Just a SNP away: The future of in vivo massively parallel reporter assay

Abstract

The human genome is largely noncoding, yet the field is still grasping to understand how noncoding variants impact transcription and contribute to disease etiology. The massively parallel reporter assay (MPRA) has been employed to characterize the function of noncoding variants at unprecedented scales, but its application has been largely limited by the in vitro context. The field will benefit from establishing a systemic platform to study noncoding variant function across multiple tissue types under physiologically relevant conditions. However, to date, MPRA has been applied to only a handful of in vivo conditions. Given the complexity of the central nervous system and its widespread interactions with all other organ systems, our understanding of neuropsychiatric disorder-associated noncoding variants would be greatly advanced by studying their functional impact in the intact brain. In this review, we discuss the importance, technical considerations, and future applications of implementing MPRA in the in vivo space with the focus on neuropsychiatric disorders.

1. Introduction

The most fundamental principle of biology is the Central Dogma, yet only 2% of the genome is protein coding (di Iulio et al., 2018). This small portion of the genome can be deciphered based on the well-established code for amino acids, start and stop codons, but there is no equivalent to the codon chart for the other 98% of the genome, hereon referred to as the noncoding genome. Cis-regulatory elements (CREs) within the noncoding genome include promoters, enhancers, silencers, and insulators, all of which can influence gene expression by regulating transcription factor (TF) binding and chromatin architecture in a cell type-specific manner (Nguyen et al., 2016; Nord & West, 2020; Segert et al., 2021; Özdemir & Gambetta, 2019). Consequently, noncoding variants within CREs can alter gene expression, but the downstream effects and clinical consequences of such variation require further investigation.

The same principle applies to neuropsychiatric disorder-associated noncoding variants, which have been traditionally identified through genome-wide association studies (GWASs). GWAS compares allele frequencies between neurotypical controls and individuals with a specific disorder to identify single nucleotide polymorphisms (SNPs) associated with that disorder. The importance of the noncoding genome in human health and disease is exemplified by the observation that about 90% of GWAS hits for any given disorder are in the noncoding genome (Maurano et al., 2012; Watanabe et al., 2019). As an additional complication, nearby SNPs on a chromosome are inherited together, leading to linkage disequilibrium (LD) which hinders identification of the putatively causal variant among GWAS-identified SNPs within a single locus. Until recently, the functional relevance of these noncoding SNPs had been poorly characterized. However, emerging evidence shows that SNPs in CREs can alter gene regulation and downstream transcription which subsequently influence the development of neuropsychiatric conditions.

The classic method to measure regulatory activity of noncoding variants is a luciferase assay which utilizes fluorescence of luciferase, a reporter gene, to indicate the activity of the upstream regulatory element that encompasses a SNP. While this method is effective at measuring the regulatory activity of one variant at a time, it does not allow the simultaneous measurement of multiple variants. As sample sizes increase, neuropsychiatric GWASs now identify thousands of variants associated with a single disorder (Flint, 2023; Horwitz et al., 2019; Uffelmann et al., 2021). For example, the largest GWAS to date for major depressive disorder (MDD) identified 14,016 genome-wide significant (GWS) hits in 178 risk loci using data from over 1.2 million individuals (Levey et al., 2021). Consequently, the need for high-throughput assays has grown to keep pace with the speed of variant discovery. To improve upon the traditional methods, the massively parallel reporter assay (MPRA) was introduced in 2009 and has only recently become widely implemented (Patwardhan et al., 2009, 2012). In this review, we focus on how MPRA can be applied to in vivo models to study the regulatory impact of neuropsychiatric disorder-associated noncoding variants in the natural physiological setting where the complex cell types, circuitry, and organ-level interactions remain intact.

2. Massively parallel reporter assay (MPRA)

The general MPRA workflow for variant characterization first involves creating a library of variants based on GWAS hits for the disorder of interest (Fig. 1A). This variant selection can be informed by other genomic resources such as expression quantitative trait loci (eQTLs) or epigenetic marks (e.g. H3K27ac, ChIP-seq, ATAC-seq) as well as computational tools like fine-mapping (Fig. 1A). The resulting library, which can contain over 100,000 oligonucleotide sequences, is then inserted into the CRE region of a plasmid, creating a pool of different plasmids, each containing a unique disease-associated variant (DAV; Fig. 1B). MPRA's high-throughput ability can be attributed to the random assignment of barcode sequences to each DAV. To test the variant effects, it is recommended that each DAV be paired with dozens of unique barcodes. This barcode representation mitigates the potential for barcode sequence effects on gene transcription (Lee et al., 2021). The resulting DAV-barcode pairs are mapped with high-throughput sequencing (Fig. 1B). The final construct contains a 150–250 base-pair (bp) sequence with the unique DAV spatially centered or within the relevant ATAC-seq peak, a generic promoter, a reporter gene, and a unique barcode sequence (Fig. 1B). Minimal promoters have traditionally been utilized in MPRA constructs so that the elemental effects are pronounced, but other generic promoters (e.g., human cytomegalovirus (CMV) promoter) or cell type-specific promoters (e.g., human synapsin (hSyn) promoter) could also be employed. Common reporter genes used in MPRA constructs include luciferase, green fluorescent protein (GFP), and red fluorescent protein (RFP; Fig. 1B). When MPRA is applied to an in vitro system, this pool of plasmids is then transfected into cells either by electroporation, lipofectamine, or viral infection (Fig. 1C). After an incubation period during which plasmids are transcribed, the cells are lysed and barcode RNA and DNA are isolated and sequenced via next-generation sequencing (NGS; Fig. 1C). The RNA barcode counts are compared to the initial plasmid DNA input, and the net barcode expression indicates the regulatory activity of the DAV paired with each barcode (Fig. 1C). The canonical MPRA design can be modified by inserting the DAV library into different regions of the plasmid, thereby studying a distinct class of noncoding elements. Details on each of these MPRA designs are beyond the scope of this review but can be found in McAfee et al., 2022. MPRA's high-throughput capacity enables it to screen the regulatory activity of thousands of noncoding variants simultaneously, aiding in the identification of putatively causal SNPs from GWAS loci riddled with SNPs in LD. Thus, MPRA promotes variant discovery and characterization to occur at a similar pace.

Fig. 1.

General massively parallel reporter assay (MPRA) pipeline.(A) Creating the MPRA library involves selecting variants from the GWAS for the disorder(s) of interest. Variant selection can be aided by statistical fine-mapping as well as other assays such as ATAC-seq, epigenetic markers, ChIP-seq, and/or eQTLs. (B) The MPRA construct is composed of the variant library, a general promoter, a reporter gene, and random barcode sequences. Each DAV should be paired with at least a dozen different barcode sequences. (C) The MPRA library can be introduced to cells via lipofectamine, electroporation, or virus. To administer the MPRA library to animal models, the construct must be packaged into virus then can be injected intravenously (IV), retro-orbitally (RO), or via direct injection into the organ of interest (i.e. stereotaxic or intramuscular). After an incubation period, the cells or tissues are lysed then RNA and DNA are isolated. The barcode sequences are then amplified off of the nucleic acids and are sequenced via NGS. The RNA barcode counts are compared to the initial plasmid DNA input, and the net barcode expression indicates the regulatory activity of the DAV paired with each barcode.

Once MPRA identifies functional regulatory variants, their regulatory mechanisms can be further investigated through integrative analyses with other genomic datasets, such as TF binding motifs (to identify the transcriptional regulators through which the variants exert their effects), cell type-specific enhancers (to determine the cell types in which the variants are active), and eQTLs (to identify target genes of the variants). Databases like FAVOR and RegulomeDB also facilitate the functional annotation of MPRA-validated variants (Dong et al., 2023; Zhou et al., 2022). Empirical fine-mapping results can be compared with computational fine-mapping approaches to assess how well variant regulatory activity correlates with the predicted causal configuration. Additionally, emerging machine learning models, such as Enformer and chromBPNet, aim to reliably predict the regulatory effects of these variants (Avsec et al., 2021; Pampari et al., 2023).

3. MPRA on neuropsychiatric disorders

While MPRA has been adopted in various fields to determine which GWAS variants exhibit functional regulatory activity, significant strides have been made in empirical characterization of variants associated with neuropsychiatric disorders. For example, schizophrenia GWAS has identified strong association signals in the CACNA1C locus that harbors over 100 SNPs reaching the GWS threshold (P < 5 × 10⁻⁸) (Pardiñas et al., 2018; Ripke et al., 2014; Trubetskoy et al., 2022). Roussos and colleagues performed luciferase assays in Neuro2A and HEK293T cells to pinpoint rs2159100 as a functional regulatory variant (Roussos et al., 2014). This SNP was located in an enhancer region that exhibits increased interaction frequency with the CACNA1C promoter in the adult prefrontal cortex. Similarly, McAfee and colleagues performed MPRAs in human neural progenitor cells (hNPCs) and identified another functional regulatory variant, rs11062170, which is located within an enhancer that interacts with the promoter of CACNA1C in adult neurons (McAfee et al., 2023). While these studies have identified different regulatory variants, the mechanism of CACNA1C-related risk for schizophrenia has been corroborated by two studies, as the risk alleles for both rs2159100 and rs11062170 were associated with decreased transcriptional activity of the reporter gene (McAfee et al., 2023; Roussos et al., 2014). Together, these results demonstrate the utility of reporter assays in pinpointing regulatory variants within complex GWAS loci. Furthermore, they suggest potential cell type-specificity of the variant function.

To date, MPRAs for neuropsychiatric DAVs have been performed using a multitude of relevant cell types including neuroblastoma cells, human embryonic stem cells (hESCs), hNPCs, induced pluripotent stem cell (iPSC)-derived NPCs and neurons, primary cultures of mouse and human neurons, microglia, and human iPSC-derived cerebral organoids (Chen et al., 2024; Deng et al., 2024; Kosicki et al., 2024; McAfee et al., 2023; Mulvey & Dougherty, 2021; Myint et al., 2020; Rummel et al., 2023) (Table 1). However, neuropsychiatric MPRAs have also been conducted in non-brain cell types such as K562 lymphoblasts and HEK293T cells which have emphasized the cell type-specificity of MPRA results (McAfee et al., 2023; Myint et al., 2020). Since DAVs are enriched in cell type-specific CREs, and the presence of cell type-specific TFs can influence MPRA results, researchers must be mindful in choosing a cell system that best captures the biology of the disease of interest (Cusanovich et al., 2015; Hu et al., 2021; Li et al., 2023; Nott et al., 2019; Sey et al., 2020, 2022; Skene et al., 2018). However, even when researchers use the most appropriate cell type, in vitro MPRA lacks tissue specificity in the sense that cell culture cannot recapitulate the cellular diversity and interactions present within a single organ, especially in the central nervous system (CNS). Cell type-specific regulatory architecture as well as the impact of cell-cell interactions on regulatory architecture highlights the need for the development of in vivo MPRA.

Table 1.

In vitro MPRAs for neuropsychiatric disorder risk variants.

Disorder(s) of interest	Variant selection criteria	Library size (# of sequences)	Cellular system(s)	Reference
Various neurodevelopmental, neurodegenerative, and psychiatric disorders.	Brain QTLs overlapping with pseudo-bulked ATAC-seq peaks that are (1) close to differentially expressed genes in schizophrenia, autism spectrum disorder (ASD), or bipolar or (2) in LD with GWAS SNPs of aforementioned disorders	102,767	Primary human cortical cells & iPSC-derived cerebral organoids	Deng et al. (2024)
Various neurodevelopmental and psychiatric disorders	All lead SNPs and SNPs in LD (r2 > 0.8) from ASD, schizophrenia, bipolar disorder, and depression GWAS	81,952	iPSC-derived excitatory neurons	Kosicki et al. (2024)
Schizophrenia	Fine-mapped variants	5,173	hNPCs & HEK293T	McAfee et al. (2023)
MDD	Variants overlapping with eQTLs or epigenetic marks	1,454	Neuroblastoma cells	Mulvey and Dougherty (2021)
Schizophrenia	Loci where SNPs have p-values <15 fold larger than that of the locus' lead SNP	2,350	K562 cells & human neuroblastoma cells	Myint et al. (2020)
Schizophrenia	Variants overlapping with eQTLs or epigenetic marks	5,118	iPSCderived NPCs and neurons; mouse NPCs andneurons	Rummel et al. (2023)

4. How MPRA has previously been applied in vivo

MPRA has been applied to a variety of mammalian tissues whether they be explanted or stereotaxically injected. Before MPRA was fully-established, Kwasnieski and colleagues introduced a similar technique, which they called CRE analysis by sequencing (CRE-seq), to measure the regulatory impact of over 1,000 single or double nucleotide mutations within a CRE associated with the Rhodopsin promoter by electroporating the library into explanted newborn mouse retinas (Kwasnieski et al., 2012). Because this promoter is specific to mammalian photoreceptors, their investigation was limited to one cell type (Kwasnieski et al., 2012). White and colleagues then followed up on these results by using the same technique to determine whether larger genomic context or local sequence features are more important in the characterization of functional CREs versus non-functional noncoding motifs (White et al., 2013). They compared the regulatory activity of Cone-rod homeobox (Crx)-bound CREs identified by ChIP-seq with unbound DNA regions that exhibited a similar amount of Crx motifs and found that high local GC content was the most distinguishing characteristic between the two (White et al., 2013). Considering these studies were conducted prior to the widespread access to NGS, their experimental designs and results exhibited impressive science and innovation.

Since then, MPRA has been adapted in vivo to study noncoding variants specifically in the context of neurodevelopmental and neuropsychiatric disorders. Unlike in vitro MPRA, the main challenge in performing MPRA in vivo is to deliver the highly complex library to living organs. One method of doing so is electroporation of neonatal mice and subsequent explantation and dissection of the brain region of interest (Shen et al., 2016, 2021) (Table 2). Another method utilizes stereotaxic injection, which is a common practice in neuroscience for viral injection into a specific brain region due to its target accuracy and reliable viral transduction. Several groups have employed stereotaxic injections to introduce MPRA libraries into distinct populations of neurons (Hrvatin et al., 2019; Lagunas et al., 2023; Mulvey et al., 2023) (Table 2). Using this technique, Mulvey and colleagues identified functional MDD-associated regulatory variants in mature hippocampal neurons (Mulvey et al., 2023). Because they used an in vivo system, they were also able to detect sex-by-gene effects such as the enrichment of regulatory SNPs in glutamatergic hippocampal neurons found in females but not males (Mulvey et al., 2023). This is especially relevant as the prevalence of MDD in females is higher than that of males, presumably due to biological as well as societal factors (Salk et al., 2017). Stereotaxic injections can also be done at any developmental stage to target the cell state during a particular critical period or to compare variant activity between juvenile and adult animals (Lagunas et al., 2023; Lambert et al., 2021; Warren et al., 2022). These comparisons are crucial in understanding the various genetic switches responsible for neuronal differentiation and will need to be pursued further in the field.

Table 2.

In vivo MPRAs for neuropsychiatric disorder risk variants.

Disorder(s) of interest	Variant selection criteria	Library size (# of sequences)	Model system & Route of administration	Reference
ASD	De novo variants within annotated 3′ UTRs from Simons Simplex Collection	1,298	Murine cortical excitatory neurons & striatal medium spiny neurons via stereotaxic injection	Lagunas et al. (2023)
Neurodevel opmental disorders & Schizophrenia	Lead SNPs and SNPs in LD (r2 > 0.8) from epilepsy and schizophrenia GWAS	408	Murine forebrain via stereotaxic injection	Lambert et al. (2021)
MDD	Variants overlapping with eQTLs or epigenetic marks	>1,000	Murine hippocampal neurons via stereotaxic injection	Mulvey et al. (2023)
Bipolar disorder	Variants at the MIR2113/POU3F2 locus	3	ex vivo electroporation of murine brain	Shen et al. (2021)

The specificity of stereotaxic injection could also be considered a disadvantage as it does not provide a holistic view of regulatory function in the entire CNS. It is also valuable to compare DAVs in multiple organs to determine tissue-specific regulatory activity. At this point, the closest attempt to a systemic MPRA model has been performed by Brown and colleagues (Brown et al., 2022). Their whole-animal MPRA (WhAMPRA) permitted the detection of tissue-specific regulatory activity between murine liver and brain as well as microglia-like HMC3 cultured cells. Their library consisted of 642 enhancers, the majority of which were expected positive and negative controls for brain, liver, and immune cells based on previous MPRA experiments. Due to the library composition, this experiment primarily served as a proof of concept exhibiting that the authors could detect differential enhancer regulation and barcode expression in the expected tissues. The mode of delivery is particularly important for systemic MPRA because MPRA libraries must be introduced to multiple organs with varying tissue tropism. By injecting AAV-PHP.eB retro-orbitally, Brown and colleagues observed transduction in the brain and liver as well as heart, kidneys, testes, and ovaries (Brown et al., 2022). The global transduction across peripheral organs is surprising as AAV-PhP.eB was developed to enhance CNS-specific tropism (Goertsen et al., 2022). However, the RNA barcode abundance was too low to perform any analysis in organs besides the brain and liver. While WhAMPRA was successful in detecting differential enhancer activity in the liver versus the brain, this design is much lower throughput compared to established in vitro protocols which are capable of screening thousands of enhancers simultaneously (Table 1). Additionally, the targeted library design for brain and liver prevented detection of differential regulatory activity in other organs. Together, WhAMPRA serves as a strong validation that a systemic in vivo MPRA is possible, and it can be further improved by retaining the high-throughput capabilities of in vitro protocols while examining regulatory activity in multiple organs and cell types.

5. Applying a systemic in vivo MPRA to study the genetic etiology of CNS disorders

As a whole, regulatory genetics is highly context specific. Gene expression is sensitive to factors like cell types, differentiation trajectories, cellular stress, disease phenotypes, and immune responses (Burke et al., 2020; Flati et al., 2020; Inoue et al., 2019; McKenzie et al., 2018; Mufson et al., 2006; Silverman et al., 2014). While traditional in vitro MPRA excels at capturing noncoding regulatory architecture, the majority of MPRA experiments have been performed in cell monocultures and thus lack sufficient resolution to capture many of these external variables. The primary goal of developing a systemic MPRA in a murine model is to elucidate the genetic etiology of human diseases while accounting for tissue specificity, sex biases, and gene-environment interactions at an organism level. When investigating CNS disorders and diseases, there are several reasons in particular as to why performing MPRA in vivo is preferable over in vitro.

The brain is a highly heterogeneous organ which itself consists of many cell types, broadly grouped into neurons and glia (Fig. 2A). NPCs mature and differentiate based on both genetic programming and the surrounding extracellular milieu which can influence a cell's fate (Janowska et al., 2019; Massirer et al., 2011). CREs are vital in determining gene expression during cellular differentiation, and the degree of CRE activity is also impacted by the cellular niche (Inoue et al., 2019; Kreimer et al., 2022; Zheng et al., 2024). Additionally, it is now appreciated that the interaction between neurons and glia is critical for brain health and function, which is exemplified by their dynamic and concerted gene expression profiles in healthy and diseased brains (Janowska et al., 2019; Ling et al., 2024). For example, microglia, once thought to simply function as a phagocyte upon injury, play a role in essential neuronal processes such as neurogenesis, neuronal differentiation, synaptic pruning, and axonal guidance (Cserép et al., 2021). Recent advances in neuroscience have facilitated the discovery of complex membrane-membrane interactions between microglia and neurons which permit rapid and dynamic communication between the two cell types (Cserép et al., 2021). Furthermore, the tripartite synapse structure between a presynaptic axon, postsynaptic dendrite, and perisynaptic astrocytic process is the main mechanism by which astrocytes regulate synapse formation and function (Tan & Eroglu, 2021). These interactions do not only impact the neurons, as synaptic activity can acutely influence the astrocytic transcriptome (Hasel et al., 2017; Ling et al., 2024).

Fig. 2.

A systemic approach to molecular biology.(A) Organs are composed of a variety of cell types that work together (only a few representative cell types shown for each organ). Here, we depict the potential organ-level interactions between the CNS and other organs which may impact the functional role of noncoding CREs. (B) A schematic visualizes how neuronal activity determines neurotransmitter release, which in turn influences formation of neural circuitry. This circuitry can impact how noncoding SNPs influence gene transcription. SNPs and synaptic transmission also have a bidirectional relationship if the regulated gene is implicated in neurotransmitter synthesis, synaptic release, or receptors. (C) Environmental factors can influence regulatory activity and impact organs in unique ways, highlighting the need to study regulatory activity across various tissue types, especially when an external stimulus is applied.

In addition to neuron-glia interactions, complex neural networks also regulate neuronal activity, which can impact transcription due to activity-dependent gene programs that require neuronal depolarization and immediate early genes (IEG) like FOS (Fig. 2B) (Malik et al., 2014). In fact, membrane depolarization can lead to cell type-specific and activity-dependent cooperative binding of TFs to noncoding regions of the genome that are enriched with heritability for neuropsychiatric disorders (Boulting et al., 2021; Sanchez-Priego et al., 2022). While variant function may vary in response to neuronal activity and glial interactions, the majority of in vitro MPRA experiments are conducted in a single cell type in a resting state, which could hinder the capture of variant function in physiologically relevant conditions. Rummel and colleagues recently exemplified that chemically-induced neuronal activity can trigger regulatory function of some schizophrenia-associated variants not present in the resting state (Rummel et al., 2023). Therefore, incorporating these cell-cell interactions and resulting neuronal activity in the MPRA design is crucial to understand how noncoding variants contribute to CNS disorders.

When neuronal networks in different brain regions consistently communicate with each other, functional circuits arise. Circuitry is often examined in the context of a behavioral phenotype such as substance use disorders (SUDs), for which the intricate neural circuitry has been extensively studied (Koob & Volkow, 2016; Lüscher & Janak, 2021). While there are still many details to be uncovered, some brain regions such as the ventral tegmental area (VTA) and nucleus accumbens (NAc) are characteristic of the circuitry involved in the pathophysiology of SUDs, primarily via dopaminergic signaling (Lüscher & Janak, 2021). When studying the regulatory function of noncoding variants associated with SUDs, it is essential that the relevant circuitry be intact (Fig. 2B). Furthermore, genetic variants were found to be associated with brain structure, indicating their potential involvement in shaping neural circuitry (Brouwer et al., 2022; Grasby et al., 2020; Satizabal et al., 2019). However, the extent to which variant function differs across brain regions and its modulation by neural circuitry remains poorly understood. Therefore, neural pathways and hallmark cell types within the circuitry must be taken into consideration when analyzing MPRA results for CNS DAVs.

In addition to the complexity of neural circuitry, the CNS communicates with other organ systems via neuroendocrine mechanisms to regulate aspects of physiology and behavior in response to internal and environmental stimuli (Fig. 2A). For example, in response to a stressor or circadian cues, the hypothalamic-pituitary-adrenal (HPA) axis, including the hypothalamus, pituitary gland, and adrenal gland, releases a combination of hormones and glucocorticoids (GCs) in a tightly coordinated manner (Herman et al., 2016; Sheng et al., 2021). This signaling cascade and the negative feedback loops largely governing this process influence multiple organ systems, as GCs bind to glucocorticoid and mineralocorticoid receptors in the CNS and periphery (Herman et al., 2016; Sheng et al., 2021). These receptors directly influence transcription as they bind to glucocorticoid response elements (GREs) in regions of accessible chromatin and interact with other TFs, exerting cell type-specific effects (Flati et al., 2020; Hudson et al., 2018; John et al., 2011; Koning et al., 2019). Shaped by critical periods in development or prolonged exposure to stress, the HPA axis can become dysregulated, leading to impaired GC signaling and a maladaptive long-term stress response (Herman et al., 2016; Sheng et al., 2021; Yehuda & Seckl, 2011; Zorn et al., 2017). This dysregulation is implicated in neuropsychiatric, immune, metabolic, and cardiovascular disorders, but the genetic mechanisms driving dysregulation of this integral system, especially noncoding and sexually dimorphic mechanisms, have yet to be fully elucidated (Arnett et al., 2016; Chatzittofis et al., 2021; Gerritsen et al., 2017; van den Berg et al., 2020). MPRA can aid in this endeavor by determining which DAVs exhibit selective regulatory activity in a stressed context as demonstrated by Penner-Goeke and colleagues, who showed a relationship between GC-responsive DAVs and risk for psychiatric disorders using self-transcribing active regulatory region sequencing (STARR-seq) (Penner-Goeke et al., 2023).

The CNS is also impacted by other organs, which further emphasizes the importance of a systems approach to genetics (Fig. 2A). One example is the gut-brain axis, as the composition of gut microbiota can influence the transcription of genes implicated in aerobic glycolysis in hippocampal astrocytes (Margineanu et al., 2020). Only recently has the neuroscience field accepted the pervasive role of the gut-brain axis in the psychopathology of CNS disorders (Agirman & Hsiao, 2021; Cenit et al., 2017; Chen et al., 2021; Fried et al., 2021; Margolis et al., 2021). Not only is the microbiota shaped by early life events, but microbiota-deficient mice also exhibit impaired social behavior and drastic changes in several neurotransmitter systems (Cenit et al., 2017; Collins et al., 2013; Neufeld et al., 2011). The gut-brain axis exemplifies how organ systems communicate in ways that can influence organ transcriptomic architecture and function. However, whether these connections affect variant function or impact downstream signaling cascades has not been investigated, underscoring the importance of performing MPRA under physiological conditions.

In addition to interrogating variant function in the brain, performing MPRA in vivo reveals the true tissue specificity of variants. Several in vitro MPRA studies have demonstrated differential allelic regulatory function in distinct cell types (McAfee et al., 2023; Myint et al., 2020; Zhao et al., 2023). However, it is difficult to distinguish whether the cell type-specific activity originates from the genetic architecture of the cell types themselves or from batch effects since the unique cell types are cultured separately with different culture conditions and in individual batches. To distinguish true biological effects from technical artifacts, it is important to test DAV effects in a condition that encompasses various cell types simultaneously. Moreover, a subset of DAVs may function in multiple tissue types contributing to disease pathogenicity. It is expected that genetic risk factors for SUDs will not only affect brain function but will also be implicated in the molecular pathways of other relevant organs such as the lung for cigarette smoking or the liver for alcohol use. For example, the rs11940694 SNP in the intron of the KLB gene has been associated with alcohol consumption and misuse (Sanchez-Roige et al., 2020). KLB encodes beta-klotho, a cofactor for FGF21 binding, which is primarily expressed in the liver, pancreas, and adipose, and is part of a brain-liver feedback loop. Beta-klotho has also been shown to control ethanol preference in mice via a brain-specific, Klb and FGF21-dependent mechanism (Schumann et al., 2016). Furthermore, rs11940694 has been identified as an eQTL for RFC1 in the cerebellum (Sanchez-Roige et al., 2019). An in vivo MPRA will help determine whether SNPs like rs11940694 exhibit tissue-specific regulation. Additionally, the regulatory activity of many DAVs may amalgamate in a particular tissue type, but not in others, which can only be detected by examining each tissue type individually (Kim et al., 2024). Therefore, the systemic approach is necessary to understand how DAVs interact and influence the organism as a whole (Fig. 2). This includes sex-specific variant activity, as Mulvey and colleagues already demonstrated that DAVs can exhibit sex-specific regulation (Mulvey et al., 2023). The activity of each DAV can also change throughout development since enhancer activity is dynamic in a cell type-specific manner (Kreimer et al., 2022; Li et al., 2018; Nord et al., 2013; Rummel et al., 2023; Wang et al., 2018; Won et al., 2016; Zhu et al., 2023). A systemic in vivo approach will permit the study of CRE-dependent regulation of transcription at different developmental time points, in a sex-specific manner, and in all relevant tissue types.

Lastly, while it is often necessary to study genetic activity in isolation, it is essential to understand how environmental factors influence genetic regulatory function (Fig. 2C). Organisms provide the best model to study gene-environment interactions, as the animal encompasses the entire system upon which the environmental factor perturbs. The traditional MPRA design is useful for assessing regulatory effects of DAVs; however, it might overlook relevant condition-specific regulatory activity. A response MPRA workflow can capture changes in regulatory activity from an added stimulus (McAfee et al., 2022; Mulvey & Dougherty, 2021; Penner-Goeke et al., 2023; Rummel et al., 2023). Results from the response condition can then be compared to baseline expression from an MPRA lacking the external manipulation. Although response MPRA is a new technique that has not yet been fully developed even in vitro, we propose a similar approach using the systemic in vivo MPRA by subjecting mice to different environmental factors. As mentioned above, organ systems work together and influence each other, which is especially true when metabolizing and reacting to foreign compounds or toxins. This technique will be particularly useful when studying the genetic etiology of SUDs since an individual cannot become addicted to a substance unless it enters their system. Additionally, substances are processed by distinct organs aside from the CNS. For example, alcohol use disorder primarily damages the liver while cannabis use disorder can induce bronchitis in the lungs (Gracie & Hancox, 2021; Leggio & Lee, 2017). By adapting response MPRA to an in vivo model, we will have the capability of truly understanding how substances or other environmental toxins impact genetic regulation and potentially contribute to disease progression.

6. Optimizing systemic in vivo MPRA

There are crucial factors to consider during development of systemic in vivo MPRA. The first is choosing a viral vector for library delivery. Because lentivirus integrates directly into the genome, lentiviral MPRA libraries incorporate higher order chromatin context (Gordon et al., 2020; Inoue et al., 2017). However, this integration happens randomly, so the placement of the library may not necessarily recapitulate that of the DAV's natural chromatin context (Cabrera et al., 2022; Demeulemeester et al., 2015). Furthermore, this positional effect can decrease the signal-to-noise ratio, which is critical for detecting regulatory effects of variants that typically have small effect sizes. It has been shown that regional chromatin interactions act independently of intrinsic enhancer activity, and relative cis-regulatory activity remains highly correlated when enhancers are placed in various chromosomal locations (Maricque et al., 2019). Furthermore, we have previously illustrated that the endogenous chromatin context of a variant can be modeled by integrating episomal MPRA-measured allelic regulatory activity with chromatin accessibility (McAfee et al., 2023). Therefore, we argue that the use of an episomal adeno-associated virus (AAV) could effectively introduce DAVs to diverse living organs in a systemic manner.

It is then crucial to determine which AAV serotype will be most effective when targeting multiple organ systems, as each serotype has unique tissue tropism (Chai et al., 2023; Wu et al., 2006). Since there is no single serotype with high transduction efficiency of all major organ systems, it may be possible to utilize a mixed capsid approach, which could result in widespread transduction. For example, this AAV serotype mixture may include common serotypes such as AAV8 as well as engineered capsids like AAV-PHP.eB, which specifically targets the CNS, or AAV-SCH9, which exhibits a fast-acting and broad neuronal tropism (Goertsen et al., 2022; Ojala et al., 2018; Zheng et al., 2024). Another approach is to utilize a more generic serotype such as AAV9, which should infect most tissue types, but this will depend on the organs of interest. Finding the ideal serotype or combination of capsids is arguably the most important yet challenging step of optimization, as it requires completion of the entire experimental process without knowing whether the virus will induce effective transduction.

Another consideration in the vector design is whether to use a single-stranded (ssAAV) or self-complementary (scAAV) vector. The latter has higher transduction efficiency as it does not rely on the host's cellular machinery to synthesize the complementary DNA strand (Fu et al., 2003; Hirsch et al., 2010; McCarty, 2008; McCarty et al., 2003; Petersen-Jones et al., 2009; Yokoi et al., 2007). However, the double stranded nature of the scAAV limits the length of sequence that can be customized between the inverted terminal repeat (ITR) sequences since the total plasmid size cannot exceed 5 kilobases (kb), meaning the vector can deliver no more than 2.5 kb of unique transgene sequence (McCarty, 2008). This restricts how large the promoter, reporter, and sequences of interest can be. Thus, the investigator must determine whether a scAAV or ssAAV is more appropriate for their MPRA design.

Traditionally, a minimal promoter (minP) has been used in MPRA designs to test the enhancer activity of genomic sequences (Brown et al., 2022; Capauto et al., 2024; Deng et al., 2024; Kreimer et al., 2022; Lambert et al., 2021; Maricque et al., 2019; McAfee et al., 2022, 2023; Mulvey et al., 2023; Myint et al., 2020; Patwardhan et al., 2012; Rummel et al., 2023; Shen et al., 2016; Siraj et al., 2024; Warren et al., 2022). However, it has been shown in vitro that reporter expression was indistinguishable when preceded by minP and without a promoter (Lalanne et al., 2024). Thus, minP might not sufficiently induce transcription in vivo after systemic administration (Lalanne et al., 2024). Rather, it might be necessary to utilize a strong core promoter such as the elongation factor 1-alpha (EF1α) promoter, human CMV promoter, chicken β-actin (CBA) promoter, or the CMV early enhancer/β-actin (CAG) promoter (Gray et al., 2011; Lagunas et al., 2023; Mulvey et al., 2021; Plassmeyer et al., 2023; Wang et al., 2017). While these promoters are considered to be ubiquitously and constitutively active, there have been reports of CMV silencing both in vitro and in vivo after prolonged expression primarily due to methylation (Brooks et al., 2004; Cabrera et al., 2022; Gray et al., 2011). These promoters are also large in size which may send the vector design over the ∼5 kb packaging capacity. This can potentially be prevented by utilizing shorter versions of these strong promoters such as EF1s, CBh, and sCAG or by implementing novel micro-promoters (Cabrera et al., 2022; Chai et al., 2023; Gray et al., 2011; Nieuwenhuis et al., 2021). The promoter-capsid combination can also impact the cell type-specificity of transgene expression, so it is crucial to be cautious when trying novel combinations (Powell et al., 2020). Together, the investigator must consider vector packaging size as well as species, organs, and cell types of interest when choosing which promoter will set the baseline MPRA expression.

Systemic administration in mice can be done via tail vein injection or retro-orbital injection. While both routes serve the same purpose, the latter requires the animal to be under anesthesia whereas tail vein injections do not (Yardeni et al., 2011). Considering every factor in the experimental design can impact gene expression, it is important to mitigate confounding variables, such as the presence of an anesthetic. Thus, the investigator should decide the appropriate injection route based on their comfort with each technique, viral serotype, and organs of interest.

The next step is determining the incubation period post-injection. When MPRA libraries are stereotaxically injected directly into the brain, investigators have waited as little as one week before sacrificing the animal and dissecting the brain (Lambert et al., 2021). With the systemic MPRA administration, a longer incubation period of three to four weeks is likely necessary as the virus must circulate and infect organs from the bloodstream. However, the incubation period cannot be too extensive as the transcription products will degrade, negatively impacting the relative barcode ratio. It is also important to determine the proper viral dosage that induces sufficient transduction systemically without harming the animal. The dosage and volume will largely depend on the administration route, but a high viral titer (>10¹² viral genomes/mouse) is recommended.

The final step of optimization concerns the tissue processing and library preparation. Once organs are harvested, it is essential to maintain the DNA and RNA integrity as much as possible. Two ways of doing so are by flash freezing the organs or by submerging them in a solution such as RNAlater. The chosen method will also influence the tissue homogenization protocol, which may be pulverization for flash frozen tissues or bead homogenization for non-frozen tissues. Due to the proposed assay's systemic approach, investigators must consider the sheer quantity of tissue to be processed, especially when working with multiple animal cohorts. Thus, it is best to utilize DNA and RNA isolation methods that can load large amounts of lysate for faster purification while generating high quality products necessary for NGS. It is also essential to mitigate DNA contamination in the RNA pool and vice versa. DNA (or RNA) contamination is usually not a significant issue for in vitro MPRA but can be substantial when organs are homogenized due to the large number of cells and their heterogeneity. This can be prevented with DNase and RNase treatment during RNA and DNA extraction, respectively. These products then undergo library preparation and NGS with the appropriate read depth.

In summary, establishment of a systemic in vivo MPRA entails careful optimizing of construct design, maximizing transduction efficiency, and ensuring proper nucleic acid preparation to maintain the library integrity. By studying variant effects under physiologically relevant conditions, reliable systemic in vivo MPRA results will revolutionize how the neurogenetics field examines the noncoding genome and its contribution to the development of many common disorders.

7. The future of in vivo scMPRA

While a systemic in vivo MPRA will expand our understanding of noncoding genetics within a complex, physiologically-relevant context, it will unfortunately suffer from organ-level cellular heterogeneity. To enhance the resolution and interpretability of in vivo MPRA, single-cell RNA-sequencing (scRNA-seq) would aptly complement the in vivo MPRA workflow by providing further evidence of how variants alter regulatory activity in a cell type-specific manner. However, the increased cellular resolution offered by in vivo single-cell (sc)MPRA comes with various challenges. When designing a scMPRA library, the investigator must determine how to accurately measure gene expression in a complex array of cell types, consider cell type-specific CRE activity, and distinguish cells that do not have reporter expression from cells that lack reporters due to inefficient transduction (Fig. 3A) (Lalanne et al., 2024; Zhao et al., 2023). Thus, scMPRA requires a more highly complex MPRA library design to address these challenges. One approach utilized by emerging in vitro scMPRA studies involves engineering a dual-barcoded library with one barcode serving as a simple CRE identifier (signifying that a CRE is present or not) and a second unique barcode downstream of the reporter sequence, which serves to quantify reporter expression (Fig. 3B) (Lalanne et al., 2024; Zhao et al., 2023).

Fig. 3.

Diagram of current challenges, strategies, and utilities for implementing a systemicin vivoscMPRA methodology.(A) Current challenges that necessitate increasing MPRA library complexity before implementing in single-cell workflows. (B) Two different library-design strategies for implementing scMPRA workflow. Each strategy includes a way to identify specific CREs and a way to quantify their activity. These designs can be modified to contain additional complex components like capture sequences to distinguish silencer effects (above; Zhao et al., 2023) and also have been validated in complex systems like heterogeneous cell culture, ex vivo tissues, and embryoids. (C) Various output data produced by the scMPRA pipeline and various types of complementary assays that can be performed alongside scMPRA.

Two recent works utilize variations of this general design to assess gene expression in heterogeneous cellular contexts and address certain challenges presented by scMPRA. The construct used by Zhao et al. features a “CRS barcode” (cBC) and a “random barcode” (rBC), both located downstream of the promoter, CRE, and reporter gene (Zhao et al., 2023). The cBC serves as an identifier of CREs, while the rBC complements the cBC by serving as a copy-number detector of the reporter gene (Zhao et al., 2023). This way, regardless of initial reporter abundance due to a heterogeneous cellular environment, copy number can be normalized to the unique cBC-rBC pairs. Zhao and colleagues effectively captured cell type-specific and cell substate-specific CRE activity with sufficient reproducibility in a mixed culture of 1:1 K562 lymphoblasts and HEK293T cells as well as in ex vivo mouse retinae (Zhao et al., 2023).

In comparison, the construct designed by Lalanne et al. features dual RNA reporters called scQers. One scQer expresses a “Tornado barcode” (oBC) based on the Tornado RNA circularization system. The oBC is positioned upstream of the CRE but downstream of a RNA Pol-III promoter, which specifically encodes for a circular RNA product to increase transcript stability (Lalanne et al., 2024). Another scQer expresses a “mRNA barcode” (mBC) downstream of a RNA Pol II-transcribed promoter and a reporter gene (Fig. 3B) (Lalanne et al., 2024). Thus, the cassette design is segmented into two units, with the oBC ​+ ​CRE transcribed into circularized RNA via RNA Pol III to serve as an identifier of CRE presence and with the reporter gene ​+ ​mBC transcribed via RNA Pol II as a measure of reporter expression (Lalanne et al., 2024). Since a unique mBC is paired to each oBC ​+ ​CRE, the original construct addresses the identification-quantification challenge by ensuring a sufficient amount of unique mBC and oBC pairs. Lalanne and colleagues then performed a piggybac transfection with this construct on both mixed culture (HepG2, K562, and HEK293T) and mouse embryoid bodies (mEBs) to map CRE activity in complex heterogeneous cell populations (Lalanne et al., 2024).

These studies show that scMPRA can help determine cell type-specific and lineage-specific CRE activity (Fig. 3C) (Lalanne et al., 2024; Zhao et al., 2023). However, scMPRA has been applied to a relatively limited number of CREs; Lalanne and colleagues tested 209 CREs in mEBs while Zhao and colleagues tested 676 core promoters in mixed culture (Lalanne et al., 2024; Zhao et al., 2023). Additionally, this method has yet to be applied to DAVs, which typically have much smaller effect sizes than CREs. Therefore, application of in vivo scMPRA to a large number of DAVs will require additional optimization.

8. Conclusion

While tremendous strides have been made in deciphering the noncoding genome, our understanding of the regulatory principles underpinning noncoding DAVs is still limited. Traditional MPRA has greatly contributed to our characterization of DAVs, albeit primarily in steady state monoculture. By establishing a systemic in vivo MPRA, we will be able to discover the tissue and sex specificity of regulatory elements as well as how environmental stimuli impact such regulation. In vivo MPRA can be further extended to a single-cell level such that regulatory activity within unique cell types can be distinguished, which will enhance our understanding of cell type-specific contributions to diseases. After characterizing such nuanced functionality, the regulatory impact of MPRA-validated DAVs can be further explored using methods such as Perturb-seq and beeSTING-seq (Morris et al., 2023; Zheng et al., 2024). While these tools are relatively low-throughput compared to MPRA, they can complement MPRA datasets by identifying the target genes of DAVs in the endogenous genomic context in a cell type-specific manner. Eventually, different experimental strategies (e.g., MPRA, CRISPR screens) and computational approaches (e.g., statistical fine-mapping, machine-learning based computational tools) will need to be integrated to understand the molecular mechanisms underlying causal DAVs. The resulting knowledge can then guide diagnosis and treatment of complex diseases including neuropsychiatric disorders.

CRediT authorship contribution statement

Katherine N. Degner: Writing – review & editing, Writing – original draft, Visualization, Conceptualization. Jessica L. Bell: Writing – review & editing, Writing – original draft, Visualization. Sean D. Jones: Writing – review & editing, Writing – original draft, Visualization. Hyejung Won: Writing – review & editing, Supervision, Funding acquisition.

Declaration of competing interest

The authors have no conflict of interest to disclose.