Advertisement

Towards Best Practices for Imaging Transcriptomics of the Human Brain

Open AccessPublished:November 04, 2022DOI:https://doi.org/10.1016/j.biopsych.2022.10.016

      Abstract

      Modern brain-wide transcriptional atlases provide unprecedented opportunities for investigating the molecular correlates of brain organization, as quantified using non-invasive neuroimaging. However, integrating neuroimaging data with transcriptomic measures is not straightforward and careful consideration is required in order to make valid inferences. In this paper we review recent work exploring how various methodological choices affect three main phases of imaging transcriptomic analyses, including (i) processing of transcriptional atlas data; (ii) relating transcriptional measures to independently derived neuroimaging phenotypes; and (iii) evaluating the functional implications of identified associations through gene enrichment analyses. Our aim is to facilitate the development of standardized and reproducible approaches for this rapidly growing field. We identify sources of methodological variability, key choices that can affect findings, and considerations for mitigating false positive and/or spurious results. Finally, we provide an overview of freely available open-source toolboxes implementing current best-practice procedures across all three analysis phases.

      Keywords

      Introduction

      Psychiatric disorders are often characterized as disorders of brain connectivity (
      • Fornito A.
      • Zalesky A.
      • Breakspear M.
      The connectomics of brain disorders.
      ,
      • Stephan K.E.
      • Friston K.J.
      • Frith C.D.
      Dysconnection in Schizophrenia: From Abnormal Synaptic Plasticity to Failures of Self-monitoring.
      ,
      • Xia C.H.
      • Ma Z.
      • Ciric R.
      • Gu S.
      • Betzel R.F.
      • Kaczkurkin A.N.
      • et al.
      Linked dimensions of psychopathology and connectivity in functional brain networks.
      ). Magnetic resonance imaging (MRI) has offered researchers a powerful tool for mapping diverse aspects of brain dysconnectivity in a range of different disorders (

      Filippi M, Basaia S, Sarasso E, Stojkovic T, Stankovic I, Fontana A, et al. Longitudinal brain connectivity changes and clinical evolution in Parkinson’s disease. Mol Psychiatry. 2021 Sep;26(9):5429–5440.

      ,
      • Fornito A.
      • Bullmore E.T.
      Connectomic Intermediate Phenotypes for Psychiatric Disorders.
      ,
      • Li S.
      • Hu N.
      • Zhang W.
      • Tao B.
      • Dai J.
      • Gong Y.
      • et al.
      Dysconnectivity of Multiple Brain Networks in Schizophrenia: A Meta-Analysis of Resting-State Functional Connectivity.
      ,
      • Mohammad-Rezazadeh I.
      • Frohlich J.
      • Loo S.K.
      • Jeste S.S.
      Brain Connectivity in Autism Spectrum Disorder.
      ,
      • Perry A.
      • Roberts G.
      • Mitchell P.B.
      • Breakspear M.
      Connectomics of bipolar disorder: a critical review, and evidence for dynamic instabilities within interoceptive networks.
      ,
      • Zeng L.L.
      • Shen H.
      • Liu L.
      • Wang L.
      • Li B.
      • Fang P.
      • et al.
      Identifying major depression using whole-brain functional connectivity: a multivariate pattern analysis.
      ), but neuroimaging, on its own, cannot shed light on the underlying molecular mechanisms of these disruptions. One fruitful way forward involves studying the genetic basis of neuroimaging-derived connectomic phenotypes. Molecular gradients play a critical role in guiding axons to their targets as the brain develops (
      • Dickson B.J.
      Molecular mechanisms of axon guidance.
      ,
      • Goodhill G.J.
      Can Molecular Gradients Wire the Brain?.
      ), and many aspects of brain structure and function, including brain connectivity, show some level of heritability (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Bellgrove M.A.
      • Fornito A.
      Where the genome meets the connectome: Understanding how genes shape human brain connectivity.
      ,
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Oldham S.
      • Tiego J.
      • Paquola C.
      • Gerring Z.
      • et al.
      Genetic influences on hub connectivity of the human connectome.
      ,
      • Elliott L.T.
      • Sharp K.
      • Alfaro-Almagro F.
      • Shi S.
      • Miller K.L.
      • Douaud G.
      • et al.
      Genome-wide association studies of brain imaging phenotypes in UK Biobank.
      ,
      • Fornito A.
      • Zalesky A.
      • Bassett D.S.
      • Meunier D.
      • Ellison-Wright I.
      • Yücel M.
      • et al.
      Genetic Influences on Cost-Efficient Organization of Human Cortical Functional Networks.
      ,

      Smit DJA, Stam CJ, Posthuma D, Boomsma DI, Geus EJC de. Heritability of “small-world” networks in the brain: A graph theoretical analysis of resting-state EEG functional connectivity. Human Brain Mapping. 2008;29(12):1368–1378.

      ,
      • Smith S.M.
      • Douaud G.
      • Chen W.
      • Hanayik T.
      • Alfaro-Almagro F.
      • Sharp K.
      • et al.
      An expanded set of genome-wide association studies of brain imaging phenotypes in UK Biobank.
      ,
      • Thompson P.M.
      • Cannon T.D.
      • Narr K.L.
      • van Erp T.
      • Poutanen V.P.
      • Huttunen M.
      • et al.
      Genetic influences on brain structure.
      ,
      • Thompson P.M.
      • Ge T.
      • Glahn D.C.
      • Jahanshad N.
      • Nichols T.E.
      Genetics of the connectome.
      ). Characterizing genetic drivers of connectomic phenotypes implicated in psychiatric disorders may thus offer a window into the complex molecular mechanisms that underpin brain organization.
      In this context, the dominant approach in imaging genetics has involved relating phenotypic variability in some connectomic or other phenotype to structural variation of the DNA through genetic association studies (
      • Elliott L.T.
      • Sharp K.
      • Alfaro-Almagro F.
      • Shi S.
      • Miller K.L.
      • Douaud G.
      • et al.
      Genome-wide association studies of brain imaging phenotypes in UK Biobank.
      ,
      • Smith S.M.
      • Douaud G.
      • Chen W.
      • Hanayik T.
      • Alfaro-Almagro F.
      • Sharp K.
      • et al.
      An expanded set of genome-wide association studies of brain imaging phenotypes in UK Biobank.
      ,
      • Brouwer R.M.
      • Klein M.
      • Grasby K.L.
      • Schnack H.G.
      • Jahanshad N.
      • Teeuw J.
      • et al.
      Genetic variants associated with longitudinal changes in brain structure across the lifespan.
      ,
      • Grasby K.L.
      • Jahanshad N.
      • Painter J.N.
      • Colodro-Conde L.
      • Bralten J.
      • Hibar D.P.
      • et al.
      The genetic architecture of the human cerebral cortex.
      ,
      • Hibar D.P.
      • Stein J.L.
      • Renteria M.E.
      • Arias-Vasquez A.
      • Desrivières S.
      • Jahanshad N.
      • et al.
      Common genetic variants influence human subcortical brain structures.
      ). While this approach can be useful to identify single nucleotide polymorphisms (SNPs) related to the phenotype, there is no guarantee that the associated variant plays a causal role in driving phenotypic variability as it may be in linkage disequilibrium with the actual causal variants (
      • Hormozdiari F.
      • Kichaev G.
      • Yang W.Y.
      • Pasaniuc B.
      • Eskin E.
      Identification of causal genes for complex traits.
      ,
      • Schaid D.J.
      • Chen W.
      • Larson N.B.
      From genome-wide associations to candidate causal variants by statistical fine-mapping.
      ). In many cases, it can also be difficult to know the molecular effects of the variant on gene expression without further investigation. While studies directly assaying transcriptional activity may shed light on molecular processes more proximally linked to pathophysiology, they have historically only been applied to relatively small patches of tissue, making it difficult to relate such measures to neuroimaging phenotypes (

      Johnson MB, Kawasawa YI, Mason CE, Krsnik Z, Coppola G, Bogdanović D, et al. Functional and evolutionary insights into human brain development through global transcriptome analysis. Neuron. 2009 May 28;62(4):494–509.

      ,
      • Kang H.J.
      • Kawasawa Y.I.
      • Cheng F.
      • Zhu Y.
      • Xu X.
      • Li M.
      • et al.
      Spatio-temporal transcriptome of the human brain.
      ,

      Keil JM, Qalieh A, Kwan KY. Brain Transcriptome Databases: A User’s Guide. J Neurosci. 2018 Mar 7;38(10):2399–2412.

      ,
      • Khaitovich P.
      • Muetzel B.
      • She X.
      • Lachmann M.
      • Hellmann I.
      • Dietzsch J.
      • et al.
      Regional Patterns of Gene Expression in Human and Chimpanzee Brains.
      ).
      The past decade has witnessed major advances in high-throughput tissue processing and analysis, making it possible to generate anatomically comprehensive transcriptomic atlases of the brain (
      • Hawrylycz M.J.
      • Lein E.S.
      • Guillozet-Bongaarts A.L.
      • Shen E.H.
      • Ng L.
      • Miller J.A.
      • et al.
      An anatomically comprehensive atlas of the adult human brain transcriptome.
      ,
      • Lein E.S.
      • Hawrylycz M.J.
      • Ao N.
      • Ayres M.
      • Bensinger A.
      • Bernard A.
      • et al.
      Genome-wide atlas of gene expression in the adult mouse brain.
      ,
      • Miller J.A.
      • Ding S.L.
      • Sunkin S.M.
      • Smith K.A.
      • Ng L.
      • Szafer A.
      • et al.
      Transcriptional landscape of the prenatal human brain.
      ). These atlases, comprising measures of the transcriptional activity of almost the entire genome quantified in many thousands of tissue samples located throughout the brain (
      • Hawrylycz M.J.
      • Lein E.S.
      • Guillozet-Bongaarts A.L.
      • Shen E.H.
      • Ng L.
      • Miller J.A.
      • et al.
      An anatomically comprehensive atlas of the adult human brain transcriptome.
      ), have opened new possibilities for identifying genes with spatial expression patterns that correlate with neuroimaging phenotypes, giving rise to the burgeoning field of imaging transcriptomics (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ,
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Bellgrove M.A.
      • Fornito A.
      Imaging Transcriptomics of Brain Disorders.
      ,

      Fornito A, Arnatkevičiūtė A, Fulcher BD. Bridging the Gap between Connectome and Transcriptome. Trends in Cognitive Sciences. 2019 Jan 1;23(1):34–50.

      ). Such analyses have already identified transcriptional correlates of a diverse range of structural and functional properties of the brain, including: inter-regional connectivity (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Oldham S.
      • Tiego J.
      • Paquola C.
      • Gerring Z.
      • et al.
      Genetic influences on hub connectivity of the human connectome.
      ,
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Pocock R.
      • Fornito A.
      Hub connectivity, neuronal diversity, and gene expression in the Caenorhabditis elegans connectome.
      ,
      • Fulcher B.D.
      • Fornito A.
      A transcriptional signature of hub connectivity in the mouse connectome.
      ,
      • Goel P.
      • Kuceyeski A.
      • LoCastro E.
      • Raj A.
      Spatial patterns of genome-wide expression profiles reflect anatomic and fiber connectivity architecture of healthy human brain.
      ,
      • Romero-Garcia R.
      • Whitaker K.J.
      • Váša F.
      • Seidlitz J.
      • Shinn M.
      • Fonagy P.
      • et al.
      Structural covariance networks are coupled to expression of genes enriched in supragranular layers of the human cortex.
      ,
      • Whitaker K.J.
      • Vértes P.E.
      • Romero-Garcia R.
      • Váša F.
      • Moutoussis M.
      • Prabhu G.
      • et al.
      Adolescence is associated with genomically patterned consolidation of the hubs of the human brain connectome.
      ); the classical sensory-fugal cortical hierarchy (

      Burt JB, Demirtaş M, Eckner WJ, Navejar NM, Ji JL, Martin WJ, et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nat Neurosci. 2018;21(9):1251–1259.

      ); canonical large-scale networks (
      • Anderson K.M.
      • Krienen F.M.
      • Choi E.Y.
      • Reinen J.M.
      • Yeo B.T.T.
      • Holmes A.J.
      Gene expression links functional networks across cortex and striatum.
      ,
      • Richiardi J.
      • Altmann A.
      • Milazzo A.C.
      • Chang C.
      • Chakravarty M.M.
      • Banaschewski T.
      • et al.
      Correlated gene expression supports synchronous activity in brain networks.
      ) and their topological (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Oldham S.
      • Tiego J.
      • Paquola C.
      • Gerring Z.
      • et al.
      Genetic influences on hub connectivity of the human connectome.
      ,
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Pocock R.
      • Fornito A.
      Hub connectivity, neuronal diversity, and gene expression in the Caenorhabditis elegans connectome.
      ,
      • Fulcher B.D.
      • Fornito A.
      A transcriptional signature of hub connectivity in the mouse connectome.
      ,
      • Whitaker K.J.
      • Vértes P.E.
      • Romero-Garcia R.
      • Váša F.
      • Moutoussis M.
      • Prabhu G.
      • et al.
      Adolescence is associated with genomically patterned consolidation of the hubs of the human brain connectome.
      ,
      • Rubinov M.
      • Ypma R.J.F.
      • Watson C.
      • Bullmore E.T.
      Wiring cost and topological participation of the mouse brain connectome.
      ,

      Vértes PE, Rittman T, Whitaker KJ, Romero-Garcia R, Váša F, Kitzbichler MG, et al. Gene transcription profiles associated with inter-modular hubs and connection distance in human functional magnetic resonance imaging networks. Philos Trans R Soc Lond B Biol Sci. 2016 Oct 5;371(1705).

      ) and temporal (
      • Cioli C.
      • Abdi H.
      • Beaton D.
      • Burnod Y.
      • Mesmoudi S.
      Differences in Human Cortical Gene Expression Match the Temporal Properties of Large-Scale Functional Networks.
      ) properties; and spatial gradients of intrinsic dynamics (
      • Shafiei G.
      • Markello R.D.
      • Vos de Wael R.
      • Bernhardt B.C.
      • Fulcher B.D.
      • Misic B.
      Topographic gradients of intrinsic dynamics across neocortex.
      ). Imaging transcriptomics has also identified the transcriptional correlates of brain dysconnectivity in diverse disorders (
      • Anderson K.M.
      • Collins M.A.
      • Chin R.
      • Ge T.
      • Rosenberg M.D.
      • Holmes A.J.
      Transcriptional and imaging-genetic association of cortical interneurons, brain function, and schizophrenia risk.
      ,
      • Anderson K.M.
      • Collins M.A.
      • Kong R.
      • Fang K.
      • Li J.
      • He T.
      • et al.
      Convergent molecular, cellular, and cortical neuroimaging signatures of major depressive disorder.
      ,

      McColgan P, Gregory S, Seunarine KK, Razi A, Papoutsi M, Johnson E, et al. Brain Regions Showing White Matter Loss in Huntington’s Disease Are Enriched for Synaptic and Metabolic Genes. Biol Psychiatry. 2018 Mar 1;83(5):456–465.

      ,
      • Rittman T.
      • Rubinov M.
      • Vértes P.E.
      • Patel A.X.
      • Ginestet C.E.
      • Ghosh B.C.P.
      • et al.
      Regional expression of the MAPT gene is associated with loss of hubs in brain networks and cognitive impairment in Parkinson disease and progressive supranuclear palsy.
      ,
      • Romme I.A.C.
      • de Reus M.A.
      • Ophoff R.A.
      • Kahn R.S.
      • van den Heuvel M.P.
      Connectome Disconnectivity and Cortical Gene Expression in Patients With Schizophrenia.
      ,
      • Seidlitz J.
      • Nadig A.
      • Liu S.
      • Bethlehem R.A.I.
      • Vértes P.E.
      • Morgan S.E.
      • et al.
      Transcriptomic and cellular decoding of regional brain vulnerability to neurogenetic disorders.
      ) [reviewed in (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Bellgrove M.A.
      • Fornito A.
      Imaging Transcriptomics of Brain Disorders.
      )], and uncovered the putative cellular correlates of brain changes observed in disorders (
      • Patel Yash
      Autism Spectrum Disorder, Bipolar Disorder, Major Depressive Disorder, Obsessive-Compulsive Disorder, and Schizophrenia ENIGMA Working Groups, et al.
      ), and during development and ageing (
      • Vidal-Pineiro D.
      • Parker N.
      • Shin J.
      • French L.
      • Grydeland H.
      • Jackowski A.P.
      • et al.
      Cellular correlates of cortical thinning throughout the lifespan.
      ). Moreover, transcriptional data has informed dynamic models of brain function (

      Deco G, Kringelbach ML, Arnatkeviciute A, Oldham S, Sabaroedin K, Rogasch NC, et al. Dynamical consequences of regional heterogeneity in the brain’s transcriptional landscape. Science Advances. 2021 Jul 1;7(29):eabf4752.

      ) and improved our understanding of disease propagation in neurodegeneration (

      Freeze B, Pandya S, Zeighami Y, Raj A. Regional transcriptional architecture of Parkinson’s disease pathogenesis and network spread. Brain. 2019 Oct 1;142(10):3072–3085.

      ,
      • Zheng Y.Q.
      • Zhang Y.
      • Yau Y.
      • Zeighami Y.
      • Larcher K.
      • Misic B.
      • et al.
      Local vulnerability and global connectivity jointly shape neurodegenerative disease propagation.
      ,

      Yan W, Ye C, Wang T, Sun J, Wu T, Ma T. Misfolded protein propagation in an integrated computational model of structural network and LRRK2 gene expression. In: 2020 42nd Annual International Conference of the IEEE Engineering in Medicine Biology Society (EMBC). 2020. p. 2368–2371.

      ).
      Despite the promise and rapid uptake of imaging transcriptomics [for reviews, see (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Bellgrove M.A.
      • Fornito A.
      Imaging Transcriptomics of Brain Disorders.
      ,

      Fornito A, Arnatkevičiūtė A, Fulcher BD. Bridging the Gap between Connectome and Transcriptome. Trends in Cognitive Sciences. 2019 Jan 1;23(1):34–50.

      )], the integration of neuroimaging with brain-wide transcriptomic-atlas data depends on numerous data-processing and analysis choices that can affect the final results. Different investigators have historically used their own bespoke analysis pipelines, making it difficult to determine the effect of such choices on reported findings. The development of best-practice workflows is therefore essential for the field to advance in a valid and reproducible way. In this review, we examine recent work exploring the impact of various processing options and attempts to develop easy-to-use pipelines that implement optimal processing choices (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ,
      • Fulcher B.D.
      • Arnatkeviciute A.
      • Fornito A.
      Overcoming false-positive gene-category enrichment in the analysis of spatially resolved transcriptomic brain atlas data.
      ,

      Markello RD, Arnatkevičiūtė A, Poline JB, Fulcher BD, Fornito A, Misic B. Standardizing workflows in imaging transcriptomics with the abagen toolbox. bioRxiv. 2021 Jul 9;2021.07.08.451635.

      ), thus building best-practice workflows for imaging transcriptomics. We outline some of the key steps in such analyses, highlight issues for careful consideration, and recommend optimal choices where they have been proposed. We focus our discussion on three key phases of any imaging transcriptomic analysis: (i) processing the transcriptional atlas data; (ii) relating the expression measures to a neuroimaging phenotype; and (iii) evaluating gene specificity and enrichment. We hope to facilitate the development of standardized processing and analysis approaches in the field, thereby facilitating comparison across studies and promoting valid and accurate inference.

      Phase 1: Processing transcriptional atlas data

      Imaging transcriptomic studies rely on brain-wide transcriptional atlases quantifying the expression of thousands of genes across multiple locations in the brain (
      • Hawrylycz M.J.
      • Lein E.S.
      • Guillozet-Bongaarts A.L.
      • Shen E.H.
      • Ng L.
      • Miller J.A.
      • et al.
      An anatomically comprehensive atlas of the adult human brain transcriptome.
      ,
      • Lein E.S.
      • Hawrylycz M.J.
      • Ao N.
      • Ayres M.
      • Bensinger A.
      • Bernard A.
      • et al.
      Genome-wide atlas of gene expression in the adult mouse brain.
      ). The methods for measuring transcriptional activity depend on a range of factors including species, desired spatial resolution, and tissue availability. The limited availability of human brain tissue means that bulk tissue microarray (

      Schulze A, Downward J. Navigating gene expression using microarrays — a technology review. Nature Cell Biology. 2001 Aug 1;3(8):E190–E195.

      ) remains the most accessible method for high-throughput spatial transcriptomics (
      • Hawrylycz M.J.
      • Lein E.S.
      • Guillozet-Bongaarts A.L.
      • Shen E.H.
      • Ng L.
      • Miller J.A.
      • et al.
      An anatomically comprehensive atlas of the adult human brain transcriptome.
      ) compared to other methods with higher spatial resolution such as single-cell RNA-seq or in situ hybridization (ISH) (see Supplementary Text S1). The Allen Human Brain Atlas (AHBA) provides an anatomically comprehensive transcriptional map of the human brain, quantifying the expression of more than 20,000 genes across 3,702 anatomical locations from six post-mortem brains (
      • Hawrylycz M.J.
      • Lein E.S.
      • Guillozet-Bongaarts A.L.
      • Shen E.H.
      • Ng L.
      • Miller J.A.
      • et al.
      An anatomically comprehensive atlas of the adult human brain transcriptome.
      ). As the donor brains were scanned with T1-weighted magnetic resonance imaging (MRI) and the scans were normalized to a standardized space, it is possible to directly compare the spatial patterning of gene expression and neuroimaging data. However, since the anatomical locations in the AHBA are sampled using spatially distributed small patches of tissue, the simplest way to achieve this spatial mapping is to apply some regional parcellation to both. In this way, summary measures of gene expression within a given brain region can be related to imaging measures quantified in the same region.
      A primary task in AHBA-like data is to construct an N×G matrix, where N is the number of parcellated brain regions and G is the number of genes assayed. The key steps of a typical workflow for obtaining this regional gene expression matrix are outlined in Figure 1. Each step requires the investigator to make choices that can affect the final results. The impact of these various choices is discussed in detail in (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ) (also see Supplementary Text S2). Here, we summarize the key aspects and recommendations for each step.
      Figure thumbnail gr1
      Figure 1Schematic representing AHBA processing steps for aggregating data into a region-by-gene matrix for further analyses. The workflow involves: Step1 - Verifying probe-to-gene annotations, where specific probe sequences are mapped to a gene; Step 2 - Filtering probes that do not exceed the background, where probes are removed if transcriptional measures in a certain proportion of samples are lower than background (each vertical stripe in the schematic correspond to a tissue sample); Step 3 - Selecting a representative probe for a gene when multiple probes for the same gene are available; Step 4 - Assigning tissue samples to parcellated brain regions (circles represent a schematic visualization of tissue samples); Step 5 - Accounting for inter-donor differences in gene expression using normalization; prior to normalization data from different donors (represented in different colors) occupy different parts of the low-dimensional principal component space. After normalization samples no longer segregate by donor; Step 6 - Selecting most relevant genes for the analysis. The example here demonstrates expression measures for two genes that show consistent (top) and inconsistent (bottom) expression across donor brains characterized by high and low differential stability (DS) respectively. After processing, data are rendered into a region-by-gene matrix to be used for further analyses. An important consideration for the analyses is accounting for spatial autocorrelation in gene expression. Figure adapted with permission from (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ).
      Step 1: Verify probe-to-gene annotations. The microarray data quantify gene expression using probe sequences that correspond to the unique portion of the DNA comprising an individual gene (
      • Liu H.
      • Bebu I.
      • Li X.
      Microarray probes and probe sets.
      ). The assignment of probes to genes is performed using available sequencing databases that are continuously updated (
      • O’Leary N.A.
      • Wright M.W.
      • Brister J.R.
      • Ciufo S.
      • Haddad D.
      • McVeigh R.
      • et al.
      Reference sequence (RefSeq) database at NCBI: current status, taxonomic expansion, and functional annotation.
      ). It is therefore essential to use the latest and most accurate probe-to-gene mapping to ensure the validity of the measures.
      Step 2: Filter probes. Microarray experiments are prone to background noise due, in part, to non-specific hybridization (
      • Liu H.
      • Bebu I.
      • Li X.
      Microarray probes and probe sets.
      ). It is therefore essential to remove noisy probes with expression levels that do not exceed the background to improve the validity of the microarray measures. Based on our analyses, retaining probes where expression levels exceed background levels in at least 50% of samples provides an appropriate balance for reducing noise while maintaining around 70% of all available probes, thus ensuring high anatomical coverage (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ).
      Step 3: Select representative probes. More than 90% of genes in the original AHBA data have more than one probe available and not all probes show consistent expression patterns. A single representative probe or a summary measure across multiple probes should be selected to reduce data complexity and aid interpretation. Several methods have been implemented in the literature (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ). We recommend selecting probes based on their correlation with RNA-seq measures collected in two (out of six) donor brains, as it provides the closest approximation to a ground-truth measurement. When this is not possible, we recommend using probes with the highest differential stability (
      • Hawrylycz M.
      • Miller J.A.
      • Menon V.
      • Feng D.
      • Dolbeare T.
      • Guillozet-Bongaarts A.L.
      • et al.
      Canonical genetic signatures of the adult human brain.
      ) which measures the consistency of a gene’s expression profile across donor brains [for a detailed rationale, see (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      )].
      Step 4: Assign tissue samples to parcellated brain regions. Each tissue sample is characterized by stereotactic coordinates and anatomical labels that can be used to map the corresponding expression measured to a region in the selected brain parcellation. We suggest mapping tissue samples to regions separately based on their broad anatomical locations (cortex/subcortex) and assigning a sample to the closest region in the parcellation (rather than the centroid of that region), while applying a 2 mm distance threshold from the original sample location to the parcellation to avoid inaccurate mapping of samples that are located too far from the region.
      Step 5: Normalize the expression measures. Since the AHBA data were collected from six donor brains, any analysis that combines samples across brains to obtain an anatomically comprehensive map must account for individual differences in donor brain expression. Normalization procedures performed by the AHBA team prior to the release of the data removed batch effects and artefactual inter-individual variation (

      Allen Institute for Brain Science. Technical white paper: microarray data normalization [Internet]. 2013. Available from: http://help.brain-map.org/display/humanbrain/Documentation

      ), but still leave substantial residual inter-individual differences (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ) (see Supplementary Text S3). Performing an additional z-score or scaled robust sigmoid (
      • Fulcher B.D.
      • Little M.A.
      • Jones N.S.
      Highly comparative time-series analysis: the empirical structure of time series and their methods.
      ) normalization step within each donor brain across regions can be used to remove this residual variability and to minimize the influence of outlying values (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ).
      Step 6: Select genes with consistent expression patterns across donor brains. Only a fraction of the > 20 000 genes in the AHBA demonstrate consistent regional variations across different donor brains, as quantified using the differential stability measure (
      • Hawrylycz M.
      • Miller J.A.
      • Menon V.
      • Feng D.
      • Dolbeare T.
      • Guillozet-Bongaarts A.L.
      • et al.
      Canonical genetic signatures of the adult human brain.
      ). Identifying those consistently expressed genes that show reproducible variation across the brain or genes that are known to be brain-expressed (

      Burt JB, Demirtaş M, Eckner WJ, Navejar NM, Ji JL, Martin WJ, et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nat Neurosci. 2018;21(9):1251–1259.

      ,
      • Fagerberg L.
      • Hallström B.M.
      • Oksvold P.
      • Kampf C.
      • Djureinovic D.
      • Odeberg J.
      • et al.
      Analysis of the Human Tissue-specific Expression by Genome-wide Integration of Transcriptomics and Antibody-based Proteomics.
      ,
      • Genovese G.
      • Fromer M.
      • Stahl E.A.
      • Ruderfer D.M.
      • Chambert K.
      • Landén M.
      • et al.
      Increased burden of ultra-rare protein-altering variants among 4,877 individuals with schizophrenia.
      ) can provide a more targeted approach for the investigation of relationships with neuroimaging phenotypes.
      The multitude of choices that can be made at each of the above six steps results in a combinatorial explosion of possible pipelines that an investigator may choose. One recent study (

      Markello RD, Arnatkeviciute A, Poline JB, Fulcher BD, Fornito A, Misic B. Standardizing workflows in imaging transcriptomics with the abagen toolbox. Jbabdi S, Makin TR, Jbabdi S, Burt J, Hawrylycz MJ, editors. eLife. 2021 Nov 16;10:e72129.

      ) extensively explored the effects of 17 decision points in the above workflow, resulting in 746,496 distinct processing pipelines. The authors used this comprehensive analysis to identify the key processing choices that affect the final results, as quantified using three outcome metrics based on three commonly used types of analysis: gene co-expression (the similarity between genes across regions), regional gene expression (spatial expression patterns of specific genes or gene sets), and correlated gene expression (the similarity between regional expression profiles across genes; all measures discussed in detail below; also see Figure 3). They found that data-processing choices involving gene normalization (Step 5) had the largest impact (Figure 2), followed by choices related to mapping tissue samples to brain regions (Step 4). The least impactful choices were related to probe selection (Step 3, Figure 2A). By combining this analysis with prior results (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ), the authors developed a recommended data-processing workflow implemented in an open source abagen toolbox (

      Markello R, Shafiei G, Zheng YQ, Mišić B. abagen: A toolbox for the Allen Brain Atlas genetics data [Internet]. Zenodo; 2021. Available from: Available from: https://doi.org/10.5281/zenodo.4646030

      ). The abagen toolbox also generates a standardized report of the selected processing options to improve transparent reporting in publications.
      Figure thumbnail gr3
      Figure 3Types of transcriptional analyses. Cortical tissue samples aggregated across six donor brains comprising AHBA (left) and the schematic representation of the connectome (right). Each sample is characterised with the expression measures across thousands of genes. Data can be aggregated into a region-by-gene expression matrix. A) Gene coexpression – correlation between regional gene expression vectors (columns in the expression matrix) resulting in a gene-by-gene similarity matrix for each pair of genes. B) Regional gene expression – associations between the spatial expression profile of each gene and regional variations in some brain property are evaluated using univariate or multivariate techniques such as partial least squares (PLS). C) Correlated gene expression (CGE) – correlation between gene expression vectors for each region (rows in the expression matrix) resulting in a region-by-region similarity matrix that can be related to pairwise measures of brain structure of function. Adapted with permission from (

      Fornito A, Arnatkevičiūtė A, Fulcher BD. Bridging the Gap between Connectome and Transcriptome. Trends in Cognitive Sciences. 2019 Jan 1;23(1):34–50.

      ).
      Figure thumbnail gr2
      Figure 2Impact of data-processing options. Panel A demonstrates the rank of the relative importance for each parameter across three analyses types. Lighter colors indicate parameters that have stronger influence of the statistical estimates. The parameter ranking for each analysis type was performed independently by calculating a distribution of difference scores in the respective statistical estimates for each parameter options while holding all other parameters constant (CGE – correlated gene expression; GCE – gene co-expression; RGE – regional gene expression). Panel B shows the distributions of statistical estimates for each of the three analysis types as kernel density plots based on the choice of gene-normalization method (most important parameter in panel A). In the case of correlated gene expression, negative values indicate a stronger spatial relationship, such that regions with higher CGE are separated by shorter physical distances. A more negative silhouette score for gene co-expression reflects poorer clustering performance. Higher correlation values for regional gene expression analyses indicate stronger associations with the neuroimaging phenotype. Reproduced with permission from (

      Markello RD, Arnatkeviciute A, Poline JB, Fulcher BD, Fornito A, Misic B. Standardizing workflows in imaging transcriptomics with the abagen toolbox. Jbabdi S, Makin TR, Jbabdi S, Burt J, Hawrylycz MJ, editors. eLife. 2021 Nov 16;10:e72129.

      ).

      Phase 2: Relating expression and neuroimaging measures

      Having converted the transcriptional data into a region-by-gene expression matrix, the next step involves relating these measures to some neuroimaging phenotype (see also Supplementary text S2). In this context, the expression data have commonly been summarized using one of three main approaches (

      Fornito A, Arnatkevičiūtė A, Fulcher BD. Bridging the Gap between Connectome and Transcriptome. Trends in Cognitive Sciences. 2019 Jan 1;23(1):34–50.

      ,

      Markello RD, Arnatkeviciute A, Poline JB, Fulcher BD, Fornito A, Misic B. Standardizing workflows in imaging transcriptomics with the abagen toolbox. Jbabdi S, Makin TR, Jbabdi S, Burt J, Hawrylycz MJ, editors. eLife. 2021 Nov 16;10:e72129.

      ) (Figure 3). The first is gene co-expression analysis, which involves analyzing spatial correlations between the expression patterns of pairs of genes (across brain regions). For all pairs of genes, the result can be represented as a (symmetric) gene-by-gene matrix to analyze the patterns of gene expression similarity (Figure 3A). This matrix can also be summarized using some aggregate value (e.g., an eigenvector of the matrix or its subcomponents, sometimes called an eigengene), yielding a highly explanatory component as a spatial map that can be linked to imaging data (
      • Forest M.
      • Iturria‐Medina Y.
      • Goldman J.S.
      • Kleinman C.L.
      • Lovato A.
      • Klein K.O.
      • et al.
      Gene networks show associations with seed region connectivity.
      ,
      • Oldham M.C.
      • Konopka G.
      • Iwamoto K.
      • Langfelder P.
      • Kato T.
      • Horvath S.
      • et al.
      Functional organization of the transcriptome in human brain.
      ,
      • Shen J.
      • Yang B.
      • Xie Z.
      • Wu H.
      • Zheng Z.
      • Wang J.
      • et al.
      Cell-Type-Specific Gene Modules Related to the Regional Homogeneity of Spontaneous Brain Activity and Their Associations With Common Brain Disorders.
      ,

      Zarkali A, McColgan P, Ryten M, Reynolds R, Leyland LA, Lees AJ, et al. Differences in network controllability and regional gene expression underlie hallucinations in Parkinson’s disease. Brain. 2020 Dec 5;143(11):3435–3448.

      ). The second type of analysis focuses on regional gene expression, where spatial correlations for selected genes or gene groups are evaluated with respect to a neuroimaging measure defined at each brain region (
      • Anderson K.M.
      • Collins M.A.
      • Kong R.
      • Fang K.
      • Li J.
      • He T.
      • et al.
      Convergent molecular, cellular, and cortical neuroimaging signatures of major depressive disorder.
      ,

      Hess JL, Radonjić NV, Patak J, Glatt SJ, Faraone SV. Autophagy, apoptosis, and neurodevelopmental genes might underlie selective brain region vulnerability in attention-deficit/hyperactivity disorder. Mol Psychiatry. 2020 Dec 18;

      ). The same principles apply to multivariate analyses such as those using partial least squares (PLS) (
      • Krishnan A.
      • Williams L.J.
      • McIntosh A.R.
      • Abdi H.
      Partial Least Squares (PLS) methods for neuroimaging: A tutorial and review.
      ), which identify weighted combinations of genes and imaging measures with maximal covariance (Figure 3B, for additional considerations regarding multivariate analysis methods, see (

      Bilenko NY, Gallant JL. Pyrcca: Regularized Kernel Canonical Correlation Analysis in Python and Its Applications to Neuroimaging. Frontiers in Neuroinformatics [Internet]. 2016 [cited 2022 Sep 28];10. Available from: https://www.frontiersin.org/articles/10.3389/fninf.2016.00049

      ,

      Helmer M, Warrington S, Mohammadi-Nejad AR, Ji JL, Howell A, Rosand B, et al. On stability of Canonical Correlation Analysis and Partial Least Squares with application to brain-behavior associations [Internet]. bioRxiv; 2021 [cited 2022 Sep 28]. p. 2020.08.25.265546. Available from: https://www.biorxiv.org/content/10.1101/2020.08.25.265546v3

      ,
      • Wang H.T.
      • Smallwood J.
      • Mourao-Miranda J.
      • Xia C.H.
      • Satterthwaite T.D.
      • Bassett D.S.
      • et al.
      Finding the needle in a high-dimensional haystack: Canonical correlation analysis for neuroscientists.
      )). The third class of analysis examines correlated gene expression (CGE), in which correlations are computed between all pairs of brain regions, quantifying the similarity of their gene-expression profiles. The result can be represented as a (symmetric) region-by-region matrix and compared directly to other types of data measured at the level of pairs of regions, like structural or functional connectivity (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Oldham S.
      • Tiego J.
      • Paquola C.
      • Gerring Z.
      • et al.
      Genetic influences on hub connectivity of the human connectome.
      ,
      • Fulcher B.D.
      • Fornito A.
      A transcriptional signature of hub connectivity in the mouse connectome.
      ,
      • Richiardi J.
      • Altmann A.
      • Milazzo A.C.
      • Chang C.
      • Chakravarty M.M.
      • Banaschewski T.
      • et al.
      Correlated gene expression supports synchronous activity in brain networks.
      ) (Figure 3C).
      Most studies in the field have either relied on analysis of regional gene expression or CGE estimates and have related these to neuroimaging measures using spatial (mass) univariate (
      • Anderson K.M.
      • Collins M.A.
      • Kong R.
      • Fang K.
      • Li J.
      • He T.
      • et al.
      Convergent molecular, cellular, and cortical neuroimaging signatures of major depressive disorder.
      ,
      • Hess J.L.
      • Akutagava-Martins G.C.
      • Patak J.D.
      • Glatt S.J.
      • Faraone S.V.
      Why is there selective subcortical vulnerability in ADHD? Clues from postmortem brain gene expression data.
      ,
      • Xie Y.
      • Zhang X.
      • Liu F.
      • Qin W.
      • Fu J.
      • Xue K.
      • et al.
      Brain mRNA Expression Associated with Cortical Volume Alterations in Autism Spectrum Disorder.
      ), connectome-wide (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Oldham S.
      • Tiego J.
      • Paquola C.
      • Gerring Z.
      • et al.
      Genetic influences on hub connectivity of the human connectome.
      ,
      • Fulcher B.D.
      • Fornito A.
      A transcriptional signature of hub connectivity in the mouse connectome.
      ), or multivariate analysis techniques (
      • Whitaker K.J.
      • Vértes P.E.
      • Romero-Garcia R.
      • Váša F.
      • Moutoussis M.
      • Prabhu G.
      • et al.
      Adolescence is associated with genomically patterned consolidation of the hubs of the human brain connectome.
      ,

      Vértes PE, Rittman T, Whitaker KJ, Romero-Garcia R, Váša F, Kitzbichler MG, et al. Gene transcription profiles associated with inter-modular hubs and connection distance in human functional magnetic resonance imaging networks. Philos Trans R Soc Lond B Biol Sci. 2016 Oct 5;371(1705).

      ,
      • Seidlitz J.
      • Nadig A.
      • Liu S.
      • Bethlehem R.A.I.
      • Vértes P.E.
      • Morgan S.E.
      • et al.
      Transcriptomic and cellular decoding of regional brain vulnerability to neurogenetic disorders.
      ,
      • Ball G.
      • Seidlitz J.
      • Beare R.
      • Seal M.L.
      Cortical remodelling in childhood is associated with genes enriched for neurodevelopmental disorders.
      ,
      • Romero-Garcia R.
      • Warrier V.
      • Bullmore E.T.
      • Baron-Cohen S.
      • Bethlehem R.A.I.
      Synaptic and transcriptionally downregulated genes are associated with cortical thickness differences in autism.
      ). The choice between the analysis approaches depends on the specific hypotheses and imaging phenotypes under investigation. For example, regional variations in neuroimaging phenotypes can be related to regional patterns of gene expression using both data-driven and hypothesis-driven approaches, whereas pairwise measures of brain connectivity derived from structural or functional data may be more naturally linked to CGE, which can capture shared transcriptional patterns between brain areas. An important property of both transcriptional and neuroimaging data is their strong spatial autocorrelation: regions in close proximity to each other tend to have more similar values than regions that are further separated in space. In the gene-expression data, the correlation between any two points decays roughly exponentially with their spatial separation (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ,
      • Fulcher B.D.
      • Fornito A.
      A transcriptional signature of hub connectivity in the mouse connectome.
      ). This exponential distance rule (EDR) in correlated gene-expression has been identified in diverse kinds of spatially resolved transcriptional data, including human cortex (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ,
      • Hawrylycz M.
      • Miller J.A.
      • Menon V.
      • Feng D.
      • Dolbeare T.
      • Guillozet-Bongaarts A.L.
      • et al.
      Canonical genetic signatures of the adult human brain.
      ), adult mouse brain (
      • Fulcher B.D.
      • Fornito A.
      A transcriptional signature of hub connectivity in the mouse connectome.
      ,
      • French L.
      • Pavlidis P.
      Relationships between Gene Expression and Brain Wiring in the Adult Rodent Brain.
      ) and across development (
      • Lau H.Y.G.
      • Fornito A.
      • Fulcher B.D.
      Scaling of gene transcriptional gradients with brain size across mouse development.
      ) and in the head of the nematode Caenorhabditis elegans (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Pocock R.
      • Fornito A.
      Hub connectivity, neuronal diversity, and gene expression in the Caenorhabditis elegans connectome.
      ). A similar EDR has been identified for connection probability (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Pocock R.
      • Fornito A.
      Hub connectivity, neuronal diversity, and gene expression in the Caenorhabditis elegans connectome.
      ,
      • Fulcher B.D.
      • Fornito A.
      A transcriptional signature of hub connectivity in the mouse connectome.
      ,

      Horvát S, Gămănuț R, Ercsey-Ravasz M, Magrou L, Gămănuț B, Essen DCV, et al. Spatial Embedding and Wiring Cost Constrain the Functional Layout of the Cortical Network of Rodents and Primates. PLOS Biology. 2016 Jul 21;14(7):e1002512.

      ,
      • Theodoni P.
      • Majka P.
      • Reser D.H.
      • Wójcik D.K.
      • Rosa M.G.P.
      • Wang X.J.
      Structural Attributes and Principles of the Neocortical Connectome in the Marmoset Monkey.
      ) and strength (

      Gămănuţ R, Kennedy H, Toroczkai Z, Ercsey-Ravasz M, Van Essen DC, Knoblauch K, et al. The Mouse Cortical Connectome, Characterized by an Ultra-Dense Cortical Graph, Maintains Specificity by Distinct Connectivity Profiles. Neuron. 2018 Feb 7;97(3):698-715.e10.

      ) in animals and for inter-regional structural connectivity in the human brain (

      Fornito A, Arnatkevičiūtė A, Fulcher BD. Bridging the Gap between Connectome and Transcriptome. Trends in Cognitive Sciences. 2019 Jan 1;23(1):34–50.

      ,
      • Roberts J.A.
      • Perry A.
      • Lord A.R.
      • Roberts G.
      • Mitchell P.B.
      • Smith R.E.
      • et al.
      The contribution of geometry to the human connectome.
      ).
      Since traditional statistical methods assume independence of observations, analyzing spatially autocorrelated (i.e., non-independent) data requires special consideration, since the autocorrelation can spuriously inflate associations between expression and imaging data. Therefore, a failure to account for spatial non-independence is likely to produce overly-optimistic estimates of the true association (
      • Alexander-Bloch A.F.
      • Shou H.
      • Liu S.
      • Satterthwaite T.D.
      • Glahn D.C.
      • Shinohara R.T.
      • et al.
      On testing for spatial correspondence between maps of human brain structure and function.
      ,
      • Burt J.B.
      • Helmer M.
      • Shinn M.
      • Anticevic A.
      • Murray J.D.
      Generative modeling of brain maps with spatial autocorrelation.
      ,
      • Markello R.D.
      • Misic B.
      Comparing spatial null models for brain maps.
      ).
      One approach to address spatial autocorrelations is to model and remove the spatial dependence of the data (e.g., via regression) and then analyze the residuals (
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Oldham S.
      • Tiego J.
      • Paquola C.
      • Gerring Z.
      • et al.
      Genetic influences on hub connectivity of the human connectome.
      ,
      • Arnatkeviciute A.
      • Fulcher B.D.
      • Fornito A.
      A practical guide to linking brain-wide gene expression and neuroimaging data.
      ,
      • Fulcher B.D.
      • Fornito A.
      A transcriptional signature of hub connectivity in the mouse connectome.
      ). This approach is well-suited to analyses of pairwise regional properties, such as CGE, but it does rest on the assumption that the modelled spatial dependence is a good approximation of the spatial autocorrelation that must be removed. An alternative approach, well-suited to regional analyses, involves the use of spatially constrained null models. These null models preserve the autocorrelation of the spatial maps, enabling inference on whether empirically observed correlations between expression and neuroimaging measures exceed expectations for two random, autocorrelated variables (Figure 4). Two broad classes of such spatially-constrained models in neuroimaging research are non-parametric spatial permutation models (
      • Alexander-Bloch A.F.
      • Shou H.
      • Liu S.
      • Satterthwaite T.D.
      • Glahn D.C.
      • Shinohara R.T.
      • et al.
      On testing for spatial correspondence between maps of human brain structure and function.
      ,
      • Alexander-Bloch A.
      • Raznahan A.
      • Bullmore E.
      • Giedd J.
      The Convergence of Maturational Change and Structural Covariance in Human Cortical Networks.
      ,
      • Gordon E.M.
      • Laumann T.O.
      • Adeyemo B.
      • Huckins J.F.
      • Kelley W.M.
      • Petersen S.E.
      Generation and Evaluation of a Cortical Area Parcellation from Resting-State Correlations.
      ,
      • Váša F.
      • Seidlitz J.
      • Romero-Garcia R.
      • Whitaker K.J.
      • Rosenthal G.
      • Vértes P.E.
      • et al.
      Adolescent Tuning of Association Cortex in Human Structural Brain Networks.
      ) and parameterized spatial models (

      Burt JB, Demirtaş M, Eckner WJ, Navejar NM, Ji JL, Martin WJ, et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nat Neurosci. 2018;21(9):1251–1259.

      ,
      • Burt J.B.
      • Helmer M.
      • Shinn M.
      • Anticevic A.
      • Murray J.D.
      Generative modeling of brain maps with spatial autocorrelation.
      ). The non-parametric models are well-suited to analyses of the cerebral cortex and leverage the fact that the cortical surface can be mapped to a sphere, allowing a simple rotation that shuffles the assignment of values to specific cortical locations while preserving the exact distance-dependence of the data. Parametrized models estimate the intrinsic spatial autocorrelation of the empirical map and use the resulting model to generate surrogate maps with randomized topography but similar spatial autocorrelation.
      Figure thumbnail gr4
      Figure 4Spatial null models. Examples of different null models – spatially naïve, spatial permutation and parameterized – applied to data with an anterior-posterior gradient. Labels under each brain represent the first appearance of the described implementation of the null model in the neuroimaging literature, if known: vázquez-rodríguez (

      Vázquez-Rodríguez B, Suárez LE, Markello RD, Shafiei G, Paquola C, Hagmann P, et al. Gradients of structure–function tethering across neocortex. Proceedings of the National Academy of Sciences. 2019 Oct 15;116(42):21219–21227.

      ), baum (

      Baum GL, Cui Z, Roalf DR, Ciric R, Betzel RF, Larsen B, et al. Development of structure–function coupling in human brain networks during youth. Proceedings of the National Academy of Sciences. 2020 Jan 7;117(1):771–778.

      ), cornblath (
      • Cornblath E.J.
      • Ashourvan A.
      • Kim J.Z.
      • Betzel R.F.
      • Ciric R.
      • Adebimpe A.
      • et al.
      Temporal sequences of brain activity at rest are constrained by white matter structure and modulated by cognitive demands.
      ), váša (
      • Váša F.
      • Seidlitz J.
      • Romero-Garcia R.
      • Whitaker K.J.
      • Rosenthal G.
      • Vértes P.E.
      • et al.
      Adolescent Tuning of Association Cortex in Human Structural Brain Networks.
      ), hungarian, burt-2018 (

      Burt JB, Demirtaş M, Eckner WJ, Navejar NM, Ji JL, Martin WJ, et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nat Neurosci. 2018;21(9):1251–1259.

      ), burt-2020 (
      • Burt J.B.
      • Helmer M.
      • Shinn M.
      • Anticevic A.
      • Murray J.D.
      Generative modeling of brain maps with spatial autocorrelation.
      ), moran (
      • Vos de Wael R.
      • Benkarim O.
      • Paquola C.
      • Lariviere S.
      • Royer J.
      • Tavakol S.
      • et al.
      BrainSpace: a toolbox for the analysis of macroscale gradients in neuroimaging and connectomics datasets.
      ). Black arrows indicate missing values due to the rotation of the medial wall into the cortical surface. Reproduced with permission from (
      • Markello R.D.
      • Misic B.
      Comparing spatial null models for brain maps.
      ).
      The main advantage of spatial permutation models is that they preserve the distance-dependence relationship of the empirical data. However, missing regions, such as the medial wall, can create problems after map rotation (i.e., the medial wall can be rotated into a cortical location), so various heuristics have been developed to address this issue, such as discarding the missing data (

      Baum GL, Cui Z, Roalf DR, Ciric R, Betzel RF, Larsen B, et al. Development of structure–function coupling in human brain networks during youth. Proceedings of the National Academy of Sciences. 2020 Jan 7;117(1):771–778.

      ,
      • Cornblath E.J.
      • Ashourvan A.
      • Kim J.Z.
      • Betzel R.F.
      • Ciric R.
      • Adebimpe A.
      • et al.
      Temporal sequences of brain activity at rest are constrained by white matter structure and modulated by cognitive demands.
      ), interpolating data for missing parcels based on nearest available regions (

      Vázquez-Rodríguez B, Suárez LE, Markello RD, Shafiei G, Paquola C, Hagmann P, et al. Gradients of structure–function tethering across neocortex. Proceedings of the National Academy of Sciences. 2019 Oct 15;116(42):21219–21227.

      ), or ignoring the medial wall (
      • Váša F.
      • Seidlitz J.
      • Romero-Garcia R.
      • Whitaker K.J.
      • Rosenthal G.
      • Vértes P.E.
      • et al.
      Adolescent Tuning of Association Cortex in Human Structural Brain Networks.
      ). Another limitation is that permutation-based methods cannot be applied to subcortical structures, which are often not adequately modelled as a 2D spherical surface. Parametrized models are not affected by missing data and can be applied equivalently to cortical and subcortical measures, but an exact matching of the empirical distance-dependence, and of the original empirical values, is not guaranteed (

      Burt JB, Demirtaş M, Eckner WJ, Navejar NM, Ji JL, Martin WJ, et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nat Neurosci. 2018;21(9):1251–1259.

      ,
      • Burt J.B.
      • Helmer M.
      • Shinn M.
      • Anticevic A.
      • Murray J.D.
      Generative modeling of brain maps with spatial autocorrelation.
      ,
      • Wagner H.H.
      • Dray S.
      Generating spatially constrained null models for irregularly spaced data using Moran spectral randomization methods.
      ). As such, analyses involving different spatial maps may be affected by variations in model fit. Comparison between various implementations of these different approaches for cortical analyses indicates that permutation-based models provide more conservative significance estimates and lower error rates when compared with parameterized models (
      • Markello R.D.
      • Misic B.
      Comparing spatial null models for brain maps.
      ). However, no single method is perfect, with error rates exceeding 40% for strongly autocorrelated data (
      • Markello R.D.
      • Misic B.
      Comparing spatial null models for brain maps.
      ), indicating that further developments are required for valid inference. An extended investigation of different null models is provided in (
      • Markello R.D.
      • Misic B.
      Comparing spatial null models for brain maps.
      ) and a toolbox for their implementation is available at https://netneurolab.github.io/neuromaps/ (

      Markello RD, Hansen JY, Liu ZQ, Bazinet V, Shafiei G, Suárez LE, et al. neuromaps: structural and functional interpretation of brain maps. bioRxiv; 2022. p. 2022.01.06.475081.

      ).

      Phase 3: Evaluating gene specificity and enrichment

      Current brain-wide transcriptional atlases quantify the expression levels of up to ∼20,000 genes (
      • Hawrylycz M.J.
      • Lein E.S.
      • Guillozet-Bongaarts A.L.
      • Shen E.H.
      • Ng L.
      • Miller J.A.
      • et al.
      An anatomically comprehensive atlas of the adult human brain transcriptome.
      ,
      • Lein E.S.
      • Hawrylycz M.J.
      • Ao N.
      • Ayres M.
      • Bensinger A.
      • Bernard A.
      • et al.
      Genome-wide atlas of gene expression in the adult mouse brain.
      ). Given the vast possible number of associations that can result from a given set of imaging phenotypes, it is critical to establish whether certain genes or gene sets are preferentially related to the phenotype of interest. Analyses can be broadly grouped as: (i) hypothesis-driven, in which a specific gene (or set of genes) of hypothetical interest is analyzed; or (ii) data-driven, where effects are computed across the whole transcriptome, after which the preferential involvement of specific genes (or functionally categorized groups of genes (
      • Ashburner M.
      • Ball C.A.
      • Blake J.A.
      • Botstein D.
      • Butler H.
      • Cherry J.M.
      • et al.
      Gene Ontology: tool for the unification of biology.
      )) is inferred (see Supplementary Text S3).
      Hypothesis-driven analyses have proven useful in understanding links between the expression profiles of putative liability genes and brain changes in different disorders [e.g., (
      • Anderson K.M.
      • Collins M.A.
      • Kong R.
      • Fang K.
      • Li J.
      • He T.
      • et al.
      Convergent molecular, cellular, and cortical neuroimaging signatures of major depressive disorder.
      ,
      • Rittman T.
      • Rubinov M.
      • Vértes P.E.
      • Patel A.X.
      • Ginestet C.E.
      • Ghosh B.C.P.
      • et al.
      Regional expression of the MAPT gene is associated with loss of hubs in brain networks and cognitive impairment in Parkinson disease and progressive supranuclear palsy.
      ,
      • Romme I.A.C.
      • de Reus M.A.
      • Ophoff R.A.
      • Kahn R.S.
      • van den Heuvel M.P.
      Connectome Disconnectivity and Cortical Gene Expression in Patients With Schizophrenia.
      ,

      Grothe MJ, Sepulcre J, Gonzalez-Escamilla G, Jelistratova I, Schöll M, Hansson O, et al. Molecular properties underlying regional vulnerability to Alzheimer’s disease pathology. Brain. 2018 Sep 1;141(9):2755–2771.

      )] as well as between the expression of certain specific genes and connectivity-related phenotypes (
      • Krienen F.M.
      • Yeo B.T.T.
      • Ge T.
      • Buckner R.L.
      • Sherwood C.C.
      Transcriptional profiles of supragranular-enriched genes associate with corticocortical network architecture in the human brain.
      ). Transcriptional profiles of cell-specific marker genes exhibit associations with cortical thickness changes in a range of disorders (
      Writing Committee for the Attention-Deficit/Hyperactivity Disorder, Autism Spectrum Disorder, Bipolar Disorder, Major Depressive Disorder, Obsessive-Compulsive Disorder, and Schizophrenia ENIGMA Working Groups. Virtual Histology of Cortical Thickness and Shared Neurobiology in 6 Psychiatric Disorders.
      ), and with age-related changes on cortical myelination (

      Paquola C, Bethlehem RA, Seidlitz J, Wagstyl K, Romero-Garcia R, Whitaker KJ, et al. Shifts in myeloarchitecture characterise adolescent development of cortical gradients. Gold JI, Satterthwaite T, Mills K, Fulcher B, editors. eLife. 2019 Nov 14;8:e50482.

      ) and thickness (
      • Vidal-Pineiro D.
      • Parker N.
      • Shin J.
      • French L.
      • Grydeland H.
      • Jackowski A.P.
      • et al.
      Cellular correlates of cortical thinning throughout the lifespan.
      ,
      • Shin J.
      • French L.
      • Xu T.
      • Leonard G.
      • Perron M.
      • Pike G.B.
      • et al.
      Cell-Specific Gene-Expression Profiles and Cortical Thickness in the Human Brain.
      ). However, as outlined by Wei et al. (2022) (
      • Wei Y.
      • de Lange S.C.
      • Pijnenburg R.
      • Scholtens L.H.
      • Ardesch D.J.
      • Watanabe K.
      • et al.
      Statistical testing in transcriptomic-neuroimaging studies: A how-to and evaluation of methods assessing spatial and gene specificity.
      ), much of this work has overlooked the question of whether the reported associations are specific to the selected set of genes, or whether a similar association could be identified using a different set of genes. Considering the low-dimensionality of gene expression data, in which common large-scale gradients explain a large proportion of transcriptional variation across genes (

      Burt JB, Demirtaş M, Eckner WJ, Navejar NM, Ji JL, Martin WJ, et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nat Neurosci. 2018;21(9):1251–1259.

      ,
      • Fulcher B.D.
      Discovering Conserved Properties of Brain Organization Through Multimodal Integration and Interspecies Comparison.
      ), it is critical to assess gene specificity by comparing identified associations between expression and neuroimaging measures to the effects observed using other sets of genes. Empirical findings indicate that, even among spatially specific associations evaluated using spatially constrained null models, only 58% survive a liberal gene specificity test using random sets of genes and only 37% survive when using randomly selected genes that are specifically expressed in the brain (
      • Wei Y.
      • de Lange S.C.
      • Pijnenburg R.
      • Scholtens L.H.
      • Ardesch D.J.
      • Watanabe K.
      • et al.
      Statistical testing in transcriptomic-neuroimaging studies: A how-to and evaluation of methods assessing spatial and gene specificity.
      ). Simulations further demonstrated that out of all associations identified between spatially autocorrelated brain phenotypes and single-gene transcriptional profiles, only 3% survive corrections for both spatial autocorrelation and gene specificity (
      • Wei Y.
      • de Lange S.C.
      • Pijnenburg R.
      • Scholtens L.H.
      • Ardesch D.J.
      • Watanabe K.
      • et al.
      Statistical testing in transcriptomic-neuroimaging studies: A how-to and evaluation of methods assessing spatial and gene specificity.
      ). This result suggests that controlling for spatial autocorrelation is not sufficient to identify specific gene—phenotype associations, and that a large proportion of seemingly meaningful associations in the literature might not be specific to the reported gene set. The selection of an appropriately matched gene set to evaluate the specificity of one’s findings should thus be made with careful consideration of the specific questions being asked of the data. As a first pass, it is always good practice to examine spatial correlations between the genes of interest and any others not included in the target set, since large sets of correlated genes may mask specificity or dominate the results, particularly in multivariate analyses.
      In contrast to hypothesis-driven studies that test associations with an a priori selected set of genes, data-driven analyses test for associations between the expression profiles of each of the thousands of genes assayed in the transcriptional data and a phenotype of interest. However, many genes are not independent, having similarly spatially correlated expression patterns or working together as part of common physiological pathways. It is thus common in data-driven analyses to test whether particularly strong associations with the phenotype are concentrated within certain sets of functionally related genes. The most common approach for these analyses involves using gene-to-function annotations based on hierarchical systems such as the gene ontology (GO) (
      • Ashburner M.
      • Ball C.A.
      • Blake J.A.
      • Botstein D.
      • Butler H.
      • Cherry J.M.
      • et al.
      Gene Ontology: tool for the unification of biology.
      ) or KEGG (
      • Kanehisa M.
      • Goto S.
      KEGG: Kyoto Encyclopedia of Genes and Genomes.
      ), which categorize genes based on their associations with molecular function, cellular components, and biological processes. Traditionally, such gene category enrichment analyses (GCEA) have been used to support the interpretation of genome-wide association results (
      • Wang L.
      • Jia P.
      • Wolfinger R.D.
      • Chen X.
      • Zhao Z.
      Gene set analysis of genome-wide association studies: Methodological issues and perspectives.
      ), or in case–control comparisons of gene expression in a select tissue sample, by evaluating whether a gene category is preferentially associated with the phenotype compared to randomly selected genes (Figure 5A).
      Figure thumbnail gr5
      Figure 5Gene category enrichment analyses. A) The standard workflow for the GCEA where scores for each gene are calculated based on the association with a spatially defined phenotype and associations for gene sets in different gene ontology (GO) categories are evaluated using a random-gene null, where gene labels are ransomized to generate a null distribution. B) Different null models to assess the significance of identified associations including randomising i) gene identities (random-gene), where gene labels are randomised destroying the inherent tructure of the gene expression data; ii) spatial brain phenotypes, where brain measure is shuffled while keeping the structure of the gene expression data; iii) spatial brain phenotypes while maintaining their spatial autocorrelation to provide an even more stringent model that maintains spatial dependencies in both gene expression and brain measures. Reproduced with permission from (
      • Fulcher B.D.
      • Arnatkeviciute A.
      • Fornito A.
      Overcoming false-positive gene-category enrichment in the analysis of spatially resolved transcriptomic brain atlas data.
      ).
      Applying GCEA to spatially embedded transcriptional data introduces additional statistical considerations that can lead to spurious evidence of enrichment. For example, Fulcher et al. (
      • Fulcher B.D.
      • Arnatkeviciute A.
      • Fornito A.
      Overcoming false-positive gene-category enrichment in the analysis of spatially resolved transcriptomic brain atlas data.
      ) noted that GCEA results, for a diverse range of neuroimaging phenotypes in the literature, implicate similar gene categories related to metabolic, neuronal, and general behavioral processes. Performing enrichment on large ensembles of randomly generated phenotypes, the authors show that the application of GCEA is associated with an average inflation of false-positive rates of around 500-fold across gene categories. Furthermore, they show that gene categories with higher false-positive rates are more frequently reported as significant in the literature, consistent with this bias affecting published findings. They show that the bias is mainly driven by the extent of gene–gene correlation within GO categories, such that categories containing genes with more similar expression profiles across the brain are more likely to be significantly enriched. The classical null models used for GCEA do not account for this inter-correlation structure (nor do they account for spatial autocorrelation), resulting in statistical inference that is biased towards false positives (Figure 5A). The authors recommend generating a null distribution by randomizing the phenotype (instead of the genes), thus preserving the gene–gene correlation structure within categories in the null samples. As evident from the prior section, the phenotype could be randomized in either a spatially constrained or unconstrained way (Figure 5B). Using such phenotype-centered null models for inference considerably reduced the number of significant enrichment results identified in an analysis of 14 different brain phenotypes measured in both mouse and human (
      • Fulcher B.D.
      • Arnatkeviciute A.
      • Fornito A.
      Overcoming false-positive gene-category enrichment in the analysis of spatially resolved transcriptomic brain atlas data.
      ). These findings suggest that a substantial fraction of enrichment results reported in the literature may be affected by false-positive bias and require further investigation for validation.
      Conclusions and ways forward
      We have outlined several key considerations when performing imaging transcriptomic analyses (Figure 6). Detailed investigation of these considerations underscores the need for care when making processing and analysis choices (see Supplementary Text S3). Several open-source toolboxes have been developed that allow the implementation of many of the best-practice procedures we outline here at each analysis phase (Table 1) (see Supplementary Text S4). However, even when relying on these tools, careful consideration on a case-by-case basis is required for valid inference and reproducible results. Moreover, any analysis using AHBA data must consider several key limitations.
      Figure thumbnail gr6
      Figure 6General recommendations for imaging transcriptomic analyses of the human brain using AHBA across three analysis phases.
      Table 1Toolboxes available for each of the analysis phases including data processing, assessing associations between gene expression and brain imaging measures and evaluating the specificity of those associations through gene enrichment analyses.
      PhaseSoftware/toolboxFunctionProgramming languageURL
      Data processingabagen (

      Markello RD, Arnatkeviciute A, Poline JB, Fulcher BD, Fornito A, Misic B. Standardizing workflows in imaging transcriptomics with the abagen toolbox. Jbabdi S, Makin TR, Jbabdi S, Burt J, Hawrylycz MJ, editors. eLife. 2021 Nov 16;10:e72129.

      ,

      Markello R, Shafiei G, Zheng YQ, Mišić B. abagen: A toolbox for the Allen Brain Atlas genetics data [Internet]. Zenodo; 2021. Available from: Available from: https://doi.org/10.5281/zenodo.4646030

      )
      AHBA data processing in preparation for imaging transcriptomic analysesPythonhttps://abagen.readthedocs.io/en/stable/
      Relating expression and neuroimaging dataImaging transcriptomics toolbox (
      • Giacomel A.
      • Martins D.
      • Frigo M.
      • Turkheimer F.
      • Williams S.C.R.
      • Dipasquale O.
      • et al.
      Integrating neuroimaging and gene expression data using the imaging transcriptomics toolbox.
      ,

      Giacomel A, Martins D, Frigo M, Turkheimer F, Williams SCR, Dipasquale O, et al. The Imaging Transcriptomics Toolbox [Internet]. Zenodo; 2022 [cited 2022 Jun 7]. Available from: https://zenodo.org/record/6364963

      )
      Integrated imaging transcriptomic analyses using mass- univariate and multivariate PLS approaches followed by enrichment analyses using phenotype-based nullPythonhttps://github.com/molecular-neuroimaging/Imaging_Transcriptomics
      BrainSMASH (
      • Burt J.B.
      • Helmer M.
      • Shinn M.
      • Anticevic A.
      • Murray J.D.
      Generative modeling of brain maps with spatial autocorrelation.
      )
      Allows generation of model-based null ensembles for correlating spatial maps to evaluate statistical significancePythonhttps://brainsmash.readthedocs.io/en/latest/
      BrainSpace (
      • Vos de Wael R.
      • Benkarim O.
      • Paquola C.
      • Lariviere S.
      • Royer J.
      • Tavakol S.
      • et al.
      BrainSpace: a toolbox for the analysis of macroscale gradients in neuroimaging and connectomics datasets.
      )
      Analysis of macroscale gradients and implementation of null models for spatially autocorrelated brain maps to evaluate statistical significancePython and Matlabhttps://github.com/MICA-MNI/BrainSpace
      Spin-test (
      • Alexander-Bloch A.F.
      • Shou H.
      • Liu S.
      • Satterthwaite T.D.
      • Glahn D.C.
      • Shinohara R.T.
      • et al.
      On testing for spatial correspondence between maps of human brain structure and function.
      )
      Implementation of the non-parametric spatial permutation test for brain maps for evaluating statistical significanceMatlabhttps://github.com/spin-test/spin-test
      Neuromaps (

      Markello RD, Hansen JY, Liu ZQ, Bazinet V, Shafiei G, Suárez LE, et al. neuromaps: structural and functional interpretation of brain maps. bioRxiv; 2022. p. 2022.01.06.475081.

      )
      Implementation of different parametric and non-parametric spatial null modelsPythonhttps://netneurolab.github.io/neuromaps/
      Gene specificity and enrichment analysesGAMBA toolbox (
      • Wei Y.
      • de Lange S.C.
      • Pijnenburg R.
      • Scholtens L.H.
      • Ardesch D.J.
      • Watanabe K.
      • et al.
      Statistical testing in transcriptomic-neuroimaging studies: A how-to and evaluation of methods assessing spatial and gene specificity.
      )
      Gene specificity testing for both hypotheses and data-driven analysesMatlabhttps://github.com/dutchconnectomelab/GAMBA-MATLAB
      GCEA using phenotype nulls (
      • Fulcher B.D.
      • Arnatkeviciute A.
      • Fornito A.
      Overcoming false-positive gene-category enrichment in the analysis of spatially resolved transcriptomic brain atlas data.
      ,

      Fulcher B, AurinaBMH. benfulcher/GeneCategoryEnrichmentAnalysis: Updated README [Internet]. Zenodo; 2021 [cited 2022 Jun 7]. Available from: https://zenodo.org/record/4470238

      )
      GO category enrichment analysis using phenotype-based nulls for data-driven analysesMatlabhttps://github.com/benfulcher/GeneCategoryEnrichmentAnalysis
      ABAnnotate toolbox (

      Lotter LD, Dukart J, Fulcher BD. ABAnnotate: A toolbox for ensemble-based multimodal gene-category enrichment analysis of human neuroimaging data [Internet]. Zenodo; 2022 [cited 2022 Jun 7]. Available from: https://zenodo.org/record/6463329

      )
      Gene category enrichment analysis using phenotype-based nulls for data-driven analysesMatlabhttps://github.com/LeonDLotter/ABAnnotate
      First, the relationship between gene expression and protein abundance is complex, and variations in transcriptional activity do not necessarily affect protein levels (

      Hansen JY, Markello RD, Tuominen L, Nørgaard M, Kuzmin E, Palomero-Gallagher N, et al. Correspondence between gene expression and neurotransmitter receptor and transporter density in the human brain. bioRxiv; 2021 p. 2021.11.30.469876.

      ). Second, the AHBA relies on microarray analysis of bulk tissue samples, and the resulting estimates may be affected by regional variations in cellular composition; results should therefore be cross-validated with single-cell RNA-seq data where possible (
      • Eze U.C.
      • Bhaduri A.
      • Haeussler M.
      • Nowakowski T.J.
      • Kriegstein A.R.
      Single-cell atlas of early human brain development highlights heterogeneity of human neuroepithelial cells and early radial glia.
      ,
      • Armand E.J.
      • Li J.
      • Xie F.
      • Luo C.
      • Mukamel E.A.
      Single cell sequencing of brain cell transcriptomes and epigenomes.
      ). Third, whereas the AHBA was designed to quantify canonical transcriptional patterns that are conserved across individuals and provides unparalleled spatial coverage compared to other post-mortem tissue banks (Table 2), the AHBA measures are nonetheless derived from a small sample of six adult donors. Variability in gene expression in the AHBA is much greater across brain regions than across these six individuals (
      • Hawrylycz M.
      • Miller J.A.
      • Menon V.
      • Feng D.
      • Dolbeare T.
      • Guillozet-Bongaarts A.L.
      • et al.
      Canonical genetic signatures of the adult human brain.
      ), suggesting that the atlas can be used to investigate robust regional expression profiles, but concerns about the representativeness of the donor brains are nonetheless valid. Fourth, the AHBA only assays gene expression patterns in adult brains, but many neural phenotypes may depend on developmentally complex and dynamically changing patterns of gene expression (
      • Kang H.J.
      • Kawasawa Y.I.
      • Cheng F.
      • Zhu Y.
      • Xu X.
      • Li M.
      • et al.
      Spatio-temporal transcriptome of the human brain.
      ,
      • Chaudhari N.
      • Hahn W.E.
      Genetic Expression in the Developing Brain.
      ,
      • Ip B.K.
      • Wappler I.
      • Peters H.
      • Lindsay S.
      • Clowry G.J.
      • Bayatti N.
      Investigating gradients of gene expression involved in early human cortical development.
      ,
      • Somel M.
      • Guo S.
      • Fu N.
      • Yan Z.
      • Hu H.Y.
      • Xu Y.
      • et al.
      MicroRNA, mRNA, and protein expression link development and aging in human and macaque brain.
      ). Finally, as AHBA data are based on post-mortem measurements from a small sample, any relationships found between gene-expression patterns and imaging phenotypes (or other outcomes) are purely correlational and do not directly reveal causal mechanisms. This consideration is particularly salient for clinical applications.
      Table 2Human brain transcriptomic databases.
      AtlasTypeAgeMethodSamplesURLMain limitations
      Human Brain Transcriptome (

      Johnson MB, Kawasawa YI, Mason CE, Krsnik Z, Coppola G, Bogdanović D, et al. Functional and evolutionary insights into human brain development through global transcriptome analysis. Neuron. 2009 May 28;62(4):494–509.

      ,
      • Kang H.J.
      • Kawasawa Y.I.
      • Cheng F.
      • Zhu Y.
      • Xu X.
      • Li M.
      • et al.
      Spatio-temporal transcriptome of the human brain.
      )
      SpatiotemporalLifespanMicroarray16 brain structureshttps://hbatlas.org/Limited spatial coverage, not suitable for brain-wide analyses
      BrainSpan (
      • Miller J.A.
      • Ding S.L.
      • Sunkin S.M.
      • Smith K.A.
      • Ng L.
      • Szafer A.
      • et al.
      Transcriptional landscape of the prenatal human brain.
      )
      SpatialPrenatalMicroarrayMultiple structureshttp://www.brainspan.org/lcmNon-continuous spatial coverage, not suitable for brain-wide analyses
      GTEx (
      • Lonsdale J.
      • Thomas J.
      • Salvatore M.
      • Phillips R.
      • Lo E.
      • Shad S.
      • et al.
      The Genotype-Tissue Expression (GTEx) project.
      ,
      GTEx Consortium
      Human genomics. The Genotype-Tissue Expression (GTEx) pilot analysis: multitissue gene regulation in humans.
      )
      SpatialAdultRNA-seq13 brain structures and other tissueshttps://www.gtexportal.org/Limited spatial coverage, not suitable for brain-wide analyses
      Human Protein Atlas (
      • Sjöstedt E.
      • Zhong W.
      • Fagerberg L.
      • Karlsson M.
      • Mitsios N.
      • Adori C.
      • et al.
      An atlas of the protein-coding genes in the human, pig, and mouse brain.
      )
      SpatialAdultRNA-seq13 main brain regionshttps://www.proteinatlas.org/humanproteome/brainRelatively low resolution in the cortex
      PsychENCODE(121)Spatial bulk tissue and cell-typeHC/disordersRNA-seqMultiple structureshttp://resource.psychencode.org/Low resolution in the cortex
      Brain RNA-Seq (
      • Zhang Y.
      • Sloan S.A.
      • Clarke L.E.
      • Caneda C.
      • Plaza C.A.
      • Blumenthal P.D.
      • et al.
      Purification and Characterization of Progenitor and Mature Human Astrocytes Reveals Transcriptional and Functional Differences with Mouse.
      )
      Cell-typeFetal/adultRNA-seqTCtx, Hphttp://www.brainrnaseq.org/Very limited spatial coverage, not suitable for brain-wide analyses, limited data access options
      UCSC Cell Browser (
      • Nowakowski T.J.
      • Bhaduri A.
      • Pollen A.A.
      • Alvarado B.
      • Mostajo-Radji M.A.
      • Di Lullo E.
      • et al.
      Spatiotemporal gene expression trajectories reveal developmental hierarchies of the human cortex.
      ,
      • Speir M.L.
      • Bhaduri A.
      • Markov N.S.
      • Moreno P.
      • Nowakowski T.J.
      • Papatheodorou I.
      • et al.
      UCSC Cell Browser: visualize your single-cell data.
      )
      Single-cellFetalRNA-seqCtx, Mgehttps://cells.ucsc.edu/?ds=cortex-devVery limited spatial coverage, not suitable for brain-wide analyses
      GBMseq (
      • Darmanis S.
      • Sloan S.A.
      • Croote D.
      • Mignardi M.
      • Chernikova S.
      • Samghababi P.
      • et al.
      Single-Cell RNA-Seq Analysis of Infiltrating Neoplastic Cells at the Migrating Front of Human Glioblastoma.
      )
      Single-cellAdultRNA-seqGBM, core and peripheryhttp://gbmseq.org/Tumor-derived tissue, not suitable for brain-wide analyses
      HC – healthy controls, TCtx – temporal cortex, Ctx – cortex, Hp – hippocampus, Mge - medial ganglionic eminence, GBM – glioblastoma; Spatial – expression profiles in spatially distributed samples; Bulk tissue – samples containing different cell types; Cell-type – expression profiles for different cell types; Single-cell – expression profiles for individual cells.
      Given these limitations, the AHBA should be viewed as a useful resource for generating hypotheses that should then be further tested using in vitro (
      • Luo L.
      • Callaway E.M.
      • Svoboda K.
      Genetic Dissection of Neural Circuits: A Decade of Progress.
      ) or in vivo (
      • Mitchell K.J.
      • Huang Z.J.
      • Moghaddam B.
      • Sawa A.
      Following the genes: a framework for animal modeling of psychiatric disorders.
      ) models; through the analysis of expression quantitative trait loci (eQTLs) (
      • Zhu Z.
      • Zhang F.
      • Hu H.
      • Bakshi A.
      • Robinson M.R.
      • Powell J.E.
      • et al.
      Integration of summary data from GWAS and eQTL studies predicts complex trait gene targets.
      ); or via supplemental analyses in complementary, although less anatomically comprehensive, datasets that measure gene expression in more people (
      • Lonsdale J.
      • Thomas J.
      • Salvatore M.
      • Phillips R.
      • Lo E.
      • Shad S.
      • et al.
      The Genotype-Tissue Expression (GTEx) project.
      ), across different developmental time points (
      • Akbarian S.
      • Liu C.
      • Knowles J.A.
      • Vaccarino F.M.
      • Farnham P.J.
      • Crawford G.E.
      • et al.
      The psychENCODE project.
      ), or in a given patient group of interest (
      • Anderson K.M.
      • Collins M.A.
      • Kong R.
      • Fang K.
      • Li J.
      • He T.
      • et al.
      Convergent molecular, cellular, and cortical neuroimaging signatures of major depressive disorder.
      ). Genetic disorders can also provide a powerful validation tool (
      • Seidlitz J.
      • Nadig A.
      • Liu S.
      • Bethlehem R.A.I.
      • Vértes P.E.
      • Morgan S.E.
      • et al.
      Transcriptomic and cellular decoding of regional brain vulnerability to neurogenetic disorders.
      ). The increasing precision and sophistication of anatomically comprehensive transcriptional atlases (
      • Chaudhari N.
      • Hahn W.E.
      Genetic Expression in the Developing Brain.
      ,
      • Ip B.K.
      • Wappler I.
      • Peters H.
      • Lindsay S.
      • Clowry G.J.
      • Bayatti N.
      Investigating gradients of gene expression involved in early human cortical development.
      ,
      • Somel M.
      • Guo S.
      • Fu N.
      • Yan Z.
      • Hu H.Y.
      • Xu Y.
      • et al.
      MicroRNA, mRNA, and protein expression link development and aging in human and macaque brain.
      ,
      • Luo L.
      • Callaway E.M.
      • Svoboda K.
      Genetic Dissection of Neural Circuits: A Decade of Progress.
      ) — that would ideally progress towards population-scale, voxel-resolution, single-cell RNAseq databanks — will continue to provide new opportunities to identify the molecular correlates of macroscale neuroimaging phenotypes measured non-invasively in living humans.

      Acknowledgements

      AF was supported by the Sylvia and Charles Viertel Charitable Foundation and the National Health and Medical Research Council (IDs: 1149292 and 1197431).
      Disclosures
      RDM is currently employed by Octave Bioscience. The remaining authors report no biomedical financial interests or potential conflicts of interest.

      References

        • Fornito A.
        • Zalesky A.
        • Breakspear M.
        The connectomics of brain disorders.
        Nature Reviews Neuroscience. 2015 Mar; 16: 159-172
        • Stephan K.E.
        • Friston K.J.
        • Frith C.D.
        Dysconnection in Schizophrenia: From Abnormal Synaptic Plasticity to Failures of Self-monitoring.
        Schizophrenia Bulletin. 2009 May 1; 35: 509-527
        • Xia C.H.
        • Ma Z.
        • Ciric R.
        • Gu S.
        • Betzel R.F.
        • Kaczkurkin A.N.
        • et al.
        Linked dimensions of psychopathology and connectivity in functional brain networks.
        Nature Communications. 2018; 9
      1. Filippi M, Basaia S, Sarasso E, Stojkovic T, Stankovic I, Fontana A, et al. Longitudinal brain connectivity changes and clinical evolution in Parkinson’s disease. Mol Psychiatry. 2021 Sep;26(9):5429–5440.

        • Fornito A.
        • Bullmore E.T.
        Connectomic Intermediate Phenotypes for Psychiatric Disorders.
        Front Psychiatry. 2012; 3
        • Li S.
        • Hu N.
        • Zhang W.
        • Tao B.
        • Dai J.
        • Gong Y.
        • et al.
        Dysconnectivity of Multiple Brain Networks in Schizophrenia: A Meta-Analysis of Resting-State Functional Connectivity.
        Front Psychiatry. 2019; 10
        • Mohammad-Rezazadeh I.
        • Frohlich J.
        • Loo S.K.
        • Jeste S.S.
        Brain Connectivity in Autism Spectrum Disorder.
        Curr Opin Neurol. 2016 Apr; 29: 137-147
        • Perry A.
        • Roberts G.
        • Mitchell P.B.
        • Breakspear M.
        Connectomics of bipolar disorder: a critical review, and evidence for dynamic instabilities within interoceptive networks.
        Mol Psychiatry. 2019 Sep; 24: 1296-1318
        • Zeng L.L.
        • Shen H.
        • Liu L.
        • Wang L.
        • Li B.
        • Fang P.
        • et al.
        Identifying major depression using whole-brain functional connectivity: a multivariate pattern analysis.
        Brain. 2012 May 1; 135: 1498-1507
        • Dickson B.J.
        Molecular mechanisms of axon guidance.
        Science. 2002 Dec 6; 298: 1959-1964
        • Goodhill G.J.
        Can Molecular Gradients Wire the Brain?.
        Trends Neurosci. 2016 Apr; 39: 202-211
        • Arnatkeviciute A.
        • Fulcher B.D.
        • Bellgrove M.A.
        • Fornito A.
        Where the genome meets the connectome: Understanding how genes shape human brain connectivity.
        NeuroImage. 2021; 244118570
        • Arnatkeviciute A.
        • Fulcher B.D.
        • Oldham S.
        • Tiego J.
        • Paquola C.
        • Gerring Z.
        • et al.
        Genetic influences on hub connectivity of the human connectome.
        Nat Commun. 2021 Jul 9; 12: 4237
        • Elliott L.T.
        • Sharp K.
        • Alfaro-Almagro F.
        • Shi S.
        • Miller K.L.
        • Douaud G.
        • et al.
        Genome-wide association studies of brain imaging phenotypes in UK Biobank.
        Nature. 2018 Oct; 562: 210-216
        • Fornito A.
        • Zalesky A.
        • Bassett D.S.
        • Meunier D.
        • Ellison-Wright I.
        • Yücel M.
        • et al.
        Genetic Influences on Cost-Efficient Organization of Human Cortical Functional Networks.
        J Neurosci. 2011 Mar 2; 31: 3261-3270
      2. Smit DJA, Stam CJ, Posthuma D, Boomsma DI, Geus EJC de. Heritability of “small-world” networks in the brain: A graph theoretical analysis of resting-state EEG functional connectivity. Human Brain Mapping. 2008;29(12):1368–1378.

        • Smith S.M.
        • Douaud G.
        • Chen W.
        • Hanayik T.
        • Alfaro-Almagro F.
        • Sharp K.
        • et al.
        An expanded set of genome-wide association studies of brain imaging phenotypes in UK Biobank.
        Nat Neurosci. 2021 May; 24: 737-745
        • Thompson P.M.
        • Cannon T.D.
        • Narr K.L.
        • van Erp T.
        • Poutanen V.P.
        • Huttunen M.
        • et al.
        Genetic influences on brain structure.
        Nature Neuroscience. 2001 Dec; 4: 1253-1258
        • Thompson P.M.
        • Ge T.
        • Glahn D.C.
        • Jahanshad N.
        • Nichols T.E.
        Genetics of the connectome.
        NeuroImage. 2013 Oct 15; 80: 475-488
        • Brouwer R.M.
        • Klein M.
        • Grasby K.L.
        • Schnack H.G.
        • Jahanshad N.
        • Teeuw J.
        • et al.
        Genetic variants associated with longitudinal changes in brain structure across the lifespan.
        Nat Neurosci. 2022 Apr; 25: 421-432
        • Grasby K.L.
        • Jahanshad N.
        • Painter J.N.
        • Colodro-Conde L.
        • Bralten J.
        • Hibar D.P.
        • et al.
        The genetic architecture of the human cerebral cortex.
        Science. 2020 Mar 20; : 367
        • Hibar D.P.
        • Stein J.L.
        • Renteria M.E.
        • Arias-Vasquez A.
        • Desrivières S.
        • Jahanshad N.
        • et al.
        Common genetic variants influence human subcortical brain structures.
        Nature. 2015 Apr 9; 520: 224-229
        • Hormozdiari F.
        • Kichaev G.
        • Yang W.Y.
        • Pasaniuc B.
        • Eskin E.
        Identification of causal genes for complex traits.
        Bioinformatics. 2015 Jun 15; 31: i206-i213
        • Schaid D.J.
        • Chen W.
        • Larson N.B.
        From genome-wide associations to candidate causal variants by statistical fine-mapping.
        Nat Rev Genet. 2018 Aug; 19: 491-504
      3. Johnson MB, Kawasawa YI, Mason CE, Krsnik Z, Coppola G, Bogdanović D, et al. Functional and evolutionary insights into human brain development through global transcriptome analysis. Neuron. 2009 May 28;62(4):494–509.

        • Kang H.J.
        • Kawasawa Y.I.
        • Cheng F.
        • Zhu Y.
        • Xu X.
        • Li M.
        • et al.
        Spatio-temporal transcriptome of the human brain.
        Nature. 2011 Oct; 478: 483-489
      4. Keil JM, Qalieh A, Kwan KY. Brain Transcriptome Databases: A User’s Guide. J Neurosci. 2018 Mar 7;38(10):2399–2412.

        • Khaitovich P.
        • Muetzel B.
        • She X.
        • Lachmann M.
        • Hellmann I.
        • Dietzsch J.
        • et al.
        Regional Patterns of Gene Expression in Human and Chimpanzee Brains.
        Genome Res. 2004 Aug; 14: 1462-1473
        • Hawrylycz M.J.
        • Lein E.S.
        • Guillozet-Bongaarts A.L.
        • Shen E.H.
        • Ng L.
        • Miller J.A.
        • et al.
        An anatomically comprehensive atlas of the adult human brain transcriptome.
        Nature. 2012 Sep 20; 489: 391-399
        • Lein E.S.
        • Hawrylycz M.J.
        • Ao N.
        • Ayres M.
        • Bensinger A.
        • Bernard A.
        • et al.
        Genome-wide atlas of gene expression in the adult mouse brain.
        Nature. 2007 Jan; 445: 168-176
        • Miller J.A.
        • Ding S.L.
        • Sunkin S.M.
        • Smith K.A.
        • Ng L.
        • Szafer A.
        • et al.
        Transcriptional landscape of the prenatal human brain.
        Nature. 2014 Apr; 508: 199-206
        • Arnatkeviciute A.
        • Fulcher B.D.
        • Fornito A.
        A practical guide to linking brain-wide gene expression and neuroimaging data.
        NeuroImage. 2019 Apr 1; 189: 353-367
        • Arnatkeviciute A.
        • Fulcher B.D.
        • Bellgrove M.A.
        • Fornito A.
        Imaging Transcriptomics of Brain Disorders.
        Biological Psychiatry Global Open Science. 2021;
      5. Fornito A, Arnatkevičiūtė A, Fulcher BD. Bridging the Gap between Connectome and Transcriptome. Trends in Cognitive Sciences. 2019 Jan 1;23(1):34–50.

        • Arnatkeviciute A.
        • Fulcher B.D.
        • Pocock R.
        • Fornito A.
        Hub connectivity, neuronal diversity, and gene expression in the Caenorhabditis elegans connectome.
        PLOS Computational Biology. 2018 Feb 12; 14e1005989
        • Fulcher B.D.
        • Fornito A.
        A transcriptional signature of hub connectivity in the mouse connectome.
        PNAS. 2016 Feb 2; 113: 1435-1440
        • Goel P.
        • Kuceyeski A.
        • LoCastro E.
        • Raj A.
        Spatial patterns of genome-wide expression profiles reflect anatomic and fiber connectivity architecture of healthy human brain.
        Human Brain Mapping. 2014; 35: 4204-4218
        • Romero-Garcia R.
        • Whitaker K.J.
        • Váša F.
        • Seidlitz J.
        • Shinn M.
        • Fonagy P.
        • et al.
        Structural covariance networks are coupled to expression of genes enriched in supragranular layers of the human cortex.
        NeuroImage. 2018 May 1; 171: 256-267
        • Whitaker K.J.
        • Vértes P.E.
        • Romero-Garcia R.
        • Váša F.
        • Moutoussis M.
        • Prabhu G.
        • et al.
        Adolescence is associated with genomically patterned consolidation of the hubs of the human brain connectome.
        PNAS. 2016 Aug 9; 113: 9105-9110
      6. Burt JB, Demirtaş M, Eckner WJ, Navejar NM, Ji JL, Martin WJ, et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nat Neurosci. 2018;21(9):1251–1259.

        • Anderson K.M.
        • Krienen F.M.
        • Choi E.Y.
        • Reinen J.M.
        • Yeo B.T.T.
        • Holmes A.J.
        Gene expression links functional networks across cortex and striatum.
        Nature Communications. 2018 Apr 12; 9: 1428
        • Richiardi J.
        • Altmann A.
        • Milazzo A.C.
        • Chang C.
        • Chakravarty M.M.
        • Banaschewski T.
        • et al.
        Correlated gene expression supports synchronous activity in brain networks.
        Science. 2015 Jun 12; 348: 1241-1244
        • Rubinov M.
        • Ypma R.J.F.
        • Watson C.
        • Bullmore E.T.
        Wiring cost and topological participation of the mouse brain connectome.
        PNAS. 2015 Aug 11; 112: 10032-10037
      7. Vértes PE, Rittman T, Whitaker KJ, Romero-Garcia R, Váša F, Kitzbichler MG, et al. Gene transcription profiles associated with inter-modular hubs and connection distance in human functional magnetic resonance imaging networks. Philos Trans R Soc Lond B Biol Sci. 2016 Oct 5;371(1705).

        • Cioli C.
        • Abdi H.
        • Beaton D.
        • Burnod Y.
        • Mesmoudi S.
        Differences in Human Cortical Gene Expression Match the Temporal Properties of Large-Scale Functional Networks.
        PLOS ONE. 2014 Dec 29; 9e115913
        • Shafiei G.
        • Markello R.D.
        • Vos de Wael R.
        • Bernhardt B.C.
        • Fulcher B.D.
        • Misic B.
        Topographic gradients of intrinsic dynamics across neocortex.
        Elife. 2020 Dec 17; 9e62116
        • Anderson K.M.
        • Collins M.A.
        • Chin R.
        • Ge T.
        • Rosenberg M.D.
        • Holmes A.J.
        Transcriptional and imaging-genetic association of cortical interneurons, brain function, and schizophrenia risk.
        Nature Communications. 2020 Jun 8; 11: 2889
        • Anderson K.M.
        • Collins M.A.
        • Kong R.
        • Fang K.
        • Li J.
        • He T.
        • et al.
        Convergent molecular, cellular, and cortical neuroimaging signatures of major depressive disorder.
        Proc Natl Acad Sci U S A. 2020 Oct 6; 117: 25138-25149
      8. McColgan P, Gregory S, Seunarine KK, Razi A, Papoutsi M, Johnson E, et al. Brain Regions Showing White Matter Loss in Huntington’s Disease Are Enriched for Synaptic and Metabolic Genes. Biol Psychiatry. 2018 Mar 1;83(5):456–465.

        • Rittman T.
        • Rubinov M.
        • Vértes P.E.
        • Patel A.X.
        • Ginestet C.E.
        • Ghosh B.C.P.
        • et al.
        Regional expression of the MAPT gene is associated with loss of hubs in brain networks and cognitive impairment in Parkinson disease and progressive supranuclear palsy.
        Neurobiology of Aging. 2016 Dec 1; 48: 153-160
        • Romme I.A.C.
        • de Reus M.A.
        • Ophoff R.A.
        • Kahn R.S.
        • van den Heuvel M.P.
        Connectome Disconnectivity and Cortical Gene Expression in Patients With Schizophrenia.
        Biological Psychiatry. 2017 Mar 15; 81: 495-502
        • Seidlitz J.
        • Nadig A.
        • Liu S.
        • Bethlehem R.A.I.
        • Vértes P.E.
        • Morgan S.E.
        • et al.
        Transcriptomic and cellular decoding of regional brain vulnerability to neurogenetic disorders.
        Nature Communications. 2020 Jul 3; 11: 1-14
        • Patel Yash
        Autism Spectrum Disorder, Bipolar Disorder, Major Depressive Disorder, Obsessive-Compulsive Disorder, and Schizophrenia ENIGMA Working Groups, et al.
        JAMA Psychiatry. 2021 Jan 1; 78: 47-63
        • Vidal-Pineiro D.
        • Parker N.
        • Shin J.
        • French L.
        • Grydeland H.
        • Jackowski A.P.
        • et al.
        Cellular correlates of cortical thinning throughout the lifespan.
        Sci Rep. 2020 Dec 11; 1021803
      9. Deco G, Kringelbach ML, Arnatkeviciute A, Oldham S, Sabaroedin K, Rogasch NC, et al. Dynamical consequences of regional heterogeneity in the brain’s transcriptional landscape. Science Advances. 2021 Jul 1;7(29):eabf4752.

      10. Freeze B, Pandya S, Zeighami Y, Raj A. Regional transcriptional architecture of Parkinson’s disease pathogenesis and network spread. Brain. 2019 Oct 1;142(10):3072–3085.

        • Zheng Y.Q.
        • Zhang Y.
        • Yau Y.
        • Zeighami Y.
        • Larcher K.
        • Misic B.
        • et al.
        Local vulnerability and global connectivity jointly shape neurodegenerative disease propagation.
        PLOS Biology. 2019 Nov 21; 17e3000495
      11. Yan W, Ye C, Wang T, Sun J, Wu T, Ma T. Misfolded protein propagation in an integrated computational model of structural network and LRRK2 gene expression. In: 2020 42nd Annual International Conference of the IEEE Engineering in Medicine Biology Society (EMBC). 2020. p. 2368–2371.

        • Fulcher B.D.
        • Arnatkeviciute A.
        • Fornito A.
        Overcoming false-positive gene-category enrichment in the analysis of spatially resolved transcriptomic brain atlas data.
        Nat Commun. 2021 May 11; 12: 2669
      12. Markello RD, Arnatkevičiūtė A, Poline JB, Fulcher BD, Fornito A, Misic B. Standardizing workflows in imaging transcriptomics with the abagen toolbox. bioRxiv. 2021 Jul 9;2021.07.08.451635.

      13. Schulze A, Downward J. Navigating gene expression using microarrays — a technology review. Nature Cell Biology. 2001 Aug 1;3(8):E190–E195.

        • Liu H.
        • Bebu I.
        • Li X.
        Microarray probes and probe sets.
        Front Biosci (Elite Ed). 2010 Jan 1; 2: 325-338
        • O’Leary N.A.
        • Wright M.W.
        • Brister J.R.
        • Ciufo S.
        • Haddad D.
        • McVeigh R.
        • et al.
        Reference sequence (RefSeq) database at NCBI: current status, taxonomic expansion, and functional annotation.
        Nucleic Acids Research. 2016 Jan 4; 44: D733-D745
        • Hawrylycz M.
        • Miller J.A.
        • Menon V.
        • Feng D.
        • Dolbeare T.
        • Guillozet-Bongaarts A.L.
        • et al.
        Canonical genetic signatures of the adult human brain.
        Nature Neuroscience. 2015 Dec; 18: 1832-1844
      14. Allen Institute for Brain Science. Technical white paper: microarray data normalization [Internet]. 2013. Available from: http://help.brain-map.org/display/humanbrain/Documentation

        • Fulcher B.D.
        • Little M.A.
        • Jones N.S.
        Highly comparative time-series analysis: the empirical structure of time series and their methods.
        Journal of The Royal Society Interface. 2013 Jun 6; 1020130048
        • Fagerberg L.
        • Hallström B.M.
        • Oksvold P.
        • Kampf C.
        • Djureinovic D.
        • Odeberg J.
        • et al.
        Analysis of the Human Tissue-specific Expression by Genome-wide Integration of Transcriptomics and Antibody-based Proteomics.
        Molecular & Cellular Proteomics. 2014 Feb 1; 13: 397-406
        • Genovese G.
        • Fromer M.
        • Stahl E.A.
        • Ruderfer D.M.
        • Chambert K.
        • Landén M.
        • et al.
        Increased burden of ultra-rare protein-altering variants among 4,877 individuals with schizophrenia.
        Nat Neurosci. 2016 Nov; 19: 1433-1441
      15. Markello RD, Arnatkeviciute A, Poline JB, Fulcher BD, Fornito A, Misic B. Standardizing workflows in imaging transcriptomics with the abagen toolbox. Jbabdi S, Makin TR, Jbabdi S, Burt J, Hawrylycz MJ, editors. eLife. 2021 Nov 16;10:e72129.

      16. Markello R, Shafiei G, Zheng YQ, Mišić B. abagen: A toolbox for the Allen Brain Atlas genetics data [Internet]. Zenodo; 2021. Available from: Available from: https://doi.org/10.5281/zenodo.4646030

        • Forest M.
        • Iturria‐Medina Y.
        • Goldman J.S.
        • Kleinman C.L.
        • Lovato A.
        • Klein K.O.
        • et al.
        Gene networks show associations with seed region connectivity.
        Human Brain Mapping. 2017; 38: 3126-3140
        • Oldham M.C.
        • Konopka G.
        • Iwamoto K.
        • Langfelder P.
        • Kato T.
        • Horvath S.
        • et al.
        Functional organization of the transcriptome in human brain.
        Nature Neuroscience. 2008 Nov; 11: 1271-1282
        • Shen J.
        • Yang B.
        • Xie Z.
        • Wu H.
        • Zheng Z.
        • Wang J.
        • et al.
        Cell-Type-Specific Gene Modules Related to the Regional Homogeneity of Spontaneous Brain Activity and Their Associations With Common Brain Disorders.
        Front Neurosci. 2021; 15639527
      17. Zarkali A, McColgan P, Ryten M, Reynolds R, Leyland LA, Lees AJ, et al. Differences in network controllability and regional gene expression underlie hallucinations in Parkinson’s disease. Brain. 2020 Dec 5;143(11):3435–3448.

      18. Hess JL, Radonjić NV, Patak J, Glatt SJ, Faraone SV. Autophagy, apoptosis, and neurodevelopmental genes might underlie selective brain region vulnerability in attention-deficit/hyperactivity disorder. Mol Psychiatry. 2020 Dec 18;

        • Krishnan A.
        • Williams L.J.
        • McIntosh A.R.
        • Abdi H.
        Partial Least Squares (PLS) methods for neuroimaging: A tutorial and review.
        NeuroImage. 2011 May 15; 56: 455-475
      19. Bilenko NY, Gallant JL. Pyrcca: Regularized Kernel Canonical Correlation Analysis in Python and Its Applications to Neuroimaging. Frontiers in Neuroinformatics [Internet]. 2016 [cited 2022 Sep 28];10. Available from: https://www.frontiersin.org/articles/10.3389/fninf.2016.00049

      20. Helmer M, Warrington S, Mohammadi-Nejad AR, Ji JL, Howell A, Rosand B, et al. On stability of Canonical Correlation Analysis and Partial Least Squares with application to brain-behavior associations [Internet]. bioRxiv; 2021 [cited 2022 Sep 28]. p. 2020.08.25.265546. Available from: https://www.biorxiv.org/content/10.1101/2020.08.25.265546v3

        • Wang H.T.
        • Smallwood J.
        • Mourao-Miranda J.
        • Xia C.H.
        • Satterthwaite T.D.
        • Bassett D.S.
        • et al.
        Finding the needle in a high-dimensional haystack: Canonical correlation analysis for neuroscientists.
        NeuroImage. 2020 Aug 1; 216116745
        • Hess J.L.
        • Akutagava-Martins G.C.
        • Patak J.D.
        • Glatt S.J.
        • Faraone S.V.
        Why is there selective subcortical vulnerability in ADHD? Clues from postmortem brain gene expression data.
        Mol Psychiatry. 2018 Aug; 23: 1787-1793
        • Xie Y.
        • Zhang X.
        • Liu F.
        • Qin W.
        • Fu J.
        • Xue K.
        • et al.
        Brain mRNA Expression Associated with Cortical Volume Alterations in Autism Spectrum Disorder.
        Cell Rep. 2020 Sep 15; 32108137
        • Ball G.
        • Seidlitz J.
        • Beare R.
        • Seal M.L.
        Cortical remodelling in childhood is associated with genes enriched for neurodevelopmental disorders.
        NeuroImage. 2020 Jul 15; 215116803
        • Romero-Garcia R.
        • Warrier V.
        • Bullmore E.T.
        • Baron-Cohen S.
        • Bethlehem R.A.I.
        Synaptic and transcriptionally downregulated genes are associated with cortical thickness differences in autism.
        Molecular Psychiatry. 2019 Jul; 24: 1053-1064
        • French L.
        • Pavlidis P.
        Relationships between Gene Expression and Brain Wiring in the Adult Rodent Brain.
        PLOS Computational Biology. 2011 Jan 6; 7e1001049
        • Lau H.Y.G.
        • Fornito A.
        • Fulcher B.D.
        Scaling of gene transcriptional gradients with brain size across mouse development.
        Neuroimage. 2021 Jan 1; 224117395
      21. Horvát S, Gămănuț R, Ercsey-Ravasz M, Magrou L, Gămănuț B, Essen DCV, et al. Spatial Embedding and Wiring Cost Constrain the Functional Layout of the Cortical Network of Rodents and Primates. PLOS Biology. 2016 Jul 21;14(7):e1002512.

        • Theodoni P.
        • Majka P.
        • Reser D.H.
        • Wójcik D.K.
        • Rosa M.G.P.
        • Wang X.J.
        Structural Attributes and Principles of the Neocortical Connectome in the Marmoset Monkey.
        Cerebral Cortex. 2022 Jan 1; 32: 15-28
      22. Gămănuţ R, Kennedy H, Toroczkai Z, Ercsey-Ravasz M, Van Essen DC, Knoblauch K, et al. The Mouse Cortical Connectome, Characterized by an Ultra-Dense Cortical Graph, Maintains Specificity by Distinct Connectivity Profiles. Neuron. 2018 Feb 7;97(3):698-715.e10.

        • Roberts J.A.
        • Perry A.
        • Lord A.R.
        • Roberts G.
        • Mitchell P.B.
        • Smith R.E.
        • et al.
        The contribution of geometry to the human connectome.
        NeuroImage. 2016 Jan 1; 124: 379-393
        • Alexander-Bloch A.F.
        • Shou H.
        • Liu S.
        • Satterthwaite T.D.
        • Glahn D.C.
        • Shinohara R.T.
        • et al.
        On testing for spatial correspondence between maps of human brain structure and function.
        NeuroImage. 2018; 178: 540-551
        • Burt J.B.
        • Helmer M.
        • Shinn M.
        • Anticevic A.
        • Murray J.D.
        Generative modeling of brain maps with spatial autocorrelation.
        NeuroImage. 2020 Oct 15; 220117038
        • Markello R.D.
        • Misic B.
        Comparing spatial null models for brain maps.
        NeuroImage. 2021 Aug 1; 236118052
        • Alexander-Bloch A.
        • Raznahan A.
        • Bullmore E.
        • Giedd J.
        The Convergence of Maturational Change and Structural Covariance in Human Cortical Networks.
        J Neurosci. 2013 Feb 13; 33: 2889-2899
        • Gordon E.M.
        • Laumann T.O.
        • Adeyemo B.
        • Huckins J.F.
        • Kelley W.M.
        • Petersen S.E.
        Generation and Evaluation of a Cortical Area Parcellation from Resting-State Correlations.
        Cerebral Cortex. 2016 Jan 1; 26: 288-303
        • Váša F.
        • Seidlitz J.
        • Romero-Garcia R.
        • Whitaker K.J.
        • Rosenthal G.
        • Vértes P.E.
        • et al.
        Adolescent Tuning of Association Cortex in Human Structural Brain Networks.
        Cereb Cortex. 2018 Jan 1; 28: 281-294
      23. Baum GL, Cui Z, Roalf DR, Ciric R, Betzel RF, Larsen B, et al. Development of structure–function coupling in human brain networks during youth. Proceedings of the National Academy of Sciences. 2020 Jan 7;117(1):771–778.

        • Cornblath E.J.
        • Ashourvan A.
        • Kim J.Z.
        • Betzel R.F.
        • Ciric R.
        • Adebimpe A.
        • et al.
        Temporal sequences of brain activity at rest are constrained by white matter structure and modulated by cognitive demands.
        Commun Biol. 2020 May 22; 3: 1-12
      24. Vázquez-Rodríguez B, Suárez LE, Markello RD, Shafiei G, Paquola C, Hagmann P, et al. Gradients of structure–function tethering across neocortex. Proceedings of the National Academy of Sciences. 2019 Oct 15;116(42):21219–21227.

        • Wagner H.H.
        • Dray S.
        Generating spatially constrained null models for irregularly spaced data using Moran spectral randomization methods.
        Methods in Ecology and Evolution. 2015; 6: 1169-1178
      25. Markello RD, Hansen JY, Liu ZQ, Bazinet V, Shafiei G, Suárez LE, et al. neuromaps: structural and functional interpretation of brain maps. bioRxiv; 2022. p. 2022.01.06.475081.

        • Ashburner M.
        • Ball C.A.
        • Blake J.A.
        • Botstein D.
        • Butler H.
        • Cherry J.M.
        • et al.
        Gene Ontology: tool for the unification of biology.
        Nature Genetics. 2000 May; 25: 25-29
      26. Grothe MJ, Sepulcre J, Gonzalez-Escamilla G, Jelistratova I, Schöll M, Hansson O, et al. Molecular properties underlying regional vulnerability to Alzheimer’s disease pathology. Brain. 2018 Sep 1;141(9):2755–2771.

        • Krienen F.M.
        • Yeo B.T.T.
        • Ge T.
        • Buckner R.L.
        • Sherwood C.C.
        Transcriptional profiles of supragranular-enriched genes associate with corticocortical network architecture in the human brain.
        PNAS. 2016 Jan 26; 113: E469-E478
      27. Writing Committee for the Attention-Deficit/Hyperactivity Disorder, Autism Spectrum Disorder, Bipolar Disorder, Major Depressive Disorder, Obsessive-Compulsive Disorder, and Schizophrenia ENIGMA Working Groups. Virtual Histology of Cortical Thickness and Shared Neurobiology in 6 Psychiatric Disorders.
        JAMA Psychiatry. 2021 Jan 1; 78: 47-63
      28. Paquola C, Bethlehem RA, Seidlitz J, Wagstyl K, Romero-Garcia R, Whitaker KJ, et al. Shifts in myeloarchitecture characterise adolescent development of cortical gradients. Gold JI, Satterthwaite T, Mills K, Fulcher B, editors. eLife. 2019 Nov 14;8:e50482.

        • Shin J.
        • French L.
        • Xu T.
        • Leonard G.
        • Perron M.
        • Pike G.B.
        • et al.
        Cell-Specific Gene-Expression Profiles and Cortical Thickness in the Human Brain.
        Cereb Cortex. 2018 Sep 1; 28: 3267-3277
        • Wei Y.
        • de Lange S.C.
        • Pijnenburg R.
        • Scholtens L.H.
        • Ardesch D.J.
        • Watanabe K.
        • et al.
        Statistical testing in transcriptomic-neuroimaging studies: A how-to and evaluation of methods assessing spatial and gene specificity.
        Human Brain Mapping. 2022; 43: 885-901
        • Fulcher B.D.
        Discovering Conserved Properties of Brain Organization Through Multimodal Integration and Interspecies Comparison.
        J Exp Neurosci. 2019 Jan 1; 131179069519862047
        • Kanehisa M.
        • Goto S.
        KEGG: Kyoto Encyclopedia of Genes and Genomes.
        Nucleic Acids Res. 2000 Jan 1; 28: 27-30
        • Wang L.
        • Jia P.
        • Wolfinger R.D.
        • Chen X.
        • Zhao Z.
        Gene set analysis of genome-wide association studies: Methodological issues and perspectives.
        Genomics. 2011 Jul 1; 98: 1-8
      29. Hansen JY, Markello RD, Tuominen L, Nørgaard M, Kuzmin E, Palomero-Gallagher N, et al. Correspondence between gene expression and neurotransmitter receptor and transporter density in the human brain. bioRxiv; 2021 p. 2021.11.30.469876.

        • Eze U.C.
        • Bhaduri A.
        • Haeussler M.
        • Nowakowski T.J.
        • Kriegstein A.R.
        Single-cell atlas of early human brain development highlights heterogeneity of human neuroepithelial cells and early radial glia.
        Nat Neurosci. 2021 Apr; 24: 584-594
        • Armand E.J.
        • Li J.
        • Xie F.
        • Luo C.
        • Mukamel E.A.
        Single cell sequencing of brain cell transcriptomes and epigenomes.
        Neuron. 2021 Jan 6; 109: 11-26
        • Chaudhari N.
        • Hahn W.E.
        Genetic Expression in the Developing Brain.
        Science. 1983 May 27; 220: 924-928
        • Ip B.K.
        • Wappler I.
        • Peters H.
        • Lindsay S.
        • Clowry G.J.
        • Bayatti N.
        Investigating gradients of gene expression involved in early human cortical development.
        Journal of Anatomy. 2010; 217: 300-311
        • Somel M.
        • Guo S.
        • Fu N.
        • Yan Z.
        • Hu H.Y.
        • Xu Y.
        • et al.
        MicroRNA, mRNA, and protein expression link development and aging in human and macaque brain.
        Genome Res. 2010 Jan 9; 20: 1207-1218
        • Luo L.
        • Callaway E.M.
        • Svoboda K.
        Genetic Dissection of Neural Circuits: A Decade of Progress.
        Neuron. 2018 Apr 18; 98: 256-281
        • Mitchell K.J.
        • Huang Z.J.
        • Moghaddam B.
        • Sawa A.
        Following the genes: a framework for animal modeling of psychiatric disorders.
        BMC Biology. 2011 Nov 11; 9: 76
        • Zhu Z.
        • Zhang F.
        • Hu H.
        • Bakshi A.
        • Robinson M.R.
        • Powell J.E.
        • et al.
        Integration of summary data from GWAS and eQTL studies predicts complex trait gene targets.
        Nat Genet. 2016 May; 48: 481-487
        • Lonsdale J.
        • Thomas J.
        • Salvatore M.
        • Phillips R.
        • Lo E.
        • Shad S.
        • et al.
        The Genotype-Tissue Expression (GTEx) project.
        Nature Genetics. 2013 Jun; 45: 580-585
        • Akbarian S.
        • Liu C.
        • Knowles J.A.
        • Vaccarino F.M.
        • Farnham P.J.
        • Crawford G.E.
        • et al.
        The psychENCODE project.
        Nature Neuroscience. 2015 Dec; 18: 1707
        • Vos de Wael R.
        • Benkarim O.
        • Paquola C.
        • Lariviere S.
        • Royer J.
        • Tavakol S.
        • et al.
        BrainSpace: a toolbox for the analysis of macroscale gradients in neuroimaging and connectomics datasets.
        Commun Biol. 2020 Mar 5; 3: 1-10
        • Giacomel A.
        • Martins D.
        • Frigo M.
        • Turkheimer F.
        • Williams S.C.R.
        • Dipasquale O.
        • et al.
        Integrating neuroimaging and gene expression data using the imaging transcriptomics toolbox.
        STAR Protocols. 2022 Jun 17; 3101315
      30. Giacomel A, Martins D, Frigo M, Turkheimer F, Williams SCR, Dipasquale O, et al. The Imaging Transcriptomics Toolbox [Internet]. Zenodo; 2022 [cited 2022 Jun 7]. Available from: https://zenodo.org/record/6364963

      31. Fulcher B, AurinaBMH. benfulcher/GeneCategoryEnrichmentAnalysis: Updated README [Internet]. Zenodo; 2021 [cited 2022 Jun 7]. Available from: https://zenodo.org/record/4470238

      32. Lotter LD, Dukart J, Fulcher BD. ABAnnotate: A toolbox for ensemble-based multimodal gene-category enrichment analysis of human neuroimaging data [Internet]. Zenodo; 2022 [cited 2022 Jun 7]. Available from: https://zenodo.org/record/6463329

        • Miller J.A.
        • Ding S.L.
        • Sunkin S.M.
        • Smith K.A.
        • Ng L.
        • Szafer A.
        • et al.
        Transcriptional landscape of the prenatal human brain.
        Nature. 2014 Apr; 508: 199-206
        • GTEx Consortium
        Human genomics. The Genotype-Tissue Expression (GTEx) pilot analysis: multitissue gene regulation in humans.
        Science. 2015 May 8; 348: 648-660
        • Sjöstedt E.
        • Zhong W.
        • Fagerberg L.
        • Karlsson M.
        • Mitsios N.
        • Adori C.
        • et al.
        An atlas of the protein-coding genes in the human, pig, and mouse brain.
        Science. 2020 Mar 6; 367eaay5947
        • Zhang Y.
        • Sloan S.A.
        • Clarke L.E.
        • Caneda C.
        • Plaza C.A.
        • Blumenthal P.D.
        • et al.
        Purification and Characterization of Progenitor and Mature Human Astrocytes Reveals Transcriptional and Functional Differences with Mouse.
        Neuron. 2016 Jan 6; 89: 37-53
        • Nowakowski T.J.
        • Bhaduri A.
        • Pollen A.A.
        • Alvarado B.
        • Mostajo-Radji M.A.
        • Di Lullo E.
        • et al.
        Spatiotemporal gene expression trajectories reveal developmental hierarchies of the human cortex.
        Science. 2017 Dec 8; 358: 1318-1323
        • Speir M.L.
        • Bhaduri A.
        • Markov N.S.
        • Moreno P.
        • Nowakowski T.J.
        • Papatheodorou I.
        • et al.
        UCSC Cell Browser: visualize your single-cell data.
        Bioinformatics. 2021 Dec 1; 37: 4578-4580
        • Darmanis S.
        • Sloan S.A.
        • Croote D.
        • Mignardi M.
        • Chernikova S.
        • Samghababi P.
        • et al.
        Single-Cell RNA-Seq Analysis of Infiltrating Neoplastic Cells at the Migrating Front of Human Glioblastoma.
        Cell Rep. 2017 Oct 31; 21: 1399-1410