The goals of this project were to compare and contrast transcriptional and metabolic changes in the following experimental contexts: (1) T21 vs. D21, (2) ±IFNγ stimulation, (3) ±Mediator kinase inhibition, and (4) IFNγ stimulation + Mediator kinase inhibition. For T21 vs. D21 comparisons, we selected lymphoblastoid cells from the Nexus Biobank that were derived from siblings matched in age (3- or 5-year old) and sex (male). We used the natural product cortistatin A (CA) to inhibit CDK8 and CDK19 (Pelish et al., 2015). CA is the most potent and selective Mediator kinase inhibitor available (Clopper and Taatjes, 2022); for example, kinome-wide screens showed no off-target kinase inhibition even at 1 µM, a concentration 5000 times greater than its measured K_D of 0.2 nM (Pelish et al., 2015). Throughout this project, we used CA at a concentration of 100 nM, to minimize potential off-target effects. Immunoblots confirmed CA-dependent inhibition of STAT1 S727 phosphorylation, a known CDK8 substrate (Figure 1—figure supplement 1A).

Prior to completion of RNA-seq experiments, we probed the timing of the IFNγ response at IFN target genes GBP1 and IRF1 by RT-qPCR (Figure 1—figure supplement 1B). Based upon these results, we chose the 4-hr time point for RNA-seq (see Methods). We also completed experiments in T21 cells after 24 hr ±CA. An overview of the RNA-seq experiments is shown in Figure 1A, with data for all genes across all conditions in Figure 1—source data 1.

Large-scale, untargeted metabolomic analyses were completed 24 hr post-IFNγ stimulation in D21 and T21 cells. This time point was chosen to approximate a ‘steady state’ following IFNγ treatment and to allow time for metabolic adaptations to occur; this time point was also consistent with prior metabolite analyses in IFN-treated cells (Wang et al., 2018; Lee et al., 2021b). An overview of the metabolomics experiments is shown in Figure 1A, with raw data for all identified metabolites across conditions in Figure 1—source data 2.

DS has been described as a ‘cytokinopathy’ based upon evaluation of blood plasma from individuals with or without T21. Arrays screening 29 different cytokines showed elevated levels in DS individuals (Malle et al., 2023). Based upon these and other results, we sought to directly measure cytokine levels to assess IFNγ- and CA-dependent effects. We completed a series of cytokine screens (n = 105 different cytokines) in T21 and D21 cells (Figure 1—figure supplement 1C), using the same experimental design as the metabolomics experiments (basal conditions, +CA, +IFNγ, or +CA+IFNγ). Biological replicate cytokine screens were completed for each condition after 24 hr treatment; measurements for all cytokines in both replicates can be found in Figure 1—source data 3.

For RNA-seq, biological triplicate experiments were completed, and a normalization (Hunter et al., 2023) was performed to account for the potential 1.5-fold expression differences from chromosome 21 in T21 cells (Hwang et al., 2021). Comparison of gene expression in T21 vs. D21 cells revealed massive differences, as expected (Figure 1—figure supplement 1D; Figure 1—source data 1). Gene set enrichment analysis (GSEA)(Subramanian et al., 2005) revealed significant upregulation of inflammatory pathways (e.g. IFNγ, IFNα, Complement, TNFα) in T21 vs. D21 cells (Figure 1—figure supplement 1E; Figure 1—source data 4), consistent with prior reports (Sullivan et al., 2016; Sullivan et al., 2017; Malle et al., 2023). Furthermore, the ‘upstream regulators’ identified from Ingenuity Pathway Analysis (IPA) (Krämer et al., 2014) predicted activation of numerous inflammatory markers in T21 cells (e.g. IFNγ, IFNα, TNF, TGFβ1, NFκB), based upon the RNA-seq results (Figure 1—figure supplement 1F; Figure 1—source data 5). Negatively enriched pathways in T21 cells (GSEA Hallmarks, Figure 1—figure supplement 1E, Figure 1—source data 4) reflected proliferative gene expression programs (e.g. MYC targets, G2M checkpoint); in agreement, we observed slower growth rates for T21 cells compared with their D21 counterparts.

Untargeted metabolomics experiments identified and quantified over 600 biochemicals representing every major metabolic pathway (Figure 1—source data 2). We observed evidence that inflammatory pathways were activated in T21 cells under basal conditions (IPA results, Figure 1—figure supplement 1G; Figure 1—source data 5), and 196 biochemicals increased and 214 decreased in T21 vs. D21 (Figure 1—figure supplement 2A, B; Figure 1—source data 2). Pathways related to fatty acid transport, growth, and energy metabolism were reduced in T21 cells (vs. D21; Figure 1—figure supplement 1G). Supporting these general trends, the metabolomics data revealed T21-specific changes in glycolysis, nucleotide and lipid biosynthesis, and other fundamental cellular processes (Figure 1—figure supplement 2B; Figure 1—source data 2). Metabolic distinctions between T21 and D21 cells will be described in more detail later, in the context of IFNγ treatment and/or Mediator kinase inhibition. An overview of metabolic pathways that are relevant for this study is shown in Figure 1B.

As with the RNA-seq and metabolomics data, the cytokine screen results implicated chronic activation of the IFN response in T21 cells under basal conditions; for example, levels of many cytokines were elevated in T21 vs. D21 cells (Figure 1C, left column). Furthermore, treatment with IFNγ appeared to ‘normalize’ cytokine levels in T21 vs. D21 cells (Figure 1C, right column), such that many cytokines were detected at roughly equal levels in IFNγ-stimulated T21 and D21 cells.

For practical reasons, the transcriptomics, metabolomics, and cytokine screening experiments completed here precluded parallel examination of additional genetically distinct donor-derived T21 and D21 cell lines (but see below). A basic observation from past studies was that the transcriptomes of T21 individuals are not only tissue- and cell type-specific, but also reflect individual-to-individual differences within the same cell types (Sullivan et al., 2016; Hwang et al., 2021). However, we hypothesized that the general trends in our sibling-matched T21 vs. D21 lines would show some commonality with large-scale cohort studies.

We analyzed RNA-seq data from whole blood samples from individuals with DS (n = 304) or euploid (i.e. D21; n = 96) controls (Waugh et al., 2023), generated by the Human Trisome Project (https://www.trisome.org/, NCT02864108). The data revealed upregulation of the IFNγ response, Complement, cytokine production, and other inflammatory pathways in T21 individuals (vs. D21; Figure 1—figure supplement 2C; Figure 1—source data 4). Approximately 10% of differentially expressed genes were shared between the Human Trisome Project cohort study (T21 vs. D21; whole blood transcriptomes from individuals varying in age and sex) and our RNA-seq results from sibling-matched T21 and D21 lymphoblastoid cell lines (Figure 1—figure supplement 2D), consistent with individual-to-individual variation within the human population (Sullivan et al., 2016; Hwang et al., 2021). However, the shared ‘core sets’ of genes reflected prominent trends in our T21/D21 RNA-seq comparisons. For example, upregulated genes represented cytokine production and other inflammatory pathways, whereas downregulated genes involved nucleotide and fatty acid metabolism, splicing, G-protein-coupled receptor (GPCR) signaling, and proliferation. These pathway trends are well-represented in our RNA-seq and metabolomics data (e.g. Figure 1—figure supplement 1E–G; Figure 1—source data 6) and additional data supporting these T21-specific changes (vs. D21) will be described in more detail later.

We also used the IFN score metric (Galbraith et al., 2023) to evaluate our sibling-matched T21 and D21 lines. The IFN score considers expression of 18 genes from a set of 382 genes comprising the IFNγ, IFNα, and inflammatory response GSEA Hallmarks gene sets. It has been used to evaluate differences between whole blood transcriptomes in a cohort of 502 participants (n = 356 T21, n = 146 D21). As shown in Figure 1—figure supplement 2E, the sibling-matched T21 line had an elevated IFN score, consistent with comparative trends from a large cohort of whole blood transcriptomes of T21 or D21 individuals (Galbraith et al., 2023).

A recent analysis of blood plasma cytokine levels, measuring 29 cytokines from DS individuals (n = 21) or age-matched euploid controls (n = 10) identified elevated levels of approximately 10 to as many as 22 of these 29 cytokines (Malle et al., 2023). In agreement, at least 13 of the 22 cytokines (59%) were elevated in our T21 vs. D21 comparison (red font in Figure 1C). Collectively, the RNA-seq and cytokine data from a large cohort of DS individuals indicate that the two model cell lines evaluated here, derived from age- and sibling-matched D21 and T21 individuals, broadly reflect population-wide trends.

We next asked whether Mediator kinases influenced hyperactive IFNγ signaling in T21 vs. D21 cells. We first describe how Mediator kinase inhibition impacted T21 or D21 cells under basal conditions.

We next assessed longer-term effects of Mediator kinase inhibition on the transcriptome in T21 cells. We completed biological triplicate RNA-seq experiments in T21 cells 24 hr after CA treatment (or DMSO controls; Figure 2—figure supplement 2A). As shown in Figure 2—figure supplement 2B, GSEA Hallmarks showed little evidence for suppression of anti-inflammatory pathways, in contrast to the results in T21 cells after 4.5 hr CA treatment. This was further supported by GSEA with the GO Biological Processes (GOBP) pathway set (Figure 2—figure supplement 2C), which includes more gene sets (n = 7641 for GOBP vs. n = 50 for Hallmarks; see Methods) and more with relevance to IFN signaling and inflammation.

Gene sets related to cholesterol and lipid metabolism remained activated in CA-treated cells after 24 hr, consistent with the RNA-seq data at the shorter 4.5 hr timeframe. Moreover, the t = 24 hr RNA-seq results showed evidence for sustained activation of SREBP as well as PPAR and LXR in CA-treated cells, a result that was inferred from the metablomics data (Figure 2D). For instance, IPA upstream regulators showed activation of ‘PPARA:RXRA coactivator’ and NR1H3/LXRA (Figure 2—figure supplement 2D). This coincided with elevated levels of PPAR and LXR ligands (e.g. OEA, oleamide, desmosterol), as measured by metabolomics 24 hr after CA treatment (Figure 2C, D; Figure 2—source data 2). We also generated GOBP heatmaps comparing the 4.5 vs. 24 hr ±CA RNA-seq data (Figure 2—figure supplement 2E). The results are distinct from the dot plots shown in Figure 2—figure supplement 2C because the heatmaps are not narrowly focused on selected GOBP terms and because a cutoff (padj ≤ 0.1) was applied. The gene sets change over time, underscoring the importance of timing in evaluation of kinase inhibitors. Also noteworthy were the many pathways related to RNA processing and splicing in the CA-treated cells at 24 hr. Splicing defects in CA-treated cells are described in a later section, but the results summarized in Figure 2—figure supplement 2E suggest that CA-dependent splicing changes are caused by reduced expression of splicing regulators, at least at longer timeframes.

GSEA results from RNA-seq data in T21 cells after 24 hr CA treatment (Figure 2—figure supplement 2B) showed downregulation of oxidative phosphorylation, a result consistent with a prior proteomics study completed in HCT116 cells ±CA (Poss et al., 2016). To probe further, we measured the oxygen consumption rate (OCR) in D21 or T21 cells ±CA (t = 24 hr). As shown in Figure 2—figure supplement 3A, B, OCR was reduced in CA-treated D21 or T21 cells, and reached statistical confidence in T21 cells. The reduced OCR in CA-treated T21 cells tracked with reduced mRNA levels for oxidative phosphorylation genes, measured by RNA-seq (Figure 2—figure supplement 3C). We also measured extracellular acidification rate (ECAR) in D21 or T21 cells ±CA (t = 24 hr). Extracellular acidification is elevated by lactate, an end-product of glycolysis. As shown in Figure 2—figure supplement 3D, E, CA treatment reduced ECAR in both D21 and T21 cells, a result consistent with our metabolomics data (lactate reduced in CA-treated cells; Figure 2—source data 2) and prior CDK8-dependent links to expression of glycolysis enzymes (Galbraith et al., 2017).

A comparison of mitochondrial function in sibling-matched D21 and T21 cells is shown in Figure 2—figure supplement 4A, B and ECAR is shown in Figure 2—figure supplement 4C. The data show significant differences in T21 cells; these results are consistent with a T21 mouse model study, although we note that T21/D21 OCR differences may be tissue-specific (Sarver et al., 2023). Because CA treatment lowered OCR and ECAR, and reduced mitochondrial respiration in T21 cells (Figure 2—figure supplement 3B), we next compared D21 cells with CA-treated T21 cells. The results suggest that Mediator kinase inhibition shifts T21 mitochondrial function and ECAR toward D21 levels (Figure 2—figure supplement 4D–F); however, we emphasize that this represents only one pair of cell lines and more work is needed to rigorously assess this topic.

Although completing parallel metabolomic, cytokine, and transcriptomics analyses across many cell lines was impractical, we did complete RNA-seq experiments in a different T21 donor cell line. Biological triplicate RNA-seq experiments were performed in a non-sibling T21 line ±CA (t = 4.5 hr). Similar to the ‘sibling-matched’ T21 line, CA-treated cells showed elevated expression of genes in the cholesterol homeostasis, fatty acid metabolism, and peroxisome pathways (Figure 2—figure supplement 5A, B). Moreover, GSEA completed with the GOBP set showed suppression of pathways related to IFN signaling and inflammation (Figure 2—figure supplement 5C, D). In fact, the CA-dependent suppression was amplified in this ‘non-sibling’ T21 line, which may reflect higher basal IFN signaling compared with the sibling-matched T21 line. IPA upstream regulators results (Figure 2—figure supplement 5E) also showed evidence for CA-dependent suppression of IFN signaling, with negative Z-scores for pro-inflammatory TFs (e.g. STAT2, IRF1, NFKB1) and cytokines (e.g. IFNG, IFNA2, IFNL1). Conversely, SREBP TFs were activated, as were LXR and PPAR, similar to the sibling-matched T21 line (Figure 2—figure supplement 5E). A heatmap summarizing transcriptome changes upon CA treatment in the two different T21 donor cell lines is shown in Figure 2—figure supplement 5F. Taken together, the transcriptomics data from this ‘non-sibling’ T21 cell line showed robust CA-dependent suppression of inflammatory pathways, with evidence for CA-dependent activation of SREBP, PPAR, and LXR TFs, consistent with the results from the different sibling-matched T21 line.

We next evaluated the transcriptomic, metabolomic, and cytokine data from IFNγ-treated cells. As shown in Figure 3—figure supplement 1A, the transcriptional response to IFNγ was broadly similar in the sibling-matched T21 and D21 cells; however, T21 cells showed higher basal expression of inflammatory genes. This can be seen upon comparison of gene expression in T21 vs. D21 cells under basal (Figure 3—figure supplement 1B) or IFNγ-stimulated conditions (Figure 3—figure supplement 1C–E). Only 10 genes showed greater induction in T21 cells in response to IFNγ, whereas 68 genes had reduced induction in T21 compared with D21 (Figure 3—figure supplement 1F; Figure 1—source data 1). Pathway analysis (GOBP) of these 68 genes showed predominant roles in IFN signaling and inflammatory responses (Figure 3—figure supplement 1G; Figure 1—source data 6). This result is consistent with hyperactive IFN signaling in T21 cells (Sullivan et al., 2016), such that induction upon exogenous IFNγ stimulation was reduced compared with D21 cells. Next, we focus on how Mediator kinase inhibition influenced transcriptional, metabolic, and cytokine responses in IFNγ-treated D21 and T21 cells.

As Mediator-associated kinases, CDK8 and CDK19 are established regulators of RNAPII transcription (Luyties and Taatjes, 2022) however, their potential impact on splicing has not been examined. Over 95% of human mRNAs are alternatively spliced and defects in this process contribute to errors in gene expression in human diseases (Scotti and Swanson, 2016). We compared alternative splicing events using rMATS (Shen et al., 2014), separately evaluating all RNA-seq replicates (biological triplicates; Figure 1A). The data revealed 741 differential exon skipping events in T21 cells (vs. sibling-matched D21) under basal conditions (Figure 5A, B; Figure 5—source data 1), with approximately equal numbers of increased exon inclusion (n = 382) and increased exon skipping events (n = 359). A smaller number of other alternative splicing events (e.g. intron retention) were detected (Figure 5—figure supplement 1A). We observed similar trends in IFNγ-treated T21 and D21 cells, in which a greater number of exon skipping events (n = 418 vs. n = 296) occurred in IFNγ-stimulated T21 vs. D21 cells (Figure 5—figure supplement 1B, C; Figure 5—source data 1).

To determine whether Mediator kinase activity might influence splicing, we next evaluated CA-treated D21 and T21 cells. We found that CA treatment had a substantial impact on splicing, in both D21 and T21 cells (Figure 5C, Figure 5—figure supplement 1D). We identified 432 or 444 sites with altered exon skipping events (D21 or T21, respectively) in CA-treated cells compared with controls (ΔPSI [percent spliced in] ≥0.2, padj < 0.05; Figure 5C, Figure 5—figure supplement 1E; Figure 5—source data 1), and CA treatment increased inclusion of alternative exons more frequently than it increased their skipping in both lines (n = 239 vs. 193 in D21; n = 256 vs. 188 in T21). Similarly in IFNγ-treated D21 and T21 cells (Figure 5D, Figure 5—figure supplement 1F; Figure 5—source data 1), Mediator kinase inhibition favored exon inclusion (n = 178 in D21; 213 in T21) over exon skipping (n = 130 in D21; 148 in T21).

To assess whether the exon inclusion events sensitive to Mediator kinase function were selective for specific gene sets, we completed pathway analyses (IPA) across all conditions tested (Figure 5E–G, Figure 5—figure supplement 1H, I; Figure 1—source data 5). To increase statistical power, we grouped exon skipping events together (i.e. alternative exon inclusion events +alternative exon skipping events). Thus, the T21 vs. D21 comparison included 741 events, D21 +CA vs. Ctrl included 432 events, and so on. The results revealed that splicing changes triggered by Mediator kinase inhibition were cell type-specific (T21 vs. D21) and selectively impacted inflammatory and metabolic pathway genes. As shown in Figure 5E, genes associated with inflammatory signaling (red highlighted pathways) showed evidence of differential splicing in CA-treated T21 cells (vs. D21) during IFNγ stimulation. In addition, pyrimidine salvage and biosynthesis genes were alternatively spliced in CA-treated T21 cells (vs. DMSO; Figure 5G). These pathway-specific effects in CA-treated cells were consistent with gene expression changes (e.g. GSEA Hallmark gene sets, Figure 1—source data 4) and metabolic changes observed in CA-treated cells. For instance, the pyrimidine biosynthesis intermediates N-carbamoylaspartate, dihydro-orotate, and orotate were markedly depleted (1.4- to 3-fold) in CA-treated T21 cells (vs. DMSO; Figure 1—source data 2). The data in Figure 5G implicate CA-dependent splicing changes as a contributing factor.

Finally, we completed parallel splicing analyses for (1) the different donor T21 cell line ±CA treatment for 4.5 hr, as well as for (2) the sibling-matched T21 line with 24 hr CA treatment (vs. DMSO controls). We identified 349 sites with altered exon skipping events in the 4.5 hr CA-treated non-sibling T21 cells, and 309 sites in the sibling-matched T21 line at 24 hr ±CA (Figure 5—figure supplement 2A, B; Figure 5—source data 2). We next completed a pathway analysis (IPA) of genes with differential exon skipping (i.e. alternative exon inclusion events + alternative exon skipping events) ± CA (Figure 5—source data 3). For the sibling-matched line with 24 hr CA treatment, IPA identified various metabolic and signaling cascades, including JAK–STAT, whereas the IPA results from the different donor T21 line (t = 4.5 hr ±CA) yielded results more similar to the other T21 line after 4.5 hr CA treatment (Figure 5—figure supplement 2C, D). In particular, IFN signaling pathways were represented, further implicating Mediator kinases as context-specific regulators of RNAPII transcription and splicing.

Sequence-specific, DNA-binding TFs are a major target of Mediator kinases, based upon quantitative phosphoproteomics data (Poss et al., 2016). To broadly assess how TF activity changed in response to Mediator kinase inhibition, we completed PRO-seq experiments in control vs. CA-treated T21 cells. When sequenced at high depth (≥60 M reads/replicate), PRO-seq allows reliable detection of short, bidirectional transcripts that center around consensus TF-binding sites. Studies from numerous labs have shown that these bidirectional transcripts represent ‘signatures’ of active TFs bound to these sites (Azofeifa et al., 2018; Wang et al., 2019; Kristjánsdóttir et al., 2020; Rubin et al., 2021). Similarly, loss of bidirectional transcription indicates reduced TF activity.

Because our primary interest was CA-dependent suppression of hyperactive IFN signaling in DS, we completed PRO-seq experiments in T21 cells under basal conditions, ±CA treatment (t = 75 min). We then applied TFEA (Transcription Factor Enrichment Analysis), a well-tested method to globally assess TF activity from PRO-seq data (Rubin et al., 2021). As shown in Figure 6A, CA treatment broadly shut down pro-inflammatory TFs in T21 cells (e.g. IRF4, IRF8, SPIB, BACH1, RELA; examples shown in Figure 6B), consistent with the ‘downstream’ RNA-seq (4.5 hr), cytokine (24 hr), and metabolomics data (24 hr). Furthermore, it was evident that CA treatment rapidly repressed the p53 family of TFs (Figure 6A, B, Figure 6—source data 1). As a tumor suppressor, p53 represses cholesterol and lipid biosynthesis pathways, in part by blocking activation of SREBP (Moon et al., 2019). Rapid suppression of p53 by Mediator kinase inhibition is consistent with the metabolic changes observed in CA-treated T21 cells (Figure 2C). Moreover, the RNA-seq data implicated SREBP TF activation in CA-treated T21 cells vs. DMSO controls at later time points (4–24 hr), consistent with de-repression of SREBP following CA-dependent inhibition of p53 activity. From these results, we conclude that altered TF function is a major contributor to (1) CA-dependent metabolic changes and (2) suppression of pro-inflammatory IFN signaling in T21 cells.

Metabolomics has been described as the most direct readout of cell state, yielding a ‘front line’ assessment of the biochemicals that drive all cellular processes (Rinschen et al., 2019). The role of Mediator kinases in human cell metabolism is poorly understood. As a whole, our metabolomics data revealed the most striking CDK8/CDK19 effects on nucleotide biosynthesis (especially in D21) and lipid homeostasis. Nucleotide levels (purines and pyrimidines) and their intermediates were broadly reduced in CA-treated cells whereas different classes of lipids were generally elevated (Figure 2—source data 2).

The gene expression changes (RNA-seq, 4.5 or 24 hr) triggered by Mediator kinase inhibition tracked with these ‘downstream’ (24 hr) metabolic changes (e.g. Figure 2B, Figure 2—figure supplement 1E). For example, nucleotide biosynthesis mRNAs were downregulated whereas lipid metabolism mRNAs were upregulated in CA-treated cells; this was reflected in the GSEA Hallmarks, with positive enrichment for pathways such as cholesterol homeostasis, bile acid metabolism, and FA metabolism and negative enrichment for DNA repair pathways, which contains many nucleotide biosynthesis genes. The differential timing of the transcriptomics and metabolomics experiments helps link CA-dependent metabolic changes to gene expression changes; however, we cannot exclude other factors and pathways. For example, reduced levels of nucleotides can trigger cellular compensation through upregulation of FA levels (Schoors et al., 2015). Consequently, elevated levels of FA and other lipids may result, in part, from compensatory mechanisms triggered by CA-dependent reductions in nucleotide levels.

Despite widespread study of Mediator kinases as regulators of RNAPII transcription, their potential function in splicing has not been addressed. We hypothesized that CDK8 and/or CDK19 may impact splicing based upon prior results that identified NAB2, SRRM2, and KDM3A as high-confidence CDK8/CDK19 substrates in human cells (Poss et al., 2016) each of these proteins has been linked to regulation of splicing (Gautam et al., 2015; Soucek et al., 2016; Baker et al., 2021). Our analysis revealed that exon skipping was dependent upon the kinase function of CDK8 and/or CDK19; increased inclusion or skipping of hundreds of alternative exons was observed in CA-treated cells (D21 or T21), and this was observed under both basal and IFNγ-stimulated conditions (e.g. Figure 5C, D).

Pathway analysis of the CA-dependent alternative splicing events identified gene sets associated with signaling pathways known to be regulated by Mediator kinases, such as TGFβ and IFN signaling (Figure 5E; Figure 5—figure supplement 2). Interestingly, CA-dependent alternative splicing of IFN-related gene sets occurred specifically in T21 cells (not D21); moreover, other pathways were selectively affected by Mediator kinase inhibition in cell type and context-specific ways. For instance, mTOR pathway genes in CA-treated D21 cells and pyrimidine metabolism genes in T21 cells (Figure 5F, G).

Taken together, these results suggest that Mediator kinases regulate pre-mRNA splicing, which adds to the mechanisms by which CDK8 and CDK19 can influence gene expression. Our data further suggest that kinase-dependent splicing regulation occurs in cell type- and context-specific ways, similar to other known regulatory functions for CDK8 and CDK19 (Luyties and Taatjes, 2022). We emphasize, however, that the effects of CDK8/CDK19 on splicing are modest, especially compared with CDK7, a different transcription-associated kinase (Rimel et al., 2020).

The mechanisms by which CDK8 and/or CDK19 influence pre-mRNA splicing remain unclear, but are likely to involve both direct kinase-dependent regulation of splicing factors (e.g. through phosphorylation of NAB2 or SRRM2) and indirect regulation through phosphorylation of DNA-binding TFs, which can regulate splicing (Rambout et al., 2018) and are common Mediator kinase substrates (Poss et al., 2016). Differential expression of splicing regulatory proteins in CA-treated cells may also contribute, based upon the reduced expression of splicing regulators at longer timeframes (t = 24 hr CA treatment; Figure 2—figure supplement 2E).

The anti-inflammatory lipids elevated in CA-treated cells included oleamide, desmosterol, endocannabinoids such as OEA, and PUFAs such as EPA, DPA, DHA, and various other ω3 or ω6 PUFAs. These metabolites act as signaling molecules, at least in part, through binding nuclear receptors or GPCRs such as FFAR1, FFAR4, and GPR119, to initiate a cascade of anti-inflammatory responses (Fu et al., 2003; Basil and Levy, 2016; Husted et al., 2017; Kimura et al., 2020; Jordan and Werz, 2022). Our RNA-seq results (4.5 and 24 hr) confirmed CA-dependent activation of nuclear receptors PPAR and LXR, but only under conditions with elevated IFN signaling (e.g. T21 cells or D21 +IFNγ). Furthermore, metabolomics data revealed elevated levels of PPAR and LXR ligands (e.g. oleamide, endocannabinoids, desmosterol) only in CA-treated cells in the context of hyperactive IFN signaling. These results reveal an unexpected mechanistic link between Mediator kinase activity and nuclear receptor function, and suggest potential regulation of GPCR function as well (Figure 7). Metabolic changes are generally not considered when evaluating the biological function of transcription-associated kinases such as CDK8 and CDK19. Our results demonstrate that kinase inhibition can activate TFs through metabolic changes, and suggest that during IFNγ stimulation, Mediator kinases normally suppress metabolic pathways that would activate PPAR and LXR nuclear receptors.

Independently and collectively, our transcriptomics, metabolomics, and cytokine screen data revealed that Mediator kinase inhibition suppressed pro-inflammatory signaling. This was most evident during IFNγ stimulation but was also apparent under basal conditions, especially in T21 cells due to their high ‘baseline’ activation of IFN signaling pathways. We acknowledge that metabolites and cytokines can have variable, dynamic, and context-specific effects on inflammatory signaling. For example, sphingolipids (Maceyka and Spiegel, 2014) and bile acids may serve pro- or anti-inflammatory roles depending upon cell type (Chiang and Ferrell, 2019), and cytokine responses are complex, in part because cytokines regulate each other (e.g. MIP3α/CCL20 is downregulated by IL10). For this study, a more straightforward metric for an anti-inflammatory change was whether CA treatment countered the effect of IFNγ stimulation alone. This was observed in a large number of cases across all experiments.

CDK8 and CDK19 are established regulators of RNAPII transcription (Luyties and Taatjes, 2022). Accordingly, the majority of CA-dependent effects that we have characterized here result from gene expression changes, consistent with our TFEA results (Figure 7) and prior phosphoproteomics data that identified TFs and other RNAPII regulatory factors as high-confidence targets of CDK8/CDK19 in human cells (Poss et al., 2016). The staggered timing of our PRO-seq (75 min), RNA-seq (4.5 or 24 hr), metabolomics (24 hr), and cytokine screen experiments (24 hr) allowed inference of direct vs. indirect effects of Mediator kinase inhibition. As expected, we observed changes in anti-inflammatory transcriptional responses over time, which likely reflects compensatory responses to Mediator kinase inhibition.

Taken as a whole, our results suggest that under conditions of IFN pathway stimulation, CDK8/CDK19 inhibition initiates a cascade of events that mediate anti-inflammatory responses (Figure 7). At short timeframes (75 min), Mediator kinase inhibition suppressed activation of IFN-responsive, pro-inflammatory TFs in T21 cells (Figure 6). Consistent with these results, mRNA levels of pro-inflammatory genes were subsequently reduced in CA-treated T21 cells (Figure 2—figure supplement 5C, D), as were protein levels of pro-inflammatory cytokines (Figure 2E). Similar CA-dependent effects were observed with exogenous IFNγ stimulation in both T21 and D21 cells (Figures 3 and 4). TFEA also identified p53 as rapidly suppressed in CA-treated T21 cells (t = 75 min); p53 blocks SREBP activation and this contributes partially or completely to p53-dependent suppression of cholesterol and lipid biosynthesis (e.g. to suppress growth) (Moon et al., 2019). CA-dependent suppression of p53 function in T21 cells could therefore activate SREBP to elevate cholesterol and lipid biosynthesis. In agreement, RNA-seq data (t = 4.5 or 24 hr) identified SREBP (IPA upstream regulators), cholesterol homeostasis and fatty acid metabolism (GSEA Hallmarks) as activated upon Mediator kinase inhibition. SREBP TFs activate cholesterol and lipid biosynthesis pathways, and metabolomics data revealed elevated levels of anti-inflammatory lipids in CA-treated cells during IFNγ signaling (e.g. D21 cells +IFNγ; T21 cells at basal conditions). Among these elevated lipids, oleamide, desmosterol, and PUFAs are nuclear receptor agonists that activate PPAR and LXR (Yang et al., 2006; Sun and Bennett, 2007; Spann et al., 2012; Roy et al., 2016); PPAR and LXR, in turn, broadly control fatty acid biosynthesis (Schultz et al., 2000), including PUFAs such as DPA, DHA, and EPA, whose levels were consistently elevated in CA-treated cells. In this way, Mediator kinase inhibition may robustly and durably elevate lipid metabolites, with potential downstream consequences for inflammatory signaling and nuclear receptor function (Figure 7).

Although the factors that contribute to the DS condition are complex (Antonarakis et al., 2020), it has been described as both an interferonopathy (Sullivan et al., 2016) and a cytokinopathy (Malle et al., 2023). An emerging theme in DS research is that individuals with DS could benefit from therapeutic strategies that diminish chronic, hyperactive IFN signaling (Sullivan et al., 2017; Tuttle et al., 2020; Waugh et al., 2023). Through multi-omics evaluation of CA-dependent effects in T21 cells, we have demonstrated that Mediator kinase inhibition antagonizes hyperactive IFN signaling, via diverse mechanisms that extend beyond transcription (Figure 7). Consequently, selective inhibition of Mediator kinase function represents a potential therapeutic strategy to address chronic inflammation and its associated co-morbidities in DS.

Immortalized lymphoblastoid cell lines derived from age- and sex-matched sibling pairs (male; one D21, age 5.7 years: TIC0000259, one T21, age 3.1 years: TIC0000177) were obtained from the Nexus Biobank. The non-sibling T21 line was also obtained from the Nexus Biobank (male, age 10.7 years; TIC0001672). Cell lines were authenticated (STR profiling) by Nexus Biobank and tested negative for mycoplasma. Cells were cultured in RPMI medium (Gibco, 23400062) supplemented with 20% fetal bovine serum (Peak Serum Inc, PS-FB3), 1× GlutaminePlus (R&D Systems, R90210), and 1× penicillin–streptomycin (Gibco, 15140-163), on low-attachment flasks at 37°C and 5% CO₂. Due to the lower proliferation rate of T21 cells compared to D21 cells, T21 cells were seeded at twice the density of D21 cells for all experiments unless otherwise noted. For all CA treatments, the compound remained in the media during IFNγ stimulation.

RIPA lysis buffer (Thermo Scientific, 89900) was used to generate cell lysates; all lysis was performed at 4°C in the presence of Halt Protease and Phosphatase Inhibitor Cocktail, EDTA-free (100×) (Thermo Scientific, 78441) and Benzonase endonuclease (Millipore Sigma, 101697). The BCA Protein Assay Kit (Pierce, 23227) was used to measure and normalize protein concentrations across samples. Protein samples were boiled for 5 min at 95°C in 4× Laemmli Sample Buffer (Bio-Rad, 1610747) and run on SDS-acrylamide gels, followed by transfer to nitrocellulose membranes. Blocking was performed using either 5% milk in TBS-T or in 5% bovine serum albumin for phospho-antibodies.

Antibodies used are as follows: anti-STAT1 (RRID:AB_2198300; Cell Signaling Technology, 9172), anti-phospho STAT1 S727 (RRID:AB_2197983; Cell Signaling Techology, 9177), and anti-GAPDH (RRID:AB_1080976; GeneTex, GTX100118). HRP-conjugated anti-rabbit secondary antibody was used for detection (Invitrogen, 31480) and protein was visualized using Immunobilon Western Chemilumiescent HRP Substrate (Millipore, WBKLS0500) on an ImageQuant LAS 4000 (GE Healthcare). Intensity values were quantified and normalized using ImageJ.

Cells were seeded and allowed to recover for 24 hr prior to being treated with 10 ng/ml interferon gamma (Gibco PHC4031) or vehicle (40 mM Tris pH 7.4). Total RNA was isolated using TRizol (Invitrogen, 15596026) according to the manufacturer’s instructions, quantified with a Qubit 3.0 using the RNA High Sensitivity (HS) kit (Invitrogen, Q32855), and 100 ng of RNA from each sample was converted to cDNA using High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, 4368813) following the manufacturer’s instructions. cDNA was diluted to 0.25 ng/μl for all samples, and amplification was performed using Sybr Select Master Mix (Thermo Fisher Scientific, 4472908) on a Bio-Rad CFX384 Real Time PCR System. ΔΔCT values were calculated using GAPDH as a loading control. Treatments, RNA isolation, and RT-qPCR were performed in biological duplicate. The primers used for qPCR are as follows – GAPDH Forward: ACCACAGTCCATGCCATCAC, GAPDH Reverse: TCCACCACCCTGTTGCTGTA, GBP1 Forward: GTGCTAGAAGCCAGTGCTCGT, GBP1 Reverse: TGGGCCTGTCATGTGGATCTC, IRF1 Forward: GAGGAGGTGAAAGACCAGAGC, IRF1 Reverse: TAGCATCTCGGCTGGACTTCGA.

To probe the timing of the RNA-seq experiments, we completed RT-qPCR at various time points as shown in Figure 1—figure supplement 1B. The 4 hr time point (post-IFNγ) was chosen because robust induction of GBP1 and IRF1 mRNA was observed. Whereas mRNA levels remained high at 6 hr, we selected the 4 hr time point because it would better represent the primary transcriptional response; moreover, the 4 hr time point was consistent with prior transcriptomics studies of IFNγ response in mammalian cells (Steinparzer et al., 2019).

For each treatment condition, six flasks of cells were seeded and allowed to recover for 24 hr prior to treatment. Cells were pre-treated with 0.1% DMSO or 100 nM CA for 1 hr, then treated with 40 mM Tris, pH 7.4 or 10 ng/ml interferon gamma (Gibco, PHC4031). After 24 hr of interferon gamma treatment, cells were harvested and snap-frozen in liquid nitrogen. Sample preparation was carried out by Metabolon (Durham, North Carolina, USA) using a previously published workflow (Ford et al., 2020). Samples were prepared using the automated MicroLab STAR system from Hamilton Company, and several recovery standards were added prior to the first step in the extraction process for QC purposes. To remove protein, dissociate small molecules bound to protein or trapped in the precipitated protein matrix, and to recover chemically diverse metabolites, proteins were precipitated with methanol under vigorous shaking for 2 min (Glen Mills GenoGrinder 2000) followed by centrifugation. The resulting extract was divided into five fractions: two for analysis by two separate reverse phase (RP)/UPLC–MS/MS methods with positive ion mode electrospray ionization (ESI), one for analysis by RP/UPLC–MS/MS with negative ion mode ESI, one for analysis by HILIC/UPLC–MS/MS with negative ion mode ESI, and one sample was reserved for backup. Samples were placed briefly on a TurboVap (Zymark) to remove the organic solvent.

All methods utilized a Waters ACQUITY ultra-performance liquid chromatography (UPLC) and a Thermo Scientific Q-Exactive high resolution/accurate mass spectrometer interfaced with a heated electrospray ionization (HESI-II) source and Orbitrap mass analyzer operated at 35,000 mass resolution. The sample extract was dried then reconstituted in solvents compatible to each of the four methods. Each reconstitution solvent contained a series of standards at fixed concentrations to ensure injection and chromatographic consistency. One aliquot was analyzed using acidic positive ion conditions, chromatographically optimized for more hydrophilic compounds. In this method, the extract was gradient eluted from a C18 column (Waters UPLC BEH C18-2.1 × 100 mm, 1.7 µm) using water and methanol, containing 0.05% perfluoropentanoic acid (PFPA) and 0.1% formic acid (FA). Another aliquot was also analyzed using acidic positive ion conditions, however it was chromatographically optimized for more hydrophobic compounds. In this method, the extract was gradient eluted from the same afore mentioned C18 column using methanol, acetonitrile, water, 0.05% PFPA, and 0.01% FA and was operated at an overall higher organic content. Another aliquot was analyzed using basic negative ion optimized conditions using a separate dedicated C18 column. The basic extracts were gradient eluted from the column using methanol and water, however with 6.5 mM Ammonium Bicarbonate at pH 8. The fourth aliquot was analyzed via negative ionization following elution from a HILIC column (Waters UPLC BEH Amide 2.1 × 150 mm, 1.7 µm) using a gradient consisting of water and acetonitrile with 10 mM Ammonium Formate, pH 10.8. The MS analysis alternated between MS and data-dependent MSn scans using dynamic exclusion. The scan range varied slighted between methods but covered 70–1000 m/z.

Identification of metabolites was performed through automated comparison of the ion features in experimental samples to a reference library of chemical standard entries (Dehaven et al., 2010), and are based on three criteria: retention index within a narrow RI window of the proposed identification, accurate mass match to the library ±10 ppm, and the MS/MS forward and reverse scores between the experimental data and authentic standards. The MS/MS scores are based on a comparison of the ions present in the experimental spectrum to the ions present in the library spectrum. While there may be similarities between these molecules based on one of these factors, the use of all three data points can be utilized to distinguish and differentiate biochemicals.

Two types of statistical analyses were performed: (1) significance tests and (2) classification analysis. Standard statistical analyses were performed in Array Studio on log-transformed data. For analyses not standard in Array Studio, the R program (http://cran.r-project.org/) was used. Following log transformation and imputation of missing values, if any, with the minimum observed value for each compound, Welch two-sample t test was used as a significance test to identify biochemicals that differed significantly (p < 0.05) between experimental groups. An estimate of the false discovery rate (q-value) was calculated to take into account the multiple comparisons that normally occur in metabolomic-based studies. Classification analyses included principal component analysis, hierarchical clustering, and random forest.

Pathway analysis of metabolomic data was performed using the Ingenuity Pathway Analysis software (QIAGEN) (Krämer et al., 2014). For metabolomic analysis, a total of 566 out of 675 metabolites could be mapped using one of the following identifiers: CAS registry number, Human Metabolome Database (HMDB) identification, KEGG, PubChem CID, or RefSeq number; metabolites not used in IPA analysis can be found in Figure 1—source data 2. Metabolomic pathway analysis used only mapped metabolites with a p-value of <0.1 when making pathway predictions.

Cells for metabolic assays were seeded 24 hr prior to treatments and allowed to recover, then were treated with either vehicle or CA at a final concentration of 100 nM. After 24 hr of treatment, cells were collected, washed, and resuspended in Seahorse XF RPMI Assay Media (Agilent, 103576) supplemented with 10 mM glucose (Agilent, 103577-100), 1 mM pyruvate (Agilent, 103578-100), and 2 mM glutamine (Agilent, 103579-100). Cells were counted and for each cell line 100,000 cells were seeded per well on XF24 cell culture plates that had been pre-coated with poly-D-lysine (Gibco, A3890401). The plates were spun down at 300 × g for 3 min using the minimum brake setting, and cells were then allowed to recover for 30–60 min in a humidified, non-CO₂ incubator set to 37°C. Compounds from the Mito Stress Test Kit (Agilent, 103015-100) were used at the following final concentrations: 1.5 µM Oligomycin, 1 µM FCCP, and 5 µM Rotenone/Antimycin A. All assays were performed on an Agilent Seahorse XFe24 Analyzer (XFe version 2.6.3.5), and data were processed using the Agilent Wave software (version 2.6.3). Two to three technical replicates of each treatment condition were plated per assay, and three biological replicates of each assay were completed on different days. Across all three biological replicates, measurements from 2 wells that did not exhibit a drop in OCR of at least 30% following Oligomycin injection were considered outliers and were removed from analysis. Technical replicates from each assay were averaged, and the average OCR or ECAR values for each treatment condition or genotype at each time point were normalized relative to the first baseline measurement taken of vehicle-treated cells (for treatment comparisons) or D21 cells (for the cross-genotype comparisons). Relative values for OCR or ECAR at each time point, for each treatment condition, were averaged between all three biological replicates.

Cells were pre-treated with 0.1% DMSO or 100 nM CA for 30 min, then treated with 40 mM Tris, pH 7.4 or 10 ng/ml interferon gamma (Gibco, PHC4031). After 4 hr, total RNA was isolated from D21 or T21 cells using TRizol (Invitrogen, 15596026) as specified by the manufacturer and quantified with a Qubit 3.0 using the RNA High Sensitivity (HS) kit (Invitrogen, Q32855). 1 μg of total RNA with an RIN number of ≥8 was used for RNA-seq library prep. Libraries were constructed using Universal Plus mRNA-Seq library preparation kit with NuQuant (Tecan, 0520). Library construction and sequencing were performed at the Genomics Shared Resource (CU Anschutz). Paired-end libraries (151 bp × 151 bp) were sequenced on the Illumina NextSeq 6000 platform (Firmware version 1.26.1, RTA version: v3.4.4, Instrument ID: A00405).

For the non-sibling T21 and sibling T21 t = 24 hr RNA-seq experiments, cells were treated with 0.1% DMSO or 100 nM CA for 4.5 hr (non-sibling T21) or 24 hr (sibling T21). After treatment, total RNA was isolated using TRizol (Invitrogen, 15596026) as specified by the manufacturer and quantified with a Qubit 3.0 using the RNA High Sensitivity (HS) kit (Invitrogen, Q32855). 200 ng of total RNA with an RIN number of ≥8 was used for RNA-seq library prep. Libraries were constructed using the Zymo-Seq RiboFree Total RNA Library Kit (Zymo Research). Paired-end sequencing reads of 150 nucleotides were generated on a NovaSeq X (RTA v4.6.7, Serial Number LH00407, Firmware v1.2.0.28691) and de-multiplexed using bcl2fastq.

For differential gene expression analysis, the workflow was as follows: adaptor sequences were trimmed from the RNA-seq raw fastq files using BBDuk (https://sourceforge.net/projects/bbmap/) and mapped using HISAT2 (Kim et al., 2019). Gene counts were generated using featureCounts (Liao et al., 2014), and differential expression analysis was performed using DESeq2 (Anders and Huber, 2010). When there were multiple transcripts per gene, those with the highest FPKM were kept for analysis. For inter-genotype (T21 vs. D21) comparisons, expression of genes on chromosome 21 was normalized to ploidy. While ERCC RNA Spike-In Mix (Invitrogen, 4456740) was added to isolated RNA samples prior to sequencing, ERCC gene counts were variable across vehicle-treated samples even after accounting for read depth, and when using multiple normalization methods and a linear regression. As such, the median of ratios method native to DESeq2 was used to generate size factors and normalize samples. QIAGEN Ingenuity Pathway Analysis (IPA) version 90348151 (Krämer et al., 2014), GSEA 4.2.3 (Subramanian et al., 2005), and Gene Ontology analysis (Ashburner et al., 2000) were used for identification of activated and inhibited pathways. For IPA analysis, a total of 27,647 out of 28,265 genes could be mapped using Entrez Gene Symbols. For cross-genotype (i.e. T21 vs. D21) pathway analysis, only genes with a fold-change > |1.25| and adjusted p-value <0.1 were used. For within-genotype (i.e. treatment effects) pathway analysis, only genes with an adjusted p-value <0.1 were used.

For analysis of RNA-seq data collected from sibling T21 cell line at 24 hr ±CA, and non-sibling T21 at 4.5 hr ±CA, batch correction was applied to gene counts using Surrogate Variable Analysis (Leek et al., 2019) as initial principal component analysis revealed inter-replicate variability as a stronger source of variation than kinase inhibition. However, differential expression analysis demonstrated that the same pathways were predicted to be up- or downregulated regardless of whether batch correction was performed. GSEA was performed by generating a list of genes with differential expression across treatments/genotypes, ranked by the negative log10 adjusted p-value and the sign (positive or negative) of log2(fold-change). Ranked gene lists were loaded onto GSEA (v4.2.3) and GSEAPreranked was run with the following settings – Number of permutations: 1000, Collapse/Remap: No_collapse, Enrichment statistic: weighted, Max size: 500, Min size: 15.

For splicing analysis, the workflow was as follows: duplicates were removed and adaptors trimmed using the bbTools function (v39.01), trimmed reads were mapped uniquely against the hg38 genome using HISAT2 (v2.1.0) (Kim et al., 2019), mapped reads were processed with rMATS (v4.0.1) (Shen et al., 2014), and results were filtered based on FDR <0.05, absolute(IncLevelDifference) <0.2, and ≥2 reads/replicate. Sashimi plots were generated from rMATS results using a modified script based on ggsashimi.py.

The IFN score was calculated as described (Galbraith et al., 2023) using normalized gene counts from DESeq2. The score is calculated with 18 genes: BPGM, CCL5, IFI27, GMPR, IRF7, CMKLR1, IFITM3, GZMA, FCGR1A, USP18, PLSCR1, CXCL9, IFI44, CD274, CXCL10, ISG15, IFI44L, and RSAD2. For each gene, the mean and standard deviation of the D21 replicates was calculated. Then the measured level of each sample was subtracted by the mean of the D21 replicates and divided by the standard deviation of the D21 replicates to create a standardized gene value. For each sample, the sum of the standardized values was taken to get the IFN score.

The data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus and are accessible through GEO Series accession number GSE220652 or GSE263239. Code used to process and visualize data can be found at https://github.com/kira-alia/Cozzolino2023 (copy archived at Cozzolino, 2024).

Cells were pre-treated with 0.1% DMSO or 100 nM CA for 30 min, then treated with 40 mM Tris, pH 7.4 or 10 ng/ml interferon gamma (Gibco, PHC4031). After 24 hr, cells were lysed in RIPA buffer (Thermo Scientific, 89900) supplemented with Halt Protease and Phosphatase Inhibitor Cocktail, EDTA-free (100×) (Thermo Scientific, 78441) and Benzonase endonuclease (Millipore Sigma, 101697). Lysate concentrations were determined using the Pierce BCA Protein Assay Kit (Thermo Scientific, 23225), and 250 μg of protein from each condition was incubated with a membrane array from the Proteome Profiler Human XL Cytokine Array Kit (R&D Systems, ARY022B). Membranes were processed according to the manufacturer’s instructions and imaged using an ImageQuant LAS 4000 (GE Healthcare). Background-subtracted technical replicate values for each cytokine in each condition were quantified using ImageJ and averaged giving a normalized intensity value. The relative intensity was compared on a per cytokine basis and was statistically assessed using a one-way ANOVA test between the eight total conditions across two biological replicates.

Cells were treated with 0.1% DMSO or 100 nM CA for 75 min before being transferred into a lysis buffer consisting of final concentrations of 10 mM Tris pH 7.4, 2 mM MgCl₂, 3 mM CaCl₂, 0.5% NP-40, 10% glycerol, 1 mM DTT, 1× Halt Protease and Phosphatase Inhibitor Cocktail (Thermo Scientific, 78441), and 2 U/μl SUPERase•In RNase Inhibitor (Invitrogen, AM2696). Nuclei were isolated and cleaned through repeated centrifugation and washes, and were counted, aliquoted, and snap-frozen in using liquid nitrogen in a freezing buffer consisting of 50 mM Tris pH 8.3, 5 mM MgCl₂, 40% glycerol. 0.1 mM EDTA, and 4 U/μl SUPERase•In RNase Inhibitor (Invitrogen, AM2696). Nuclear run-ons were performed for 3 min at 37°C on a Groovin’ Tubes thermoshaker (Boekel Scientific, 270500), on a mixture of 10 million human and 100,000 Drosophila melanogaster nuclei per replicate. PRO-seq library preparation was performed as described (Fant et al., 2020). Library sequencing was performed at the Cornell Genomics Facility (Cornell Institute of Biotechnology) on the Illumina NextSeq 2000 platform (Instrument ID VH00231).

Raw fastq files from the sequencing pipeline described above were trimmed and mapped using the NascentFlow pipeline (DOI 10.17605/OSF.IO/NDHJ2). Bidirectional sites of transcription in the human genome (assembly hg38) were identified using the BidirectionalFlow pipeline (https://github.com/Dowell-Lab/Bidirectional-Flow; Sanford, 2023), which runs TFIT (Azofeifa and Dowell, 2017). Bidirectional calls across all replicates and conditions relevant to this study were merged using muMerge (https://github.com/Dowell-Lab/mumerge; Stanley, 2022) and used as an annotation file for running TFEA (Rubin et al., 2021). Example traces of differentially active sites of bidirectional transcription were plotted using pyGenomeTracks (Lopez-Delisle et al., 2021).

The data in this publication have been deposited in NCBI’s Gene Expression Omnibus and are accessible through GEO Series accession number GSE249549. Code used to process and visualize data can be found at https://github.com/kira-alia/Cozzolino2023 (copy archived at Cozzolino, 2024).