A massively parallel reporter assay dissects the influence of chromatin structure on cis-regulatory activity

Abstract

A gene’s position in the genome can profoundly affect its expression because regional differences in chromatin modulate the activity of locally acting cis-regulatory sequences (CRSs). Here we study how CRSs and regional chromatin act in concert on a genome-wide scale. We present a massively parallel reporter gene assay that measures the activities of hundreds of different CRSs, each integrated at many specific genomic locations. Although genome location strongly affected CRS activity, the relative strengths of CRSs were maintained at all chromosomal locations. The intrinsic activities of CRSs also correlatedwith their activities in plasmid-based assays. We explain our data with a quantitative model in which expression levels are set by independent contributions from local CRSs and the regional chromatin environment, rather than by more complex sequence- or protein-specific interactions between these two factors. The methods we present will help investigators determine when regulatory information is integrated in a modular fashion and when regulatory sequences interact in more complex ways.

Two types of cis-regulatory information control the production of mRNAs: local cis-regulatory CRSs, such as promoters, enhancers and insulators^1–3, and the regional chromatin surrounding a gene’s location in the genome^4–6. Although short CRSs are often sufficient to drive reporter genes in patterns that reflect their activity in the genome^7–12, their activities are also heavily influenced by regional differences in chromatin state. Regional differences in chromatin underlie chromosome position effects, the phenomenon whereby the same gene is expressed at dramatically different levels depending on its location in the genome^6,13–25. To identify the chromatin features that make different chromosomal locations permissive for CRS activity, investigators have produced detailed maps of the epigenetic and topological properties of the genome^26–31. By integrating a reporter gene at several—or even thousands of—different chromosomal locations, investigators have tried to identify the chromatin features that correlate with a region being permissive or repressive for the function of a particular CRS^{6,15,19,21–23,32}.

A key unanswered question is whether local CRSs and regional chromatin interact in complex sequence-specific ways or whether the two effects contribute independently to gene expression. Strong and weak CRSs in one chromatin state might change their relative activities in other chromatin environments, depending on the specific molecular interactions that occur at different genomic locations. Alternatively, different chromatin states might simply scale the activities of strong and weak elements without altering their relative activities. This question remains unresolved because to separate the effects of local and regional regulatory features we must measure the activity of many CRSs integrated at diverse genomic locations. Despite the rich history^{6,13,15,17–21,23,24,33–36} of chromosome position effect studies, previous experiments only measured small numbers of CRSs. It therefore remains an open question whether the effects of chromosome position depend heavily on the type of CRS that is present or whether chromosome effects are largely independent of the sequence of CRS.

To address this question we devised a massively parallel reporter assay (MPRA)^8,37–41, called patchMPRA (parallel targeting of chromosome positions by MPRA), to measure the activity of hundreds of CRSs integrated at several distinct genomic locations. The key innovation was to create a system in which integrated reporter genes produce mRNAs with two distinct barcodes: a CRS barcode (cBC) that specifies the identity of its CRS and a genomic barcode (gBC) that specifies the location of the integrated reporter gene. patchMPRA allows us to measure the same set of CRSs at different chromosomal locations and thereby interrogate how local CRSs and regional chromatin properties set gene expression levels across the human genome.

We first created ‘landing pad’ cell lines, each of which carries a single barcoded landing pad for site-directed recombination of an MPRA library into a single genomic location. A landing pad consists of asymmetric lox sites⁴², flanking a CMV-GFP cassette, and a unique gBC that specifies the location of the landing pad in the genome (Fig. 1a). To facilitate integration of landing pads into many genomic locations in a range of cell types, we placed the landing pad cassette on a self-inactivating lentivirus and cloned a library of gBCs. Every unique lentiviral insertion delivers a landing pad with a unique gBC to a different genomic location. We created a pool of landing pad lines in human K562 cells via low-multiplicity lentiviral transductions to promote single-copy genomic integrations. K562 is an immortalized myelogenous leukemia cell line and a tier 1 ENCODE cell line with genome-wide annotations based on an array of functional genomics data²⁶. We generated dozens of clonal cell lines derived from singly transduced K562 cells and mapped the integration sites of gBCs for 15 lines (Supplementary Table 1). The landing pads integrated into diverse epigenetic landscapes as determined by the constellation of epigenetic marks associated with different integration sites (Supplementary Fig. 1 and Supplementary Table 2), including into repressed regions with the heterochromatin marks Lys27-trimethylated histone H3 (H3K27me3) and Polycomb.

Figure 1.

patchMPRA experimental design. (a) loxFAS and loxP landing pads are integrated into random genomic locations using lentiviral transduction. The inserted construct encodes eGFP and a hygromycin resistance and thymidine kinase (Hyg/TK) cassette under the control ofa CMV promoter. Each landing pad contains a unique sequence barcode (gBC) downstream of the loxP site. Singly transduced cells are sorted into 96-well plates and clonal cell lines are expanded. The genomic locations of landing pad gBCs in clonal lines are mapped with an inverse-PCR-based approach. (b) Libraries of regulatory DNA sequences are cloned upstream of uniquely barcoded (cBC) dsRed reporter genes in a transfer vector containing loxFAS and loxP sites. Transfer vector libraries are transfected with Cre recombinase. (c) After recombination every integrated reporter produces an mRNA with both a cBC and a gBC that specify the identity and location of the reporter gene. DNA and RNA are extracted from cells that undergo Cre-mediated recombination, and cBC–gBC pairs are quantified by sequencing.

In our first experiment we pooled eight landing pad lines and introduced an MPRA library in which every reporter gene contains a unique cBC in its 3′ untranslated region (UTR) that identifies its upstream CRS (Fig. 1b). To provide redundancy in the measurements, every CRS in the library was present 25 times, each time fused with a different cBC in its 3′ UTR. Asymmetric lox sites flanked the MPRA library to facilitate directional Cre recombinase–mediated cassette exchange into the landing pads. We then isolated cells that lost GFP fluorescence using fluorescence-activated cell sorting (FACS), enriching for cells in which barcoded reporter genes integrated into barcoded landing pads. Each reporter gene in these pools transcribes an mRNA that contains a cBC specifying the identity of the CRS that produced the mRNA and a gBC specifying the location of its host landing pad (Fig. 1c). By quantifying barcodes with RNA sequencing, we simultaneously measured the expression driven by every CRS in the library at every landing pad in the pool.

The CRS library in this first experiment contained sequences predicted to be either strong enhancers, weak enhancers or repressed according to ENCODE segmentation algorithms^43,44. These sequences were previously screened for regulatory activity in K562 cells using an episomal MPRA⁴⁵. We selected high- and low-activity sequences from each segmentation class (Supplementary Fig. 2 and Supplementary Table 3) and cloned a total of 310 elements into a transfer vector containing the Hsp68 minimal promoter and dsRed coding sequence flanked by asymmetric lox sites. These elements were cloned in the same orientation with respect to the minimal promoter in which they were assayed by Kwasnieski et al.⁴⁵. It was this library that was integrated into a pool of eight landing pad cell lines (above)

We achieved a high number of CRS integrations, totaling more than 30,000 unique site-directed recombination events as based on the number of unique gBC–cBC combinations we observed in the data. This high coverage ensured that CRS and genomic locations were well represented in the experiment. On average each CRS was represented by 11 unique barcode integrations per landing pad. The correlation between replicate experiments was R = 0.66 (Spearman correlation, R_s = 0.71) (Fig. 2a), which allowed us to make quantitative comparisons between CRSs within and between landing pads. Replicate experiments conducted months apart produced the same patterns of CRS activity across four landing pads subject to drastically different position effects (Supplementary Fig. 3), demonstrating the robustness of our assay.

Figure 2.

Genomic positions have large effects on CRS activity. (a) cBC and gBC measurements (log₂(RNA reads/DNA reads)) are averaged for each CRS within a landing pad. The mean expression for each CRS at each landing pad is plotted for two biological replicate experiments (Pearson product-moment correlation R = 0.66; Spearman’s rank-order correlation coefficient R_s = 0.71). (b) The distributions of expression for the same set of 310 CRSs are plotted for each landing pad (LP). Box plot elements: center line, median; limits, 25% quartile (Q1) and 75% quartile (Q3); whiskers, from Q1 – 1.5 × IQR to Q3 + 1.5 × IQR, where IQR is the interquartile range.

The genomic location of landing pads had large effects on reporter gene levels (Fig. 2b). For example, the entire collection of CRSs at landing pad 3 had higher levels of activity than the same CRSs integrated at landing pad 4. These data are consistent with the absence of anti-repressor and insulator elements in the landing pads that would mitigate chromosome position effects⁴⁶, as well as with data demonstrating position effects of CRSs integrated with inverted-repeat-containing PiggyBac vectors⁶. Thus, there are strong chromosomal position effects in our experiment, which are not insulated by the transcriptionally inert long terminal repeats⁴⁷ flanking the landing pads.

Although genomic location had a large effect on average CRS activity, we also found that relative CRS activities were largely preserved between landing pad locations. When we divided the 300 CRSs into low, medium and high groups based on their activity at landing pad 1, we found that these designations were largely preserved at all locations (Fig. 3a). Activity designations were preserved regardless of which landing pad was used as the reference (Supplementary Fig. 4). This was true even when the position effect was strong enough to suppress high-activity CRSs below the levels of low-activity CRSs at other genomic locations. In several cases the correlation between CRSs at different landing pads approached the correlation between replicate experiments (Fig. 3b and Supplementary Fig. 5).

Figure 3.

Intrinsic CRS activity is linearly scaled by genomic position. (a) CRSs are determined to have low, medium or high activity at landing pad 1 and plotted for each landing pad according to their classification at landing pad 1 (high, orange; medium, gray; low, blue). (b) CRS expression data at landing pads 1 and 4 (left) and landing pads 1 and 6 (right) are shown (Pearson product-moment correlation R = 0.71 and 0.65, respectively, and Spearman’s rank correlation coefficient R_s = 0.65 and 0.60, respectively). Box plot elements: center line, median; limits, 25% quartile (Q1) and 75% quartile (Q3); whiskers, from Q1 – 1.5 × IQR to Q3 + 1.5 × IQR, where IQR is the interquartile range.

These results suggest that the local DNA sequence of a CRS determines its intrinsic cis-regulatory activity and that this intrinsic activity is then scaled up or down by the regional chromosome environment, independent of the specific CRS sequence. To test this prediction, we fit a model in which gene expression is the product of intrinsic activity and chromosome position effect. We fit the model

$$E_{ij}=C_i{L}_j+\epsilon ,\epsilon ~N(0,{\sigma }^2)$$

to our patchMPRA data, averaged across replicates. Here E_ij is the expression of CRS i at genomic location j, C_i is the intrinsic activity of CRS i, L_j is the multiplicative scaling effect of genomic location j, and ε is random error. The fitted model explained our data with R = 0.9 and R_s = 0.9 (Fig. 4a) and cross-validated correlations of R = 0.83 and R_s = 0.87 (Supplementary Fig. 6a,b) An ANOVA of this model suggested that genomic location explains 64% of the total variance in gene expression and that CRS sequence explains another 16%, with roughly 20% attributable to biological replicate variance (Supplementary Fig. 6c,d). Part of the unexplained variance could also be attributed to specific interactions between CRS and genomic location; however, given the replicate variance, the contribution of any specific chromatin–CRS interactions to activity must be small compared to the independent effects of genomic location or CRS sequence. The close agreement between the model and the data supports the hypothesis that regional properties of the chromosome scale the activity of CRSs in a sequence-independent manner.

Figure 4.

A linear model captures the effects of local and regional cis-regulatory effects. (a) CRS expression data across landing pads are described by a linear model (R = 0.90; R_s = 0.91). (b) Coefficients (C_i) for each CRS from the linear model fit on CRS expression data across landing pads are plotted according to the CRS ENCODE segmentation (R, repressed; SE, strong enhancer; WE, weak enhancer) and the activity of the CRS in episomal MPRA. Wilcoxon test between groups (*P < 0.1, **P < 0.05). (c) Activities of 11 previously tested CRSs are shown at 7 more landing pads (LP9–LP15) (high, orange; low, blue) (d) 24 CRSs were assayed with either the Hsp68 or MinP promoter across landing pad locations. (e) Comparison of longer, literature-validated enhancers to shorter cis-regulatory sequences. All sequences greater than 130 bp are literature-validated enhancers. Box plot elements: center line, median; limits, 25% quartile (Q1) and 75% quartile (Q3); whiskers, from Q1 – 1.5 × IQR to Q3 + 1.5 × IQR, where IQR is the interquartile range.

The values of intrinsic activity factors C_i are estimated from the fitted model as the average activity of each element across all locations. These values are estimated without the use of episomal MPRA data. Nevertheless, the C_i values we fit from patchMPRA data corresponded well to the expression values we measured from plasmid-based MPRA, with more active CRSs having higher C_i values (Fig. 4b). Overall, the expression on plasmids was moderately correlated to expression at genomic landing pads (average R = 0.35, data not shown), but groups of CRSs defined as having low, medium or high activity on the basis of plasmid expression maintained their relative activity levels across all landing pads (Supplementary Fig. 7). This result suggests that the activity of a CRS measured on a plasmid, in the absence of regional chromosome effects, is a reasonable representation of its intrinsic cis-regulatory activity.

The fitted values L_j represent the modular effects of different chromatin regions on the activity of CRSs. Because we only assayed a small number of landing pads from different ENCODE segmentation classes, we cannot yet draw general conclusions about what specific chromosomal properties associate with permissive versus restrictive landing pad sites. We do note, however, that in our experiment some genomic regions annotated by ENCODE as ‘repressed’ permitted high landing pad activity and that some genomic regions annotated as ‘transcribed’ restricted activity (Supplementary Table 1). In an experiment wherein a single reporter gene was integrated at thousands of different genomic locations, Akhtar et al.⁶ also failed to identify a strong relationship between reporter activity and specific chromatin states or topological domains, but did note attenuated reporter gene activity in chromosomal regions associated with the nuclear lamina. While the specific chromatin features that dictate permissiveness remain to be identified, our data strongly support a model in which chromatin functions independently of local CRSs.

We performed a second patchMPRA experiment in which we pooled all 15 mapped landing pad cell lines to determine whether the independence we observed between local and regional effects holds across other genomic locations. We also included library members in which the Hsp68 minimal promoter was swapped for the MinP promoter^40,45 to test whether linear scaling depends on the specific promoter present in the local CRS. Finally, we included new library members representing full length, literature-validated enhancer sequences (Supplementary Table 4) to test whether linear scaling holds when longer enhancer sequences are present.

We observed several clear trends (Supplementary Table 5). Local CRSs that drove either high or low expression in our first experiment maintained their relative activities in the new landing pad locations (Fig. 4c). This suggests that modularity of local CRSs is a general property of many chromosomal locations. Linear scaling of CRS activity was also preserved across genomic locations when the Hsp68 promoter was swapped with the MinP promoter (Fig. 4d). At all locations the Hsp68 promoter drove higher expression than the MinP promoter, but CRSs with both promoters showed the same relative strengths across genomic locations. A linear model including terms for local CRS, regional chromosome effect and promoter identity explains the data well (R = 0.82). An ANOVA of this model shows that the local CRS explains 9.0%, the regional chromosome effect explains 26%, and the promoter identity explains 26%. This result suggests that promoters are strong but modular elements that function independently of their local CRS and regional chromosome environment.

We also directly compared the activities of full length, literature-validated enhancers against the activities of CRSs in our original library. Although the literature-validated enhancers were generally longer than the CRSs in our first library, they did not show higher cis-regulatory activities, and they showed the same scaling across landing pad locations (Fig. 4e). Literature-validated enhancers do not appear to carry additional sequences that buffer them from chromosome position effects. Taken together, our results suggest a modular organization of the genome in which the effects of promoters, enhancers and regional chromatin all contribute independently to set levels of gene expression.

Here we have presented a method to jointly study the contributions of local DNA sequences and regional genomic features to gene expression. We demonstrated that local DNA sequences determine the intrinsic activity of genome-integrated CRSs and that genomic position acts as a multiplicative amplifier of gene expression that scales the entire collection of CRSs equally. The CRSs we assayed in K562 cells tended to be enriched for binding sites for transcription factors such as AP-1, ETS-1 and GATA-1. Whether CRSs with other types of motifs (for example, insulator binding sites) or promoters show the same scaling effects is an important open question.

While the current reproducibility of individual replicates in patchMPRA is good (R_s = 0.71), it is not as high as the reproducibility of episomal MPRA. Currently this limits the number of CRSs and landing pads that can be assayed in a single experiment because it is important to average over several barcodes per CRS. One reason for this is that patchMPRA measures DNA and RNA barcodes produced from a single integration per cell. Measurement error is primarily driven by difficulty in quantifying barcodes from the DNA fraction and by the low efficiency of Cre-lox recombination (12%). As a result, most FACS-sorted landing pad cells (~70%) have not recombined, further complicating the quantification of DNA barcodes. Counterselection methods could be used to increase the fraction of cells with recombinations, or in vitro transcription could be used to linearly amplify DNA barcodes before sequencing. Additional gains in throughput could also come by allowing multiple landing pad integrations per cell. Finally, if linear scaling is a general property of most genomic regions, throughput could be increased by reducing the number of CRSs per library and increasing the number of locations assayed. This approach could also be combined with TRIP⁶, which assays a single CRS at many thousands of locations.

The multiplicative scaling relationship we observe suggests that local and regional cis-regulatory elements are modular and affect the same step in gene expression. For example, local CRSs might influence transcription factor occupancy by carrying transcription factor binding sites with different affinities for transcription factors, whereas the regional chromatin environment might influence transcription factor occupancy by determining the accessibility of CRSs. A chromosomal region that is relatively inaccessible scales the activity of all CRSs down by the same amount, regardless of which particular transcription factors they bind, because all transcription factors must diffuse through the same densely packed chromatin. Other molecular mechanisms that affect chromatin, such as long-range looping or histone modifications, might also account for the regional effect if they influence the diffusion of all transcription factors to the same degree. This model does not, therefore, suggest sequence-specific or protein-specific interactions between regional chromatin and local CRSs.

A future direction of this work will be to determine the chromosomal features that determine the regional scaling factors of genomic locations. Addressing this question will require assaying a larger number of landing pad locations that cover a diverse set of chromatin environments and using the CRISPR–Cas9 system to direct landing pads to genomic regions with particular epigenetic properties.

METHODS

Methods, including statements of data availability and any associated accession codes and references, are available in the online version of the paper.

ONLINE METHODS

Lentiviral landing pad vector construction.

The lentiviral landing pad vector was constructed as follows. A synthetic DNA fragment, referred to as the “landing pad cassette,” was inserted between the long terminal repeats of a modified, self-inactivating lentiviral construct using NheI and BamHI restriction sites. The self-inactivating nature of this lentiviral construct ensures that the flanking long terminal repeats are transcriptionally inert after integration. The landing pad cassette itself is composed of the following components. As a transcription unit for imaging and selection of transduced cells, we used a bicistronic gene encoding a Hyg/TK fusion protein⁴⁸ (RV-L3-HyTK-2L, Addgene 11684) and green fluorescent protein (eGFP), linked by a P2A sequence⁴⁹ and driven by a CMV promoter. This cassette is flanked by a loxFAS site⁴² on the 5′ end and a loxP site on the 3′ end for unidirectional recombination using Cre recombinase. Hyg/TK provides positive selection when transduced cells are treated with hygromycin, and counterselection following Cre-lox recombination when cells are treated with ganciclovir. A library of 12-bp DNA sequence barcodes (gBC) was cloned immediately downstream of the loxP site. These barcodes serve as unique identifiers for mapping the genomic locations of landing pad integrations. A WHP post-transcriptional response element (WPRE) was included downstream of the loxP site in the 3′ UTR. This lentiviral landing pad library was used to produce high-titer lentiviral library for transduction of mammalian cells.

K562 cell culture and generation of landing pad cell lines.

K562 cells (see Life Sciences Reporting Summary) were maintained in Iscove’s Modified Dulbecco′s Medium (IMDM) + 10% FBS + 1% non-essential amino acids. The lentiviral library carrying the barcoded landing pad cassettes was used to transduce K562 cells at a multiplicity of infection of one. 2.5 × 10⁴ K562 cells were exposed to 2.5 × 10⁴ infectious units of viral particles in suspension for 20 h. After 20 h, cells were collected by gentle centrifugation (100g for 5 min) and the medium was replaced with fresh, supplemented IMDM. Cells were expanded for 7 d under selection with hygromycin, resulting in a relatively pure population of transduced K562 cells. To obtain clonal cell lines harboring distinct landing pad locations, we used FACS to sort single GFP-positive K562 cells into 96-well plates. Individual clones were then expanded in culture for 10–14 d.

Mapping genomic locations of landing pads.

We modified an inverse PCR-based protocol^6,50 to map the genomic locations of landing pad integration sites. Briefly, 2 μg of genomic DNA from landing pad cell lines was digested with Csp6I (Thermo Fisher) in a volume of 50 μl for 3 h at 37 °C. Purified, digested DNA was ligated in a volume of 400 μl at 4 °C for 16 h to promote self-ligation. Ligation reactions were concentrated down to 50 μl and purified using Amicon Ultra concentrator tubes (Millipore), and 25 μl was used as the template for inverse PCR. We performed inverse PCR with primers LP01 5′-TTAAGGCGCGCCCCCTCAGACCCTTTTAGTCAGTGTGG-3′ and LP02 5′-TTAAGCATGCCGATAAGCTTGATATCGAATTC-3′ (Supplementary Table 6), which amplified a fragment containing the landing pad barcode and a region of genomic DNA flanking the insertion, and added AscI and SphI restriction sites for Illumina library preparation. Inverse PCR products were purified (Qiagen PCR Purification kit), digested with AscI and SphI, subjected to ligation with Illumina sequencing adapters Supplementary Table 6), and used as the template for Illumina enrichment PCR (Supplementary Table 6). PCR fragments were purified (Qiagen PCR Purification kit), diluted to 10 nM, and sequenced on the Illumina MiSeq machine.

Selection and validation of landing pad locations.

Genomic sites of landing pads were identified by aligning sequenced genomic fragments from inverse PCR to the human reference genome (hg19). ENCODE segmentation data were used to select landing pad cell lines with a diverse set of integration locations (Supplementary Table 1). For validation, we used genomic DNA from each clonal cell line as the template for PCR with an internal landing pad primer and a location-specific primer (Supplementary Table 6).

pGL transfer vector construction.

The pGL transfer vector for MPRA cassette exchange was constructed as follows. The landing pad cassette was subcloned into the pGL4.23 vector backbone using NheI and BamHI restriction sites. Cis-regulatory elements, the Hsp68 minimal promoter and a dsRed open reading frame containing cotranscribed sequence barcodes (cBC) were cloned between the loxFAS and loxP sites using XhoI and AgeI restriction sites.

Library design.

Our library of regulatory DNA sequences was previously studied using episome-based MPRA⁴⁵. Each sequence is predicted to be a strong enhancer, weak enhancer or repressed region in K562 cells based on extensive chromatin profiling data⁴³. Kwasnieski et al.⁴⁵ measured the activity of 600 sequences from the strong enhancer and weak enhancer classes and 300 from the repressed class in K562 cells. We filtered this set for sequences whose cis-regulatory activities were tightly measured (s.e.m. < 0.1). Within each class we selected a set of sequences with high activity and a set of sequences with low activity (Supplementary Table 3). We also included control sequences that we expected to drive high or low reporter expression. Each sequence was tagged by 25 unique cBCs. The library was synthesized by Agilent Technologies through a limited licensing agreement as a pool of 130-bp oligonucleotides and cloned directionally into the pGL transfer vector in the same orientation as assayed in Kwasnieski et al.⁸.

For the second library, we cloned 13 literature-validated CRSs amplified from K562 genome (primers LP34–LP59) and 11 CRSs amplified from the first library (primers LP60–LP81) into a pGL transfer vector using HiFi DNA Assembly (NEB) (Supplementary Tables 4 and 5). We amplified Hsp68:dsRed and MinP:dsRed minimal promoter cassettes⁴⁵ from plasmid constructs with a 10-bp barcode in the reverse primer (four fixed bases, six random bases; primers LP82–LP85). Barcoded promoter–reporter fragments were then cloned 3′ of CRSs in the pGL transfer vector with restriction-ligation cloning. 25 colonies per construct were pooled to create the second library, so that each CRS–promoter combination was tagged by 25 cBCs. The library was sequenced on the Illumina MiSeq platform (2 × 150 bp, 300× coverage per cBC) to associate every barcode to a unique combination of CRS and promoter.

Cre recombinase-mediated genomic integration and cell sorting.

Libraries of barcoded reporter genes were integrated into pools of landing pad cell lines using Cre recombinase, which enabled reporter gene libraries to recombine with multiple landing pad locations in the K562 genome. We electroporated 4 μg of transfer vector library with 1 μg of plasmid encoding Cre recombinase (pBS185 CMV-Cre, Addgene 11916) into 1.2 million K562 cells representing eight unique landing pad lines using the Neon Transfection System (Life Technologies). We performed four replicate electroporations and pooled two of them into each of two replicates to assess the biological reproducibility of expression measurements. For this experiment, we estimated we needed to transfect 0.9 million cells to observe an average of 10 barcodes for each of 300 regulatory elements at each of 8 landing pads. Cells were cultured for 1 week after electroporation to allow eGFP signal from the landing pad cassette to degrade before sorting. We used FACS to remove cells that did not undergo Cre-lox recombination by setting a threshold on the lower tail of eGFP expression in landing pad cells that were not exposed to reporter gene libraries or Cre recombinase, and excluding cells with higher eGFP signal than this threshold. 28% of landing pad cells had eGFP levels below the empirically established threshold before Cre-lox recombination. After recombination, 40% of cells had eGFP levels below the threshold. We estimate 12% efficiency of Cre-lox recombination for the pool of landing pads used in this study. Thus, FACS provided a 2.5-fold enrichment of the landing pad cells that underwent recombination (30% of sorted cells vs. 12% of total cells).

For the second library, we electroporated transfer vector library and Cre-recombinase into 1.2 million K562 cells representing 14 unique landing pad lines. We performed six replicate electroporations and pooled two of them into each of three replicates.

Expression measurements for integrated cis-regulatory element libraries.

After sorting, cells were grown for 5 d to increase the number of cells containing regulatory elements at landing pads. RNA and gDNA were harvested from 17.5 million cells for each replicate following manufacturer’s protocols (ToTALLY RNA kit, Ambion). RNA was treated with two rounds of DNase following the “Rigorous DNase treatment” procedure in the manufacturer’s protocol (Turbo DNase, Ambion). cDNA was synthesized in four replicates using oligo(dT) primers (SuperScript III First-Strand Synthesis System, Invitrogen). CRS barcodes (cBC) and landing pad barcodes (gBC) were then amplified from cDNA (2 replicates per sample) and gDNA (10 replicates with 700 ng, 40 replicates with 75 ng and 40 replicates with 500 ng) using primers LP32 and LP33 (98 °C for 1 min, 20 cycles: 98 °C for 30 s, 57 °C for 30 s, 72 °C for 30 s, and 72 °C for 5 min; NEB HF Phusion MM). All replicates were pooled and digested with EcoRI and SphI. We then ligated Illumina adapters containing multiplexing barcodes. Ligated fragments were further amplified using LP31 and LP30 (98 °C for 1 min, 20 cycles: 98 °C for 30 s, 57 °C for 30 s, 72 °C for 30 s, and 72 °C for 5 min; NEB HF Phusion MM) and gel purified. Barcodes were sequenced on the Illumina NextSeq platform (1 × 150 bp). Reads that contained both a cBC and gBC in the proper sequence context were included in subsequent analysis. Sequencing resulted in an average of 40.04 million reads for two replicates of DNA and an average of 12.25 million reads for two replicates of RNA. The expression of each barcode at each landing pad is calculated as log₂(RNA reads/DNA reads). We averaged all the cBC measurements for each regulatory element at a given landing pad for both replicates and used these averages to compare expression across landing pads.

Modeling of MPRA data.

We fit log₂-transformed expression with an additive linear model of CRS and landing pad identities using the lm function in R. Coefficients were fit in reference to CRS₁ and LP₁.

Reporting Summary.

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability.

The raw barcode counts for each library are available in Supplementary Tables 3 and 5. Illumina sequencing data are deposited in the NCBI Short Read Archive (SRA) under accession code PRJNA394868.