To evaluate the efficiency of delivering our sysMPRA libraries to diverse mouse tissues, we employed a delivery system optimized for enhanced transduction and reproducibility. We designed the sysMPRA plasmid (pAAV-MPRAe) with an Hsp68 minimal promoter, as previously described by Lambert and colleagues (Lambert et al., 2021). This ensures maximum inducibility of transcription without driving high levels at baseline, a crucial feature for assessing library delivery. The candidate enhancer sequences and barcodes (MPRA insert) were cloned downstream of both the minimal promoter and the mCherry reporter. We introduced cloning sites within the plasmid backbone allowing for easy modular change of both promoter (promoter cloning sites) and enhancer sequences (MPRA cloning sites) (Supplementary Figure S1).

First, we assessed the transduction and expression of our mCherry reporter system by testing 3 cross-tissue positive control enhancer sequences each with a unique barcode (Supplementary Table S1). These sequences were cloned into the sysMPRA plasmid to create a test MPRA preliminary library (MPRAct) which allowed us to provide an easy means to confirm the functionality of our sysMPRA approach. We transduced the MPRAct library into wildtype adult mouse tissue, and we evaluated the transduction and transcription in targeted tissues, including neurons in multiple brain regions, by measuring mCherry fluorescence in tissue sections (Figure 2A). Indeed, our data clearly indicates the effective performance of the MPRAct library and, in turn, validates the sysMPRA approach.

Next, we implemented our designed library of 461 enhancers each paired with 20 unique barcodes (MPRAi) to assess the ability of sysMPRA to relate differences in genome sequences to regulatory differences (Figure 1A; Supplementary Table S2). The relatively large number of barcodes ensures that each enhancer is well-represented, even in the event of dropout at the cloning stage. We speculate that potential dropout rate can likely be induced by the barcode sequences integrating into the mRNA sequence template, thereby potentially impacting mRNA stability and translation directly. To establish a comprehensive library for our study, we designed the MPRAi library with the following components: (1) We included a collection of anticipated positive and negative controls derived from brain, liver, and immune cells, referencing their documented regulatory activity in prior MPRA experiments (Kheradpour et al., 2013; Nguyen et al., 2016). (2) We incorporated 144 candidate enhancers into the MPRAi library that are highly conserved and have mouse cortex H3K27ac chromatin immunoprecipitation sequencing (ChIP-seq) regions near genes implicated in vocal learning as well as their orthologs across species (see Methods). This aims to evaluate the potential of these enhancers to regulate genes associated with vocal learning and to improve the overall signal from mouse brain tissue (mouse brain specific candidate enhancers). (3) We introduced a set of 28 sequences known to bind the transcription factor MEF2C, implicated in transcriptional regulation across multiple brain regions (Harrington et al., 2016; Chen et al., 2016) as well as Alzheimer’s disease (AD) predisposition (Karch and Goate, 2015). This addition enables testing our sysMPRA technology to detect the impact of disrupting transcription factor binding sites. (4) We integrated a set of 27 candidate enhancers containing both the risk and the non-risk alleles of candidate regulatory AD-associated variants from GWAS (Lambert et al., 2013). This inclusion allows for the detection of the impact of SNPs.

The designed MPRAi library was cloned into the plasmid backbone to create the plasmid library (pAAV-MPRAi) (Figure 1B) and subsequently, we delivered the constructed MPRAi library into the brains of mice using retro-orbital injection of AAV-PHP. eB as described previously by our group (Lawler et al., 2022) (Figure 1C). We collected the mouse tissues and performed sectioning on the liver, frontal cortex, and the striatum. Next, we analyzed the mCherry fluorescence of the nuclear-associated reporter relative to NeuN (labels neurons, used to evaluate brain) or DAPI (labels nuclei, used to evaluate liver) levels by using confocal imaging (Figure 2B). Indeed, mCherry fluorescence was detected in liver, brain striatum and brain cortex cells, a good indication of systemic viral transduction of the MPRAi library in multiple tissues (Figure 2B).

Then, we analyzed the MPRAi plasmid library complexity itself to assess possible drop-out effects of the MPRAi library assembly caused by the cloning procedure. This allowed us to ascertain how many of the originally synthesized 20 unique barcodes were still present in the final library. This is an important metric, as it gives means to correct the number of barcodes actually present in the MPRAi library and therefore a more accurate assessment of the viral transduction rates of the sysMPRA technology. We performed plasmid DNA sequencing by using Illumina MiSeq and discovered that the MPRAi plasmid library complexity is 43% of the input library complexity. Thus, our MiSeq data show a 57% drop-out of barcodes introduced by the cloning procedure, but we still have 100% coverage of the candidate enhancers. Despite the unexpected and significant drop-out in the MPRAi library, we reasoned that an adequate number of barcodes per enhancer would be still available if our sysMPRA is functioning efficiently. Our sequence data analysis of the barcodes per enhancer showed a range of 1–18 barcodes, with an average of 8.6 barcodes (Supplementary Figure S2). Over 90% of the enhancers were associated with at least 5 unique barcodes in the final library (Supplementary Figure S2). In the future, adapting the cloning procedure is highly likely to yield a higher barcode library coverage.

Subsequently, we further investigated the viral transduction efficiency as well as the MPRAi library expression across various tissues by measuring the library complexity based on the number of unique barcodes detected at DNA and RNA level. The ability of the sysMPRA to detect transcriptional differences across tissues depends on the complexity of the library that can be transduced into each tissue. Enhancer regulatory activity is highly cell type- and tissue-specific (Jindal and Farley, 2021). Thus, the unique DNA barcodes recovered should be a function of the libraries ability to transduce. The unique RNA barcodes recovered will be a function of the set of designed enhancers to drive barcode expression. To measure transduced library complexity in each tissue, we counted the number of barcodes present in the viral DNA reads (Figure 2C). We found that most samples (excluding heart and muscle) across all tissues contained greater than 89% (ranging from 89% to 96%) of all measured barcodes (Figure 2C). The complexity of the brain regions and other tissues were only slightly less than complexity measured in transfected HMC3 cells (Figure 2C). This demonstrates our ability to efficiently transduce a complex library across a broad set of mouse tissues using AAV-PHP. ebb. The proportion of RNA barcodes recovered dropped, most likely due to poorly expressed candidate enhancers (Figure 2D). As expected, there was a greater drop in RNA barcodes relative to DNA barcodes for tissues outside the brain due to the library construction being heavily brain enhancer-focused.

Next, we cataloged the transduction within our MPRAi library by measuring the levels of unique barcodes across the various mouse tissues. Similarly, as before, we injected (retro-orbital) the library into 10 mice and collected the samples from multiple tissues (Supplementary Figure S3). Furthermore, we introduced the MPRAi library into the microglia-like HMC3 cell line (Dello et al., 2018), aiming to compare our sysMPRA in vivo technology with cell culture technologies. This approach simultaneously enabled us to investigate the potential function of MEF2 binding sites and AD-associated genetic variants, as previous studies have implicated both factors to microglia (Deczkowska et al., 2017; Gjoneska et al., 2015a). We employed a custom program, arrayProc.2.1.1. py, to analyze unique barcodes in each sequenced sample, enabling the quantification of barcode reads at the DNA level (see Supplemental Methods, computational analysis). Then, we refined the sequence data, retaining only high-quality barcode reads, which we defined as reads that matched with both the designed restriction enzyme site within the viral plasmid and the adjacent bases of the barcode. Our results demonstrate the identification of 3,983 high-quality unique barcodes across multiple tissues (Supplementary Table S3). This corresponds to an overall MPRAi library transduction rate of 95.6%, calculated as 3,983 (identified barcode sequences) divided by the total MPRAi library input barcodes 4,149 (461 × 9 barcodes). In other words, the MPRAi library drop-out was only 4.4%, meaning that we were able to detect, on average, the majority (8.6) of the 9 barcodes for each candidate enhancer.

Next, we evaluated the Spearman Rho (ρ) correlation of the plasmid DNA barcode measurements from each pair of samples. The ρ coefficients ranged from ρ = 0.737 to ρ = 0.991, with a median of ρ = 0.951 ( Supplementary Figure S4A; Supplementary Table S3). These findings clearly demonstrate that the identified barcodes of each sample are highly correlated across the various tissues. To get a clear overview of the number of detected barcodes within the various tissues, we displayed each tissue type (sample) and calculated the proportion of high-quality unique barcodes detected in the viral DNA ranging from no barcodes detected (0.0) to all barcodes detected (1.00, corresponding to 3,983 barcodes) (Supplementary Figure S4B). We show that each tissue sample and the HMC3 cell line has a high proportion of high quality unique barcodes detected, ranging from 0.82 (muscle tissue 10_2) to 0.99 (HMC3_C2), with an average of 0.94 ± 0.035. Thus, at minimum, we were able to detect 0.82 × 3,983 = 3,266 unique barcodes and at maximum 3,944 barcodes, a strong indication of widespread transduction (Supplementary Figure S4B). Collectively, these results unequivocally showcase the widespread transduction of our MPRAi library throughout all mouse tissues. Thus, our sysMPRA technology demonstrates its efficacy in facilitating robust in vivo transduction of the library.

To investigate the capability of sysMPRA to efficiently detect tissue-specificity of the candidate enhancers, we measured the RNA barcode expression across all the different tissues. This would allow us a direct comparison of the candidate enhancer activity across the various tissues. Leveraging a wealth of expertise accumulated over nearly a decade in studying neurobiology and non-coding regions within neural tissues, our special focus was directed to candidate enhancers in brain tissues. Additionally, we incorporated candidate enhancers for liver and immune cells to explore diverse tissues, evaluating the broad applicability of sysMPRA in detecting potential enhancer activity across animal tissues. This aimed to assess the robust functionality of the sysMPRA technology. It is crucial to recognize that the chosen panels of candidate enhancers tailored for the brain, liver, and immune-like cells (HMC3) are tissue-specific. This implies that the potential enhancer activity of these candidates will be most pronounced where the gene regulatory machinery is prevalent in the respective tissues. In simpler terms, enhancers designed specifically for the brain are likely to be active in the brain tissues, with minimal or no activity anticipated in the liver, immune cells or other tissues and vice versa.

To measure the barcode RNA expression, we extracted RNA from the various mouse tissues and performed RNA sequencing on the samples that passed our rigorous quality control with the Illumina NovaSeq (Figure 1D; Supplementary Figure S3). The barcode RNA expression levels were assessed by the amount of RNA barcodes detected in the tissue (RNA counts) of interest as it is a direct measurement of mRNA level expression. First, we used the RNA barcode counts (Supplementary Table S5) relative to the DNA barcode counts (Supplementary Table S4) to estimate the activity of all candidate enhancers with MPRAnalyze (Ashuach et al., 2019). We found that the tissue-specific candidate enhancers, including the likely positive control enhancers for these tissues (HMC3, liver, M1, cortex, hippocampus and striatum), had a strong tendency to be expressed relative to the negative control sequences (Figure 3A). The p-values of the MAD score for the candidate enhancers is extremely low, nearing zero, strongly suggesting high transcriptional activity as compared to the negative controls, which exhibit a shift in MAD score peak p-values toward higher values (between 0.35 and 0.80) (Figure 3A). Moreover, we directly compared the cortical candidate enhancers to the positive and negative controls, demonstrating that these enhancers are likely to drive substantial activity in the cortex (Supplementary Figure S5). Notably, two-thirds of positive controls and half of our candidate enhancers with MEF2 motifs activated transcription in the cortex. Compared to the positive control, the candidate cortical and MEF2 enhancers have similar distribution of MAD scores and have lower median MAD scores (Supplementary Figure S5A). Similarly, the proportion of candidate cortical and MEF2 enhancers were disproportionately enriched to have significant transcriptional activity (P < 8.210^-8, Supplementary Figure S5B). These findings align with expectations for cortical candidate enhancers in comparison to the positive controls. Subsequently, we investigated the candidate enhancers for activity in cultured HMC3 cells (Figure 3B) and for the brain tissues ( Figure 3C) by ratioing the DNA versus RNA reads, as it is a good proxy for the transcription activity of the enhancers. We show that, among the panel of candidate enhancers (461), numerous exhibit transcriptional activity in cultured HMC3 cells (Figure 3B) and in the brain tissues (cortical tissue) (Figure 3C). This is demonstrated by the significant detection of mRNA levels (RNA barcodes) expressed for several enhancers (red dots, p < 0.01) in both the HMC3 cells and the brain tissues ( Figures 3B, C). These findings also reveal that the most of our active candidate enhancers are predominantly identified in the brain and indeed align strongly with our expectations, considering the majority of candidate enhancers are primarily designed for brain tissues. Furthermore, when we compared the quality control metrics between the in vivo and HMC3 version of the experiments (Supplementary Table S3), we found that the data quality and signal distributions were similar. For example, the RNA:DNA ratios from the HMC3 cells (Figure 3B) showed comparable spread to the RNA:DNA ratios from brain tissue, like cortex (Figure 3C), a strong indication that the quality control metrics between the in vivo and in vitro experiments are consistent.

Next, we dissected the panel of positive control candidate enhancers in the MPRAi library in more depth by analyzing the MAD score of enhancer activity in the various tissues (HMC3, liver and brain) relative to the negative control candidate enhancers. This allowed us to evaluate our sysMPRA experimental approach to identify tissue-specific differences in enhancer activity in living mice. We tested the set of candidate enhancers active in both HEPG2 (liver-like) and K562 (immune) cells (Kheradpour et al., 2013), which showed a nominal trend toward expression in HMC3 cells (one-sided t-test p = 0.078), the liver (one-sided t-test p = 0.012), and the brain tissues (one-sided t-test p = 0.019) (Figure 3D). As expected, HEPG2-specific enhancers (liver) tended to be transcribed in only the liver (one-sided t-test p = 0.039). The set of control enhancers for cortical tissue (brain tissue) and MEF2+ enhancers (brain tissue) revealed the highest enhancer regulatory activity in the brain (one-sided t-test p = 0.00025 and p = 0.0000328, respectively) (Figure 3D). These results clearly demonstrate that the overall expression patterns of control candidate enhancers in the MPRAi library align with our expectations. Consequently, it proves that our sysMPRA technology identifies tissue-specific differences in enhancer activity in vivo, at least for the designed positive controls, and validates our experimental approach.

Subsequently, we evaluated the expression patterns across all candidate enhancers by calculating the Spearman’s Rho correlation between various sysMPRA tissues. We found a statistically significant correlation between enhancer activity across different brain tissues (Spearman Rho ρ = 0.348 to ρ = 0.433) (Figure 3E) and little correlation between brain versus liver enhancer activity (Spearman Rho ρ = 0.0018 to ρ = 0.0971) (Figure 3E). Indeed, this data aligns with our expectations, given that the vast majority of the candidate enhancers in our MPRAi library are designed for brain tissues, with only a few targeting the liver tissue. Furthermore, we included the microglia-like cell line (HMC3) in our calculations and showed significant correlations between enhancer activity across brain tissues (Spearman Rho ρ = 0.225 to ρ = 0.407) (Figure 3E). This can be explained by the fact that microglia cells are specialized immune cells residing in the central nervous system that play critical roles in the brain during the development, homeostasis, and pathologies and therefore can overlap with the MPRAi library candidate enhancer activities (Figure 3E). Interestingly, our data suggest that the microglia-like cell line has the strongest correlation with the hippocampus and cortex and indicates a moderate correlation with the MPRAi library candidate enhancer activity in the liver (Spearman Rho ρ = 0.225) (Figure 3E). This may be because many of our candidate enhancers are bound by the transcription factor MEF2C, which is known to play an important role in microglia (Deczkowska et al., 2017) Overall, our findings suggest that the estimated enhancer activity levels can reliably gauge regulatory activity across various tissues by our sysMPRA technology.

To verify the relevance of the enhancer activity we measured (Figure 3E) to the tissue-specific regulatory code, we compared the activity measured across all enhancers to machine learning model predictions of open chromatin (Kaplow et al., 2022; Zhou and Troyanskaya, 2015). Open chromatin prediction models are known to be correlated with enhancer activity (Roadmap et al., 2015) (Supplementary Table S6). To ensure that our MPRA could quantify tissue-specific regulatory activity, we assessed how well a model trained in a cell type for which our candidate enhancers were not designed to be active could predict MPRA regulatory activity; poor performance would indicate that the MPRA’s quantifications were tissue-specific. We found there was a weak but mostly positive correlation with the machine learning models trained in human embryonic stem cells Spearman Rho ρ = −0.001–0.076.

We then assessed the predictions of machine learning models trained on brain tissue open chromatin and observed significant correlations with sysMPRA measured enhancer activity in brain tissue (Spearman Rho ranching from ρ = 0.121 to ρ = 0.183 with p-values from 9.39 × 10⁻³ to 7.90 × 10⁻⁵). No correlation was found in liver tissue (Spearman Rho = −0.0117; p = 3.7 × 10⁻³) (Figure 3F). Similarly, the predictions from the machine learning models trained on liver open chromatin were significantly correlated with sysMPRA enhancer activity in liver (Spearman Rho = 0.158; p = 6.68 × 10⁻⁴) but not brain (Spearman Rho from −0.0112 to 0.0341 with p-value 0.466–0.838) (Figure 3F). Given that HMC3 cells model microglia, we tested whether machine learning models for a similar cell type, CD14⁺ macrophages (London et al., 2013), were able to capture HMC3 regulatory activity. We found a weak correlation between HMC3 regulatory activity predicted CD14⁺ monocyte open chromatin (Spearman Rho = 0.084; p = 0.07). The correlation was higher with liver, a tissue known to contain a large proportion of CD14⁺ macrophages (Spearman Rho = 0.184; p = 0.0007) (Dixon et al., 2013).

Finally, we conducted a thorough analysis of the RNA barcode expression in all sysMPRA transduced tissues from the mice and HMC3 cells (Figure 1D; Supplementary Table S5). This aimed to gain comprehensive insights of the activity of candidate enhancers in a wide range of tissues. Similarly, as described above, we evaluated the Spearman Rho (ρ) correlation and show a wide variety in the Spearman Rho correlation ranging from (ρ) = 0.366 to (ρ) 0.995 with a median of (ρ) = 0.629 (Supplementary Figure S4C; Supplementary Table S3). Our results consistently demonstrate a strong correlation between various brain tissues (M1, cortex, striatum, hippocampus, hypothalamus) and HMC3 cells, but low correlation in liver tissue (Supplementary Figure S4C; distinct colored red squares). This strongly suggests that the designed candidate enhancers for brain tissues exhibit a robust tendency to be active, leading to the expression of RNA barcodes. These findings match with MAD scores p-values, including positive controls for enhancer activity in liver (HEPG2), immune cells (K562), and cortical tissue as well as candidate enhancers with MEF2 transcription factor binding sites; they also align with Spearman Rho correlations for brain tissues, HMC3 cells, and liver tissue (Figure 3). As such, sysMPRA performs effectively in identifying tissue-specific transcriptional regulation of enhancers tailored for the specific tissues (brain, HMC3, and liver). The MPRAi library’s candidate enhancers, which are not designed for tissues like kidney and muscle, show consistently lower Spearman Rho correlations (Supplementary Figure S4C; different shades of blue-colored squares). These results suggest that the majority of these candidate enhancers are unlikely to exhibit significant activity in non-target tissues. However, there are a few exceptions. For example, some MPRAi library candidate enhancers show low activity in lung tissue relative to HMC3, cortex, and hippocampus and others have low enhancer activity in kidney tissue relative to HMC3 (Supplementary Figure S4C; lighter shades of red-colored squares). This indicates that a few MPRAi candidate enhancers might have some activity in lung and kidney tissue.

In addition, our results demonstrate that the RNA barcodes are less reproducible across samples (Supplementary Figure S4C; median of (ρ) = 0.629) than the DNA barcodes (Supplementary Figure S4A; median of ρ = 0.951). It is important to realize that these two entities represent completely different dynamics. DNA barcodes represent the viral transduction of the MPRAi library, indicating its ability to transduce into the various tissues in vivo. Meanwhile, RNA barcodes directly correlate with the gene regulatory capacity of the MPRAi library’s candidate enhancers in various tissues. In other words, the expression of RNA barcode is directly influenced by the activity of candidate enhancers in the tissues. Thus, detecting RNA barcodes necessitates some level of enhancer activity, while the detection of DNA barcodes is completely unrelated. For this reason, it is not surprising that RNA barcode detection declines in reproducibility across the tissue samples. Indeed, the highest proportion of unique detected barcodes at RNA levels are demonstrated for brain, HMC3, and liver samples, aligning well with the tailored design of the MPRAi library candidate enhancers for these tissues (Supplementary Figure S4D). Our findings also reveal that most other tissues (muscle, kidney, lung) had a significant proportion of unique RNA barcodes detected with a range of 0.5–0.85 for muscle and lung tissue, respectively (Supplementary Figure S4D). This indicates a significant recovery of 50%–85% of the RNA barcodes from the MPRAi library input in these tissues.

In summary, our data strongly indicates that sysMPRA is capable of evaluating estimated enhancer activity in vivo, demonstrating its reliability in measuring tissue-specificity regulatory activity of candidate enhancers across different tissues.

Current in vivo MPRA technology lacks the sensitivity to discern subtle activity differences arising from disruptions in individual transcription factor binding sites or SNPs. This limitation hampers its ability to fully capture the dynamic intricacies of gene regulation in non-coding genome regions within brain neural networks in their natural environment. Therefore, if sysMPRA could overcome this crucial limitation, it would allow researchers to study non-coding regions within the natural environment of brain neural networks. To evaluate whether sysMPRA is sensitive enough to detect subtle activity differences arising from disruptions in individual transcription factor binding sites or SNPs, we devised two strategies: (1) examining the influence of disruptions in transcription factor binding site motifs on enhancer activity in vivo, and (2) assessing the impact of disease-related SNPs on candidate enhancers and their effects on regulatory activity in vivo.

First, we assessed how disruption in transcription factor binding site motif MEF2 disrupts enhancer activity in vivo. As part of the MPRAi library, we designed a set of 28 candidate enhancer sequences based on binding the MEF2 transcription factor in the mouse cortex (see methods). Additionally, we created versions of each enhancer where the transcription factor binding site MEF2 itself was shuffled as well as a version where this motif together with the surrounding 5 nucleotides were shuffled (Figure 4A). We found that the non-disrupted MEF2 motif-containing enhancers had the strongest activity in brain tissue, some activity in HMC3 cells, and no activity in liver tissue (Figure 4B, MEF2 candidate enhancer (MEF2 OCR), panels cortex, HMC3 and liver). This aligns seamlessly with MEF2C’s function in both the brain and microglia (Harrington et al., 2016; Deczkowska et al., 2017; Telese et al., 2015) and by its absence in the liver (Baldarelli et al., 2021). Furthermore, our results on the MAD-scores of enhancer activity MEF2+ as described in this study confirm these results (Figure 3D; MEF2+ enhancers).

Consistent with these observations, we show a significant reduction in enhancer activity in cortical tissue and HMC3 cells upon disruption of MEF2 motifs (Figure 4B). The MAD-scores for the MEF2 candidate enhancers with MEF2 motifs shuffled are significantly lower than those for the original candidate enhancers with MEF2 motifs (Figure 4B; bottom and middle panel; one-tailed t-tests; p-value = 0.003 and p = 0.03, respectively). Additionally, the candidate enhancers with the MEF2 motifs and surrounding sequences shuffled have even lower enhancer activity (Figure 4B; bottom and middle panel; one-tailed t-tests; p-value = 0.002). However, disrupting MEF2 motifs in liver tissue does not result in a significant decline in enhancer activity (Figure 4B; top panel, p > 0.1). Importantly, the extent to which enhancer activity was disrupted by shuffling the MEF2C binding site was strongly correlated with the original baseline expression of the enhancer (Figure 4C; Supplementary Table S7; Rho = 0.88; p = 9.3 × 10⁻⁷). We found that 7/8 enhancers with a MAD score of >1.3 were significantly disrupted (FDR-adjusted paired t-test p value <0.05) while 0/20 enhancers with MAD score <1.3 were significant. This finding strongly implies that instances where disrupting the MEF2C transcription factor binding site has no impact are situations where the enhancer is inherently inactive in the assay, rather than indicating that the MEF2C binding site is unimportant for enhancer activity.

Next, we investigated the impact of candidate Alzheimer’s Disease (AD) GWAS-derived SNPs on candidate enhancers and their effects on regulatory activity in vivo (Figure 4). We utilized our sysMPRA technology to measure the impact of these AD-related SNPs and identified eight SNPs for which the risk and non-risk alleles showed significant (p < 0.05) and divergent activity across brain tissue (Figure 4D; Supplementary Table S8). This highlights that our sysMPRA technology is able to detect subtle enhancer activity differences arising from SNPs in these candidate enhancers (Figure 4D). Two of the eight candidate SNPs (rs6498140 and rs10991386) disrupt a MEF2 motif and for these SNPs, the disrupted motif sequence also displayed a difference from the reference allele (Figure 4D; far right, light blue bars, RegionDisrupt, p < 0.05) with the strongest effect seen for rs6498140 (Figure 4). This candidate AD-associated SNP rs6498140 is proximal to the gene CLEC16A, a gene implicated in AD that is a master regulator of autoimmunity and neurodegeneration (Pandey et al., 2023). In comparison to the reference allele (Figure 4E, MAD-score, light blue bars), the alternate allele has the highest regulatory activity in both brain tissue and in HMC3, but not liver (Figure 4E, MAD-score, red bars). This allele creates a MEF2 transcription factor binding site motif in the enhancer by the mutation C (Ref) to G (Alt) (Figure 4F). These findings are consistent with members of the MEF2 transcription factor family demonstrating active transcription in both brain and microglia (Deczkowska et al., 2017; Mitchell et al., 2018). Importantly, rs6498140 displays GTEx eQTL associations in several tissues, notably in the frontal cortex, where the alternate allele correlates with higher expression of CLEC16A (Consortium et al., 2013). These findings strongly suggest sysMPRA’s capability to perceive subtle activity differences stemming from SNPs and thereby demonstrating outstanding sensitivity. This not only validates previous studies but also broadens our understanding of genetic variation and gene regulation in Alzheimer’s Disease (AD) within relevant in vivo tissues.

Our data showcase that the sysMPRA technology possesses the sensitivity to detect even the most subtle activity differences resulting from disruptions in individual transcription factor binding sites (MEF2) or AD-related SNPs. The technological advantages of sysMPRA not only surpass current limitations in the field but also represent a significant step forward towards studying the dynamic complexities of gene regulation in non-coding genome regions within brain neural networks in vivo.