogen´etica [27]. Otra opci´on es a˜nadir las muestras del hemisferio derecho de los dos donantes aplicando enfoques de normalizaci´on para las muestras procedentes de ambos hemisferios [28]. •Existe un sesgo de autocorrelaci´on entre muestras cercanas – genes m´as cercanos espacialmente, tienen expresi´on m´as parecida – lo que introduce un sesgo adicional a los valores de correlaci´on con los IDPs. Este efecto se puede reducir promediando muestras que pertenecen a las mismas regiones del cerebro y eliminando los efectos de autocorrelaci´on [25, 29, 30]. Despu´es de vincular el IDP de inter´es con los datos de AHBA, el resultado obtenido es una lista de genes que tienen asociaci´on significativa. Esta lista se puede analizar en profundidad para obtener interpretabilidad de los mecanismos neurobiol´ogicos subyacentes de los IDPs. Por ejemplo, las estrategias basadas en hip´otesis previas se centrar´an en comprobar si la lista de asociaci´on obtenida incluye los genes de inter´es relacionados con la patolog´ıa estudiada. Alternativamente, las estrategias basadas en datos est´an basadas en utilizar distintas herramientas o enfoques bioinform´aticos, sin asumir a priori que ciertos genes pueden tener o no mayor participaci´on en la patolog´ıa de estudio. De forma totalmente dirigida por datos, podemos obtener un ranking de participaci´on de genes, sobre los cuales podemos utilizar t´ecnicas bioinform´aticas para, por ejemplo, anotar las vi funciones biol´ogicas en las que participa la lista, utilizando el an´alisis de enriquecimiento (enrichment) de Gene Ontology (GO) [31]. Los datos del AHBA se han utilizado ampliamente desde su publicaci´on en 2012, vinculando la expresi´on transcript´omica con los IDPs relacionados con la poblaci´on sana [28, 32, 33], la enfermedad de Alzheimer [34, 35], del Parkinson [36, 37], demencia frontotemporal [38, 39], otros trastornos neurodegenerativos [40, 41], epilepsia [42, 43], tumores cerebrales [44, 45], trastorno del espectro autista [46, 27], esquizofrenia [47, 48], trastorno de depresi´on mayor [49, 50] y otros trastornos neuropsiqui´atricos [51, 52]. En esta Tesis, se ha analizado la neurogen´etica subyacente a las caracter´ısticas del conectoma cerebral en individuos sanos, en pacientes con Distrofia Miot´onica Tipo I (DM1) y en individuos con Trastorno del Espectro Autista (TEA). En el cap´ıtulo 2, como cap´ıtulo metodol´ogico e introductorio para el resto de la tesis, investigamos las relaciones entre redes cerebrales a m´ultiples escalas, combinando redes estructurales y funcionales. Es importante destacar que a fecha de hoy no sabemos predecir con precisi´on la conectividad estructural a partir de la funcional, ni viceversa. En realidad, este problema es un desaf´ıo actual para la neurociencia moderna. Uno de los principales impedimentos es la falta de repositorios p´ublicos que integren redes estructurales y funcionales a diversas escalas, o resoluciones. Y es precisamente aqu´ı donde hemos querido aportar en este cap´ıtulo. En concreto, para abordar esta cuesti´on, en este cap´ıtulo hemos desarrollado nuevos an´alisis de datos, de c´odigo abierto que permiten examinar la correspondencia entre las conectividades estructurales y funcionales a diferentes escalas. Hemos desarrollado una estrategia a nivel de m´odulo, o comunidad, que mejora los enfoques a nivel de regi´on para el estudio de la correspondencia entre estructura y la funci´on cerebral. Adem´as, tambi´en hemos desarrollado nuevos recursos que hacen uso de m´etricas neurogen´eticas para el estudio de redes cerebrales a nivel de m´odulo, que pueden facilitar nuevas oportunidades para seguir avanzando en la investigaci´on de neurociencia de redes, particularmente en lo que respecta a los trastornos cerebrales. vii En el Cap´ıtulo 3, nuestro objetivo fue delimitar los perfiles neurogen´eticos de los patrones de degeneraci´on cerebral en la DM1, una enfermedad monog´enica rara producida por la mutaci´on de un ´unico gen (DMPK), y que en el Pa´ıs Vasco tiene una prevalencia aproximadamente tres veces mayor que en otras regiones del mundo [53]. Para ello, cruzamos mapas cerebrales de p´erdida de volumen (VL) y d´eficits neuropsicol´ogicos (ND) con los datos transcript´omicos del AHBA. Adem´as, para validar los resultados obtenidos bas´andose en datos, hemos hecho uso de estudios neuropatol´ogicos y de ARN en una peque˜na serie de muestras de cerebro con DM1. En el Cap´ıtulo 4, hemos analizado los mecanismos moleculares que sustentan los diferentes subtipos de TEA, obtenidos mediante t´ecnicas de clustering en conectomas funcionales de estos sujetos. El subtipado en TEA es de alt´ısimo inter´es, ya que es un trastorno del neurodesarrollo con una gran heterogeneidad entre los genes alterados, los s´ıntomas de los individuos con TEA y sus comportamientos. La estrategia seguida en este cap´ıtulo para caracterizar esta heterogeneidad ha sido agrupar perfiles de conectividad cerebral funcional a gran escala (entre regiones corticales y subcorticales, formado una red que cubre al cerebro completo), en una poblaci´on de 657 individuos con autismo. Adem´as, hemos utilizado los datos del AHBA para caracterizar el mecanismo molecular detr´as de cada subtipo mediante el an´alisis de enrichment del conjunto de genes que muestran una alta similitud espacial con los perfiles de alteraciones de la conectividad funcional entre cada subtipo (alteraciones medidas con respecto a un grupo de participantes de control con neurodesarrollo normal). Finalmente, y como conclusi´on de esta tesis, he desarrollado diferentes metodolog´ıas y estrategias que se pueden aplicar para procesar datos cerebrales de resonancia magn´etica, con el objetivo de construir conectomas estructurales y funcionales a macroescala, y relacionar estos conectomas con datos transcript´omicos espaciales de expresi´on de genes y con muy alta resoluci´on. La mayor´ıa de los datos utilizados en esta Tesis son p´ublicos, al igual que las herramientas de las que he hecho uso. Combinando las diferentes herramientas, he podido xiv Contents 2.2.5 The optimal brain partition based on crossmodularity maximization . . . . . . . . . . . . 27 2.2.6 Transcriptomic data at module-level to assess brain-related disorders . . . . . . . . . . . . . 27 2.3 Results.......................... 28 2.3.1 Optimal γ-fused structure-function modular organization.................... 30 2.3.2 Multi-scale γ-fused structure-function modular organization................... 31 2.3.3 Anatomical and functional description of a given brain partition . . . . . . . . . . . . . . . . . 34 2.3.4 Neurobiological relevance of a given brain partition in brain-related disorders by making use of neurogenetics data . . . . . . . . . . . . . . 36 2.4 Discussion........................ 37 2.5 Data and code Availability . . . . . . . . . . . . . . . 41 3 Transcriptional signatures of synaptic vesicle genes define myotonic dystrophy type I neurodegeneration 43 3.1 Introduction....................... 43 3.2 Methods......................... 45 3.2.1 Participants, two cohorts . . . . . . . . . . . . 45 3.2.2 Demographic, clinical and neuropsychological variables..................... 47 3.2.3 MRI acquisition and pre-processing . . . . . . 48 3.2.4 Imaging statistical analysis . . . . . . . . . . . 50 3.2.5 Transcriptomics brain maps . . . . . . . . . . 50 3.2.6 Association between VL and transcriptomics . 52 3.2.7 Association between NDs and transcriptomics 52 3.2.8 DM1 relevant genes and GO . . . . . . . . . . 53 3.2.9 Neuropathological analyses . . . . . . . . . . . 54 3.3 Results.......................... 56 3.4 Discussion........................ 68 Contents xv 4 The Neurogenetics of Functional Connectivity Alterations in Autism: Insights From Subtyping in 657 Individuals 73 4.1 Introduction....................... 74 4.2 Methods......................... 76 4.2.1 Participants . . . . . . . . . . . . . . . . . . . 76 4.2.2 Functional Connectivity Matrices . . . . . . . 78 4.2.3 Data Harmonization . . . . . . . . . . . . . . 78 4.2.4 ASD Subtyping via Consensus Clustering . . . 79 4.2.5 Statistical Differences in Brain Morphology and Behavior Between ASD Subtypes . . . . . . . 80 4.2.6 Functional connectivity alterations of ASD subtypes with respect to TDC . . . . . . . . . . . 80 4.2.7 Transcriptomics . . . . . . . . . . . . . . . . . 81 4.2.8 The use of spatial autoregressive models for the association of transcriptomics with subtypes . 82 4.2.9 Gene Set Enrichment Analysis and Protein Interaction Analysis . . . . . . . . . . . . . . . . 83 4.3 Results.......................... 84 4.4 Discussion........................ 95 4.5 Data and code availability . . . . . . . . . . . . . . . 99 5 Conclusions 101 List of Figures 1.1 Sagittal view of a brain drawing with phrenology-based description labels . . . . . . . . . . . . . . . . . . . . 2 1.2 Technologies to acquire brain data at different spatiotemporalscales ..................... 5 1.3 Functional and structural connectomes . . . . . . . . 6 1.4 State-of-the-art rs-fMRI preprocessing pipeline . . . . 11 1.5 The Principle of DTI and Contrast Generation . . . . 13 1.6 FC and SC computation explained . . . . . . . . . . 15 1.7 Standard pipeline to preprocess transcriptomic expression data from the AHBA . . . . . . . . . . . . . . . 16 1.8 Main methods applied to study the spatial association of Image Derived Phenotypes and genetic information 18 2.1 Methodological sketch and pipeline . . . . . . . . . . 29 2.2 Selection of best initial parcellation atlas and optimal value of γparameter .................. 31 2.3 Graph-node strength is modulated by γ, the amount of interplay between structural connectivity (SC) and functional connectivity (FC). . . . . . . . . . . . . . . 32 2.4 Dendrogram measures and module analysis in the the optimal brain parcellation . . . . . . . . . . . . . . . 33 2.5 Statistical dependencies between different multi-scale metrics.......................... 35 2.6 Brain localization, Functional description and Anatomical description of the optimal brain partition . . 36 xvii xviii List of Figures 2.7 Neurogenetic module-level data for disease-related transcriptomic characterization of the optimal brain partition 38 3.1 Methodological scheme for the association between transcriptomics and atrophy in DM1, measured as brain VL 58 3.2 Data-driven strategy to determine the association between the transcriptome and VL in DM1 . . . . . . . . . . . 60 3.3 Functional description of the genes with the highest association with VL . . . . . . . . . . . . . . . . . . . 61 3.4 Data-driven strategy to define the association between the transcriptome and attention co-activation maps in DM1........................... 62 3.5 Neuropathological analyses of brain samples revealed a high heterogeneous Tau-pathology in DM1, different totheoneinAD .................... 65 3.6 RNA expression in hippocampal tissues from DM1 patients revealed protein dysfunction of synaptic vesicle processes in DM1 . . . . . . . . . . . . . . . . . . . . 67 4.1 General workflow . . . . . . . . . . . . . . . . . . . . 86 4.2 The two major ASD subtypes were subdivided hierarchically into smaller subtypes, but the significant enrichment found only occurred for the two subtypes at the highest dendrogram level. . . . . . . . . . . . . . 87 4.3 Two major stable ASD subtypes, one with hypoconnectivity and the other with hyperconnectivity . . . . 89 4.4 Association between transcriptomics and connectivity patterns for each autism spectrum disorder subtype . 91 4.5 Excitation/inhibition imbalance enrichment for only one class of participants with autism spectrum disorder (subtype2) ....................... 92 4.6 Replicability of the results including samples from the wholebrain ....................... 94 List of Tables 3.1 Demographic, clinical and neuropsychological variables 46 3.2 Age, gender and years of education between DM1 patientsandHCs ..................... 47 3.3 List of relevant genes for the neurobiological aspects of DM1 obtained after reviewing previous studies . . . . 55 3.4 Neuropathological details of cases studied . . . . . . . 64 4.1 Main data characteristics for each Institution participating in our study . . . . . . . . . . . . . . . . . . . . 77 4.2 Behavioral Characterization of ASD Subtypes . . . . 88 xix Chapter 1 Introduction 1.1 The pursuit of brain functions The brain is the most enigmatic body organ of any living animal. In the last centuries, researchers of different disciplines have tried to explain how a bunch of neurons could work together and how consciousness emerges from that tangle, developing a wide variety of theories. Not too recent, but not too far in time theories such as Phrenology [54], which nowadays is considered as pseudoscience, were the way to explain brain functions from brain shape. Dr. Gall (1758-1826) and collaborators built a full brain map which assigned brain functions to cranial areas [55], guiding the clinical diagnosis by cranial palpation (Figure 1.1). These ideas settled the basis of the Brain Localization Theory (BLT), which was defined and developed due to lesion-driving function studies. The major example of the BLT development was the discovery of the language production area (Broca’s area) by Dr. Broca (1824-1880) [56, 57]. BLT was based on the principle that isolated brain regions, produced specific functions. But, it is possible that a single brain region could codify a simple or complex behavior? In that epoch, other researchers started to question the localization theory. Dr. Flourens (1794-1867) supported that different regions interacted between them, building functional systems [58]. All these works were based on the observation of func1 2Chapter 1. Introduction tional symptoms produced after brain lesions, in humans and other animals. Figure 1.1: Sagittal view of a brain drawing with phrenology-based description labels. The image is extracted from the book [55]. Labels meaning: a. Olivary bodies; b. Annular protuberance; c. Entrance of the anterior pyramids under the annular protuberance; 1. Decussation of the anterior pyramids; 48-49. Situation of the organ of philoprogenitiveness; 49-50. Situation of the organ of inhabitiveness; 50-51. Situation of the organ of self-esteem; 51-52. Situation of the organ of firmness; 52-53. Situation of the organ of veneration; 53-54. Situation of the organ of benevolence; 54-55. Situation of the organ of comparison; 55-56. Situation of the organ of eventuality; I. Organ of Destructiveness; II. Organ of amativeness; III. Organ of Philoprogenitiveness; IV. Organ of adhesiveness; V. Organ of inhabitiveness; VI. Organ of combativeness; VII. Organ of secretiveness; VIII. Organ of acquisitiveness; IX. Organ of constructiveness; X. Organ of cautiousness; XI. Organ of approbativeness; XII. Organ of self-esteem; XIII. Organ of benevolence; XIV. Organ of reverence; XV. Organ of firmness; XVI. Organ of conscientiousness; XVII. Organ of hope; XVIII. Organ of marvellousness; XX. Organ of mirthfulness; XXI. Organ of imitation; XXVI. Organ of coloring; XXVII. Organ of locality; XXVIII. Organ of order; XXIX. Organ of calculation; XXX. Organ of eventuality; XXXII. Organ of tune. During 19th and 20th century, BLT theory was extensively used in clinical practice, and very reputed clinicians and researchers, such as Dr. Karl Wernicke (1848-1905), Dr. John Hughlings Jackson (1835- The pursuit of brain functions 3 1911), or Dr, Wilder Penfield (1891–1976), were mapping the brain areas with their functions. Patients as HM (1926-2008), who had a full memory consolidation impairment due to surgical rejection of the hippocampus, helped the establishment of BLT theory. Despite considerable advances, the brain remains a mystery, but with the actual technology, researchers advanced more quickly in the solution to this unknown and are more certain that brain functions are related to networks, but not to specific brain areas. So, for example, Broca’s area is a part (maybe the most important) of the language network. Recently, a new theory developed by Dr. Michael D. Fox and collaborators, named Lesion Network Mapping (LNM), has come up studying also brain lesions and the symptoms which produced [59]. LNM is based on the study of which brain networks are connected to the lesioned area in a large normative dataset, thus the patient with that lesion would have potentially affected or disconnected those networks. Following the same process in a group of patients with the same neurologic or psychiatric symptoms, but with different lesioned brain regions, would define the common network that has been disconnected in all the patients, which would be the one producing the symptom. LNM has been proved in numerous quantity of neurologic, motor and non-motor, behavioural, and psychiatric symptoms (see [60] and references therein). Furthermore, it has been tested for its potential use for deep brain stimulation [61]. Following this, researchers such as Dr. Maurizio Corbetta, and also our group, have also applied LNM to mapping cognitive functions to brain networks through brain lesions [62, 63, 64]. LNM is a promising technique that increases the certainty of cognitive functions are the result of network operations inside the brain. For that reason, in this Thesis, the starting point is the understanding of the brain as a complex network. 10 Chapter 1. Introduction •Distortion correction: the sequence used to acquire fMRI images is gradient-echo echoplanar imaging (EPI) and produces artefacts where air and tissue meet, e.g. the sinuses (producing artefacts in the orbitofrontal area) and the ears (producing artefacts in the temporal poles) [68]. To correct these effects, it is necessary to acquire an additional sequence called field map which characterizes the MRI scanner field. •Removal of unwanted signals and trends: despite the BOLD signal being related to neural activity produced in the GM, there are also signals in WM and CSF. These signals are usually considered as nuisances to remove [71] together with linear and quadratic trends, using a generalized linear model (GLM). •Temporal filtering: the rs-fMRI BOLD signals have their physiological characteristics at frequencies between 0.01-0.08Hz [72], so a bandpass filter will be required in that range. •Registration to a common template: this step is not mandatory, but it is needed for group analyses, registering the individual images to a common template and space. The common template could be the one built with the individual anatomical images or a normative anatomical template, e.g. MNI152. •Spatial smoothing: a spatial smoothing, usually a 3D Gaussian filter, is used to reduce the noise of this sequence and to increase the statistical power in group comparison studies, as smooth reduces individual variabilities. The filter must be smaller than the activation signal to be detected, and usually is 2 times the size of the voxel [68]. All these steps must be performed sequentially and ordered. In figure 1.4 a state-of-the-art pipeline shows an example of the different steps and the order that should be followed: There exists complete rs-fMRI preprocessing pipelines commonly used by the neuroimaging community such as fmriprep [73] https:// MRI data preprocessing 11 Figure 1.4: State-of-the-art rs-fMRI preprocessing pipeline. fmriprep.org/en/stable/,CONN [74] https://web.conn-toolbox.org/, or C-PAC [75] https://github.com/FCP-INDI/C-PAC. In this Thesis, I developed and freely shared our own fMRI preprocessing pipeline which adjusted better to our needs and lab facilities as the mentioned ones. It can be found at: https://github.com/compneurobilbao/compneuro-fmriproc 1.5.3 DTI preprocessing In this sequence, the preprocessing not only includes the artefact removal but also the fitting of the diffusion tensor and the fibre tracking. The main artefacts are produced by the distortions of the magnetic field, eddy currents, and head movement. •Eddy current correction: the strong magnetic field gradients applied to acquire the tensor imaging may lead to eddy currents which produce stretches, shear, false fibre tracking, enhaced background, image intensity loss, and image blurring. The main tool to correct these distortions and also for head motion is the eddy tool included in FSL. 12 Chapter 1. Introduction •Distortion correction: similar to the fMRI sequence (both fMRI and DTI are called Echo-Planar imaging (EPI) sequences) is affected by similar types of artefacts. To correct them, one can use field map images and also it is necessary to acquire two reference images with contrary orientations. FSL software also includes the tool topup for correcting them. •Diffusion tensor fitting: for each voxel the movement of the water molecules can be described using a tensor (Figure 1.5 panel B). A local fitting of this tensor can be computed on the corrected image. Different measures such as Fractional Anisotropy (FA, Figure 1.5 panel D), mean diffusivity (MD), Radial Diffusivity (RD), and Axial Diffusivity (AD), can be calculated using the eigenvectors and eigenvalues obtained. •Fibre tracking: knowing the direction and orientation of the tensor in each voxel (Figure 1.5, panel E) we can follow a pathway that connects two brain regions in a process called tractography. There exist two kinds of tractography algorithms: deterministic and probabilistic. The deterministic tractography process [77] starts with fixing a seed in a selected location and tracking the individual fibres by connecting the voxels with the adjacent ones toward the fibre direction. That fibre direction is determined by the leading local eigenvector of the diffusion tensor (Figure 1.5 panel C). This is an iterative process until a termination criterion is reached, such as minimum FA, fibre length, fibre curvature, or termination point arriving. In probabilistic tractography there is also a random process in the propagation direction. The way the fibre is propagated can vary depending on the algorithm used, the most popular are: probtrackx [78], iFOD1 [79], and iFOD2 [80]. Common software for DTI preprocessing are FSL,MRtrix3 (https: //mrtrix.readthedocs.io/en/dev/index.html), and DIPY (https://dipy.org/). There also exists a standard preprocessing pipeline DWIprep (https://github. com/GalKepler/dwiprep) which uses FSL and MRtrix3 for doing the steps. Brain networks 13 Figure 1.5: The Principle of DTI and Contrast Generation. This figure is extracted from [76]. ”From diffusion measurements along multiple axes (A), the shape and the orientation of a “diffusion ellipsoid” is estimated (B). This ellipsoid represents what an ink stain would be if ink were dropped within the pixel. An anisotropy map (D) can be created from the shape, in which dark regions are isotropic (spherical) and bright regions are anisotropic (elongated). From the estimated ellipsoid (B), the orientation of the longest axis can be found (C), which is assumed to represent the local fibre orientation. This orientation information is converted to a color (F) at each pixel. By combining the intensity of the anisotropy map (D) and color (F), a color-coded orientation map is created (E).” In this Thesis, I developed and freely shared our own DTI preprocessing pipeline which adjusted better to our needs and lab facilities. It can be found at: https://github.com/compneurobilbao/compneuro-dwiproc 1.6 Brain networks The human connectome is modeled as a network of networks, which are complex systems composed of similar parts, named nodes, interconnected between them by edges, allowing the information flow in the whole system. In figure 1.3, panels B and C, nodes are the colored 14 Chapter 1. Introduction dots that represent the brain regions of panel A, and edges are the links in blue-green colors connecting the nodes. The mathematical language that describes and quantifies networks is called graph theory. Selecting the nodes of the brain graph is a critical step for the analyses one wants to perform. One alternative could be to select all the voxels or small patches containing several voxels [2], but it could be computationally expensive. Alternatively, it is common the use of brain parcellations or atlases that can be either functional [81, 82], structural [83, 84], or both structural and functional [85, 86, 87]. Finally, it is also possible to apply a data-driven strategy to cluster brain micro-regions based on the homogeneity of the data or certain characteristics of the networks [2]. Once the brain-nodes are selected, the next step is to define the connectivity between them. Two main classes of connectivity exist: •Functional connectivity (FC): describes the activation profiles and dependencies among distinct, and may be distant, neural populations or brain regions [88]. The stronger the statistical dependency between both areas, the higher will be the FC value of the edge. That dependency could be measured by different methods, such as Pearson’s Correlation, partial correlation, mutual information, etc (Figure 1.6, panel A). •Structural connectivity (SC): describes the anatomical connections between brain regions through axon bundles or WM tracts [2]. Usually the measure utilized is the number of fibres connecting two regions (Figure 1.6, panel B). Another way to estimate connectivity edges is the Effective connectivity (EC) that describes causal interactions underlying the information flow [2], but this class has not been used in this Thesis. Neurogenetics and connectomics 15 Figure 1.6: FC and SC computation explained.A: two time-series (here Thalamus and Precentral) are correlated between them using Pearson’s correlation, the result is stored in the FC matrix. The rest of the FC values are from the correlations between all the regions (in pairs) included in Desikan’s parcellation. B: The number of fibres (#f ) between Thalamus and Precentral is stored in the SC matrix. As in FC, the rest of the values are from the #f between all the regions included in Desikan’s parcellation. Notice that, in the SC heatmap #f is in logscale 1.7 Neurogenetics and connectomics In the last years, the release of brain-wide and single-cell transcription datasets has enabled researchers to link connectome characteristics with genetic data, for a review see [11]. This strategy allows for identifying specific genes and cell types implicated in the development 16 Chapter 1. Introduction and progress of brain-related disorders [11]. Techniques such as Genome Wide Association Studies (GWAS) or Transcriptome Wide Association Studies (TWAS) have been broadly used to link Image Derived Phenotypes (IDPs) with Single Nucleotide Polymorphisms (SNPs), not only in neurodegenerative and neuropsychiatric variants but also in healthy populations [12, 13, 14, 15, 16, 17, 18]. Both approaches require a large number of participants and strong hypotheses on the genetic traits to be studied. Furthermore, the genetic information is extracted from blood or saliva, so it cannot give information about how the genetic variant impacts the brain tissue [11]. An alternative approach that solves the aforementioned issues of GWAS or TWAS is the use of brain-transcriptomic atlases [19, 20, 21, 22, 23, 24], which provide the expression of thousands of genes covering different brain locations. Among the available datasets, the most used is the Allen Human Brain Atlas (AHBA), which includes data from six donors without any known pathology and covers the full brain [19]. For using this atlas in MRI studies, we consider as normative data the transcriptomic profiles extracted from the brains, pulling all the samples from each donor in a unique atlas. After several preprocessing steps [25], needed to clean and standardize the data, it is possible to correlate the transcriptomic expression of the available genes to the IDPs that one wants to study. Figure 1.7: Standard pipeline to preprocess transcriptomic expression data from the AHBA. The figure is extracted from [25] Neurogenetics and connectomics 17 In this Thesis, I used the AHBA for the neurogenetic analyses in chapters 2, 3, 4. There are some issues to take into account when using this dataset: •Several genes in the database have different expression patterns across the donors. In [26], the same authors of the original dataset defined a measure accounting for that issue. The Differential Stability (DS) tells the expression reproducibility in the six donors of each gene. DS can be used for selecting the most reproducible probe annotated to a specific gene or for filtering low-reproducible genes for the analyses. •Only two of the six brains were sampled in both hemispheres. One option is to only use samples from the left hemisphere (sampled in the six brains), but it will reduce the Degrees Of Freedom (DOFs) of the association analyses between IDPs and neurogenetics, thus reducing the significance [27]. However, it is possible to add the right hemisphere samples of the two donors applying sample normalization approaches [28]. •There exists an autocorrelation bias between nearby samples, which inflates the correlation values with the IDPs. This effect can be reduced by averaging samples belonging to the same brain regions and removing the autocorrelation effects [25, 29, 30]. After linking the IDP of interest with the AHBA data, the outcome will be the list of significant associated genes. That list can be further analyzed to gain interpretability of IDP underlying neurobiological mechanisms. For example, hypothesis-driven strategies will focus on checking if genes of interest related to the pathology studied are included in the list. By contrast, data-driven strategies can be performed using distinct bio-informatics tools or approaches. For example, it is possible to annotate the biological functions that the list is participating in, using the Gene Ontology (GO) enrichment analysis [31]. Figure 1.8 shows a summary of how to link AHBA data to a IDP. 18 Chapter 1. Introduction Figure 1.8: Main methods applied to study the spatial association of Image Derived Phenotypes and genetic information. The figure is extracted from [11]. ”Transcriptome data is obtained from postmortem brain tissue from different brain regions across the cortex (A, AHBA example). The transcriptome matrix (brain samples x genes) contains the expression of all the measured genes for each sample of the brain. Computing the similarity of the spatial expression of each pair of genes in the brain the gene-coexpression matrix is obtained, which identifies groups of genes with similar spatial patterns. Single cell transcriptome data allow to classify each gene into a cell type (B). This information allows us to compute the distribution of the spatial similarity of the IDP with the genes pertaining to each cell type. The spatial similarity of the whole transcriptome can be computed with the IDPs to obtain a distribution of similarity values (C has been adapted from [28], and C-F shows the significant fold enrichment on Positive Regulation of Synapse Assembly as a case example). The top ranked genes (most similar to the IDPs) can be used in a gene-set enrichment analysis to find significantly associated functional annotations with the genes (C). The functional annotations or genes of interest can be further explored to obtain the correlation value with IDP, project their values onto the brain surface and compute the temporal trajectory along the lifespan (D-E). This top ranked list can also be used to find relationships with other genes and find the importance of our candidate genes in the obtained interactomic gene network (F). The position of each node in the network is plotted as a function of the hubness of each gene (or degree centrality, the number of connections to other genes of the network) in the bars figure in F.” Thesis overview 19 AHBA data has been broadly used since its publication in 2012, linking the transcriptomic expression with IDPs related to healthy population [28, 32, 33], Alzheimer’s disease [34, 35], Parkinson’s disease [36, 37], Fronto-temporal Dementia [38, 39], other neurodegenerative disorders [40, 41], Epilepsy [42, 43], brain tumours [44, 45], Autism Spectrum Disorder [46, 27], Schizophrenia [47, 48], Major Depression Disorder [49, 50], and other neuropsychiatric disorders [51, 52]. 1.8 Thesis overview In this Thesis, I have analysed the neurogenetics underlying brain network characteristics in healthy individuals, in patients with Myotonic Dystrophy Type I (DM1), and in patients with Autism Spectrum Disorder (ASD). Chapter 2 contains a structural-functional multi-scale approach for making brain parcellations at different network resolutions. The approach also includes a framework to obtain the gene expression of each ROI of the parcellation using the AHBA dataset. Furthermore, for the optimal brain parcellation, it is detailed the genetic relevance of each module in a set of several brain-related disorders. In Chapter 3, the objective was to delineate the neurogenetic profiles of brain degeneration patterns in DM1, a rare monogenic disorder produced by a mutation of a single gene (DMPK ), and that in the Basque Country has a prevalence about three times higher than other regions in the world [53]. To do that, we intersect brain maps of volume loss (VL) and neuropsychological deficits (NDs) with the transcriptomic data from the AHBA. Furthermore, to validate the results obtained in a data-driven way, neuropathological and RNA analyses were performed in a small series of DM1 brain samples. In Chapter 4, we analysed the molecular mechanisms behind two subtypes of ASD, which is a neurodevelopmental disorder with a large heterogeneity among the genes altered, patient’s symptoms, and behaviours. We applied a subtyping approach based on consensus clustering of functional brain connectivity patterns to a population of 657 26 Chapter 2. Linking structure, function and neuro-genetics across all equivalent links in the individual matrices. Next, and to match the sparse nature of SCp, a threshold was applied to FCp, resulting in comparable link density. Subsequently, we binarized the two matrices and fused them into a single matrix as follows: γSFC ≡γ×FCp+ (1 −γ)×SCp, where γis a real number between 0 and 1. This strategy allows us to continuously parameterize the level of interplay between SCpand FCp, enabling the recovery of each connectivity class when γ= 0 or γ= 1, while obtaining hybrid connectivities for intermediate values. 2.2.4 Multi-scale hierarchical representation at module-level For each of the 9 distinct iPAs, a hierarchical agglomerative clustering (HAC) was employed to extract different nested modules, allowing to build different network resolutions at different scales (each one determined by a cutting tree number of modules M). More specifically, HAC was performed using a weighted method on the connectivity patterns obtained from the matrix γSFC, here defined as the correlation distance between pairs of micro-regions in the γSFC matrix. Within the context of tree or dendrogram partitions, different multi-scale metrics can be established. These metrics can be defined either at the module-level or at a more granular level (with the highest spatial resolution determined by the micro-region level). In this particular study, we have defined the module size (MS), representing the count of micro-regions encompassed within a module. Additionally, we have introduced the multi-scale index (MSI), which quantifies the number of levels within the tree structure where a module remains intact without further division. In relation to micro-regions, we have selected the height (H) metric, which denotes the specific level at which a given micro-region separates from its parent module. Materials and Methods 27 2.2.5 The optimal brain partition based on crossmodularity maximization After obtaining a hierarchical partitioning of the connectivity matrix, clustering micro-regions into distinct modules M, the determination of the optimal level for cutting the tree depends on the specific metric being optimized. In this study, following the methodology introduced in [86], the metric chosen for maximization is the cross-modularity χ, defined as: χ≡(QF×QS×TF S)1 3 This metric simultaneously accounts for three different qualities: the modularity of the functional matrix (QF), the modularity of the structural matrix (QS), and their similarity (TF S ). The latter is defined as the DICE similarity between a functional module and a structural module, and averaging across all modules in the given partition. By varying M along the tree, we can calculate χfor each configuration M and select the optimal partition M∗where χis maximized. Additionally, unlike the approach in [86], we here have introduced a second parameter γ∗in the maximization of χ, which controls the amount of structure-function interplay. 2.2.6 Transcriptomic data at module-level to assess brain-related disorders In addition to multi-scale brain partitions and structure-function analysis, we processed the transcriptomic open data from the Allen Human Brain Atlas (AHBA) [19] by using the abagen tool developed in [129]. In particular, abagen allows us the generation of brain transcriptome module-level values for a specific parcellation, that in our case, we chose the optimal brain partition. Moreover, we examined the transcriptomic expression at module-level in relation to genes associated with 32 brain disorders introduced in [133]. To do this, we calculated the mean transcriptomic expression of all genes related to each disorder within each module. Subsequently, we performed a 28 Chapter 2. Linking structure, function and neuro-genetics z-score analysis for each disorder to identify modules with low transcriptomic expression (z <-2) and modules with high transcriptomic expression (z >2). The average expression of the different genes were also grouped into 7 different disease cateogories: Psychiatric disorders, Substance abuse, Movement disorders, Neurodegenerative diseases, Tumor conditions, or Developmental disorders. 2.3 Results Structural and functional connectivity matrices at various resolutions were built making use of brain images from the open dataset “Max Planck Institut Leipzig Mind-Brain-Body Dataset” – LEMON [131], well-known for having high-quality multimodal acquisitions and preprocessed MRI sequences. Rather than using different subjects across lifespan, we specifically selected 136 young participants, aged between 20 and 30 years old, who had both the preprocessed rs-fMRI sequence and DWI sequence available. We next processed the raw images following standard neuroimaging pipelines to obtain structural connectivity (SC) and functional connectivity (FC) matrices (Figure 2.1A). Subsequently, we built population connectivity matrices by choosing, for each link in the matrix, the median value across all the equivalent links in the individual connectivity matrices. Finally, we modeled the amount of interplay between structure and function connectivities using a single fusion parameter, γ. When γequals 0 or 1, we have purely structural or functional connectivity, respectively. In intermediate situations, the amount of overlapping connectivity is modulated by γ. Furthermore, and similar to the methodology described in [86], for each value of the fusion parameter γ, we applied hierarchical agglomerative clustering to the resulting γ-fused structure-function matrix, γSFC (Figure 2.1B). Results 29 Figure 2.1: Methodological sketch and pipeline.A: Participants’ raw MRI data from N=136 healthy volunteers of the MPI-LEMON dataset were used. From preprocessed rs-fMRI images, we generated an initial brain parcellation using gray-matter masks and the open repository pyClusterROI. Subsequently, we employed our own publicly available code on GitHub to process the DWI images and extract the SC matrices. All post-processed data at the participant level, including the time series BOLD signal corresponding to the different iPAs, and the SC and FC matrices are available at https://zenodo.org/record/8158914.B: We calculated population SCpand FCpmatrices, and subsequently generated γ-fused matrices as described in the methods section. The γSFC matrices were used to construct dendrogram trees for different γ parameters. Additionally, we employed abagen to generate transcriptomic expression matrices for various tree-based parcellations. The transcriptomic matrices and dendrogram trees are also accessible on Figshare. The entire pipeline and project codes can be found on GitHub at https://github.com/compneurobilbao/bha2. 30 Chapter 2. Linking structure, function and neuro-genetics 2.3.1 Optimal γ-fused structure-function modular organization The different trees of the γSFC matrices describe distinct scales, enabling the construction of networks at different resolutions. The determination of the optimal partition by cutting the tree at M∗modules depends on the specific metric we seek to maximize. In order to showcase the versatility of our data with various metrics, we specifically concentrate on a metric we have previously defined and named cross-modularity [86]. Represented as χ, it is defined as the product of three quantities: the modularity of the functional matrix, the modularity of the structural matrix, and the average similarity between structural and functional modules. Therefore, we aimed to maximize χto select the optimal value of γ for a given initial parcellation atlas (iPA). Figure 2.2 shows box-plots of different values of the cross-modularity χacross different dendrogram levels and for different iPAs with different size, ie., varying the number of micro-regions ranging from 183 to 2165. For further analyses, we selected the iPA with 2165 micro-regions since it exhibited a higher mean value together with a lower variability value of χwithin a wide range of scales, particularly within the first 120 levels where the maximum χis located. Beyond this point, there was a decreasing trend. The optimal brain partition corresponds to the iPA that maximizes χby reducing the dendrogram to M∗modules and setting the fusion parameter to γ∗. In our case, the optimal value γ∗was determined to be 0.7, and this occurred at the level of 28 modules. However, two modules were considered invalid as they consisted of only one and two micro-regions, respectively, making it impractical to analyze the community structure within them. Henceforth, any subsequent analysis referring to the optimal brain partition will be based on the iPA that maximizes χby reducing the dendrogram to M∗= 26 modules and setting the fusion parameter to γ∗= 0.7. Results 31 Figure 2.2: Selection of best initial parcellation atlas and optimal value of γparameter.Left: Box-plots of the cross-modularity χacross different dendrogram levels and for different size of the initial parcellation atlas (iPA). The different values of χ, on which the boxplots are calculated, come from the different levels in the dendrogram (here we have varied from 2 to 120). As the number of ROIs for the initial parcellation increases, the cross-modularity χ also increases. Right: The optimal value of γ∗= 0.7 is also illustrated for the same range of dendrogram levels, ie. from 2 to 120. 2.3.2 Multi-scale γ-fused structure-function modular organization As an illustrative example, and to gain a deeper understanding of how the multi-scale organization is influenced by γ, we investigated the network strength of the γSFC matrix in two network representations. At the finest spatial resolution, corresponding to the lowest level in the hierarchical tree consisting of 2165 micro-regions (represented as brain maps in Figure 2.3A), and for each of the macro-regions, including the frontal lobe, parietal lobe, occipital lobe, temporal lobe, insula, and a collection of subcortical areas (Figure 2.3B). From both analyses, it is evident that increasing γ, transitioning from structure (γ= 0) to function (γ= 1), results in higher strengths shifting from subcortical areas to cortical regions. This pattern is observed consistently across all macro-regions, where the strength progressively increases with γ, peaking around γ≈0.7. 32 Chapter 2. Linking structure, function and neuro-genetics Figure 2.3: Graph-node strength is modulated by γ, the amount of interplay between structural connectivity (SC) and functional connectivity (FC). A: Brain maps of graph-node (normalized) strength. It is observed that for γ= 0 (pure SC), high strengths are predominantly localized in subcortical regions. However, as γvaries, there is a decrease in strength within these regions and an increase in cortical regions. B: Box plots for the distribution of graph-node strength across different macro-regions: Frontal lobe, Parietal lobe, Temporal lobe, Occipital lobe, a collection of Subcortical areas, and Insula. Notably, for each macro-region, the strength shows an increasing trend from γ= 0, reaching a peak at γ≈0.7, followed by a subsequent decrease. Beyond the two selected levels of resolution for calculating strength, a hierarchical tree organization enables obtaining different metrics at different module-levels. For instance, we first colored the M∗= 26 distinct modules of the optimal brain partition for ease of visualization (Figure 2.4, top panel). Some tree metrics we computed were the module size (MS), height (H), and multi-scale index (MSI). While H is defined for each micro-region, MS and MSI are defined at the module-level. Results 33 Figure 2.4: Dendrogram measures and module analysis in the the optimal brain parcellation. Each color in the dendrogram corresponds to a distinct module (that for the optimal brain partitions correspond to 26 different modules). The y-axis of the dendrogram illustrates different levels (L) and provides visual representation of a given module and their inferior (Linf) and superior (Lsup) levels. Different multi-scale metrics can be defined, eg., higher values of the Multi-scale Index (MSI) correspond to enhanced preservation of the module throughout multiple levels in the dendrogram, indicating heightened stability across the tree structure. The module size is also represented by the width of each module in the dendrogram. The lower inset shows a marked micro-region Height indicating to which extent a given microregion is detached from the overall tree structure. Lower values of Height represent stronger connectivity of the micro-region with the other micro-regions belonging to the same module. The figure also shows brain maps depicting the three aforementioned measures for each of the modules. We also defined the module height (MH), by averaging over all individual H values within the given module (Figure 2.4, full dendro- 34 Chapter 2. Linking structure, function and neuro-genetics gram and dendrogram-inset). We represented these metrics on brain plots, where we observed distinct patterns for each module (Figure 2.4, brain plots). We conducted further analyses to examine the correlations among these metrics and the intra-module strength, as a proxy for module segregation (Figure 2.5). From the different tree measures, both MS and MH showed a high correlation (r = 0.95, p <0.001) with module segregation, as well as between themselves (r = -0.84, p <0.001). Interestingly, the first module M1 was an outlier in the metric of MSI. Located at the border between the Precuneus, Isthmus Cingulate, and Posterior Cingulate cortices (Figure 2.6), M1 showed a remarkable resilience in remaining intact without splitting across multiple levels in the tree, possibly indicating a multi-scale functional role. 2.3.3 Anatomical and functional description of a given brain partition Our methodology for defining a given partition defined by specific values of M and γis generally data-driven. At some point, it may be of interest to characterize the anatomical and functional properties of a specific partition. As an illustrative example, we made use of the optimal partition with M∗= 26 modules and γ∗= 0.7 and determined the spatial locations of these 26 modules within the brain (Figure 2.6, brain-glasses). We next obtained their functional and anatomical characterization by measuring respectively the amount of overlap within known Resting State Networks (RSNs) [81] and regions from the Desikan-Killiany atlas [83], the latter consisting of 34 cortical regions (in right and left hemispheres), 8 subcortical regions segmented from Freesurfer, and the cerebellum (Figure 2.6, heatmaps). Several of our modules exhibited strong overlapping within the RSNs. For instance, module M6 encompassed the Somato-motor Network (SMN) while also integrated parts of the Dorsal and Ventral Attention Networks (DAN, VAN). Module M18 showed an overlap with a portion of the Default Mode Network (DMN), and module M14 did it with the Visual Network (VIS). Moreover, certain modules displayed Results 35 high overlapping with specific anatomical regions. For example, module M20 included the medial and lateral Orbitofrontal cortices, while module M26 integrated the Basal Ganglia and Thalamus. Figure 2.5: Statistical dependencies between different multi-scale metrics. Crosscorrelation plots between multi-scale index (MSI), intra-strength (as a proxy for module segregation), size and height. Different color-points represent different modules of the optimal brain parcellation with a total number of 26 modules. Principal-diagonal plots show probability density functions of the different metrics. Notice that higher correlations between module size, module height, and intra-strength indicate that heightened connectivity within modules (segregation) results in larger dendrogram modules and delayed micro-region splitting in the tree. The MSI reveals a distinct outlier (M1) encompassing several structures such as a part of the precuneus, isthmus cingulate, and posterior cingulate, located at the border between DMN, fronto-parietal, and dorsal attention networks. Chapter 3 Transcriptional signatures of synaptic vesicle genes define myotonic dystrophy type I neurodegeneration In previous chapters, I have described how Image Derived Phenotypes (IDPs) can be intersected with the neurogenetic data from the Allen Human Brain Atlas (AHBA) to study the neurogenetic basis of brain-related disorders. In this chapter, we applied these techniques to describe the genes associated to two IDPs, volume loss (VL) and neuropsychological deficits (NDs) in patients diagnosed with myotonic dystrophy type I (DM1), a monogenic (or Mendelian) disorder. 3.1 Introduction DM1 is a complex multisystem disease that affects skeletal muscles [134], heart [135], lungs [136], endocrine system [137], regulation of sleep cycles [138], and other aspects of brain activity [139]. Epidemiologically, DM1 is the most common adult-onset muscular dystrophy in humans, with a reported prevalence of 1/7,400 people worldwide 43 44 Chapter 3. Transcriptional signatures of DM1 neurodegeneration [140] and is about three times higher in Gipuzkoa [53], Northern Spain (where this study was performed). Neuroimaging studies have shown the brain damage in DM1 patients, including grey matter (GM) atrophy mainly affecting the frontal and parietal lobes [141] but also, in the hippocampus [142] and other subcortical structures [143]. White matter tract alterations (WM-TAs) have been widely reported in DM1, in both cross-sectional [144, 145, 146, 147, 148, 149, 150, 151, 152] and longitudinal studies [153]. These indicate widespread WM-TA throughout the whole brain, but that is more severe in the frontal, temporal and subcortical tracts. In relation to neuropsychological performance, executive function, attention and visuoconstruction have been shown significant association with WMTA [145, 154]. To the date, the neurogenetic profiles of such patterns of WM-TA or the ones corresponding to GM damage have not been yet determined. Using an approach that combines magnetic resonance imaging (MRI) and large-scale brain transcriptomics, we aimed here to assess to what extent the structural damage in the DM1 brain represents a neurogenetic signature. In contrast to other neurodegenerative diseases in which a large number of candidate genes are implicated, for example, about 700 genes in Alzheimer’s disease (AD) [155], DM1 is a monogenic disorder caused by a mutation in the gene encoding the myotonic dystrophy protein kinase (DMPK) [156]. However, although the disease is monogenic, its phenotype is mainly due to an abnormal activity of the ribonucleic acid (RNA)-binding protein muscleblind-like 1 and 2 genes (MBNL1,MBNL2) [157, 158] and CUGBP which regulates the expression of many other genes, such as the chloride channel 1 gene (CLCN1) that regulates chloride conductance during muscle development [159], the insulin receptor gene (INSR) [160], the bridging integrator 1 gene (BIN1) [161] or other genes directly related to the main symptoms of the condition [162]. The neuropsychological profile of DM1 is mainly associated with cognitive difficulties including visuospatial processing and executive functioning [163, 164]. Strikingly, the DMPK pathogenic genotype has also been associated with other genes that encode proteins Methods 45 implicated in brain neurodegeneration, majorly to tau deposits [165], but also amyloid beta (Aβ) [166] or alpha synuclein [167, 168]. Thus, it is suspected that the NDs and brain damage found in DM1 patients might present some similarities to those in other neurodegenerative diseases, although this issue has yet to be fully addressed. Therefore, despite the monogenic origin of DM1, the gene-to-gene interactome scales up to implicate multiple systems in the brain and body. To date, a precise association between the entire transcriptome and the brain neurodegeneration and cognitive deterioration in DM1 patients remains unexplored. Some studies have assessed the relationship between genetics and structural brain alterations in DM1, confirming that the number of pathogenic repeats of the cytosine-thymine-guanine (CTG) triplet in the DMPK gene (a parameter used to quantify the molecular severity of the disease) was associated with both more GM and white matter atrophy, and to the NDs of these patients [147, 169]. Here, we performed an intersection analysis of neuroimaging phenotypes and the AHBA of large-scale transcriptional human data [19], following a similar methodology to that used previously [32, 28, 34]. Our hypothesis was that identifying the genes whose expression coincided more closely with the brain damage found in DM1 patients, we might better understand the gene relationships associated with the brain damage that arises in this pathology. Similarly, we also assessed the relationship between gene transcription in specific anatomical regions and the brain maps of NDs in these patients. 3.2 Methods 3.2.1 Participants, two cohorts A total of N=95 subjects were recruited to a cross-sectional study, 35 of whom were DM1 patients who were treated at the Neurology Department of the Donostia University Hospital (Gipuzkoa, Spain), while 60 subjects participated as Healthy controls (HCs). All patients and HCs were recruited from the vicinity of the Donostia University 46 Chapter 3. Transcriptional signatures of DM1 neurodegeneration Hospital, and the two groups were matched for age, gender and education. The imaging data from the DM1 patients and HCs was acquired at two different Institutions. At one, 19 DM1 patients (mean age 53.3 years [SD ±8.1 years]; 9 males, 10 females) and 29 HC (52.2 [±8.1] years; 12 males, 17 females) were examined and at the second, 16 DM1 patients (48.8 [±7.7] years; 7 males, 9 females) and 31 HC (47.6 [±7.6] years; 14 males, 17 females). For the mean values, and the comparisons between groups and cohorts see Tables 3.1 and 3.2. First cohort Second cohort t p Effect size (Cohen) Demographic N19 16 Age, years 53.30 (8.09) 48.75 (7.72) 1.69 0.10 0.57 Males, n (%) 9 (47.37) 7 (43.75) 0.05* 0.83 -0.07 Education, years 16.37 (4.87) 13.75 (4.85) 1.59 0.12 0.54 Clinical N19 16 CTG expansion size 522.79 (448.44) 827.13 (433.08) -2.03 0.05 -0.69 MIRS 2.47 (0.96) 3.53 (0.83) -3.37 0.002 -1.17 Time between MRI and CTG measures, years 0.14 (0.14) 9.40 (7.23) -5.6 0.000 -1.93 Neuropsychological N18 16 Visuospatial -0.35 (1.21) -1.17 (1.34) 1.86 0.07 0.64 Verbal memory -0.04 (2.33) -0.74 (1.99) 0.94 0.36 0.32 Attention -2.50 (1.92) -2.62 (2.14) 0.19 0.85 0.06 Executive functioning -0.91 (2.08) -1.84 (2.40) 1.21 0.23 0.42 Visual memory -0.34 (0.97) -1.01 (1.14) 1.83 0.08 0.64 Intelligence (IQ) 101.11 (11.51) 91.19 (13.61) 2.3 0.03 0.79 Table 3.1: Demographic, clinical and neuropsychological variables. Mean values (standard deviations in brackets) of the different variables separated by cohort. The neuropsychological variables coincide with the different composite-scores from the different cognitive domains. All the neuropsychological variables were calculated using normative data from a healthy Spanish population. All domains are represented in Z scores (Mean=0; SD=1) except IQ (Mean=100; SD=15). Smaller values indicate worse performance. Bold values indicate p < 0.05. *For gender differences, the χ2test was used. The DM1 patients were only included if they had molecular confirmation of their DM1 diagnosis, indicating the expansion from fifty to thousands of CTG repeats in the DMPK gene [156]. The diagnosis was obtained when patients were between 18 and 40 years old, and therefore categorized as adult-onset DM1, as proposed by the fourth Methods 47 DM1 HC t p Effect size (Cohen) 1st Cohort N19 29 Age, years 53.30 (8.09) 52.17 (8.05) 0.47 0.64 0.14 Males, n(%) 9 (47.37) 12 (41.38) 0.17* 0.68 -0.12 Education, years 16.37 (4.87) 14.79 (3.36) 1.33 0.19 0.39 2nd Cohort N16 31 Age, years 48.75 (7.72) 47.55 (7.54) 0.51 0.61 0.16 Males, n(%) 7 (43.75) 14 (45.16) 0.01* 0.93 -0.03 Education (years) 13.75 (4.85) - - - - Table 3.2: Age, gender and years of education between DM1 patients and HCs. Mean values of different variables are shown (standard deviations are given in brackets). *For gender differences, the Chi2 test was used. Outcome Measure for Myotonic Dystrophy Type 1 (OMMYD-4). Patients were excluded if at least one of the following criteria was met: history of a major psychiatric or somatic disorder in accordance with Diagnostic and Statistical Manual of Mental Disorders, fifth edition (DSM-V) criteria; acquired brain damage; alcohol or drug abuse; the presence of corporal paramagnetic devices like pacemakers or metal prosthesis that might compromise the MRI studies; and the presence of brain abnormalities that could affect the volumetric analysis. HCs satisfied the same exclusion criteria but the number of the CTG repeats in their DMPK gene ranged from 5 to 34 [156]. DM1 patients were recruited from the Neuromuscular Unit in the Neurology Department of the Hospital Universitario Donostia, while HCs were recruited from their healthy relatives and general population. 3.2.2 Demographic, clinical and neuropsychological variables The demographic variables of the subjects recorded were their age, gender and years of education. The clinical variables were the CTG expansion size and Muscular Impairment Rating Scale (MIRS) score 48 Chapter 3. Transcriptional signatures of DM1 neurodegeneration [170]. Neuropsychological variables corresponded to composite values from different cognitive domains obtained through a comprehensive neuropsychological evaluation performed by an experienced neuropsychologist who was blind to the patient’s clinical condition (CTG expansion size and MIRS results). The neuropsychological assessment included several subtests from the Wechsler Adult Intelligence Scale III (WAIS III) [171], including: Digit span, Vocabulary, Block design, Object assembly, Arithmetic, and Similarities. Other cognitive tests used were: Stroop test, California Computerized Assessment Package (CALCAP), Raven’s progressive matrices, Rey Auditory Verbal Learning Test (RAVLT) [172], Word Fluency [173, 174], Rey-Osterrieth Complex Figure test (ROCF) [175] and Benton’s Judgement of Line Orientation [176]. The patients’ raw scores were converted into standardized t-values based on the normative scores for the Spanish population in each test. Finally, the different neuropsychological scores were reduced into six different domains: visuospatial (Block design and ROCF copy), verbal memory (RAVLT immediate recall, RAVLT delayed recall, Total RAVLT), attention (Digit span, STROOP word, STROOP colour, Simple Reaction Time (RT), election RT, Sequential 1 RT, Sequential 2 RT), executive functioning (Total RAVEN, semantic fluency, phonemic fluency, STROOP colour-word, STROOP interference), visual memory (ROCF delayed recall), and intelligence (estimation of IQ was based on Vocabulary, Block design, Object assembly, Arithmetic, and Similarities subtests). 3.2.3 MRI acquisition and pre-processing For the first cohort, MRI was conducted on a 3 Tesla scanner (TrioTim, Siemens) using a high-resolution 3D sequence of magnetizationprepared rapid acquisition with gradient echo (MPRAGE) and applying the following parameters: Sagittal 3D T1 weighted acquisition, repetition time (TR) = 2,300 ms, echo time (TE) = 2.86 ms, inversion time = 900 ms, flip angle = 9°, matrix = 192 ×192 mm2, slice thickness = 1.25 mm, voxel dimensions = 1.25 ×1.25 ×1.25 mm3, number of signals averaged (NSA) = 1, slices = 144, no gap, total Methods 49 scan duration = 7 min and 22 s. Diffusion-weighted images were acquired using an echo planar imaging (EPI) sequence: 1.75 ×1.75 × 2 mm voxels; 77 axial slices; b value of 1,000 s/mm2; 64 direction diffusion-weighted and 1 baseline image; TR = 10,000 ms, TE = 92 ms; angle 90, acquisition matrix size = 122 ×122. For the second cohort, MRI was conducted on a 1.5 Tesla scanner (Achieva Nova, Philips), using a high-resolution volumetric turbo field echo (TFE) sequence with the following parameters: Sagittal 3D T1 weighted acquisition, TR = 7.2 ms, TE = 3.3 ms, inversion time = 0 ms, flip angle = 8°, matrix = 256 ×232 mm2, slice thickness = 1 mm, voxel dimensions = 1 ×1×1 mm3, NSA = 1, slices = 160, no gap, total scan duration = 5 min and 34 s. Diffusion-weighted images were acquired using a single shot spectral presaturation with inversion recovery (SPIR): 1.75 ×1.75 ×2 mm voxels; 60 axial slices; b value of 800 s/mm2; 32 direction diffusion-weighted and 1 baseline image; TR = 9,967 ms, TE = 66 ms; angle 90, acquisition matrix size = 128 ×128. To perform voxel-based GM morphometric comparisons between the subjects, DM1 and HCs, we performed voxel-based morphometry (VBM) following a procedure similar to that used previously [4, 177], an optimized VBM protocol [178] carried out with the FSL v6.01 software. First, skull removal was performed, followed by GM segmentation and registration to the MNI 152 standard space using non-linear registration [179]. The resulting images were averaged and flipped along the x-axis to create a left-right symmetric, study-specific GM template. Second, all native GM images were non-linearly registered to this study-specific template and ‘modulated’ to correct for local expansion (or contraction) due to the non-linear component of the spatial transformation. The modulated GM images were then smoothed with an isotropic Gaussian kernel at sigma = 3, and finally, the partial GM volume estimates normalized to the subject’s head size were compared. WM-TA were assessed with tract-based spatial statistics (TBSS) from FSL [180] using fractional anisotropy (FA) images. First, all individual FA images were normalized to a common template using 50 Chapter 3. Transcriptional signatures of DM1 neurodegeneration a non-linear transformation. Next, the mean image across all subjects was computed and skeletonized to get the mean FA skeletons, which represent regions with high confidence bundles common to all subjects, thus removing some of the individual subject-specific tractbased heterogeneities. FA images for each subject were then projected onto the mean skeleton. 3.2.4 Imaging statistical analysis For the VBM analyses, a generalized linear model was fitted for each voxel and image using the FSL software, controlling for age and head size with two different contrasts: DM1 <HC and DM1 >HC. All the results were obtained with two-tailed tests, correcting for multiple comparisons using the Monte Carlo simulation cluster-wise correction implemented in the AFNI v19.3.00 software, and using 10,000 iterations to estimate the probability of false positive clusters with a p value <0.05. Each cohort was analysed separately and in combination, as explained below, although statistical comparisons between the two cohorts were not performed. For the TBSS analyses, group comparison was performed using the randomise tool in FSL, a non-parametric permutation test for finding significant statistical differences between groups at the voxel level. For multiple comparison correction, we used thresholdfree cluster enhancement (TFCE)[181] with a number of iterations of 5,000 and family-wise error (FWE) corrected p= 0.05, thus ensuring that the chance of false positives is no more than 5%, or equivalently, ensuring 95% confidence of no false positives. Group comparison was performed using two different contrasts, DM1 >HC and HC >DM1. 3.2.5 Transcriptomics brain maps To build brain maps of transcription, we took advantage of the publicly available data from the AHBA (http://human.brain-map.org/) [19]. The dataset consisted of MRI images and a total of 58,692 microarraybased transcription profiles of about 20,945 genes sampled over 3,702 different regions across the brains of six humans. Of the total 3,702 Methods 51 sampling sites, 2,728 were located in cortical and subcortical GM, 368 in the cerebellum, 586 in the brain stem, and 15 in white matter. Our analyses here are restricted to the 2,728 sites covering the cortical and subcortical GM. To pool all the transcription data into a single brain template, we followed a similar procedure to that employed elsewhere [25]. First, to re-annotate the probes to genes we made use of the re-annotator toolkit [182]. Second, we removed those probes with insufficient signal by looking at the sampling proportion (SP), which was calculated for each brain as the ratio between the samples with a signal greater than the background noise divided by the total number of samples. Probes with a SP lower than 70% in any of the six brains were removed from the analysis, thereby ensuring sufficient sampling power in all the brains. After that, we chose the value of the probe for each gene with the maximum differential stability (DS), accounting for the reproducibility of gene expression across brain regions and individuals, and calculated using spatial correlations similar to those employed previously [26]. For this the automated anatomical labelling (AAL) atlas was used [84] from which the cerebellum was excluded, resulting in 90 different anatomical regions 1. Finally, to remove the intersubject differences, the transcription values for each gene and brain were transformed into Zscores and pooled together from the six different brains, obtaining a single map using the MNI coordinates provided in the dataset. Finally, to eliminate the spatial dependencies of the transcription values at the sampling sites (that is, to correct for the fact that nearest sites have more correlated transcription), we finally obtained a single transcription value for each region in the AAL atlas by calculating the median of all the values belonging to the given region. 1AAL regions were eroded with a Gaussian kernel with a full width at half maximum (FWHM) equal to 2 mm, thereby eliminating false-positive sampling sites, i.e.: those that do not belong to the region of interest but to one in the neighborhood. 58 Chapter 3. Transcriptional signatures of DM1 neurodegeneration Figure 3.1: Methodological scheme for the association between transcriptomics and atrophy in DM1, measured as brain VL.A: Two cohorts of DM1 patients were recruited (orange and blue) and we obtained the brain maps of the VL for each, comparing the images with a group of HCs using VBM and TBSS, correcting for multiple comparisons. We aimed to characterize the association between VL and transcriptomics, assessing the similarity in the spatial patterns of VL across brain regions and the spatial patterns of gene transcription from the AHBA dataset, pre-processed following a pipeline that is summarized in six main steps (for further details, see Methods). After running the AHBA pipeline, about 14 K genes finally had transcription values used in the analysis from the 58-K probes originally available. The red regions in the brain correspond to the sites at which transcription was sampled. *Of the total 3,702 sampling sites, 2,728 were located in cortical and subcortical grey matter, 368 in the cerebellum, 586 in the brain stem, and 15 in white matter. Our analyses here are restricted to the 2,728 sites covering the cortical and subcortical grey matter. B: SP and DS for the 27 preselected hypothesis-driven genes included in the list of candidates relevant to DM1 (obtained by reviewing the literature). The CACNA1S,CLCN1,KL and SIX5 genes did not have a mean SP value above 70% and thus, they were excluded from further analyses. Of the remaining 23 genes, the maximum DS corresponded to SNCA and the minimum DS to DMPK (see arrows). By examining the spatial similarity in the transcription values, the remaining 23 candidate genes were clustered into three groups. The blue one formed by DMD,SNCA and MAPT played a major role in the characterization of VL. Results 59 A list of the 27 most relevant hypothesis-driven genes in DM1 was drawn up (Figure 3.1, panel B; for details on references supporting the selection of each gene, see Table 3.3). Of these 27 preselected genes, four genes were discarded based on their SP and DS values (KL, SIX5, CLCN1, CACNA1S). These 4 genes had less than 70% SP (Figure 3.1, panel B), which implied insufficient transcription signal across all the sampling sites. Therefore, the final list of the most relevant hypothesis-driven genes contained: DMPK, HSPB2, INSR, CPEB4, ANK2, ARHGEF7, SOS1, PHKA1, MBNL1, KIF13A, APP, MAPT, SNCA, MBNL2, RIPK1, PRNP, TARDBP, MAP3 K7, BIN1, DMD, LDB3, TTN and CAPN3. When the pairwise gene-to-gene similarity in transcription was evaluated across the brain for these 23 genes, Silhouette maximization identified three clusters (in yellow, blue and green in Figure 3.1, panel B), indicative of functional similarities in the transcription signals among the 23 most relevant genes. We next applied a DDS that involved identifying the genes from the entire transcriptome with maximal association in the VL maps, resulting in a total of 370 N genes and 187 P genes from the first cohort and 441 N-genes and 161 NP-genes from the second one. The genes common to the two cohorts were those finally used in the analysis, a total of 251 N genes and 101 P genes (Figure 3.2, panel A). Interestingly, two of the genes in the list of the 23 most relevant hypothesis-driven genes also appeared in the list of N genes, SNCA and DMD, displaying a similar transcription pattern as the MAPT gene (blue cluster in Figure 3.1, panel B). The spatial correlation of SNCA and DMD transcription with the amount of VL in the different brain regions proved to be negative for the two genes and smaller in the two cohorts: <-0.52 for DMD and <-0.70 for SNCA (Figure 3.2, panel B). In addition to identifying the N and P genes, we also searched for the NC and PC genes (Figure 3.3, panel A) that represent hubs towards the N and P tails of the expression similarity matrix. We found 452 NC genes, 396 PC genes and 238 genes connecting both N and P tails. Remarkably, in the group of PC genes, we found LDB3,CAPN3 and HSPB2 that were in the list of hypothesisdriven genes, belonging to the same cluster of expression similarity 60 Chapter 3. Transcriptional signatures of DM1 neurodegeneration (green cluster, Figure 3.1, panel B). In addition, the preselected gene PHKA1 was also common to the groups of NC and PC genes. Figure 3.2: Data-driven strategy to determine the association between the transcriptome and VL in DM1.A: Histogram of the spatial-correlation values (measured as the Z-score) between VL and transcriptional activity for all the genes in both cohorts. For both cohorts the N genes (Z<-2) and P genes (Z>2) are coloured in blue and red, respectively. The final list of genes used for further analyses are those that are common to the two cohorts, consisting of 251 N genes and 101 P genes. From all the genes that provide a maximum association between VL and transcriptomics, only two genes were in the panel of preselected genes: SNCA and DMD.B: Brain maps of transcription in the brain regions for the two genes DMD and SNCA, which provided a high spatial correlation (r) with the VL brain maps for both cohorts. Pooling the N, P, NC and PC genes together, along with those common to both the NC and PC categories, we adopted a DDS to achieve unsupervised clustering of the expression similarity matrix, identifying two major clusters after Silhouette maximization (in blue and red in Figure 3.3, panel B). Importantly, the N and P genes fully segregated into the two differentiated clusters, with all the N genes belonging to the blue cluster and all the P genes to the red one, thereby confirming the different functional roles of the groups of genes in the N and P tails. These two clusters were used separately for gene enrichment analysis. The search for the GO biological pro- Results 61 Figure 3.3: Functional description of the genes with the highest association with VL.A: An all-to-all gene-expression similarity matrix identified the connector hub genes. A total of 1,086 genes were found, equal to the sum of the NC = 452 (blue), PC = 396 (red) and 238 common NC and PC genes (grey). B: Two clusters were finally found that pooled all gene classes, the blue one contains the original N = 251 genes and the red one containing the original p = 101 genes. C: Gene enrichment for GO biological process and Reactome pathways: in blue are the neg-corr genes and in red, the pos-corr genes. Abbreviations: Act., activation; CNS, central nervous system; Mod., modulation; Neg., negative; Pos., positive; Rec., receptor/s; Reg., regulation cesses and Reactome pathways confirmed the differentiated roles of these two clusters, with the neg-corr genes more related to neuronal and synaptic function, involving key synaptic vesicle events such as recycling, localization, endocytosis and exocytosis but also, the dynamics of serotonin and dopamine neurotransmitter release (Figure 3.3, panel C). By contrast, the cluster of pos-corr genes was more related to non-neuronal activities, such as interferon signalling, endothelial cell differentiation, angiogenesis, blood transport and cell development. To assess how the transcriptomics correlated with ND, we focused on the neuropsychological domains in which the composite score re- 62 Chapter 3. Transcriptional signatures of DM1 neurodegeneration flected strong impairment, satisfying Z<-2, which was only the case for the attention category (first cohort Z= -2.5, second cohort Z= - 2.62: see Table 1 and Figure 3.4, panel A for the distribution of all the Z-score values from the two cohorts). As indicated in the Methods, the attention composite was obtained by averaging the Z-scores for the following domains: digit span, STROOP word, STROOP colour, simple RT, election RT, sequential 1 RT and sequential 2 RT. Figure 3.4: Data-driven strategy to define the association between the transcriptome and attention co-activation maps in DM1.A: Attention scores measured as Zscores for the two cohorts. Because the Z-scores were normalized to the values in the HCs, negative values of Zindicate worse performance than the HCs. B: Attention co-activation maps built with the GingerALE tool and projected onto the atlas. C: Histograms of the spatial correlations between the Z-scores of the attention maps and the transcriptional activity for each gene. The tail of N genes (Z<-2, coloured in blue) includes the SNCA gene from the list of preselected genes, whereas the tail of the P genes (Z>2, red) does not include any of these. Following a procedure similar to that described in Figure 3.3 (panel A), we identified the PC genes (red), NC genes (blue, including MAPT), and those common to the NC and PC (grey, including PHKA1). D: After pooling all classes of genes together and clustering, two groups were defined: one including all the neg-corr genes (blue, with SNCA and MAPT) and one with the pos-corr genes (red, with PHKA1 ). Gene enrichment for the tags GO biological process and Reactome pathways. As in Figure 3.3 (panel C), the two clusters also represented two separated functions: the neg-corr one correlated with neuronal functions, while the pos-corr correlated to non-brain functions. Abbreviations: Mod., Modulation; Neg., Negativek; Pos., Positive; Reg., Regulation Results 63 We then obtained brain co-activation maps using ‘Attention’ as a keyword for the search in GingerAle (Figure 3.4, panel B). As for VL, we calculated the association between the attention co-activation maps and transcriptomics by calculating the spatial Pearson correlation between the two Z score vectors (one value per brain region in the atlas, see the histogram of all the correlations in Figure 3.4, panel C). There was a total of 217 N genes (including SNCA), 54 P genes, 336 NC genes (including MAPT from the list of preselected hypothesisdriven genes), 265 PC genes, and 156 genes common to both the NC and PC gene sets (including PHKA1). Pooling together all classes of genes, we followed a DDS of unsupervised clustering and identified two clusters after Silhouette maximization (Figure 3.4, panel C). Like VL, the two clusters were highly segregated and incorporated all neg-corr genes in one cluster, which was enriched in genes related to neuronal and synaptic function (Figure 3.4, panel D), with all the pos-corr genes in the other cluster enriched in non-brain-related activities. To validate some of the predictions done by our computational strategy linking neurodegeneration and neuropsychology with transcriptomics, we also performed neuropathological analyses of brain samples from DM1 patients. Neuropathological details of the cases are shown in Table 3.4. Overall, moderate cerebral atrophy with enlarged ventricles was found in every DM1 case. The main common microscopical alterations were pre-tangles, neurofibrillary tangles, and neuropil threads in selected regions of the brain stem and cerebrum. NFTs were stained with antibodies AT8 (P-tau Ser202/Thr205), Ptau ThR181, P-tau Ser422, 3Rtau, and 4Rtau. The distribution was variable depending on the case. Cases I09-254 and I13-132 were categorized as stages I/II of Braak; cases I09–255, I13–131, and I13–134 as stages III/IV; and cases I09–256 and I13–133 as stages V/VI. βamyloid deposition was absent in five cases, and present in cases I13–134 and I13–133 corresponding to Thal phases 1 and 3, respectively. Mild astrocytic gliosis and microgliosis were observed in the same regions with NFTs. Dystrophic neurites, reactive astrocytes and active microglia were found around β-amyloid deposits in senile 64 Chapter 3. Transcriptional signatures of DM1 neurodegeneration plaques. Lewy bodies were found in one case (I13–133) corresponding to stage 3 of Braak of Parkinson’s pathology. TDP-43 proteinopathy was absent in every case. Mild to moderate small blood vessel disease, characterized by artheriolosclerosis and arteriolar hyalinosis, was common. This was accompanied by mild status cribosus in the striatum, thalamus and white matter in four cases. The white matter of the cerebrum was reduced in size and showed mild to moderate myelin pallor in five cases (I09–255, I13–131, I13–132, I13–133 and I13–134); the corpus callosum was thinner. Myelin changes, as seen with Kl¨uver–Barrera were accompanied by loss of nerve fibres as visualized with anti-neurofilament antibodies. Axonal ballooning was absent. No inflammatory changes were observed in the white matter, with the exception of a few pigment-laden perivascular macrophages. Corpora amylacea were seldom observed in the subventricular region, perivascular spaces and subpial parenchyma. One case had suffered from a recent infarction in the left occipital cortex. Three cases had mild Purkinje cell loss (I13–131, I13–132 and I13–133), and two had mild atrophy of the inferior olives accompanied by astrocytic gliosis; representative tau pathology in Figure 3.5 lower panel. case age NFT Braak SP Thal LB Braak TDP-43 others Western blotting P-tau-Ser214, tau bands I09– 254 60 Stage II No No No — 58/59 kDa, I09– 255 58 Stage IV No No No sbvd, sc, wmd, inf oliv 58 kDa, 55 kDa I09– 256 74 Stage VNo No No inf oliv 58 kDa, 55 kDa I13– 131 67 Stage III/IV No No No sbvd, sc, wmd, Pc↓ weak 68 kDa, 64 kDa, 60 kDa, 58 kDa, 55 kDa I13– 132 58 Stage I-II No No No sbvd, wmd, Pc↓64 kDa, 60 kDa, 58 kDa, 55 kDa I13– 133 68 Stage V/VI Thal 3 LB 3 No sbvd, sc, wmd, Pc↓ 68 kDa, 64 kDa, 60 kDa, 58 kDa, 55 kDa, lower bands I13– 134 59 Stage III/IV Thal 1 No No sbvd, sc, wmd, infarct 66 kDa, 64 kDa, 58 kDa, 55 kDa Table 3.4: Neuropathological details of cases studied.Abbreviations: inf oliv: mild to moderate nerve cell loss and mild astrocytosis in the inferior olives; infarct: acute occipital infarction; LB Braak: Lewy body pathology according to Braak stages; NFT Braak: Braak stages of neurofibrillary tangle pathology; Pc↓: mild to moderate loss of Purkinje cells in the cerebellum; sbvd: small blood vessel disease (arteriolosclerosis, arteriolar hyalinosis); sc: status cribosus and dilatation of perivascular spaces; SP Thal: phases of senile plaques according to Thal scale; TDP-43: TDP-43 proteinopathy; wmd: mild to moderate demyelination of the cerebral white matter. Results 65 Figure 3.5: Neuropathological analyses of brain samples revealed a high heterogeneous Tau-pathology in DM1, different to the one in AD.Upper panel: Gel electrophoresis and Western blotting of sarkosyl-insoluble fractions of the hippocampus incubated with antibodies against 3Rtau, 4Rtau, and phosphorylated tau at serine 214 (P-tau-Ser214) in DM1 cases, and one AD for control. AD is characterized by three bands of 68, 64 and 60 kDa and an upper weak band of 73 kDa. DM1 cases are characterized by the presence of lower bands of 58 and 55 kDa, and occasional upper bands of 60 and 64 kDa; the band of 68/69 kDa is barely present. This particular pattern, including differences from one case to another, is associated with low molecular weight 3Rtau and 4Rtau isoforms suggesting complex altered Tau splicing. Lower panel: Representative images of NFTs and neuropil threads in the entorhinal cortex (EC), CA1 area of the hippocampus (CA1), temporal cortex (TC), and parietal cortex (PC), visualized with the antibody AT8, and anti-3Rtau and 4Rtau antibodies. Paraffin sections slightly counterstained with haematoxylin; bar =25 µm Western blots of sarkosyl-insoluble fractions stained with antiTau-P-Ser214 antibodies revealed three bands of 68 kDa, 64 kDa, 60 kDa and an upper band of 73 kDa in one case with AD stage V. This pattern was in striking contrast with DM1 cases I09–254, I09–255 and I09–256 processed in the same membrane that apparently showed two main bands of 58 and 55 kDa united by a smear, and no bands of upper molecular weights. The incubation of other membranes with anti-4Rtau and anti-3Rtau showed similar alterations. Unique bands of low molecular between 50 and 60 kDa were 66 Chapter 3. Transcriptional signatures of DM1 neurodegeneration in striking contrast with the bands between 60 and 70 kDa observed in AD. Next, the seven cases with DM1 were run in the same gel and blotted with anti-P-tau-Ser214 antibodies. Two bands of 58 and 55 kDa characterized cases I09–255 an I09–256; one band of about 59/60 kDa characterized case I09–254. However, four bands of about 64, 60, 58 and 55 kDa were obtained in cases I13–131, I13–132 and I13–134. Finally, an additional band of about 68 kDa and bands of molecular weight below 50 kDa were identified in case I13–133. In agreement with the different phospho-tau bands depicted with the antibody P-tau-Ser214, the blots using anti-4Rtau and anti-3Rtau equally disclosed individual patterns of bands particularly linked to 3Rtau. Biochemical details of tau bands in every case are shown in Table 3.4; representative Western blots in Figure 3.5 upper panel. We also studied the expression of selected genes from hippocampal tissue from DM1 patients. The study of messenger-RNA (mRNA) levels of genes related to synaptic vesicles revealed down-regulated expression of SNAP25 (p= 0.035) and a trend to reduced expression of SYP and SYN1 (p= 0.06 and p= 0.08, respectively) between DM1 and NFT cases. Regarding the expression of genes related to structural components of synapse, HOMER1 mRNA was significantly reduced in NFT cases compared with MA (p= 0.037). PCLO and ABLIM2 showed a tendency to decrease (p= 0.07 and p= 0.07, respectively) in NFT cases compared with MA cases. None of this group of genes was altered in DM1. The study of the expression of GABAergicand glutamatergic-related genes showed significantly reduced levels of GABRD transcript in NFT cases when compared with MA (p= 0.05) and a tendency to reduced expression of GABRG2 in NFT cases compared with MA (p= 0.10) and DM1 cases (p= 0.09) (Figure 3.6). Very striking, from the genes analysed in Figure 3.6, genes SNCA, SNAP25, SYP, SYN1, SLC17A7 (from synaptic vesicles; panel A), FRMPD4, PCLO, BSN, ARPC5L (from synapse structural components; panel B), and GABRA3, GRIA1 (from GABAergicand glutamatergic-related genes; panel C), all of them appears in the neg-corr gene list used in the enrichment obtained after linking VL or ND with transcriptomics. The tendency showed Results 67 in Figure 3.6 of this set of genes is to be downregulated with respect to MA and NFL, in agreement with the negative-association of these genes with VL. Figure 3.6: RNA expression in hippocampal tissues from DM1 patients revealed protein dysfunction of synaptic vesicle processes in DM1. mRNA expression of synapserelated genes was analysed in the hippocampus in middle-aged (MA) control cases, myotonic dystrophy 1 (DM1) cases, and cases with neurofibrillary tangles (NFT control) linked to ADrelated pathology at middle and advanced stages (III-VI) of Braak and Braak. A: Synaptic vesicles coding genes. B: Synapse structural components coding genes. C: GABAergicand glutamatergic-related coding genes. All data were expressed as mean values ±SEM. Differences between groups are statistically significant at *p <0.05; trends are indicated with * and the exact p value 74 Chapter 4. The Neurogenetics of FC Alterations in Autism typically developing control participants. 4.1 Introduction Autism encompasses multiple manifestations, from impaired social communication and language to restricted or repetitive behavior patterns, interests, and activities [239, 240, 241]. Due to the vast heterogeneity in behavior, and as recommended in DSM-5, this condition is referred to as autism spectrum disorder (ASD), in which the term “spectrum” emphasizes the variation in the type and severity of manifestations [242]. ASD is thought to result from complex interactions during development between genetic, cellular, circuit, epigenetic, and environmental factors [243, 244, 245, 246, 247]. Several researchers have suggested that an excitation/inhibition (E/I) imbalance during development [248, 249] may be an essential mechanism, yet specific factors driving the condition are not well understood. Therapeutic interventions aiming to restore the E/I balance in ASD are a major challenge [250]. Concerning neurobiology, heterogeneity in brain morphology [251] and brain networks has been found, e.g., in the frontal, default mode, and salience networks [252, 253, 254, 255, 256, 257], as well as in the social network [258] — encompassing the primary motor cortex, fusiform gyrus, amygdala, cerebellum, insula, somatosensory cortex, and anterior cingulate cortex [259, 255, 260]. ASD is also heterogeneous in relation to network characteristics; less segregation and greater efficiency [261, 262], and the opposite as well [263] or a combination of both [264, 265], have been shown. Furthermore, ASD neuroanatomical correlates are not static but undergo changes throughout development [266, 5, 267], and the same seems to occur behaviorally in social functioning and communication [268]. Altogether, accumulated evidence has shown high heterogeneity within ASD in the participation of functional brain networks and behavioral manifestations and in the longitudinal trajectories at the individual level. Moreover, with neuroimaging studies, recent work has shown additional sources of heterogeneity due to variations in diagnostic and in- Introduction 75 clusion criteria and differences in the processing neuroimaging pipeline [269, 270]. ASD is also a polygenic, highly heterogeneous condition, with 1010 genes associated with ASD as of July 8, 2022, according to the Simons Foundation Autism Research Initiative (SFARI) gene human database [see also https://gene.sfari.org/database/human-gene/]. Of those, 213 have a relevance score of 1, meaning that they have maximum published pathophysiological evidence to ASD. This high genetic complexity is another manifestation of the heterogeneity of this condition on several scales. Previous work has assessed the associations between transcriptomics and brain morphology [225], showing that genes that are downregulated and enriched for synaptic transmission in individuals with autism were associated with variations in cortical thickness. Novel strategies for ASD subtyping are needed to overcome such multiscale heterogeneity, which is the most significant challenge in the development of effective therapies. Some studies have addressed the heterogeneity in ASD to better stratify this condition [271, 272, 273, 274]. Previous work performed clustering, pooling together ASD and typically developing control (TDC) groups [271], and found 2 groups of individuals showing hyperconnected or hypoconnected patterns (each group containing both ASD and TDC participants). Stratification yields reduced interindividual differences and, therefore, could complement—and even alleviate—the need for large sample sizes in autism-based biomarker discovery [275]. Here, and following previous work [276, 277, 273], we looked at large-scale brain connectivity patterns common within groups of individuals to deploy subtyping in ASD. In particular, we applied consensus clustering strategies to multivariate connectivity patterns of brain regions [278, 279] for associating connectivity-based ASD subtypes with their neurogenetic profile. Following previous work [32, 28, 29, 223, 40, 222, 224, 34], we hypothesized different biological characterization underlying the neurodevelopmental and maturation brain connectivity profile for each subtype, unknown for this condition. For this, we used the AHBA of whole-brain transcriptional data [19] and performed subtyping on 657 individuals with ASD from the Autism Brain Ima- 76 Chapter 4. The Neurogenetics of FC Alterations in Autism ging Data Exchange (ABIDE) repository [280], all of them having passed a very strict quality assurance criterion of elimination of participants by head movement during image acquisition, thus correcting a well-known spurious excess of functional connectivity driven by head movements, which is even more pronounced in the autistic condition. Moreover, to overcome interscanner variability in the functional connectivity values across different institutions, we applied rigorous harmonization strategies to transform heterogeneous data into equivalents [281, 282, 283, 284]. 4.2 Methods 4.2.1 Participants A total of 2156 participants from the ABIDE-I [280] and ABIDEII [285] repositories were initially considered in this study, of which 1026 were individuals with ASD and 1130 were TDC participants. These data were collected across 35 different scanning cohorts. For each participant, both anatomical and functional magnetic resonance imaging (MRI) data were used. Acquisition parameters for each scanning site are found at http://fcon 1000.projects.nitrc.org/indi/abide. Additionally, we extracted several composite scores from the Autism Diagnostic Observation Schedule-Generic, Autism Diagnostic InterviewRevised, Vineland Adaptive Behavior Scales, Social Responsiveness Scale, Social Communication Questionnaire, and raw score of the Autism Quotient, and the verbal, performance, and Full Scale IQ scores to address cognitive performance and disorder severity. Data quality-assurance We discarded subjects with a scanning duration shorter than 5 minutes after scrubbing, lacking full brain coverage, and an average framewise displacement greater than 0.3 mm [69]. Subjects from the KUL sample 3 and NYU sample 2 were also omitted because they only contained ASD subjects and therefore those cohorts did not provide Methods 77 any TDC. Consequently, the number of finally included subjects was 1541 (884 TDC, 657 ASD), corresponding to 33 different scanning cohorts that were further merged into 24 institutions following the guidelines provided in the ABIDE website. The descriptive statistics per institution (number of subjects, ASD cases, mean age, sex distribution) can be found in Table 4.1. Institution name Number of subjects contributed Age (mean ±sd) Sex distribution (Female) Number of ASD cases BNI (II) 40 39.85 ±15.59 0 19 CALTECH (I) 37 27.42 ±9.76 8 18 CMU (I) 20 25.45 ±5.29 4 8 EMC (II) 27 8.40 ±1.09 4 15 ETH (II) 26 22.91 ±4.57 0 6 GU (II) 76 10.92 ±1.66 28 29 IU (II) 37 24.43 ±7.59 9 17 KKI (I and II) 205 10.29 ±1.28 70 48 LEUVEN (I) 61 18.18 ±4.97 7 26 MAX MUN (I) 44 28.77 ±11.79 7 18 NYU (I and II) 245 13.82 ±6.74 42 115 OHSU (II) 83 10.95 ±2.05 31 33 OLIN (I) 22 17.41 ±3.70 4 12 ONRC (II) 18 22.39 ±3.58 5 7 PITT (I) 43 19.53 ±6.87 7 22 SBL (I) 25 34.08 ±6.41 0 13 SDSU (I and II) 86 13.75 ±2.67 15 42 STANFORD (I) 19 10.23 ±1.48 6 11 TCD (I and II) 73 16.86 ±3.41 0 33 UCD (II) 28 14.98 ±1.81 7 15 UCLA (I and II) 70 12.98 ±2.42 8 36 UM (I) 107 14.60 ±3.26 24 41 USM (I and II) 103 23.36 ±7.67 5 52 YALE (I) 46 12.97 ±2.97 13 21 ALL 1541 16.50 ±8.82 304 657 Table 4.1: Main data characteristics for each Institution participating in our study. BNI = Barrow Neurological Institute; CALTECH = California Institute of Technology; CMU = Carnegie Mellon University; EMC = Erasmus University Medical Center Rotterdam; ETH = ETH Z¨urich; GU = Georgetown University; IU = Indiana University; KKI = Kennedy Krieger Institute; LEUVEN = University of Leuven; MAX MUN = Ludwig Maximilians University Munich; NYU = NYU Langone Medical Center; OHSU = Oregon Health and Science University; OLIN = Olin; Institute of Living at Hartford Hospital; ONRC = Olin Neuropsychiatry Research Center, Institute of Living at Hartford Hospital; PITT = University of Pittsburgh School of Medicine; SBL = Social Brain Lab, Netherlands Institute for Neurosciences; SDSU = San Diego State University; STANFORD = Stanford University; TCD = Trinity Centre for Health Sciences; UCD = University of California Davis; UCLA = University of California Los Angeles; UM = University of Michigan; USM = University of Utah School of Medicine; YALE = Yale Child Study Center. I: ABIDE 1. II: ABIDE 2. 78 Chapter 4. The Neurogenetics of FC Alterations in Autism 4.2.2 Functional Connectivity Matrices A state-of-the-art pre-processing pipeline was adopted using FSL 5.0.9, AFNI 16.0.01 [286] and MATLAB 2020b. We first applied slice-time correction, volume alignment to the average one to correct for head motion artifacts, which was followed by intensity normalization. We next regressed out 24 motion parameters, as well as the average cerebrospinal fluid (CSF) and average white matter signal. A band-pass filter was applied between 0.01 and 0.08 Hz, and linear and quadratic trends were removed. All voxels were spatially smoothed with a 6 mm FWHM. After processing the rs-fMRI images, FreeSurfer version 5.3.0 was used for brain segmentation and cortical parcellation. A total of 82 regions were generated from the Desikan-Killiany atlas, with 68 cortical regions (34 in each hemisphere) and 14 subcortical regions segmented from FreeSurfer (left/right thalamus, caudate, putamen, pallidum, hippocampus, amygdala, and accumbens). For each participant, the parcellation was projected to the individual functional data and the mean functional time series of each region was obtained. Finally, one connectivity matrix for each participant was built by Fisher’s Z-transformation of the Pearson correlation coefficients between the region pairs of the time series. 4.2.3 Data Harmonization We verified the presence of heterogeneity related to scanning institutions in our connectivity matrices by applying to each link a KruskalWallis test, for testing median location differences, and a Lavene’s test for differences in variances. The former yielded all the existing links significantly different across institutions after FDR correction (3321 links in total), whereas the latter gave 1236 surviving links. To harmonize our multi-institution functional connectivity data, and before performing subtyping, we used an in-house implementation of Combat https://pypi.org/project/pycombat, adjusting these multiinstitution batch effects by linear mixed modeling and the use of empirical Bayes methods [282]. We also included in this model the Methods 79 diagnosis label (TDC or ASD) as a biological variable of interest, ensuring that group-level connectivity differences were preserved after harmonization. The scanning institution heterogeneity disappeared after Combat harmonization, while retaining the between-group variability of our data. 4.2.4 ASD Subtyping via Consensus Clustering We regressed out from the ASD harmonized connectivity matrices the effects of age, sex and head motion. Then, these matrices were used to define, for each brain region i, a matrix of Euclidean distances between uand vASD subjects, i.e Di≡di uv =v u u t M X j=1 (yu ij −yv ij)2,(4.1) where Mreflects the number of brain regions, and yij the wholebrain connectivity pattern for a given region i(a vector of dimension equal to the number of regions, where each component is defined as the amount of connectivity between the given region and any other in the parcellation). Then, each distance matrix Diwas partitioned into kgroups of subjects using a k-medoids clustering method [287], and the resulting clustering information encoded into an adjacency matrix, whose entries are 1 if a pair of subjects belongs to the same cluster and zero otherwise. Subsequently, a N×Nconsensus matrix Cwas evaluated by averaging this information across the nodes. Hence, the entries of Cuv indicate the number of partitions in which subjects uand vare assigned to the same group, divided by the number of partitions. Eventually, the consensus matrix is averaged over the k range in the interval (2-20), so that information about the underlying structure at different resolutions is combined in the final consensus matrix, for more details see [278]. The consensus matrix Cwas further used to define a Newman and Girvan-like modularity matrix [288]: 80 Chapter 4. The Neurogenetics of FC Alterations in Autism B=C−P, (4.2) where Pis the expected co-assignment matrix, uniform as a consequence of the null ensemble strategy obtained by repeating the permutation of labels 1000 times. Such a modularity matrix Bencodes all the information about the interaction between subjects at different levels. As a result, one could now define any distance quantity applied to this matrix for assessing clustering. Instead, we directly fed this Bmatrix into a generalized Louvain method for community detection (https://github.com/GenLouvain/GenLouvain), yielding an optimal output partition that maximizes the network modularity. The stability of each subtype and the 95% CIs of the estimated maximum modularity were assessed by bootstrapping [289]. 4.2.5 Statistical Differences in Brain Morphology and Behavior Between ASD Subtypes We applied multiple linear regression to assess statistical differences between ASD subtypes in region-wise volume and thickness from FreeSurfer, while controlling for age, sex, and total intracranial volume and a one-way analysis of variance for differences in behavior. Multiple testing was corrected by controlling the false discovery rate (FDR). 4.2.6 Functional connectivity alterations of ASD subtypes with respect to TDC To assess the functional connectivity alterations between each ASD subtype and the TDC group, we performed Multivariate Distance Matrix Regression (MDMR) [290, 291]. Specifically, MDMR regressed each distance matrix per region given by Eq. 4.1 onto a design matrix Xformed by a set of mpredictors, yielding a pseudo-F statistic Fi x that reads: Methods 81 Fi x=tr(HxGi)/(mx) tr[(I−H)Gi]/(N−m),(4.3) where tr indicates the trace operator, Nthe number of observations,mx the degrees of freedom of predictor x,Hxthe isolated effect of predictor xfrom the usual matrix H=X(XTX)−1XT, and Githe so-called Gower matrix built from the distance matrix Di[292]. In our case, the predictors consisted of the ASD vs TDC group factor as the variable of interest, and sex, age, mean framewise displacement and FIQ as covariates. Like the F-estimator in a standard ANOVA analysis, Eq. 4.3 assesses the variance explained by a predictor variable with respect to the unexplained variance. Finally, to estimate how much variability can be attributed to each predictor, a pseudoR2effect size can be computed by dividing the numerator without the degrees of freedom in Eq. 4.3 by the total sum of squared pairwise distances in the Gower matrix R2 x=tr(HxG) tr(G),(4.4) Similar to standard linear models, this effect size quantifies the proportion of the total sum of squares that can be explained by the predictors. 4.2.7 Transcriptomics To build brain transcription maps, we took advantage of the publicly available data in the AHBA [19]. The dataset consisted of MRI images, and a total of 58,692 microarray-based transcription profiles of about 20,945 genes sampled from 3,702 different regions across the brains of six humans. To pool all the transcription data into a single brain template, we followed a similar procedure to that employed elsewhere [25], and previously applied in chapter 3, which includes: (1) Probe re-annotation using a re-annotator toolkit [182]; (2) Removal of probes whose sampling proportion in any of the six brains did not exceed the 70%; (3) Unique probe to gene assignment using 82 Chapter 4. The Neurogenetics of FC Alterations in Autism the maximum differential stability (DS) criterion [26]; (4) Removal of the inter-subject differences by pooling together the Z-scores of the transcription values for each gene and brain; (5) Computation of a single transcription value for each region in the Desikan-Killiany atlas [83] by calculating the median of all the values belonging to the given region; (6) Gene filtering considering only brain-specific genes relative to other tissues using the Human Protein Atlas [11, 293] (https://www. proteinatlas.org). The precise list of brain-specific genes was that extracted on 27th September 2021, and it is available on the https://github. com/compneurobilbao/asd-subtyping-enrichment/tree/main/data/brain-specific genes. txt Some methodological considerations need to be clarified. First, the choice SP > 70%, used in [40] (and chapter 3) was taken following the rationale of choosing the threshold as high as possible, but without being too restrictive, and keeping as many genes as possible for the final study. Second, the choice of the Desikan-Killiany partition was taken as Donor’s brains can be directly segmented using FreeSurfer, which incorporates by default the Desikan-Killiany partition. To use a different partition, it is necessary to transform into that template the donor’s brain, whose images were acquired ex-vivo and as such, it is very likely to contain more misalignments in the brain samples. Finally, restricting our analyses to genes that are relevant in the study of the brain was done to preselect a subset of all genes, as AHBA includes genes expressed in any body tissue. After all the aforementioned considerations, the resulting total number of brain-specific genes was 1882. 4.2.8 The use of spatial autoregressive models for the association of transcriptomics with subtypes The association between brain maps might be inflated due to the existence of spatial autocorrelations. To control for this excess of biased correlations, we introduced a spatially lagged dependent variable to our regression models, i.e. we considered a model of the form Methods 83 y=ρWy+Xβ+v, where y is the brain pattern of transcriptomics for a particular gene, X the input matrix that contains the intercept and the pseudo-R2maps for a particular subtype, Wa weight matrix that determines the spatial autocorrelation structure, and vthe normally distributed model’s noise. Both βand ρare the estimates, with the former quantifying the degree of effect of Xin y, and the latter the amount of spatial autocorrelation. Such a model was implemented using the function ML Lag in the pysal package (https://pysal.org/). To define the weight matrix W, a distance-based approach was adopted, such that spatial autocorrelations were largely driven by nearest brain regions. The cortico-cortical distance was calculated by using their geodesic distance implemented in BrainSmash package [185], while to determine the sobcortico-cortical distance, and similar to [29], the Euclidean distance was calculated [29, 185, 294]. As a result, for each gene we obtained one t-statistic and one pvalue, which allowed us to assess the association with the pseudo-R2 maps while accounting for possible spatial autocorrelations. Among the significantly associated genes, we identified as relevant those genes included in the SFARI database. 4.2.9 Gene Set Enrichment Analysis and Protein Interaction Analysis We only considered for the analyses such genes with FDR-corrected p (pFDR) value < .05 in each subtype. After that, we performed a gene set enrichment analysis using WebGestalt [295] (http://www.webgestalt.org/), introducing as the input the list of the corrected genes and the tstatistic from the association analysis. We computed the gene set enrichment analysis for gene ontology (GO) biological process [31] and Reactome pathways [221] and only considered enriched categories pFDR value < .05. We further applied an ensemble-based enrichment analysis, similar to the one developed in [296], to evaluate whether significant enrichment annotations were affected by inflation or false positive bias [296]. First, we generated 10,000 surrogate brain maps with the same spatial autocorrelation as the original pseudo-R2maps 90 Chapter 4. The Neurogenetics of FC Alterations in Autism Next, we assessed the differences in connectivity patterns between each ASD subtype and the TDC group, measured by regionwise normalized pseudo-R2brain maps, resulting from multivariate distance matrix regression (Figure 4.4). The spatial similarity between these maps was very low (r80 = 0.09, permutation-based p= .67, after using 5000 surrogates that preserved spatial autocorrelation), indicating that each subtype exhibited a distinct neurobiological profile of brainwide connectivity, as expected since the subtypes were obtained by clustering the functional connectivity profiles. Specifically, for subtype 1, higher differences as compared with TDC were found in the superior temporal gyrus, posterior cingulate cortex, and the insula, covering the functional networks of default mode and salience. For subtype 2, higher differences existed in the thalamus, similar to previous work [301], putamen, and precentral gyrus. Thus, alterations affecting the default mode network were common to both subtypes, but one (subtype 1) also showed specific disruptions involving the salience network and the other (subtype 2) in the somatomotor network. For the biological characterization of each subtype, we set out to identify which genes had an expression across brain regions significantly associated (pFDR <.05) with the differences in connectivity measured by the normalized R2brain maps (Figure 4.4, histograms), whereby larger R2values correspond to larger functional connectivity alterations. For subtype 1, a total of 195 negative-associated (NEG) genes and 364 positive-associated (POS) genes existed. Significant NEG genes, also present in the SFARI gene human database with a relevance score of 1, were GFAP, CHD7, SKI, SHANK3, ANK3, and CACNA1E, while POS genes were ASXL3, MAP1A, STXBP1, DPYSL2, KNCB1, SCN8A, RIMS1, and CDKL5. Similarly, for subtype 2, we found 142 NEG genes, of which GRIA2, RFX3, SHANK2, GRIN2B, DLG4, LRRC4C, ARX, and GABRB3 were also present in the SFARI list, and 180 POS genes, including MAGEL2 and IQSEC2. We next applied gene enrichment to the list of significant genes within each subtype, finding no significant enrichment for subtype 1, the type Results 91 Figure 4.4: Association between transcriptomics and connectivity patterns for each autism spectrum disorder subtype. For subtypes 1 and 2, we calculated the pseudoR2map, considering the differences in the connectivity pattern that each subtype has from typically developing control participants. (Right) Brain maps of normalized pseudo-R2. (Left) Histograms of association values between pseudo-R2and gene transcription activity (different values correspond to association with different genes). This procedure was repeated using the pseudo-R2map for each subtype. The tail of the negative genes (false discovery rate–corrected p<.05 and t-statistic <0) is marked with a blue rectangle and the tail of the positive genes (false discovery rate–corrected p<.05 and t-statistic >0) with a red one for both subtypes. Significance limits (t) are also shown. For each distribution tail, we also show the relevant genes present in the SFARI autism spectrum disorder genes with a score = 1. with brain hypoconnectivity. However, for subtype 2, the enrichment of the NEG genes included GO biological processes and Reactome pathways related to glutamate signaling (affecting both AMPA and NMDA receptors) and synapse organization in relation to the E/I imbalance occurring during the development of brain circuits (Figure 4.5, panel A). We also assessed which NEG genes participated in each biological process and pathway (Figure 4.5, panel B), finding that genes DLG4, GRIN2B, GRIA2, and SHANK2 were participating in most of them; and, the gene DLG4 plays a role in all of them. Additionally, the DLG4 gene was the one with the highest degree in the protein interaction network. We found a significant enrichment with biological processes related to E/I imbalance for subtype 2 but not for subtype 1. To test 92 Chapter 4. The Neurogenetics of FC Alterations in Autism Figure 4.5: Excitation/inhibition imbalance enrichment for only one class of participants with autism spectrum disorder (subtype 2).A: GSEA characterization of the false discovery rate–significant genes in subtype 2, including the GO biological processes (dark gray) and Reactome pathways (light gray) enrichments. We further tested whether the enrichment findings were affected by their reporting rate in the literature, and for all cases reported here, we obtained pFDR <.05. B: Participation count for each gene in the processes shown in (A) ranging from 4 to 10 (corresponding to a participation in all processes that only occurred for DLG4). C: Protein-protein interaction physical network from the list of false discovery rate–significant genes. For ease of visualization, only subnetworks with a minimum of 10 genes are depicted. D: Node degree of the genes participating in the network shown in (C). DLG4 is the gene with the highest degree. (B–D) Bars corresponding to genes with SFARI score = 1 are colored in red; SFARI score = 1S in dark red; SFARI score = 2 in orange; and SFARI score = 2S in dark orange, and the same color code was used in (C) and (D) for network nodes. whether these findings suggested that the functional connectivity in subtype 1 was different from that in previous studies of ASD, we calculated the similarity of the connectivity profiles of our 2 subtypes with typical connectivity alterations in ASD, represented by brain maps in [302] and calculated from 4 different ASD databases of resting functional MRI data. We calculated for each subtype the average spatial similarity across the 4 existing brain maps of connectivity alterations in [302]. For subtype 1 the average similarity was not significant (r80 = 0.19, p =.45), but for subtype 2 it was significant (r80 = 0.46, p =.02), indicating that subtype 2 more closely resembled the typical connectivity alterations reported in ASD, which Results 93 in fact is the subtype for which we found significant enrichment toward E/I imbalance. Robustness of transcriptomic-connectivity association results We also compared the results obtained from the generalized Louvain algorithm to those found by multiresolution clustering. As a measure of similarity between the 2 solutions, Dice index values of the solutions were 0.99 and 0.89, respectively, which indicated a high level of reproducibility of the gene expression association with brain alterations between the 2 clustering strategies. Additionally, we studied the effects of considering a different brain partition on the results of the transcription-connectivity association. Using the functionally defined Schaefer [82] brain partition with 100 different regions, the association results obtained from the DesikanKilliany atlas as compared to those from the Schaefer partition had very low similarity for subtype 1 (r1880 =−0.11, p < .001), and slightly higher results were found for subtype 2 (r1880 = 0.40, p < .001). By adding the same subcortical regions to the Schaefer partition, the gene association became very similar for the 2 brain partitions and for the 2 subtypes (subtype 1, r1880 = 0.87, p < .001; subtype 2, r1880 = 0.92, p < .001), suggesting a strong contribution of the subcortical alterations to the robustness of our association results. These results were obtained by using left-hemisphere transcription sites 2, but the results were also preserved when we repeated the analysis for the 2 brain hemispheres (Figure 4.6). 2We focused on the left hemisphere, as all donors provided sampling sites of genes in this hemisphere, and only 2 of the 6 donors from the AHBA dataset were sampled in both left and right hemispheres. 94 Chapter 4. The Neurogenetics of FC Alterations in Autism Figure 4.6: Replicability of the results including samples from the whole brain. Similar to Figure 4.5, but now including genetic information from sampling sites in both brain hemispheres. These new results show replicability of the reported enrichment, although more genes are now implicated in addition to an increased complexity of the protein-protein interaction network. Finally, it is important to note that no significant enrichment was found for subtypes lower in the dendrogram level corresponding to the 2 subtypes described above (Figure 4.2). The significant enrichment did not exist after repeating the same analysis using the entire ASD group, indicating the need for subtyping first in the entire Discussion 95 population to reveal our findings. To prove that our gene enrichment findings were specific to the ASD condition we also repeated the same procedure but only using the TDC population. To do that, we first divided the entire TDC cohort into two subgroups, half-sized and randomly chosen, so that they were well-matched for age, sex, movement, FIQ, and overall connectivity (mean across positive entries in the connectivity matrices). Next, we applied in one subgroup the same subtyping procedure (including the same previous denoising step to avoid clusters driven by effects of no interest), obtaining again two subtypes, one representing hyper-connectivity and the other hypoconnectivity. We then used the other TDC subgroup to calculate the pseudo-R2statistical maps, which were subsequently associated with the transcriptomics data. As a result, no gene survived FDR correction in any subtype, thus indicating that the E/I imbalance found in the hyperconnected autistic subtype is specific to the autistic condition. Likewise, although the subtyping performed in both ASD and TDC groups resulted in solutions with similar overall connectivity separation, resulting in 2 sets of hypoand hyperconnected brains, a multivariate distance matrix regression analysis applied to the functional correlation patterns showed that the hyperconnected subtype found in ASD was statistically different from that in TDC (p<.001). The hypoconnected subtypes in ASD and TDC were also different from each other (p<.001). This might explain why no similar findings in the enrichment were found for the hyperconnected TDC subgroup. In summation, the significant association between E/I imbalance and altered functional connectivity was observed when subtyping in ASD, and only in the ASD group characterized by overall hyperconnectivity, demonstrating the specificity of the reported enrichment. 4.4 Discussion Two significant subtypes result from functional connectivity–based subtyping in a cohort of 657 individuals with ASD. The two are indistinguishable by behavioral scores, and also by morphometric com- 96 Chapter 4. The Neurogenetics of FC Alterations in Autism parisons based on structural neuroimaging, in agreement with recent results [275]. Compared with the TDC group, the first subtype is characterized by hypoconnectivity, with major implications in the superior temporal gyrus, posterior cingulate cortex, and insula, showing connectivity alterations in the default mode and salience networks with no significant gene enrichment after correcting for multiple comparisons. The second subtype, representing 43% of participants with autism, is characterized by hyperconnectivity, with major implications in the thalamus, putamen, and precentral gyrus and showing network alterations in somatomotor and default mode networks. In a recent analysis linking genomics and resting functional connectivity in 32,726 individuals with psychiatric conditions, significant ASD contributions were shown in the thalamic and somatomotor networks [301], consistent with our results for subtype 2. Only subtype 2 had a significant gene enrichment toward glutamate signaling (affecting both AMPA and NMDA receptors), consistent with the E/I imbalance that occurs during brain development and one of the most accepted hypotheses in the pathophysiology of autism [303]. Indeed, it is thought that in the development of ASD, there is an increase in the ratio between excitation and inhibition, leading to hyperexcitability of cortical circuits [249]. It is also possible that differential E/I alteration of selective brain circuits might result in an unaltered E/I ratio at the network level [248]. Our work maps patterns of functional connectivity alterations with genes that are involved in E/I balance. While it is true that perturbation in these genes in animal models strongly affects E/I imbalance in brain networks [304], the participant data that we analyzed in this study do not directly address the E/I imbalance, and this is a limitation of our methodology. It is also important to emphasize that the E/I enrichment found in our study is specific to the ASD condition and, as such, does not occur in the TDC group. Moreover, the connectivity profile in the entire autistic population, i.e., if no subtyping is performed, does not have significant enrichment, indicating the need for subtyping first to find the connection with E/I imbalance in one subtype of individuals with ASD. Discussion 97 Our subtyping approach was based on patterns of functional connectivity alterations. There are 3 major reasons supporting our choice not to use structural features for our subtyping analysis. First, it would require a different clustering approach to the one adopted here, which is based on the consensus of connectivity patterns. Second, and based on recent data-driven results from an international autism imaging biomarker challenge [275] with more than 146 institutions submitting prediction algorithms, the 10 best-performing algorithms (with ASD prediction accuracies having area under the curve >0.80) showed a dominant contribution of the functional modality, with a much higher discriminative power than the structural MRI data. Third, our main goal was to study the origin of functional connectivity–based heterogeneity in autism, and structural features (representing different brain aspects) give rise to a different kind of heterogeneity. As a result, the proper combination of these 2 diverse sources of heterogeneity would require a multimodal approach different from the one developed here. Our approach is unique in several ways. First, our study is based on a large cohort of individuals with ASD (N = 657) from the ABIDE initiative, all of them having passed the rigorous criteria of motion removal, and it combines anatomical and functional neuroimaging data from 24 different institutions. Second, we used Combat, a rigorous data harmonization method to eliminate the variability between MRI scans across the 24 institutions, one of the largest sources of variability when combining imaging data from multiple institutions [305]. Third, our analysis of brain connectivity was carried out on a large scale, in which each brain region is represented by its connectivity pattern across the entire brain. Therefore, we did not consider a priori any brain region as more dominant or relevant than the others. Fourth, we made use of a consensus clustering approach that we developed [6, 279], and that has been successfully tested by others [306], to group participants in the same subtype if the connectivity profiles are similar across all the analyzed regions. Finally, we made use of the AHBA to describe the neurogenetic profiles of each subtype, which has been used before for morphometric information in 98 Chapter 4. The Neurogenetics of FC Alterations in Autism ASD [225] but never for characterizing subtypes based on functional connectivity patterns of this condition. Due to the heterogeneity and diversity reported in ASD genetics, the use of AHBA may shed new light, because it provides information on the transcriptome across the brain in unprecedented detail, accounting for 3702 sampling sites with transcription information on 20,500 genes as a specific signature for each anatomical region. Moreover, the use of AHBA is complementary to other techniques, such as GWAS [226], that simultaneously address genotypephenotype associations from hundreds of thousands to millions of genetic variants in a data-driven manner. Indeed, genome-wide association studies have previously been used for ASD subtyping [307, 16] using behavioral scores as traits and, therefore, the subtypes obtained were more closely related to symptom severity and not to functional connectivity. Our enrichment results for subtype 2 show that DLG4, also known as PSD95, is a gene with major implications in the protein interaction network of subtype 2. DLG4 mediates NMDA and AMPA receptor clustering and function; it affects glutamatergic transmission and has been shown to have an aberrant function in ASD [308, 309, 310, 311, 312]. DLG4 also influences the size and density of dendritic spines during brain development, having strong effects on synaptic connectivity and activity, e.g., reduced DLG4 activity leads to increased dendritic spine numbers [313]. Some limitations should be noted. First, our transcriptomic analysis was based on AHBA, which is derived from healthy, and not from ASD, brain tissues. Therefore, the relations studied here between ASD-dependent connectivity patterns and healthy transcriptomics highlight large-scale organization aspects of the connectivity alterations to gene expression. Future studies should confirm our findings using gene expression data from a pathological cohort, which is not currently available. Second, the number of donors from the AHBA data is very limited (n = 6) and the sampling sites available do not cover the full brain. Third, our subtyping method found 2 subtypes of ASD participants who were hypoconnected and hyperconnected Data and code availability 99 at the network level. The same classes of subtypes were found by subtyping the TDC group. However, when comparing first the hypoconnectivity subtypes between the ASD and TDC groups and then the hyperconnectivity subtypes, connectivity patterns were significantly different in both cases, which justifies the significant enrichment found for the hyperconnectivity ASD subtype, but not for TDC. Finally, our main neurogenetic finding in one ASD subtype, involving genes largely affecting the E/I imbalance, is based only on the statistical association between transcriptome activity and patterns of functional connectivity alterations. Future studies should explicitly test the causal link between E/I imbalance and functional connectivity alterations in ASD. In summary, our novel approach, which includes data harmonization, multivariate distancing in large-scale functional connectivity patterns, and transcriptome brain maps, reveals strong enrichment for glutamate signaling (affecting both AMPA and NMDA receptors) and synapse organization in one subgroup of participants with ASD, reinforcing the hypothesis of an E/I imbalance occurring during brain development of individuals with ASD. 4.5 Data and code availability The data employed in this study belong to the ABIDE-I and ABIDEII repositories. Their IDs can be found in https://github.com/compneurobilbao/ asd-subtyping-enrichment, as well as the codes used for the analyses. The initial multicenter ABIDE dataset consisted of 35 different scanning cohorts and two groups of subjects, TDC (N=1130) and ASD (N=1026). After following our data quality-assurance, we obtained connectivity matrices of 884 TDC and 657 ASD, belonging to 24 institutions, which were ultimately used for the subtyping analysis. These matrices are available at https://doi.org/10.6084/m9.figshare.21901821. 106 Bibliography [8] M. Gatica et al. High-order interdependencies in the aging brain. Brain Connectivity, 11(9):734–744, 2021. [9] M. Gatica et al. High-order functional redundancy in ageing explained via alterations in the connectome in a whole-brain model. PLOS Computational Biology, 18(9):e1010431, September 2022. [10] A. Jimenez-Marin et al. Brain connectivity and cognitive functioning in individuals six months after multiorgan failure. NeuroImage. Clinical, 25:102137, 2020. [11] I. Diez and J. Sepulcre. Unveiling the neuroimaging-genetic intersections in the human brain. Current Opinion in Neurology, Publish Ahead of Print, 5 2021. [Online; accessed 2022-02-02]. [12] L. Shen et al. Whole genome association study of brain-wide imaging phenotypes for identifying quantitative trait loci in MCI and AD: A study of the ADNI cohort. NeuroImage, 53(3):1051–1063, November 2010. [13] The Alzheimer’s Disease Neuroimaging Initiative et al. Common genetic variants influence human subcortical brain structures. Nature, 520(7546):224–229, April 2015. [14] D. P. Hibar et al. Novel genetic loci associated with hippocampal volume. Nature Communications, 8(1):13624, January 2017. [15] L. T. Elliott et al. Genome-wide association studies of brain imaging phenotypes in UK Biobank. Nature, 562(7726):210– 216, October 2018. [16] A. Yousaf et al. Quantitative genome-wide association study of six phenotypic subdomains identifies novel genome-wide significant variants in autism spectrum disorder. Translational Psychiatry, 10(1):215, 12 2020. Bibliography 107 [17] B. Zhao et al. Transcriptome-wide association analysis of brain structures yields insights into pleiotropy with complex neuropsychiatric traits. Nature Communications, 12(1):2878, May 2021. [18] X. Bledsoe and E. R. Gamazon. A Transcriptomic Atlas of the Human Brain Reveals Genetically Determined Aspects of Neuropsychiatric Health. medRxiv: The Preprint Server for Health Sciences, pp. 2023.03.10.23287072, June 2023. [19] M. J. Hawrylycz et al. An anatomically comprehensive atlas of the adult human brain transcriptome. Nature, 489(7416):391– 399, 2012. [20] J. M. Keil et al. Brain Transcriptome Databases: A User’s Guide. The Journal of Neuroscience: The Official Journal of the Society for Neuroscience, 38(10):2399–2412, March 2018. [21] H. J. Kang et al. Spatio-temporal transcriptome of the human brain. Nature, 478(7370):483–489, October 2011. [22] C. Colantuoni et al. Temporal dynamics and genetic control of transcription in the human prefrontal cortex. Nature, 478(7370):519–523, October 2011. [23] J. A. Miller et al. Transcriptional landscape of the prenatal human brain. Nature, 508(7495):199–206, April 2014. [24] GTEx Consortium et al. Genetic effects on gene expression across human tissues. Nature, 550(7675):204–213, October 2017. [25] A. Arnatkeviciute et al. A practical guide to linking brain-wide gene expression and neuroimaging data. NeuroImage, 189:353– 367, April 2019. [26] M. Hawrylycz et al. Canonical genetic signatures of the adult human brain. Nature Neuroscience, 18(12):1832–1844, 12 2015. PMID: 26571460 PMCID: PMC4700510. 108 Bibliography [27] J. Rasero et al. The neurogenetics of functional connectivity alterations in Autism: Insights from subtyping in 657 patients. Biological Psychiatry, pp. S0006–3223(23)01230–1, April 2023. [28] I. Diez and J. Sepulcre. Neurogenetic profiles delineate largescale connectivity dynamics of the human brain. Nature Communications, 9(1):3876, September 2018. [29] J. B. Burt et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nature Neuroscience, 21(9):1251–1259, 9 2018. [30] A. Altmann et al. Analysis of brain atrophy and local gene expression in genetic frontotemporal dementia. Brain Communications, 2(2):fcaa122, July 2020. [31] M. Ashburner et al. Gene ontology: tool for the unification of biology. the gene ontology consortium. Nature Genetics, 25(1):25–29, 5 2000. PMID: 10802651 PMCID: PMC3037419. [32] E. Bueichek´u et al. Central neurogenetic signatures of the visuomotor integration system. Proceedings of the National Academy of Sciences of the United States of America, 117(12):6836–6843, March 2020. [33] C. Paquola et al. Shifts in myeloarchitecture characterise adolescent development of cortical gradients. eLife, 8:e50482, November 2019. [34] J. Sepulcre et al. Neurogenetic contributions to amyloid beta and tau spreading in the human cortex. Nature Medicine, 24(12):1910–1918, December 2018. [35] M. J. Grothe et al. Molecular properties underlying regional vulnerability to Alzheimer’s disease pathology. Brain: A Journal of Neurology, 141(9):2755–2771, September 2018. Bibliography 109 [36] Y.-Q. Zheng et al. Local vulnerability and global connectivity jointly shape neurodegenerative disease propagation. PLoS biology, 17(11):e3000495, November 2019. [37] S. Basaia et al. Neurogenetic traits outline vulnerability to cortical disruption in Parkinson’s disease. NeuroImage. Clinical, 33:102941, 2022. [38] H. Santamar´ıa-Garc´ıa et al. Neurocognitive patterns across genetic levels in behavioral variant frontotemporal dementia: a multiple single cases study. BMC neurology, 22(1):454, December 2022. [39] G. Shafiei et al. Network structure and transcriptomic vulnerability shape atrophy in frontotemporal dementia. Brain, 146(1):321–336, January 2023. [40] A. Jimenez-Marin et al. Transcriptional signatures of synaptic vesicle genes define myotonic dystrophy type I neurodegeneration. Neuropathology and Applied Neurobiology, 47(7):1092– 1108, December 2021. [41] J. Sun et al. Genes associated with grey matter volume reduction in multiple sclerosis. Journal of Neurology, 269(4):2004– 2015, April 2022. [42] A. Buchin et al. Multi-modal characterization and simulation of human epileptic circuitry. Cell Reports, 41(13):111873, December 2022. [43] Y. Yin et al. Structural and functional changes in drug-na¨ıve benign childhood epilepsy with centrotemporal spikes and their associated gene expression profiles. Cerebral Cortex (New York, N.Y.: 1991), 33(10):5774–5782, May 2023. [44] A. S. Mandal et al. Genetic, cellular, and connectomic characterization of the brain regions commonly plagued by glioma. Brain: A Journal of Neurology, 143(11):3294–3307, December 2020. 110 Bibliography [45] R. Romero-Garcia et al. Transcriptomic and connectomic correlates of differential spatial patterning among gliomas. Brain: A Journal of Neurology, 146(3):1200–1211, March 2023. [46] Y. Xie et al. Brain mRNA Expression Associated with Cortical Volume Alterations in Autism Spectrum Disorder. Cell Reports, 32(11):108137, September 2020. [47] L.-B. Cui et al. Transcriptional level of inflammation markers associates with short-term brain structural changes in firstepisode schizophrenia. BMC medicine, 21(1):250, July 2023. [48] X. Xu et al. Genetic mechanisms underlying gray matter volume changes in patients with drug-naive first-episode schizophrenia. Cerebral Cortex (New York, N.Y.: 1991), 33(5):2328– 2341, February 2023. [49] K. M. Anderson et al. Convergent molecular, cellular, and cortical neuroimaging signatures of major depressive disorder. Proceedings of the National Academy of Sciences, 117(40):25138– 25149, October 2020. [50] J. Li et al. Cortical structural differences in major depressive disorder correlate with cell type-specific transcriptional signatures. Nature Communications, 12(1):1647, March 2021. [51] P. Manza et al. Brain structural changes in cannabis dependence: association with MAGL. Molecular Psychiatry, 25(12):3256–3266, December 2020. [52] I. Diez et al. Early-life trauma endophenotypes and brain circuit-gene expression relationships in functional neurological (conversion) disorder. Molecular Psychiatry, 26(8):3817–3828, August 2021. [53] A. L. de Munain et al. Prevalence of myotonic dystrophy in guipuzcoa (basque country, spain). Neurology, 43(8):1573–1573, 8 1993. Bibliography 111 [54] F. J. Gall. Letter from Dr. F. J. Gall, to Joseph Fr von Retzer, upon the Function of the Brain, in Man and Animals. Der neue Teutsche Merkur, 3:311–332, 1798. [55] J. G. Spurzheim. The Anatomy of the Brain, with a General View of the Nervous System. Boston: Marsh, Capen & Lyon, 1834, 1826. [56] P. P. Broca. Sur le principe des localisations c´er´ebrales. Bull. de la Soci´et´e d”Anthropologie, 2:190–204, 1861. [57] P. P. Broca. Localisations des fonctions c´er´ebrales. si`ege de la facult´e du langage articul´e. Bull. de la Soci´et´e d”Anthropologie, 4:200–208, 1863. [58] J. P. Flourens. Recherches exp´erimentales sur les propri´et´es et les fonctions du syst`eme nerveux dans les animaux vert´ebr´es. Paris: JB Balli`ere, 1842. [59] A. D. Boes et al. Network localization of neurological symptoms from focal brain lesions. Brain: A Journal of Neurology, 138(Pt 10):3061–3075, 2015. [60] J. Joutsa et al. Lesion network mapping for symptom localization: recent developments and future directions. Current Opinion in Neurology, 35(4):453–459, aug 2022. [61] C. Ganos et al. A neural network for tics: insights from causal brain lesions and deep brain stimulation. Brain, 145(12):4385– 4397, 01 2022. [62] J. S. Siegel et al. Disruptions of network connectivity predict impairment in multiple behavioral domains after stroke. Proceedings of the National Academy of Sciences, 113(30), July 2016. [63] A. Salvalaggio et al. Post-stroke deficit prediction from lesion and indirect structural and functional disconnection. Brain, 143(7):2173–2188, July 2020. 112 Bibliography [64] A. Jimenez-Marin et al. Multimodal and multidomain lesion network mapping enhances prediction of sensorimotor behavior in stroke patients. Scientific Reports, 12(1):22400, December 2022. [65] O. Sporns et al. The human connectome: A structural description of the human brain. PLoS Comput. Biol, 1:245–251, 2005. [66] T. J. Sejnowski et al. Putting big data to good use in neuroscience. Nature Neuroscience, 17(11):1440–1441, November 2014. [67] K. Krupa and M. Bekiesi´nska-Figatowska. Artifacts in magnetic resonance imaging. Polish Journal of Radiology, 80:93– 106, 2015. [68] R. A. Poldrack. Handbook of Functional MRI Data Analysis. Cambridge University Press, 2011. [69] J. D. Power et al. Spurious but systematic correlations in functional connectivity mri networks arise from subject motion. NeuroImage, 59(3):2142–2154, 2 2012. [70] R. H. R. Pruim et al. ICA-AROMA: A robust ICAbased strategy for removing motion artifacts from fMRI data. NeuroImage, 112:267–277, May 2015. [71] C. Caballero-Gaudes and R. C. Reynolds. Methods for cleaning the BOLD fMRI signal. NeuroImage, 154:128–149, July 2017. [72] D. Cordes et al. Frequencies contributing to functional connectivity in the cerebral cortex in ”resting-state” data. AJNR. American journal of neuroradiology, 22(7):1326–1333, August 2001. [73] O. Esteban et al. fMRIPrep: a robust preprocessing pipeline for functional MRI. Nature Methods, 16(1):111–116, January 2019. Bibliography 113 [74] S. Whitfield-Gabrieli and A. Nieto-Castanon. Conn : A Functional Connectivity Toolbox for Correlated and Anticorrelated Brain Networks. Brain Connectivity, 2(3):125–141, June 2012. [75] C. Cameron et al. Towards Automated Analysis of Connectomes: The Configurable Pipeline for the Analysis of Connectomes (C-PAC). Frontiers in Neuroinformatics, 7, 2013. [76] S. Mori and J. Zhang. Principles of diffusion tensor imaging and its applications to basic neuroscience research. Neuron, 51(5):527–539, September 2006. [77] S. Mori et al. Three-dimensional tracking of axonal projections in the brain by magnetic resonance imaging. Annals of Neurology, 45(2):265–269, February 1999. [78] T. E. J. Behrens et al. Characterization and propagation of uncertainty in diffusion-weighted MR imaging. Magnetic Resonance in Medicine, 50(5):1077–1088, November 2003. [79] J.-D. Tournier et al. Robust determination of the fibre orientation distribution in diffusion MRI: Non-negativity constrained super-resolved spherical deconvolution. NeuroImage, 35(4):1459–1472, May 2007. [80] J. Tournier et al. Improved probabilistic streamlines tractography by 2nd order integration over fibre orientation distributions. Proceedings of the International Society for Magnetic Resonance in Medicine, 1670, 2010. [81] B. T. T. Yeo et al. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology, 106(3):1125–1165, 2011. [82] A. Schaefer et al. Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity mri. Cerebral Cortex (New York, N.Y.: 1991), 28(9):3095–3114, 9 2018. PMID: 28981612 PMCID: PMC6095216. 114 Bibliography [83] R. S. Desikan et al. An automated labeling system for subdividing the human cerebral cortex on mri scans into gyral based regions of interest. NeuroImage, 31(3):968–980, 2006. [84] N. Tzourio-Mazoyer et al. Automated anatomical labeling of activations in spm using a macroscopic anatomical parcellation of the mni mri single-subject brain. NeuroImage, 15(1):273– 289, 1 2002. PMID: 11771995. [85] M. F. Glasser et al. A multi-modal parcellation of human cerebral cortex. Nature, 536(7615):171–178, August 2016. [86] I. Diez et al. A novel brain partition highlights the modular skeleton shared by structure and function. Scientific Reports, 5(1):10532, 2015. [87] A. Jimenez-Marin et al. Open datasets and code for multiscale relations on structure, function and neuro-genetics in the human brain. biorXiv preprint, aug 2023. [88] K. J. Friston. Functional and effective connectivity in neuroimaging: A synthesis. Human Brain Mapping, 2(1-2):56–78, 1994. [89] M. A. Koch et al. An investigation of functional and anatomical connectivity using magnetic resonance imaging. NeuroImage, 16(1):241–250, 2002. [90] C. J. Honey et al. Network structure of cerebral cortex shapes functional connectivity on multiple time scales. Proceedings of the National Academy of Sciences of the United States of America, 104(24):10240–10245, June 2007. [91] J. J. Crofts and D. J. Higham. A weighted communicability measure applied to complex brain networks. Journal of the Royal Society, Interface, 6(33):411–414, April 2009. [92] G. Deco et al. Key role of coupling, delay, and noise in resting brain fluctuations. Proceedings of the National Academy of Bibliography 115 Sciences of the United States of America, 106(25):10302–10307, June 2009. [93] C. J. Honey et al. Can structure predict function in the human brain? NeuroImage, 52(3):766–776, 2010. [94] Y. Adachi et al. Functional connectivity between anatomically unconnected areas is shaped by collective network-level effects in the macaque cortex. Cerebral Cortex (New York, N.Y.: 1991), 22(7):1586–1592, July 2012. [95] H.-J. Park and K. Friston. Structural and functional brain networks: from connections to cognition. Science (New York, N.Y.), 342(6158):1238411, November 2013. [96] J. Go˜ni et al. Resting-brain functional connectivity predicted by analytic measures of network communication. Proceedings of the National Academy of Sciences of the United States of America, 111(2):833–838, January 2014. [97] A. Mess´e et al. Predicting functional connectivity from structural connectivity via computational models using MRI: an extensive comparison study. NeuroImage, 111:65–75, May 2015. [98] M. Breakspear. Dynamic models of large-scale brain activity. Nature Neuroscience, 20(3):340–352, March 2017. [99] K. Bansal et al. Personalized brain network models for assessing structure-function relationships. Current Opinion in Neurobiology, 52:42–47, October 2018. [100] B. V´azquez-Rodr´ıguez et al. Gradients of structure-function tethering across neocortex. Proceedings of the National Academy of Sciences of the United States of America, 116(42):21219–21227, October 2019. [101] L. E. Su´arez et al. Linking structure and function in macroscale brain networks. Trends in Cognitive Sciences, 24(4):302–315, 2020. 122 Bibliography [158] J. Lee and T. Cooper. Pathogenic mechanisms of myotonic dystrophy. Biochemical Society Transactions, 37(6):1281–1286, 12 2009. [159] J. D. Lueck et al. Chloride channelopathy in myotonic dystrophy resulting from loss of posttranscriptional regulation for clcn1. American Journal of Physiology-Cell Physiology, 292(4):C1291–C1297, 4 2007. [160] M. Santoro et al. Alternative splicing of human insulin receptor gene (insr) in type i and type ii skeletal muscle fibers of patients with myotonic dystrophy type 1 and type 2. Molecular and Cellular Biochemistry, 380(1-2):259–265, 8 2013. [161] C. Fugier et al. Misregulated alternative splicing of bin1 is associated with t tubule alterations and muscle weakness in myotonic dystrophy. Nature Medicine, 17(6):720–725, 6 2011. PMID: 21623381. [162] A. Mankodi et al. Myotonic dystrophy in transgenic mice expressing an expanded cug repeat. Science (New York, N.Y.), 289(5485):1769–1773, 9 2000. PMID: 10976074. [163] A. Sistiaga et al. Cognitive/personality pattern and triplet expansion size in adult myotonic dystrophy type 1 (dm1): Ctg repeats, cognition and personality in dm1. Psychological Medicine, 40(3):487–495, 3 2010. [164] S. Winblad et al. Cognitive deficits and ctg repeat expansion size in classical myotonic dystrophy type 1 (dm1). Behavioral and brain functions: BBF, 2:16, 5 2006. PMID: 16696870 PMCID: PMC1475858. [165] C. Carpentier et al. Tau exon 2 responsive elements deregulated in myotonic dystrophy type i are proximal to exon 2 and synergistically regulated by mbnl1 and mbnl2. Biochimica Et Biophysica Acta, 1842(4):654–664, 4 2014. PMID: 24440524. Bibliography 123 [166] H. Jiang et al. Myotonic dystrophy type 1 is associated with nuclear foci of mutant rna, sequestration of muscleblind proteins and deregulated alternative splicing in neurons. Human Molecular Genetics, 13(24):3079–3088, 12 2004. PMID: 15496431. [167] A. Annic et al. Early dopasensitive parkinsonism related to myotonic dystrophy type 2. Movement Disorders: Official Journal of the Movement Disorder Society, 23(14):2100–2101, 10 2008. PMID: 18759330. [168] K. Itoh et al. Neuropathology does not correlate with regional differences in the extent of expansion of ctg repeats in the brain with myotonic dystrophy type 1. Acta Histochemica Et Cytochemica, 43(6):149–156, 12 2010. PMID: 21245981 PMCID: PMC3015052. [169] L. Serra et al. How genetics affects the brain to produce higherlevel dysfunctions in myotonic dystrophy type 1. Functional Neurology, 30(1):21–31, 3 2015. PMID: 26214024 PMCID: PMC4520669. [170] J. Mathieu et al. Assessment of a disease-specific muscular impairment rating scale in myotonic dystrophy. Neurology, 56(3):336–340, 2 2001. [171] D. Wechsler and N. Seisdedos Cubero. WAIS-III escala de inteligencia de Wechsler para adultos-III. TEA Ediciones, 1999. OCLC: 981374932. [172] M. D. Lezak et al. Neuropsychological assessment, 4th ed. Neuropsychological assessment, 4th ed. Oxford University Press, New York, NY, US, 2004. page: xiv, 1016. [173] M. Casals-Coll et al. Estudios normativos espa˜noles en poblaci´on adulta joven (proyecto neuronorma j´ovenes): normas para los test de fluencia verbal. Neurolog´ıa, 28(1):33–40, 1 2013. 124 Bibliography [174] J. Pena-Casanova et al. Spanish multicenter normative studies (neuronorma project): Norms for verbal fluency tests. Archives of Clinical Neuropsychology, 24(4):395–411, 6 2009. [175] A. Rey. FTest de Copia y Reproducci´on de una Figura Compleja. Tea Ediciones, S.A., 2009. [176] N. Adams. Contributions to neuropsychological assessment: A clinical manual, 2nd ed. Neurology, 45(8):1637–1637, 8 1995. [177] G. Douaud et al. Anatomically related grey and white matter abnormalities in adolescent-onset schizophrenia. Brain, 130(9):2375–2386, 2007. [178] C. D. Good et al. A voxel-based morphometric study of ageing in 465 normal adult human brains. NeuroImage, 14(1 I):21–36, 2001. [179] J. L. R. Andersson et al. Non-linear registration, aka spatial normalisation. fmrib technial report tr07ja2. Oxford Centre for Functional Magnetic Resonance Imaging of the Brain, Department of Clinical Neurology, Oxford University, Oxford, UK, (June):22, 2007. [180] S. M. Smith et al. Tract-based spatial statistics: voxelwise analysis of multi-subject diffusion data. NeuroImage, 31(4):1487– 1505, 7 2006. PMID: 16624579. [181] S. M. Smith and T. E. Nichols. Threshold-free cluster enhancement: addressing problems of smoothing, threshold dependence and localisation in cluster inference. NeuroImage, 44(1):83–98, 1 2009. PMID: 18501637. [182] J. Arloth et al. Re-annotator: Annotation pipeline for microarray probe sequences. PLOS ONE, 10(10):e0139516, 10 2015. [183] A. R. Laird et al. Brainmap: the social evolution of a human brain mapping database. Neuroinformatics, 3(1):65–78, 2005. PMID: 15897617. Bibliography 125 [184] S. B. Eickhoff et al. Coordinate-based activation likelihood estimation meta-analysis of neuroimaging data: a randomeffects approach based on empirical estimates of spatial uncertainty. Human Brain Mapping, 30(9):2907–2926, 9 2009. PMID: 19172646 PMCID: PMC2872071. [185] J. B. Burt et al. Generative modeling of brain maps with spatial autocorrelation. NeuroImage, 220:117038, 10 2020. [186] J. Buxton et al. Detection of an unstable fragment of dna specific to individuals with myotonic dystrophy. Nature, 355(6360):547–548, 2 1992. PMID: 1346924. [187] Y. H. Fu et al. An unstable triplet repeat in a gene related to myotonic muscular dystrophy. Science (New York, N.Y.), 255(5049):1256–1258, 3 1992. PMID: 1546326. [188] H. G. Harley et al. Expansion of an unstable dna region and phenotypic variation in myotonic dystrophy. Nature, 355(6360):545–546, 2 1992. PMID: 1346923. [189] M. Mahadevan et al. Myotonic dystrophy mutation: an unstable ctg repeat in the 3’ untranslated region of the gene. Science (New York, N.Y.), 255(5049):1253–1255, 3 1992. PMID: 1546325. [190] A. Suzuki et al. Mkbp, a novel member of the small heat shock protein family, binds and activates the myotonic dystrophy protein kinase. The Journal of Cell Biology, 140(5):1113–1124, 3 1998. PMID: 9490724 PMCID: PMC2132705. [191] R. S. Savkur et al. Aberrant regulation of insulin receptor alternative splicing is associated with insulin resistance in myotonic dystrophy. Nature Genetics, 29(1):40–47, 9 2001. PMID: 11528389. [192] J. Shin et al. Impaired neurodevelopment by the low complexity domain of cpeb4 reveals a convergent pathway with neurode- 126 Bibliography generation. Scientific Reports, 6:29395, 2016. PMID: 27381259 PMCID: PMC4933966. [193] G. Ayalon et al. An ankyrin-based mechanism for functional organization of dystrophin and dystroglycan. Cell, 135(7):1189– 1200, 12 2008. PMID: 19109891. [194] K.-Y. Lee et al. Compound loss of muscleblind-like function in myotonic dystrophy. EMBO molecular medicine, 5(12):1887– 1900, 12 2013. PMID: 24293317 PMCID: PMC3914532. [195] M. Nakamori et al. Splicing biomarkers of disease severity in myotonic dystrophy. Annals of Neurology, 74(6):862–872, 12 2013. PMID: 23929620 PMCID: PMC4099006. [196] A. Wojtkowiak-Szlachcic et al. Short antisense-locked nucleic acids (all-lnas) correct alternative splicing abnormalities in myotonic dystrophy. Nucleic Acids Research, 43(6):3318–3331, 3 2015. PMID: 25753670 PMCID: PMC4381072. [197] W. Dansithong et al. Mbnl1 is the primary determinant of focus formation and aberrant insulin receptor splicing in dm1. The Journal of Biological Chemistry, 280(7):5773–5780, 2 2005. PMID: 15546872. [198] R. N. Kanadia et al. A muscleblind knockout model for myotonic dystrophy. Science (New York, N.Y.), 302(5652):1978–1980, 12 2003. PMID: 14671308. [199] M. Nakamori et al. Aberrant myokine signaling in congenital myotonic dystrophy. Cell Reports, 21(5):1240–1252, 10 2017. PMID: 29091763 PMCID: PMC5689469. [200] H. Du et al. Aberrant alternative splicing and extracellular matrix gene expression in mouse models of myotonic dystrophy. Nature Structural & Molecular Biology, 17(2):187–193, 2 2010. PMID: 20098426 PMCID: PMC2852634. Bibliography 127 [201] M. Hao et al. Muscleblind-like 2 (mbnl2) -deficient mice as a model for myotonic dystrophy. Developmental Dynamics: An Official Publication of the American Association of Anatomists, 237(2):403–410, 2 2008. PMID: 18213585. [202] J. E. Morgan et al. Necroptosis mediates myofibre death in dystrophin-deficient mice. Nature Communications, 9(1):3655, 2018. PMID: 30194302 PMCID: PMC6128848. [203] M. Wehling-Henricks et al. Klotho gene silencing promotes pathology in the mdx mouse model of duchenne muscular dystrophy. Human Molecular Genetics, 25(12):2465–2482, 2016. PMID: 27154199 PMCID: PMC5181628. [204] M. Wehling-Henricks et al. Macrophages escape klotho gene silencing in the mdx mouse model of duchenne muscular dystrophy and promote muscle growth and increase satellite cell numbers through a klotho-mediated pathway. Human Molecular Genetics, 27(1):14–29, 2018. PMID: 29040534 PMCID: PMC5886268. [205] G. T. Corbett et al. Prp is a central player in toxicity mediated by soluble aggregates of neurodegeneration-causing proteins. Acta Neuropathologica, 139(3):503–526, 3 2020. [206] E. Kabashi et al. Tardbp mutations in individuals with sporadic and familial amyotrophic lateral sclerosis. Nature Genetics, 40(5):572–574, 5 2008. PMID: 18372902. [207] Z. Simmons. What’s in the literature? Journal of Clinical Neuromuscular Disease, 10(4):202–207, 6 2009. PMID: 19494733. [208] C. A. Boucher et al. A novel homeodomain-encoding gene is associated with a large cpg island interrupted by the myotonic dystrophy unstable (ctg)n repeat. Human Molecular Genetics, 4(10):1919–1925, 10 1995. PMID: 8595416. 128 Bibliography [209] S. K. Heath et al. Characterisation of expression of mdmahp, a homeodomain-encoding gene at the murine dm locus. Human Molecular Genetics, 6(5):651–657, 5 1997. PMID: 9158137. [210] J. P. Orengo et al. Expanded ctg repeats within the dmpk 3’ utr causes severe skeletal muscle wasting in an inducible mouse model for myotonic dystrophy. Proceedings of the National Academy of Sciences of the United States of America, 105(7):2646–2651, 2 2008. PMID: 18272483 PMCID: PMC2268190. [211] Y. Kino et al. Mbnl and celf proteins regulate alternative splicing of the skeletal muscle chloride channel clcn1. Nucleic Acids Research, 37(19):6477–6490, 10 2009. PMID: 19720736 PMCID: PMC2770659. [212] N. Bresolin et al. Cognitive impairment in duchenne muscular dystrophy. Neuromuscular Disorders, 4(4):359–369, 7 1994. [213] J. M. Ervasti et al. Deficiency of a glycoprotein component of the dystrophin complex in dystrophic muscle. Nature, 345(6273):315–319, 5 1990. PMID: 2188135. [214] M. Naidoo and K. Anthony. Dystrophin dp71 and the neuropathophysiology of duchenne muscular dystrophy. Molecular Neurobiology, 57(3):1748–1767, 3 2020. [215] M. Thangarajh et al. Relationships between DMD mutations and neurodevelopment in dystrophinopathy. Neurology, pp. 10.1212/WNL.0000000000008363, 10 2019. [216] R. Worton. Muscular dystrophies: diseases of the dystrophin-glycoprotein complex. Science (New York, N.Y.), 270(5237):755–756, 11 1995. PMID: 7481760. [217] L. Machuca-Tzili et al. Flies deficient in muscleblind protein model features of myotonic dystrophy with altered splice forms of z-band associated transcripts. Human Genetics, 120(4):487– 499, 11 2006. PMID: 16927100. Bibliography 129 [218] X. Lin et al. Failure of mbnl1-dependent post-natal splicing transitions in myotonic dystrophy. Human Molecular Genetics, 15(13):2087–2097, 7 2006. PMID: 16717059. [219] Y. Yamashita et al. Four parameters increase the sensitivity and specificity of the exon array analysis and disclose 25 novel aberrantly spliced exons in myotonic dystrophy. Journal of Human Genetics, 57(6):368–374, 6 2012. PMID: 22513715. [220] Z. Z. Tang et al. Muscle weakness in myotonic dystrophy associated with misregulated splicing and altered gating of ca(v)1.1 calcium channel. Human Molecular Genetics, 21(6):1312–1324, 3 2012. PMID: 22140091 PMCID: PMC3284119. [221] A. Fabregat et al. Reactome pathway analysis: a highperformance in-memory approach. BMC Bioinformatics, 18(1):142, 12 2017. [222] J. Ritchie et al. Transcriptomic characterization of mri contrast with focus on the t1-w/t2-w ratio in the cerebral cortex. NeuroImage, 174:504–517, 7 2018. [223] A. Fornito et al. Bridging the gap between connectome and transcriptome. Trends in Cognitive Sciences, 23(1):34–50, 1 2019. [224] R. Romero-Garcia et al. Structural covariance networks are coupled to expression of genes enriched in supragranular layers of the human cortex. NeuroImage, 171:256–267, 5 2018. [225] R. Romero-Garcia et al. Synaptic and transcriptionally downregulated genes are associated with cortical thickness differences in autism. Molecular Psychiatry, 24(7):1053–1064, 7 2019. [226] V. Tam et al. Benefits and limitations of genome-wide association studies. Nature Reviews Genetics, 20(8):467–484, 8 2019. [227] N. Sergeant et al. Dysregulation of human brain microtubuleassociated tau mrna maturation in myotonic dystrophy type 1. 130 Bibliography Human Molecular Genetics, 10(19):2143–2155, 9 2001. PMID: 11590131. [228] P. Vermersch et al. Specific tau variants in the brains of patients with myotonic dystrophy. Neurology, 47(3):711–717, 9 1996. PMID: 8797469. [229] O. Leroy et al. Brain-specific change in alternative splicing of tau exon 6 in myotonic dystrophy type 1. Biochimica Et Biophysica Acta, 1762(4):460–467, 4 2006. PMID: 16487687. [230] C. M. Dhaenens et al. Mis-splicing of tau exon 10 in myotonic dystrophy type 1 is reproduced by overexpression of celf2 but not by mbnl1 silencing. Biochimica Et Biophysica Acta, 1812(7):732–742, 7 2011. PMID: 21439371. [231] D. Ghanem et al. Altered splicing of tau in dm1 is different from the foetal splicing process. FEBS letters, 583(4):675–679, 2 2009. PMID: 19166838. [232] R. C. Florek et al. Electrocardiographic abnormalities in patients with myotonic dystrophy. The Western Journal of Medicine, 153(1):24–27, 7 1990. PMID: 2202157 PMCID: PMC1002461. [233] P. V. Fragola et al. Cardiac involvement in myotonic dystrophy. The American Journal of Cardiology, 74(10):1070–1072, 11 1994. PMID: 7977052. [234] J. D. Rhodes et al. Activation of the innate immune response and interferon signalling in myotonic dystrophy type 1 and type 2 cataracts. Human Molecular Genetics, 21(4):852–862, 2 2012. [235] M. Nakamori et al. Aberrantly spliced -dystrobrevin alters - syntrophin binding in myotonic dystrophy type 1. Neurology, 70(9):677–685, 2 2008. Bibliography 131 [236] A. Kiuchi et al. Presenile appearance of abundant alzheimer’s neurofibrillary tangles without senile plaques in the brain in myotonic dystrophy. Acta Neuropathologica, 82(1):1–5, 6 1991. [237] O. Hern´andez-Hern´andez et al. Myotonic dystrophy ctg expansion affects synaptic vesicle proteins, neurotransmission and mouse behaviour. Brain, 136(3):957–970, 3 2013. [238] O. Hern´andez-Hern´andez et al. Synaptic protein dysregulation in myotonic dystrophy type 1: Disease neuropathogenesis beyond missplicing. Rare Diseases, 1(1):e25553, 1 2013. [239] H. Hodges et al. Autism spectrum disorder: definition, epidemiology, causes, and clinical evaluation. Translational Pediatrics, 9(S1):S55–S65, 2 2020. [240] M.-C. Lai et al. Autism. Lancet (London, England), 383(9920):896–910, 3 2014. PMID: 24074734. [241] A. Masi et al. An overview of autism spectrum disorder, heterogeneity and treatment options. Neuroscience Bulletin, 33(2):183–193, 4 2017. [242] M. H. Mehling and M. J. Tass´e. Severity of autism spectrum disorders: Current conceptualization, and transition to dsm-5. Journal of Autism and Developmental Disorders, 46(6):2000– 2016, 6 2016. [243] R. Bhandari et al. Neuropsychopathology of autism spectrum disorder: Complex interplay of genetic, epigenetic, and environmental factors. In M. M. Essa and M. W. Qoronfleh, editors, Personalized Food Intervention and Therapy for Autism Spectrum Disorder Management, volume 24, pp. 97–141. Springer International Publishing, Cham, 2020. collection-title: Advances in Neurobiology DOI: 10.1007/978-3-030-30402-7 4. [244] V. Eapen. Genetic basis of autism: is there a way forward?:. Current Opinion in Psychiatry, 24(3):226–236, 5 2011.