Skip to main content
NCBI home page

This is a preprint.

It has not yet been peer reviewed by a journal.

The National Library of Medicine is running a pilot to include preprints that result from research funded by NIH in PMC and PubMed.

bioRxiv logoLink to bioRxiv
[Preprint]. 2026 Jan 5:2025.11.21.689830. Originally published 2025 Nov 24. [Version 2] doi: 10.1101/2025.11.21.689830

Machine Learning Analysis of the Human Initiator Region Reveals Key Features of Different Types of Core Promoters

Torrey E Rhyne-Carrigg 1, Long Vo ngoc 1, Claudia Medrano 1, Kassidy E Gillespie 1, James T Kadonaga 1,2
PMCID: PMC12700081  PMID: 41394737

Abstract

The initiator (Inr) is the starting point for the transcription of many genes. Here, we generated highly predictive machine learning models of the human Inr region, and determined that the Inr is present in about 60% of natural promoters, identified a novel TATA-specific Inr, and detected the overlapping but functionally distinct TCT motif. Quantitative genome-wide analyses revealed a strict and synergistic interaction between the Inr and DPR, a duality between the TATA and DPR, a flexible and sometimes independent function of the TATA box in relation to the Inr, and different properties of the TCT motif in humans and Drosophila.

Introduction

The RNA polymerase II transcription system is responsible for the expression of protein-coding genes as well as many noncoding RNAs in eukaryotes. The initiation of RNA polymerase II transcription is mediated by the core promoter, which is the stretch of DNA that is approximately −40 to +40 nucleotides relative to the +1 transcription start site (TSS) (for reviews, see Smale and Kadonaga 2003; Vo ngoc et al. 2017a; Haberle et al. 2018; Roeder 2019; Schier and Taatjes 2020; Sloutskin et al. 2021; Archuleta et al. 2024). The core promoter, the “gateway to transcription”, is at a key strategic position in the transcription process, as it is the site of convergence of the signals that lead to initiation.

All core promoters do not function by the same mechanism, but rather, different core promoters can have different activities that are determined by the presence or absence of DNA sequence elements such as the TATA box, downstream promoter region (DPR), initiator (Inr), and TCT motif (Supplemental Fig. S1). [The DPR is a unified version of the MTE and DPE elements (Vo ngoc et al. 2020). We will use the term DPR when referring to the DPE or DPR.] In human promoters, the estimated abundances of the TATA, Inr, and DPR are 15–23%, 40–56%, and 25–34%, respectively (Vo ngoc et al. 2017b, 2020). There are no universal core promoter elements, and many human promoters lack any of the known elements. Importantly, the individual core promoter elements confer specific transcriptional regulatory properties. For instance, in Drosophila, TATA-specific and DPR-specific enhancers have been identified (Butler and Kadonaga 2001; Juven-Gershon et al. 2008).

The Inr is the central focal point for the initiation of transcription of many genes. In an early analysis of eukaryotic promoters, a YYCA+1YYYYY consensus was identified at the TSS (Corden et al. 1980). This motif was incisively demonstrated to be a core promoter element and termed the Inr by Smale and Baltimore (1989). The Inr is recognized by the TFIID complex (Kaufman and Smale 1994; Purnell et al. 1994; Verrijzer et al. 1995), which also binds to the TATA box and DPR. Because of its abundance and correspondence with the +1 TSS, it is essential to have an accurate and reliable predictive model of the Inr.

In our previous studies of the core promoter, we used a combination of high-throughput analysis of randomized promoter elements (HARPE) and machine learning (SVR, support vector regression) to generate SVR models of the human TATA box and the human and Drosophila DPR (Vo ngoc et al. 2020, 2023). These SVR models provide objective, quantitative, data-based predictions of the activity of core promoter motifs and have been found to be more effective at predicting transcriptional activity than consensus sequence-based methods (Vo ngoc et al. 2020). One key advantage of the HARPE-SVR approach is that it involves the analysis of several hundred thousand different DNA sequence variants, which provide much more information than sequences from ~10,000–30,000 natural promoters. In addition, in each HARPE assay, we vary the DNA sequences in only the target region of interest in an otherwise constant promoter backbone. This approach provides high quality information on a specific segment of the promoter. Thus, in this work, we focused on the Inr, which is the key core promoter element between the TATA and DPR, and generated multiple different SVR models of the human Inr region in different promoter and transcriptional (i.e., biochemical vs. cell-based) contexts. The Inr SVR models, in conjunction with SVR models of the TATA and DPR, revealed key features of different types of core promoters that contain the Inr or the overlapping but functionally distinct TCT motif.

Results and Discussion

SVR models of the human Inr

To generate predictive models of the human Inr, we performed HARPE (Fig. 1A) and SVR analyses by the method of Vo ngoc et al. (2020). To this end, we first measured the biochemical transcription activity of ~500,000 Inr sequence variants in which a 10-nt region (from −4 to +6 relative to the +1 TSS in the SCP1 promoter backbone) was randomized in a DPR-driven promoter as well as in a separate TATA-driven promoter (Fig. 1A). Because DPR- and TATA-dependent transcription occur by distinct mechanisms (see, for example: Willy et al. 2000; Hsu et al. 2008; Vo ngoc et al. 2020), we analyzed the properties of the Inr in the two different contexts. The HARPE data from independent replicates were found to be reproducible (Supplemental Table 1). A small subset of the sequence variants exhibited high transcriptional activity (Fig. 1B), and HOMER (Heinz et al. 2010) analysis of the top 0.1% of variants revealed a motif that strongly and somewhat surprisingly (in relation to previous human Inr consensus sequences; see, for example, Vo ngoc et al. 2017b) resembles the Drosophila Inr consensus, TCA+1KTY or TCA+1GTY (Fig. 1C, Supplemental Fig. S2; see, for example: Hultmark et al. 1986; Ohler et al. 2002, Fitz-Gerald et al. 2006). It is important to note, however, that TCA+1KTY is the consensus of the most active Inr sequences and should not be interpreted to be representative of typical Inr elements in natural promoters.

Figure 1.

Figure 1.

Open in a new tab

HARPE and SVR analyses of the human Inr in DPR- and TATA-driven promoter backbones. (A) Use of the HARPE method for the analysis of the human Inr region. The SCP1-based promoter backbones are identical except for the presence of a DPR (mutTATA) versus a TATA box (mutDPR). The 10-nt region from −4 to +6 (relative to the SCP1 +1 TSS; designated as “Inr10”) was randomized to generate approximately 500,000 Inr sequence variants in each promoter backbone. Each Inr sequence variant was identified with a corresponding 14-nt barcode from +53 to +67. (B) Most Inr sequence variants exhibit low transcriptional activity, but a small subset of the variants is highly active. Shown is the transcription strength (with biochemical assays, in order of decreasing activity) of each of the approximately 500,000 Inr variants with a DPR (left) or TATA box (right). (C) HOMER analysis of the 0.1% most active Inr variants (in biochemical assays) with a DPR (left) or TATA box (right) reveals a distinct motif that closely resembles the Drosophila Inr. Shown is the web logo of the top HOMER motif. (D) The SVRb (trained with biochemical data) models of the human Inr accurately predict the transcription strengths of Inr sequence variants with a DPR (left) or TATA box (right). Each SVRb model was trained with 200,000 variants and then tested with approximately 7,000–8,000 independent variants (i.e., not used for training) from the HARPE dataset. For each of the independent test sequences, the predicted SVRb score was compared with the observed transcription strength. (PCC) Pearson’s correlation coefficient, (rho) Spearman’s rank correlation coefficient. SVR scores are relative (i.e., not comparable between models) and are not equivalent to the observed transcription strengths. All panels show average transcription strengths from two independent biological replicates. The SVRb models were trained on the averaged values.

To provide a generally useful means of evaluating the human Inr, we used the HARPE data to train and optimize SVR models of the human Inr region (Supplemental Fig. S3), and obtained two SVR models, which we termed Inr10 (with DPR) SVRb and Inr10 (with TATA) SVRb. [Inr10 specifies the 10-nt randomized region; (with DPR) and (with TATA) denote the use of DPR- and TATA-driven core promoters, respectively; and SVRb indicates biochemical transcription data.] With these SVR models, we observed an excellent correlation between the observed transcription strengths of independent (i.e., not used for training) test sequences and their predicted Inr activities (Fig. 1D). Thus, the Inr10 SVRb models of the human Inr provide accurate predictions of transcriptional activity.

We next investigated whether different libraries with a longer stretch of randomized Inr sequences would yield HARPE data and SVR models that are similar to those obtained with the Inr10 experiments. We therefore constructed and analyzed 14-nt randomized Inr region libraries (from −5 to +9 relative to the SCP +1 TSS) and found that the properties of the top 0.1% of variants and the resulting SVR models, termed Inr14 (with DPR) SVRb and Inr14 (with TATA) SVRb (Supplemental Fig. S4), are nearly identical to those obtained with the 10-nt Inr libraries (Supplemental Fig. S5). We also tested a third library in which both the DPR and the TATA box were mutated in the promoter backbone. This library exhibited much lower overall transcription levels but nevertheless yielded the same top HOMER motif (Supplemental Fig. S5; Heinz et al. 2010) and a good SVR model (Supplemental Fig. S4). Therefore, both the Inr10 and Inr14 SVRb models provide reliable predictions for the activity of the human Inr with the DPR as well as the TATA box.

Third, we tested whether transcription of the Inr region HARPE libraries in cells would yield results that are similar to those obtained in biochemical transcription reactions. We thus performed HARPE analyses by transfection of the 14-nt randomized Inr libraries into HeLa cells. The 0.1% most active cell-transcribed variants exhibited, as did the biochemical experiments (Supplemental Figs. S2,S5), a top HOMER motif that strongly resembles the Drosophila Inr (Supplemental Fig. S6). We generated cell-based Inr14 SVRc models (where SVRc indicates a cell-based SVR model; Supplemental Fig. S4) and found that the predictions of the SVRc models strongly correlate with those of their corresponding SVRb models (Fig. 2A,B). Thus, the biochemical and cell-based experiments lead to nearly identical conclusions about the human Inr.

Figure 2.

Figure 2.

Open in a new tab

Analysis of the Inr and TCT motifs with biochemical (SVRb) and cell-based (SVRc) models of the human Inr. SVR models were generated with biochemical as well as cell-based HARPE data with a 14-nt randomized region (from −5 to +9 relative to the SCP +1 TSS; termed “Inr14”), which encompasses both the Inr and TCT motifs. (A,B) SVRc models of the human Inr with a DPR (A) or TATA box (B) show strong correlation with the corresponding SVRb models. SVRc and SVRb scores of 30,642 natural human promoters are compared. (PCC) Pearson’s correlation coefficient, (rho) Spearman’s rank correlation coefficient. (C,D) The TCT motif can be identified more distinctly with the cell-based Inr14 SVRc models than with the biochemical Inr14 SVRb models. SVR scores are shown for 30,642 natural human promoters, with sequences matching the Inr (CA+1NW) or TCT (YC+1TYTYY) motifs highlighted in blue or red, respectively. The red dots are plotted on top of the blue dots for visibility.

Analysis of the Inr and TCT motif

To estimate the abundance of the Inr in natural human promoters, it was first necessary to determine the optimal SVR score thresholds for classifying sequences as active versus inactive. To this end, we carried out a performance assessment for each SVR model to identify the score thresholds at which the models are most accurate (Supplemental Figs. S7−S10). For example, with the Inr10 (with DPR) SVRb model, we found that 10-nt sequences with an SVR score ≥1.3 are predicted to be active as an Inr (Supplemental Fig. S7B). With the SVR score thresholds, we found that approximately 60% of natural human focused promoters contain an active Inr (Supplemental Fig. S11, Supplemental Table 2). [In these analyses, we examined focused promoters in which transcription initiates at a single site or in a narrow cluster of sites.] This observation is comparable to a recent estimate of ~40–56% based on a consensus sequence analysis (Vo ngoc et al. 2017b). In addition, the different SVR models of the Inr (with a DPR, TATA box, or neither element) mostly agree on their predictions of which sequences are active Inr elements (Supplemental Fig. S11). These findings provide an independent revised assessment of the high abundance of the Inr in human promoters.

We additionally examined the shared features of the Inr sequences in natural promoters that are predicted to be active. To this end, we used the Inr10 SVRb models to generate sequence logos of all of the predicted active Inr sequences in human GM12878 cells, and identified a minimal CA+1NW motif (Supplemental Fig. S11E,F), which resembles the BBCA+1BW consensus that was based on the analysis of natural promoters and estimated to be present in about 40–56% of human promoters (Vo ngoc et al. 2017b). The CA+1NW consensus serves as a minimal Inr reference sequence, but, unlike the SVR models, CA+1NW does not provide any quantitative information on the transcriptional activity of any Inr that matches this motif. It is useful to note, however, that the Inr14 SVRb and SVRc models predict that most of the human Inr sequences with a match to the minimal CA+1NW Inr motif are active (Fig. 2C,D, Supplemental Fig. S12).

Importantly, in addition to the Inr, the randomized Inr region encompasses the overlapping but functionally distinct TCT motif (Supplemental Fig. S1), which is a rare but biologically important core promoter element that mediates the transcription of ribosomal protein genes and other genes involved in translation in animals (Parry et al. 2010; Wang et al. 2014). We therefore investigated whether the SVR models of the Inr region also recognize the activity of the TCT motif (YC+1TYTYY in humans; Parry et al. 2010). Unlike the Inr, the TCT motif is detected much more distinctly with the Inr14 SVRc models than with the Inr14 SVRb models (Fig. 2C,D, Supplemental Fig. S12). To determine the basis of this effect, we examined the observed transcriptional activities of TCT sequences in the HARPE assays and found that TCT sequences are much more active in cells than in biochemical experiments (Supplemental Fig. S13). Hence, the biochemical transcription extracts weakly detect the activity of TCT-dependent promoters. In contrast, the cell-based HARPE experiments effectively incorporate the activities of both the Inr and TCT motif. Notably, most of the identified TCT motifs are predicted to be active with the SVRc models (Fig. 2C,D, Supplemental Fig. S12). In addition, consistent with previous analyses of the TCT motif (Parry et al. 2010; Wang et al. 2014), a large fraction of the TCT-containing promoters are associated with ribosomal protein genes or genes that encode proteins that are involved in translation (Supplemental Table 3A,B). Thus, the Inr14 SVRc models can be used to predict the activity of both the Inr and the TCT motif in humans. It should also be noted that the Inr element alone, separately from the TCT motif, could be analyzed with the SVRb models, which provide strong predictions of Inr activity but only weakly detect the TCT motif.

An Inr variant that functions specifically with the TATA box

We next compared the properties of the Inr in a DPR-driven versus a TATA-driven promoter. By using the Inr10 (with DPR) and Inr10 (with TATA) SVRb models, we determined the predicted Inr activities of all of the 1,048,576 possible 10-nt sequences (Supplemental Fig. S14A). This analysis revealed three Inr variants that differ in the positioning of the core Inr CAKT motif within the 10-nt randomized region. We then examined the occurrence of the three Inr variants in natural promoters and found that they are conserved from Drosophila to humans, as seen in human GM12878 cells (Fig. 3A), human K562 cells (Supplemental Fig. S14B), and Drosophila embryos (Fig. 3B). In the natural promoters, the A nucleotide in the core CAKT Inr motif is located at different positions (A+1, A+2, or A+3) relative to the observed +1 TSS. In these analyses, we used datasets that were generated by different methods to ensure that the TSS positions are not biased to any peculiarities of a single method. Hence, the TSSs in the GM12878 cells were identified by the GRO-cap method (Kruesi et al. 2013; Core et al. 2014), and the TSSs in the K562 cells and Drosophila embryos were determined with the csRNA-seq method (Duttke et al. 2019).

Figure 3.

Figure 3.

Open in a new tab

The positioning of the Inr sequence motif relative to the +1 TSS is different in DPR-driven promoters than in TATA-driven promoters. (A) Comparison of the predicted Inr activities of 30,642 natural human promoters with the Inr (with DPR) SVR model versus the Inr (with TATA) SVR model reveals three distinct variants, which are labeled according to the position of the A nt in the core CAKT Inr motif relative to the observed +1 TSS. The web logos are based on the sequences in the highlighted dashed ovals. (B) The three Inr variants are conserved from Drosophila to humans. Comparison of the predicted Inr activities of 4,489 natural Drosophila promoters with the Inr (with DPR) SVR model versus Inr (with TATA) SVR model reveals the same three Inr variants that are seen in human promoters.

The predicted activities of these three Inr variants differ in a DPR- versus a TATA-driven promoter (Fig. 3, Supplemental Fig. S14). With the Inr10 (with DPR) SVRb model, the canonical A+1 Inr is most active, the A+2 Inr is moderately active, and the A+3 Inr is inactive or very weakly active. This preference is consistent with the previously observed spacing requirement between the Inr and DPR (Burke and Kadonaga 1997; Kutach and Kadonaga 2000; Vo ngoc et al. 2020; Vo ngoc et al. 2023). In contrast, with the Inr10 (with TATA) SVRb model, the A+1, A+2, and A+3 Inr variants are predicted to be of comparable activity. The A+3 variant exhibits distinctly different properties with the TATA box relative to the DPR.

To examine the Inr variants from a different perspective, we compared their occurrence in natural TATA- versus DPR-containing promoters (Supplemental Fig. S15). Strikingly, we found that an active A+3 Inr is present in 8.3% (368/4,448) of TATA-containing human promoters, but only in 0.1% (10/7,630) of DPR-containing human promoters. Even more notably, an active A+3 Inr is present in 4.6% (46/1,009) of TATA-containing Drosophila promoters and is completely absent (0/3,070) in the DPR-containing Drosophila promoters. These findings collectively indicate that the A+3 Inr variant functions specifically with the TATA box relative to the DPR.

A strong TATA box can drive transcription initiation in the absence of an Inr or DPR

By using the SVR models of the human Inr in conjunction with our previously generated models of the human TATA box (Vo ngoc et al. 2020), human DPR (Vo ngoc et al. 2020), and Drosophila DPR (Vo ngoc et al. 2023), we examined the co-occurrence of these elements in natural human and Drosophila promoters. A key feature of these analyses is that they provide quantitative genome-wide assessments rather than anecdotal observations of the relationships between the elements. When the promoters were ranked according to their Inr SVR scores, we observed a striking inverse relationship between predicted Inr and TATA activities (Fig. 4A,B, Supplemental Fig. S16). The duality between the TATA and DPR (see, for example: Willy et al. 2000; Hsu et al. 2008; Vo ngoc et al. 2020) can also be clearly seen in the heat maps that are ranked by the DPR SVR scores or the TATA SVR scores (Supplemental Fig. S16).

Figure 4.

Figure 4.

Open in a new tab

Predicted Inr activity positively correlates with predicted DPR activity and negatively correlates with predicted TATA box activity. Natural promoters were ranked according to their human Inr (with DPR) scores and divided into 30 bins of approximately equivalent size. For each bin, the mean SVR scores for human Inr (with DPR), human Inr (with TATA), human TATA box (Vo ngoc et al. 2020), and human or Drosophila DPR (Vo ngoc et al. 2020, 2023) were determined. (A,B) Positive correlation between the DPR and Inr and negative correlation between the TATA box and Inr in both humans (A) and Drosophila (B). (C,D) The predicted frequency of active Inr elements sharply declines with the predicted strength of active TATA motifs in human (C) and Drosophila (D) promoters. (E,F) The predicted frequency of active Inr elements increases with the predicted strength of DPR motifs in human (E) and Drosophila (F) promoters. In C–F, active Inr elements were defined to be those that are predicted to be active with both the Inr10 (with DPR) and Inr10 (with TATA) SVR models.

Further analysis of the Inr and TATA revealed that the predicted frequency of active Inr elements sharply declines with the predicted strength of active TATA motifs (Fig. 4C,D, Supplemental Fig. S17A,B). This effect was seen in two different human cell lines as well as in Drosophila embryos and S2 cells. In contrast, the predicted Inr and DPR activities are positively correlated (Fig. 4A,B, Supplemental Fig. S16), with the predicted frequency of active Inr elements increasing with the predicted strength of DPR motifs (Fig. 4E,F, Supplemental Fig. S17C,D). The positive correlation tween these elements (Burke and Kadonaga 1996). A negative correlation between the TATA and Inr was previously observed, as assessed with an Inr consensus sequence (Vo ngoc et al. 2020). The SVR models of the Inr further revealed the sharp decline of the Inr frequency with increasing strength of active TATA elements.

To examine the interrelationships between the TATA, Inr, and DPR elements, we carried out permutation testing with natural human and Drosophila promoters (Supplemental Fig. S18). This analysis revealed two significantly enriched core promoter types – those containing only a TATA box [TATA+, Inr−, DPR−] and those containing both an Inr and DPR [TATA−, Inr+, DPR+]. The enrichment of the [TATA−, Inr+, DPR+] promoters is in accord with the previously noted synergy between the Inr and DPR as well as the co-occurrence of these elements (Kutach and Kadonaga 2000; Vo ngoc et al. 2020). The enrichment of the [TATA+, Inr−, DPR−] promoters is consistent with the inverse correlation between the TATA and Inr (Fig. 4, Supplemental Figs. S16,S17) and the duality between the TATA box and DPR. These findings collectively indicate that the Inr appears to play a more important role in DPR-driven transcription than in TATA-driven transcription and that a strong TATA box can drive transcription initiation in the absence of an Inr or DPR.

It is also notable that we observed a much larger dynamic range of Inr activity with a DPR than with a TATA box (Supplemental Fig. S19). With a DPR-driven promoter, the strongest Inr variants are ~100-fold more active than the median inactive sequence, whereas with a TATA-driven promoter, the strongest Inr variants are only ~10-fold more active than the median inactive sequence. These findings further indicate that there is a stronger functional relationship between the Inr and DPR than between the Inr and TATA. Thus, the high throughput HARPE data, the SVR models of the Inr, TATA, and DPR, and the genome-wide human and Drosophila promoter analyses collectively lead to a model in which there is a strict and synergistic interaction between the Inr and DPR, a duality between the TATA and DPR, and a flexible and sometimes independent function of the TATA box in relation to the Inr.

Analysis and perspectives

The Inr is a widely-used transcriptional element that is present in about 60% of human promoters. Here, we combined high-throughput transcriptional analyses with machine learning to generate highly predictive models of the human Inr. The SVR models are effective at predicting active Inr elements, and it is thus reasonable to ask, anthropomorphically, “What are the SVR models thinking?” To address this question, we first determined the predicted Inr activity of all of the 1,048,576 possible 10-nt sequences and identified the predicted optimal Inr motif (Supplemental Fig. S20A). Second, to evaluate the importance of each position within the Inr region, we randomized the nucleotides at each position and measured the resulting decrease in model performance (Supplemental Fig. S20B). This analysis revealed that the positions corresponding to the minimal CANW Inr sequence are the most important for Inr activity, but also that all four nucleotides in this motif contribute to Inr function. Third, we evaluated the effect of every nucleotide substitution at each position across all 10-nt sequences, and identified position- and nucleotide-specific effects on predicted transcriptional activity (Supplemental Fig. S20C). Collectively, these model interpretation analyses uncover the key sequence features that determine Inr activity.

There are different types of core promoters that have different properties based on the presence or absence of motifs such as the TATA box and DPR. For instance, in Drosophila, TATA-specific and DPR-specific transcriptional enhancers have been identified (Butler and Kadonaga 2001; Juven-Gershon et al. 2008). The SVR models have enabled us to gain a better understanding of the relationships between the TATA box, Inr, and DPR elements, and have thus provided new insights into the nature of the different types of core promoters. In this regard, we further investigated the TCT motif. Because the TCT motif is rare and overlaps the Inr, we were not able to generate a TCT-specific SVR model. Therefore, for the analysis of the TCT motif, we used sequences that match the human TCT consensus (YC+1TYTYY), and found that the relation between the TATA box and TCT motif in humans is, strikingly, opposite from that seen in Drosophila (Supplemental Fig. S21). Specifically, we observed that the TATA box is overrepresented in human TCT promoters and underrepresented in Drosophila TCT promoters. We likewise found that the TCT motif is enriched in human TATA-containing promoters and depleted in Drosophila TATA-containing promoters. These findings indicate that the properties of the TCT motif are different in humans than in Drosophila. This point should be taken into consideration in the analysis of the TCT motif in humans.

It is also important to consider whether different types of core promoter elements are associated with specific biological functions. For instance, the TCT motif is used by the ribosomal protein gene network (Parry et al. 2010). To explore the possible biological functions that may be associated with the two enriched core promoter types (TATA-only and Inr-DPR; Supplemental Fig. S18), we performed gene ontology (GO) analyses with human promoters (Supplemental Table 4A,B). We found that genes associated with TATA-only promoters are enriched for many functions related to signaling, immune response, and stress response, which suggests that genes that mediate responses to changing conditions may frequently employ TATA-driven transcription. In contrast, the Inr-DPR promoters did not yield any significantly enriched GO terms. These results suggest that there are different biological functions of TATA-only and Inr-DPR core promoters in human gene regulation.

Somewhat remarkably, we found that the optimal human Inr, TCA+1KTY, is essentially identical to the Drosophila Inr consensus (see, for example: Hultmark et al. 1986; Ohler et al. 2002, FitzGerald et al. 2006). Similarly, in Vo ngoc et al. (2023), we found that the DPR consensus sequence is nearly the same in humans and Drosophila. It thus appears that the Inr and DPR motifs have remained largely unchanged over the ~700 million years of species divergence time between Drosophila and humans (Kumar et al. 2022), despite differences in promoter G+C content (Deaton and Bird 2011).

Thus, in conclusion, we were able to generate SVR models of the human Inr and to gain key insights into its occurrence as well as its relation to other core promoter elements at the genome-wide level. There remains, however, much to be learned. In the future, we will seek to identify the DNA motifs and protein factors that drive transcription from the roughly 30% of human promoters that lack any of the known core promoter elements. We also plan to integrate our understanding of the core promoter with the functions of promoter- and enhancer-binding proteins. In this manner, it may be possible to uncover new unified perspectives into the specific molecular interactions that mediate the regulation of transcription by RNA polymerase II.

Materials and Methods

HARPE analysis of the Inr

The HARPE plasmid libraries for the Inr were generated as described in Vo ngoc et al. (2020). The HARPE vector and SCP1-based promoter backbones, which include constructs with mutations in the DPR and/or the TATA box, are identical to those used previously (Vo ngoc et al. 2020, 2023). For HARPE analysis of the Inr, the 14-nt region from −5 to +9 and the 10-nt region from −4 to +6 (relative to the SCP1 +1 TSS) were randomized to generate ~500,000 Inr sequence variants within SCP1 promoter backbones containing a DPR (mutTATA), a TATA box (mutDPR), or neither element (mutTATA, mutDPR). In vitro transcription with HeLa nuclear extracts and transfection into HeLa cells were performed as described in Vo ngoc et al. (2020). Sample and data processing were performed as described in Vo ngoc et al. (2020, 2023). Illumina sequencing of NGS PCR amplicons was performed on NovaSeq 6000 or NovaSeq X Plus sequencing systems at the IGM Genomics Center, University of California, San Diego. Additional information on HARPE library generation, transcription, RNA processing, sequencing, and data analysis is provided in the Supplemental Material.

Generation and use of the SVR models

Machine learning analyses were performed as described in Vo ngoc et al. (2020, 2023), except that we used functions from the Python package scikit-learn (version 1.3.0; Pedregosa et al. 2011) instead of the R package e1071 (version 1.7–2). Each SVR model was trained with 200,000 HARPE variants and then evaluated with ~7,000–8,000 independent variants (i.e., not used for training). Grid search was used to optimize the hyperparameters cost and gamma, and cross validation was carried out with two independent test sets. Additional information on model training and optimization is provided in the Supplemental Material. We used SVR models of the human Inr (this study), human TATA box (“SVRTATA”; Vo ngoc et al. 2020; GSE139635), human DPR (“SVRb”; Vo ngoc et al. 2020; GSE139635, which is the same as “SVRh”; Vo ngoc et al. 2023; GSE225570) and Drosophila DPR (“SVRd”, Vo ngoc et al. 2023; GSE225570).

Supplementary Material

Supplement 1

Acknowledgments

We thank Grisel Cruz-Becerra and George Kassavetis for critical reading of the manuscript. J.T.K. is the Amylin Chair in the Life Sciences. This publication includes data generated at the UC San Diego IGM Genomics Center utilizing Illumina NovaSeq 6000 and NovaSeq X Plus systems that were purchased with funding from National Institutes of Health SIG grant (#S10 OD026929). This work used Expanse CPU at the San Diego Supercomputer Center through the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program (project BIO230152), which is supported by NSF grants 2138259, 2138286, 2138307, 2137603, and 2138296. T.E.R.-C. received support from NIH grant T32 GM133351 to UCSD Biological Sciences Pathways in Biological Sciences (PiBS) Training Program and NSF Graduate Research Fellowship (GRFP) DGE-2545911. This work was supported by NIH grant R35 GM118060 to J.T.K. This paper was typeset with the bioRxiv word template by @Chrelli: www.github.com/chrelli/bio-Rxiv-word-template.

Footnotes

Competing Interest Statement

The authors declare no competing interests.

References

  1. Archuleta SR, Goodrich JA. Kugel JF. 2024. Mechanisms and functions of the RNA polymerase II general transcription machinery during the transcription cycle. Biomolecules 14: 176. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Burke TW, Kadonaga JT. 1996. Drosophila TFIID binds to a conserved downstream basal promoter element that is present in many TATA-box-deficient promoters. Genes Dev 10: 711–724. [DOI] [PubMed] [Google Scholar]
  3. Burke TW, Kadonaga JT. 1997. The downstream core promoter element, DPE, is conserved from Drosophila to humans and is recognized by TAFII60 of Drosophila. Genes Dev 11: 3020–3031. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Butler JE, Kadonaga JT. 2001. Enhancer-promoter specificity mediated by DPE or TATA core promoter motifs. Genes Dev 15: 2515–2519. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Corden J, Wasylyk B, Buchwalder A, Sassone-Corsi P, Kedinger C, Chambon P. 1980. Promoter sequences of eukaryotic protein-coding genes. Science 209: 1406–1414. [DOI] [PubMed] [Google Scholar]
  6. Core LJ, Martins AL, Danko CG, Waters CT, Siepel A, Lis JT. 2014. Analysis of nascent RNA identifies a unified architecture of initiation regions at mammalian promoters and enhancers. Nat Genes 46: 1311–1320. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Deaton AM, Bird A. 2011. CpG islands and the regulation of transcription. Genes Dev 25: 1010–1022. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Duttke SH, Chang MW, Heinz S, Benner C. 2019. Identification and dynamic quantification of regulatory elements using total RNA. Genome Res 29: 1836–1846. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. FitzGerald PC, Sturgill D, Shyakhtenko A, Oliver B, Vinson C. 2006. Comparative genomics of Drosophila and human core promoters. Genome Biol 7: R53. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Haberle V., Stark A. 2018. Eukaryotic core promoters and the functional basis of transcription initiation. Nat Rev Mol Cell Biol 19: 621–637. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Heinz S, Benner C, Spann N, Bertolino E, Lin YC, Laslo P, Cheng JX, Murre C, Singh H, Glass CK. 2010. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol Cell 38: 576–589. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Hsu JY, Juven-Gershon T, Marr MT II, Wright KJ, Tjian R, Kadonaga JT. 2008. TBP, Mot1, and NC2 establish a regulatory circuit that controls DPE-dependent versus TATA-dependent transcription. Genes Dev 22: 2353–2358. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Hultmark D, Klemenz R, Gehring WJ. 1986. Translational and transcriptional control elements in the untranslated leader of the heat-shock gene hsp22. Cell 44: 429–438. [DOI] [PubMed] [Google Scholar]
  14. Juven-Gershon T, Cheng S, Kadonaga JT. 2006. Rational design of a super core promoter that enhances gene expression. Nat Methods 3: 917–922. [DOI] [PubMed] [Google Scholar]
  15. Juven-Gershon T, Hsu JY, Kadonaga JT. 2008. Caudal, a key developmental regulator, is a DPE-specific transcription factor. Genes Dev 22: 2823–2830. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Kaufmann J, Smale ST. 1994. Direct recognition of initiator elements by a component of the transcription factor IID complex. Genes Dev 8: 821–829. [DOI] [PubMed] [Google Scholar]
  17. Kruesi WS, Core LJ, Waters CT, Lis JT, Meyer BJ. 2013. Condensin controls recruitment of RNA polymerase II to achieve nematode X-chromosome dosage compensation. Elife 2: e00808. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Kumar S, Suleski M, Craig JM, Kasprowicz AE, Sanderford M, Li M, Stecher G, Hedges SB. 2022. TimeTree 5: an expanded resource for species divergence times. Mol Biol Evol 39: msac174. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Kutach AK, Kadonaga JT. 2000. The downstream promoter element DPE appears to be as widely used as the TATA box in Drosophila core promoters. Mol Cell Biol 20: 4754–4764. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Ohler U, Liao GC, Niemann H, Rubin GM. 2002. Computational analysis of core promoters in the Drosophila genome. Genome Biol 3: RESEARCH0087. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Parry TJ, Theisen JWM, Hsu JY, Wang YL, Corcoran DL, Eustice M, Ohler U, Kadonaga JT. 2010. The TCT motif, a key component of an RNA polymerase II transcription system for the translational machinery. Genes Dev 24: 2013–2018. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Purnell BA, Emanuel PA, Gilmour DS. 1994. TFIID sequence recognition of the initiator and sequences farther downstream in Drosophila class II genes. Genes Dev 8: 830–842. [DOI] [PubMed] [Google Scholar]
  23. Roeder RG. 2019. 50+ years of eukaryotic transcription: an expanding universe of factors and mechanisms. Nat Struct Mol Biol 26: 783–791. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Schier AC, Taatjes DJ. 2020. Structure and mechanism of the RNA polymerase II transcription machinery. Genes Dev 34: 465–488. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Sloutskin A, Shir-Shapira H, Freiman RN, Juven-Gershon T. 2021. The core promoter is a regulatory hub for developmental gene expression. Front Cell Dev Biol 9: 666508. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Smale ST, Baltimore D. 1989. The “initiator” as a transcription control element. Cell 57: 103–113. [DOI] [PubMed] [Google Scholar]
  27. Smale ST, Kadonaga JT. 2003. The RNA polymerase II core promoter. Annu Rev Biochem 72: 449–479. [DOI] [PubMed] [Google Scholar]
  28. Verrijzer CP, Chen JL, Yokomori K, Tjian R. 1995. Binding of TAFs to core elements directs promoter selectivity by RNA polymerase II. Cell 81: 1115–1125. [DOI] [PubMed] [Google Scholar]
  29. Vo ngoc L, Wang YL, Kassavetis GA, Kadonaga JT. 2017a. The punctilious RNA polymerase II core promoter. Genes Dev 31: 1289–1301. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Vo ngoc L, Cassidy CJ, Huang CY, Duttke SHC, Kadonaga JT. 2017b. The human initiator is a distinct and abundant element that is precisely positioned in focused core promoters. Genes Dev 31: 6–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Vo ngoc L, Huang CY, Cassidy CJ, Medrano C, Kadonaga JT. 2020. Identification of the human DPR core promoter element using machine learning. Nature 585: 459–463. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Vo ngoc L, Rhyne TE, Kadonaga JT. 2023. Analysis of the Drosophila and human DPR elements reveals a distinct human variant whose specificity can be enhanced by machine learning. Genes Dev. 37: 377–382. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Wang YL, Duttke SHC, Chen K, Johnston J, Kassavetis GA, Zeitlinger J, Kadonaga JT. 2014. TRF2, but not TBP, mediates the transcription of ribosomal protein genes. Genes Dev 28: 1550–1555. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Willy PJ, Kobayashi R, Kadonaga JT. 2000. A basal transcription factor that activates or represses transcription. Science 290: 982–984. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement 1
NIHPP2025.11.21.689830v2-supplement-1.pdf (18.7MB, pdf)

Articles from bioRxiv are provided here courtesy of Cold Spring Harbor Laboratory Preprints

RESOURCES