Promoter-mediated diversification of transcriptionalbursting dynamics following gene duplicationEdward Tunnacliffea,b, Adam M. Corrigana,b,1, and Jonathan R. Chubba,b,2

aMedical Research Council Laboratory for Molecular Cell Biology, University College London, WC1E 6BT London, United Kingdom; and bDepartment of Celland Developmental Biology, University College London, WC1E 6BT London, United Kingdom

Edited by Joseph S. Takahashi, Howard Hughes Medical Institute and University of Texas Southwestern Medical Center, Dallas, TX, and approved July 6, 2018(received for review January 21, 2018)

During the evolution of gene families, functional diversification ofproteins often follows gene duplication. However, many genefamilies expand while preserving protein sequence. Why do cellsmaintain multiple copies of the same gene? Here we haveaddressed this question for an actin family with 17 genes encodingan identical protein. The genes have divergent flanking regionsand are scattered throughout the genome. Surprisingly, almostthe entire family showed similar developmental expression pro-files, with their expression also strongly coupled in single cells. Usinglive cell imaging, we show that differences in gene expression wereapparent over shorter timescales, with family members displayingdifferent transcriptional bursting dynamics. Strong “bursty” behav-iors contrasted steady, more continuous activity, indicating differentregulatory inputs to individual actin genes. To determine the sourcesof these different dynamic behaviors, we reciprocally exchanged theupstream regulatory regions of gene family members. This revealedthat dynamic transcriptional behavior is directly instructed by up-stream sequence, rather than features specific to genomic context.A residual minor contribution of genomic context modulates thegene OFF rate. Our data suggest promoter diversification followinggene duplication could expand the range of stimuli that regulate theexpression of essential genes. These observations contextualize thesignificance of transcriptional bursting.

transcriptional bursting | stochastic gene expression | single-celltranscriptomics | Dictyostelium | gene family

Gene duplication is recognized as an important process forgenerating complexity in evolution (1). Following duplica-

tion, gene sequences are present in at least two copies in thegenome. Assuming these sequences are identical and subject tothe same regulatory constraints, they will perform the samefunction—they are redundant. Over time, duplicate genes typi-cally diverge in sequence and function; however, in some cases,strong selection acts to maintain identical amino acid or nucle-otide sequences over long periods of evolution. Examples in-clude histones, where humans have 14 genes for histone H4,each encoding the same protein (2). Similarly, ribosomal RNAgenes are present in hundreds to thousands of copies in eu-karyotes with extremely high sequence conservation betweenfamily members (3).Why does an organism require so many genes encoding an

apparently identical end product? One explanation is that a largeamount of gene product is required, and multiple genes allowmore transcription. However, while histone genes can be undercoordinate control during the cell cycle (4), they have differentpromoter elements and show varying contributions to total his-tone content in normal and cancer cells, suggesting regulatorydifferences between family members.To understand how differences in gene regulation have

influenced the evolution of multigene families, we investigatedthe actin gene family of the amoeba Dictyostelium discoideum.This organism has more than 30 actin genes, of which 17 (theact8 group) encode an identical amino acid sequence (5). Act8family genes produce more than 95% of total actin (6) and aredispersed throughout the genome (5). This actin family organization

is a broadly applicable evolutionary strategy, shared by species thatdiverged more than 400 Mya (SI Appendix, Table S1) (7).Dictyostelium cells are highly motile, so may require lots of

actin, perhaps beyond the production capacity of a single gene.However, estimates of their actin content are of the same orderas skeletal muscle, which derives its actin from only one gene(8, 9). Divergent flanking sequences (10, 11) and different ge-nomic contexts of the act8 genes suggest different regulatorydynamics and responses—for example, during development.The expansion of the family may also buffer against gene ex-pression noise—it may be undesirable for the expression of anessential protein to be unpredictable, and additional genes mayaverage out noise.Here, we evaluate the potential for different regulatory dy-

namics within the gene family. The family shows comparativelysimilar expression profiles over development and strong couplingbetween genes in single cells. However, the genes differ in thedynamics of their transcriptional bursts and show differentbursting responses upon induction of development. Switchingpromoters of actin genes demonstrates that transcriptional dy-namics are instructed predominantly by upstream sequence,rather than genomic context.

ResultsDevelopmental Dynamics of Actin Gene Expression. Having multiplegenes encoding the same protein dispersed throughout the genomemay have enabled diversification and refinement of actin expres-sion. Consistent with this view, there is considerable diversity of

Significance

Gene transcription occurs in discontinuous bursts. Althoughbursts are conserved in all forms of life, the causes and impli-cations of bursting are not clear. Here we delineate a specificcause of bursts and contextualize the significance of bursting,using analysis of a gene family encoding 17 identical actin pro-teins. Although the genes show similar developmental ex-pression, which is coupled in single cells, they show strongdifferences in bursting dynamics. These distinct bursting pat-terns indicate that different signals regulate the individual genesand suggest expansion of the family may have allowed di-versification of actin gene regulation. By exchanging the pro-moters of genes, we show that the dominant driver of burstingdynamics is the gene promoter, not the genome context.

Author contributions: E.T. and J.R.C. designed research; E.T. performed research; E.T. andA.M.C. contributed new reagents/analytic tools; E.T. and J.R.C. analyzed data; and E.T.and J.R.C. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Published under the PNAS license.1Present address: AstraZeneca, Discovery Sciences, Cambridge Science Park, CB4 0WGCambridge, United Kingdom.

2To whom correspondence should be addressed. Email: j.chubb@ucl.ac.uk.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1800943115/-/DCSupplemental.

Published online July 30, 2018.

8364–8369 | PNAS | August 14, 2018 | vol. 115 | no. 33 www.pnas.org/cgi/doi/10.1073/pnas.1800943115

Dow

nloa

ded

by g

uest

on

Oct

ober

3, 2

020

upstream regulatory sequence between act8 family members,with TATA and 3′ UAS motifs conserved across most of thefamily, while other motifs such as a G-box are found in only somepromoters (SI Appendix, Fig. S1A). While some promoterscontain several elements, others, such as act8, show little com-plexity, with large runs of A and T. Conserved elements werealso found at the 3′ end of act genes, which could enable furtherregulatory diversification, although earlier studies showed nostrong differences in RNA turnover within the family (12).To test whether actin genes are differentially regulated, we

used RNA-sequencing data to determine developmental profilesof gene expression (13). Fig. 1A shows the developmental ex-pression patterns of all 17 act8 genes. All genes are induced upondifferentiation onset with a peak between 1 and 3 h. Most genesshow decreased expression during middevelopment (6–10 h)and, by 16 h, show little expression. Promoter differences mayexplain the subtle variations in developmental expression—geneswith similar promoters, such as act9, act13, and act14, show moresimilar expression during development. An exception to thegeneral pattern is act1, which has additional peaks in expressionlater in development, although the gene shows comparatively fewread-counts. Despite minor differences, the remaining 16 genesdisplay broadly similar expression, suggesting the generation ofdifferent expression profiles during development is unlikely tohave been a major influence in the expansion of the family.

Single-Cell Coupling of Actin Gene Expression. To obtain single-cellresolution and more accurate quantitation of actin gene ex-pression, we measured the relative abundance of act8 familytranscripts using single-molecule RNA FISH (smFISH). Tosample genes with contrasting promoter architecture, we choseact1, act5, act6, and act8. Twenty-four MS2 stem loops (14) weretargeted at the 5′ of coding sequences, causing the MS2 loops tobe included in the transcribed RNA. Cells with MS2-taggedgenes were probed with a fluorescent oligonucleotide comple-mentary to the MS2 array. Cytoplasmic particles correspondingto single RNAs were counted (Fig. 1B). The act5, act6, and act8genes were all strongly expressed, with act8 the strongest andact6 the weakest, but each gene showed tens to hundreds ofRNAs per cell. In contrast, act1 expression was close to back-ground (cells without MS2), indicating that in undifferentiatedcells, actin expression at full capacity is not required. In line withthis, we found that disruption of up to four act8 family genes hadno effect on cell-doubling times, and a six-gene mutant showedonly a weak growth defect (SI Appendix, Fig. S1 B and C).Each act gene showed considerable expression variability (Fig.

1B). Although variability is useful in cellular decision-making(15), it can also be disruptive, effectively making some cellsoverexpress the gene, while others are deficient. Noisy gene ex-pression may create variability for one gene, but uncorrelatedfluctuations in the 16 other genes could dilute this noise,allowing an optimal actin level for each cell. To test this rea-soning, we used single-cell RNA-sequencing (scRNAseq) data(16) to compare the expression of act genes in individual cells.Single-cell read-counts for all act8 genes were clustered asheatmaps showing correlation values between pairs of genes(Fig. 1C). Some genes showed only weak correlations, consistentwith the possibility of multiple family members diluting outstochastic variation. However, most genes were strongly corre-lated in their expression in undifferentiated single cells andacross multiple developmental time points (SI Appendix, Fig.S1D). These data imply most act8 family genes are coordinatelyregulated in single cells, with cell–cell variability (high versus lowexpression) a decision taken over most of the family. This sug-gests the expansion of the family was not selected primarily forbuffering molecular noise in gene expression.We infer no strong differences in act gene posttranscriptional

regulation. Protein expression from act genes (monitored usingmNeonGreen knock-ins) reflected the differences in transcriptlevel seen by smFISH, with no differences observed in the vari-ability of protein expression or localization to actin structures (SIAppendix, Fig. S1E).

Actin Genes Show Different Transcriptional Bursting Patterns. Withonly small differences in act gene developmental expression,despite divergent control sequences, we tested whether differ-ential regulation is apparent over smaller timescales, using MS2-tagged act genes. Upon transcription, nascent MS2-tagged RNAcan be detected with a fluorescent MS2 coat protein (MCP-GFP) as a fluorescent spot at the transcription site. In live cells,the spot intensity varies over time (Fig. 2A), reflecting the fluc-tuating transcriptional activity of the gene (12).Different actin genes exhibit very different transcriptional

behaviors in undifferentiated cells (Fig. 2A). No act1 spots weredetected, indicating the gene is inactive or active below the de-tection threshold [5 RNAs (17)]. Both act5 and act6 showedbursts of activity followed by periods of inactivity, with spot in-tensity greater for act5 than act6. In contrast, act8 activity fluc-tuations were small compared with the other genes. Theseexample images were representative of hundreds of cells (Fig.2B). Act8 was strongly ON in most cells, most of the time,whereas act6 was mostly OFF, with the occasional moderateintensity burst. Act5 was between these extremes, retaining fre-quent switching between ON and OFF. These dynamics aresummarized in spot-intensity distributions (Fig. 2C). The burst-ing of act5 and act6 is apparent in the thin upper tails of thedistributions and a large proportion of cells in the OFF state. In

A

B C

Fig. 1. Developmental regulation and single-cell coupling of actin geneexpression. (A) Uniquely mapped RNAseq read-counts [reads per kilobaseper million mapped reads (RPKM)] for all 17 act8 genes over development.(B) Transcript counts of act8 family genes in undifferentiated cells measuredby smFISH. Each dot is a cell. Mean shown as cross in box plot. (C) Highcorrelations between act8 genes in single cells. Panel shows heatmap of allpairwise comparisons of act8 family expression levels in undifferentiatedcells from scRNAseq data.

Tunnacliffe et al. PNAS | August 14, 2018 | vol. 115 | no. 33 | 8365

CELL

BIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

3, 2

020

contrast, act8 was mostly ON, with an approximate normal dis-tribution of spot intensities in the active state. Act5 and act6transcription showed significantly higher variance than act8 (SIAppendix, Fig. S2A).To assess the type of control mechanisms regulating act genes,

we determined how different measures of variability change withincreasing gene activity. Increasing mean expression while de-creasing noise (Cv

2, ðσ=μÞ2) indicates burst frequency control ofgene expression (18, 19). In contrast, increasing noise strength(Fano factor, σ2=μ) with mean expression indicates burst sizecontrol. All three genes showed a strong negative correlationbetween spot intensity and noise (Spearman’s rank correlation,act5: r = −0.92, P = 0, act6: r = −0.79, P = 5 × 10−4, act8: r =−0.83, P = 1 × 10−4) (Fig. 3A). In contrast, only act6 intensitycorrelated with noise strength (act5: r = −0.13, P = 0.62; act6: r =0.83, P = 1 × 10−4; act8: r = −0.36, P = 0.17), suggesting act6modulates both burst frequency and size to change gene ex-pression, while act5 and act8 modulate frequency alone.Noise decreases with increasing gene expression (20). The

difference in noise between act5 and act8 is unlikely to be causedby differences in expression level, as for a given spot intensity,act5 variability was higher than act8 (Fig. 3A). The noise–meanrelationship of act5 and act6 can be explained by the same ex-ponential function, as the data will lie on the same linear re-gression line (Fig. 3A), indicating the difference in variancebetween act5 and act6 arises because of reduced noise at higherexpression.

We directly measured bursting dynamics from live cell data byimposing a threshold separating ON and OFF phases, using act1traces to estimate the measurement noise (SI Appendix, Fig. S2 Band C). ON and OFF durations are represented as cumulativefrequency plots (Fig. 3B), with act8 showing short OFF and longON phases, act6 the opposite, with act5 intermediate.

Signal Regulation of Bursting. What generates the different actgene bursting patterns? One view is that the likelihood of a burstrelates to the concentration or activity of a transcription factor(TF), with the TF responding to signaling. Different burstingpatterns would result from different genes responding to dif-ferent signals. Alternatively, all genes respond to the same sig-nals, with different TF binding properties generating differentburst dynamics. The complete lack of homology between the act8and act5 promoters argues against this second view. However, if

00:00 07:30 15:00 22:30 30:00 37:30

00:00 07:30 15:00 22:30 30:00 37:30

00:00 07:30 15:00 22:30 30:00 37:30

act5

act6

act8

A

B C

Time (min)

Spotintensity

(a.u.) Spotintensity(a.u.)

Cellno.

15 30 4515

20

40

60

80

100

1

2

3

4

5

6

7

30 45 15 30 45

0

-10

-1

1

2

3

4

5

6

act1act5act6act8

act5 act6 act8

Fig. 2. Distinct bursting patterns of actin genes. (A) Live imaging of tran-scription dynamics of three different actin genes. Intensity of transcriptionspots fluctuates over time to varying extents for each gene (time in min:sec).(B) Transcription spot-intensity traces for populations of cells. Each panelrepresents an individual experiment for each actin gene. Each row repre-sents the intensity trace for a single cell. Black indicates the cell was outsidethe field-of-view. Each block in a panel is a field-of-view. Imaging was car-ried out over three experimental days for each gene, with multiple fields-of-view per experiment: act1 (152 cells), act5 (476), act6 (384), and act8 (275).(C) Probability density functions of spot-intensity distributions for all ex-periments. Horizontal lines represent quartiles and median.

A

B

C

Fig. 3. Differential regulation of actin transcription bursts. (A) Correlationsbetween spot intensity and different measures of variability (noise/Cv

2 andnoise strength/Fano). Each data point represents a field-of-view (FOV), withshades of each color showing data from different experimental days. (B)Burst ON and OFF durations, with the threshold imposed at spot intensity of5,000 AU. Data are shown as cumulative frequency plots. (C) Different re-sponses of act genes to induction of development (starvation). Each line isthe mean spot intensity per gene averaged across four independent data-sets: act5 (415 cells) act6 (467), and act8 (375). Shaded areas show SD.

8366 | www.pnas.org/cgi/doi/10.1073/pnas.1800943115 Tunnacliffe et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

3, 2

020

the first model is valid, there should be stimuli with different effectson different act genes.Cell size and cell speed both correlate with act5, act6, and act8

transcription, suggesting larger, faster cells are more likely toexpress actin (SI Appendix, Fig. S3 A and B). However, thesedata do not distinguish between the genes. For undifferentiatedcells, the major requirements for actin are in cytokinesis,phagocytosis, and macropinocytosis. To identify signals thatmight differentially regulate actin bursts, we screened a panel ofdifferent culture environments to test how these and other cellularprocesses relate to actin protein expression (SI Appendix, TableS2). Although subtle effects were observed, the relative ordering ofexpression level for act5, act6, and act8 did not change, so thesedifferent culture cues are unlikely to explain strong differencesin bursting.Differentiation in Dictyostelium is induced by starvation, fol-

lowed by extracellular cAMP and other signals. The increase inrelative mRNA for all act8 genes during early differentiation(Fig. 1A) is not suggestive of differences in the regulation of acttranscription by these signals. However, global mRNA synthesisrates during differentiation are only ∼15% of rates for undifferen-tiated cells (21), so the relative actin increase might reflect continuingact transcription while much of the genome falls silent, not a suddentranscription surge. To directly visualize transcription during the actinmRNA spike, we imaged act-MS2 cells during early starvation.Starvation triggered different responses for different act genes.

At the onset of starvation, relative levels of transcription weresimilar to those of undifferentiated cells (Fig. 3C). However, act5showed a strong increase in the average transcription site intensity(Fig. 3C and SI Appendix, Fig. S4 A and B). In contrast, act8showed a decline in output, with act6 transcription remaininglow. The stronger output of act5 occurred with more cells showinglonger, more frequent bursts (SI Appendix, Fig. S4C). These dataindicate different act genes respond differently to specific stimuli.

Evaluating the Contributions of Promoter and Genomic Context toBursting. What are the nuclear determinants of bursting dynamics?Many factors acting over a wide range of length scales can controltranscription (22–25). Cis-acting elements, chromatin structure andmodification, genomic context, and nuclear organization are allpotential inputs. However, the ordering and magnitude of theseinputs to transcription are unclear, due to extensive crosstalk be-tween the multiple levels of regulation.

Given the diversity in promoter architecture and genome po-sition across the act8 family, we specifically tested the role ofpromoter sequences and genome context in regulating burstingdynamics. We exchanged the promoters of act6 and act8, thegenes with the most distinct bursting behaviors, by replacing thepromoter of one gene with that of the other at the endogenouslocus. The sequences exchanged were the proximal 388 bp (act8)and 543 bp (act6) of promoters. Unlike earlier promoter re-placement studies (26), we used live cells to directly monitortranscription dynamics. The act6 and act8 genes are exactly thesame length, are on different chromosomes, have no introns, andcan be deleted without phenotype. This reciprocal switch en-abled us to determine the effects of promoter sequence on geneactivity, independently of native genomic context.Analysis of the transcription dynamics of these switched cell

lines identified the promoter as the dominant factor in the reg-ulation of transcription. Fig. 4A shows that the “A8P-A6G” cellline, with the act8 promoter upstream of the act6 gene, had re-markably similar transcription dynamics to endogenous act8—steady gene activity with low cell-to-cell variability. Similarly, the“A6P-A8G” gene, with the act6 promoter upstream of act8, wasmostly OFF with infrequent short bursts of activity, resemblingnormal act6. Spot-intensity distributions were similar betweenendogenous and promoter-switched genes with the same pro-moter (SI Appendix, Fig. S5A). There may be some residual controlfrom the genomic context, apparent when comparing act8 and A8P-A6G, where the spot-intensity distribution of A8P-A6G is moreskewed toward zero than the act8 distribution and there is a small,but significant, difference in variance for these genes (SI Appendix,Fig. S5B). However, at a coarse level, the promoter is more instructivefor nascent transcript levels.To determine the drivers of transcription dynamics, we used

cooccurrence matrices to assess changes in gene activity overshort time lags, with each point representing the spot intensity attime t and t + l, where l is a lag (two frames). Matrices for act6and act8 both show thin bands parallel to both axes represen-tative of significant short-term changes in activity (Fig. 4B). Datapoints closer to the diagonal represent more slowly fluctuatinggene activity, and most act8 data are here. Plots for both genesshow a clear separation between data where the gene slowlyfluctuates and switches rapidly. For the promoter-switchedgenes, the overall shape of the distributions indicates the pro-moter underpins most of the structure of the data; however, for

15

20

40

80

60

100

-1

0

1

2

3

4

5

7

6

30 45 15 30 45

Spotintensity

(a.u.)

Cellno.

A B

100

80

60

40

20

10080604020 1008060402000

St+2(%)

100

80

60

40

20

St+2(%)

0

10

20

30

40

50+

Counts

15 30 45 15 30 45

104act6 act8 A8P-A6G A6P-A8G

Fig. 4. Promoter regulation of actin transcription bursts. (A) Transcription dynamics of endogenous and promoter-switched genes. Each panel shows oneexperiment for act6, act8, and the promoter-switched genes A6P-A8G and A8P-A6G. Each row represents the spot-intensity trace for a single cell. Promoter-switched cell lines were imaged over three experimental days: A6P-A8G (714 cells) and A8P-A6G (760). (B) Cooccurrence matrices for endogenous andpromoter-switched cell lines representing transitions between imaging frames. The xy coordinates determined by spot intensity at time t (St) and at time t + 2(St+2), expressed as a percentage of the maximum range of spot intensities across all experiments.

Tunnacliffe et al. PNAS | August 14, 2018 | vol. 115 | no. 33 | 8367

CELL

BIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

3, 2

020

A8P-A6G, the separation between slow and fast switching statesis less clear, indicating A8P-A6G and act8 dynamics are notidentical.To further evaluate the contributions of the promoter and

genomic locus to bursting, we used an unbiased method ofclassification based on dynamic features of time-series data. Themethod, “bag-of-patterns” (27), simplifies time-series into col-lections of discrete symbolic “words” representing the localstructure of the data. The “letters” within words represent dif-ferent bins of signal intensity, and the sequences of letters withinwords reflect the spot behavior as the intensity fluctuates betweenbins. Individual time-series are defined by the relative word usage,tallied as histograms (SI Appendix, Fig. S6C). Comparison of time-series involves calculation of Euclidean distances between wordhistograms, with final representation as a dendrogram. Three pa-rameters, word length (w), number of bins (α), and duration oftime-series subsequence (n), were optimized. We used act5, act6,and act8 data as training sets to identify parameter combinationscapable of clustering the data according to gene identity (Fig. 5Aand SI Appendix, Fig. S6D).To test the contributions of promoter and genomic context to

bursting, we used the optimized parameters to classify act6, act8,A8P-A6G, and A6P-A8G according to their transcription dy-namics. Genes sharing a promoter rather than a genomic locusdisplay more similar dynamic behaviors. Act6 and A6P-A8Gclustered more closely together compared with act8 and A8P-A6G(Fig. 5B and SI Appendix, Fig. S6E). Two A8P-A6G datasets couldnot be clustered; however, all four that were clustered fit within theact8 branch. Overall, the majority of the data clustered according topromoter identity, indicating the promoter is the dominant driver ofbursting dynamics.Comparisons between endogenous and promoter-switched

genes showed that OFF periods of transcription were not dif-ferent between pairs of genes sharing a promoter (KS test: act6vs. A6P-A8G, P = 0.18; act8 vs. A8P-A6G, P = 0.16) (Fig. 5C andSI Appendix, Fig. S7). However, A6P-A8G and A8P-A6G bothspend significantly less time in the ON state than act6 and act8,respectively (KS test: act6 vs. A6P-A8G, P = 0.01; act8 vs. A8P-A6G, P = 3.9 × 10−5). This suggests that while promoter

sequence controls the majority of the dynamic behavior, featuresspecific to the genomic locus may influence burst duration.

DiscussionTo gain insight into potential processes driving expansion of anactin gene family in the absence of protein sequence di-versification, we evaluated the contribution of differential geneexpression. Our data reveal the expansion of the Dictyosteliumact8 gene family is unlikely to have occurred to allow differenttemporal patterns of actin expression during development. Nei-ther is expansion likely to have had a large contribution from aneed to buffer noisy gene expression. Instead, we found thegenes differ greatly in the dynamics of their transcriptionalbursting. We interpret this effect as meaning that the expansionmay have occurred, at least in part, to enable actin expression tobe regulated by an expanded set of signaling cues. In support ofthis hypothesis, we show that act8 family members show differentbursting responses to starvation, a trigger for the onset of dif-ferentiation. In this view, genes such as act8 would act as the“workhorses” of the family, delivering much of the cellular actin,with genes such as act5 showing thermostat-like control, to topup levels as appropriate. Expansion may represent a solution todiversifying regulation of a gene with a high transcript load, in agenome too compact for extensive upstream control by multipleenhancers. We then tested whether bursting is driven by thespecific promoter sequence of each gene, or via other features ofgenomic context, by switching promoters of actin genes on dif-ferent chromosomes. The dominant contribution to burstingdynamics came from the promoter.Residual effects on bursting dynamics that might be attributed

to genomic context were detected in an increased OFF rate ob-served when a promoter operates at a nonnative site. This couldconceivably be due to destabilization of local chromatin confor-mation by insertion of a nonlocal DNA sequence. For example,looping between the 5′ and 3′ ends of a gene can facilitate tran-scription reinitiation, which may contribute to the repetitive tran-scription events of a burst (28). Incompatibility between 5′ and 3′ends could inhibit looping and increase the OFF rate. This in-compatibility scenario assumes specificity provided by the 5′ end ofthe gene, returning the emphasis on control back to the promoter.

A

C

Bact5act6act8

0.5

1

1.5

2

2.5 act6act8A8P-A6GA6P-A8G

0.40.50.60.70.80.91

1.1

K-S testp = 0.18p = 0.16

act6act8A8P-A6GA6P-A8G

act6act8A8P-A6GA6P-A8G

K-S testp = 0.01p = 3.9 x 10-5

0.4

0.2

0 15 300

0.6

0.8

1

Cumulativefrequency

On duration (min)

0.4

0.2

0 15 300

0.6

0.8

1

Cumulativefrequency

Off duration (min)

Fig. 5. Actin transcriptional bursting is predominantlypromoter-driven. (A) Classification of the training setdata of act5, act6, and act8. Dendrogram showingclustering by Euclidean distance using “bag-of-pat-terns.” (B) Transcription dynamics of endogenousand promoter-switched genes were clustered usingthe same parameters as in A. (C) Cumulative distri-butions of burst ON and OFF durations for endoge-nous and promoter-switched genes.

8368 | www.pnas.org/cgi/doi/10.1073/pnas.1800943115 Tunnacliffe et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

3, 2

020

We therefore propose that the increased OFF rate of the switchedgenes represents an upper limit on the contribution to burstingdynamics from the genomic context.Our data do not imply chromatin structure and nuclear or-

ganization are unimportant for bursting. We propose that theeffects of chromatin and nuclear structure on a gene are firstinstructed by DNA sequence. This view is consistent with manystudies showing that chromatin modifications and nuclear orga-nization are imposed by transcription itself (29, 30). We em-phasize that these dominant effects of the promoter have beenobserved for two genes in an apparently simple developmentaleukaryote. This view may have to be modified when consideringa metazoan genome with long-range transcriptional control.Dictyostelium has a small genome size with short intergenic re-gions—over 60% of the genome encodes protein. However, theorganism has DNA and H3K9 methylation, mediator, a nuclearlamin, and late-replicating peripheral heterochromatin—stan-dard features of metazoan chromatin modification, topology,and organization. Metazoan genes for which genomic context isperhaps most strongly implicated are the globin and Hox clusters(25, 31). Here the genomic context is presumably maintainedunder strong selective pressure to keep the family together forcoordinate control. The idea that the act8 family has undergonedispersal to sample the diversity of different genomic contexts isnot strongly supported by our data.

MethodsMolecular Biology and Cell Line Generation. For live cell imaging of tran-scription, we targeted an MS2 cassette to the act1, act6, and act8 genes,respectively. Targeting vectors were designed so that the MS2 sequence wasat a similar position in the coding sequence of all genes, 18–24 bp down-stream of the ATG. Act5-MS2 cells have been described previously (17). Toswitch promoters, the same promoter fragments used as targeting arms inthe act6 and act8-MS2 vectors were cloned next to the MS2 repeats in thetargeting vector for the other gene. To ensure targeting to the correct locus,we cloned regions upstream of the promoters as the 5′ homology arms oftargeting vectors. In MS2 cell lines, the selectable marker was removed by

transient CRE expression to allow MS2 transcription to use endogenousterminators. For visualizing nascent RNA, MS2-tagged cells were transfectedwith an extrachromosomal vector expressing the MCP-GFP fusion protein(17). To monitor protein levels we targeted codon-optimized mNeonGreen(32) to the 3′ end of the endogenous genes in AX3 cells, followed by removalof the selectable marker. For transcription imaging, we used DictyosteliumAX3 cells with a stably expressed red fluorescent nuclear marker, H2Bv3-mCherry (17).

Imaging and Data Analysis. Cells were imaged in a low fluorescent medium in8-well chambers (Nunc Lab-Tek II) and imaged on anUltraVIEWVoX spinning-disk confocal microscope (Perkin-Elmer) with an EM-CCD camera (C9100-13;Hamamatsu) using a 60× 1.4 NA objective. For imaging during development,cells were washed free of media and viewed under nonnutrient agar (12).Cell-tracking, spot identification, and motility/size analysis used custom-builtMATLAB software (17). For smFISH, we used the protocol of ref. 33. We useda single-probe, end-labeled with Quasar 670, which binds the spacer be-tween each MS2 stem loop (17). Cells were imaged using the UltraVIEW witha 100× objective and 640-nm laser. Cytoplasmic mRNA counts were de-termined using FISH-quant (34).

To implement bag-of-patterns for our transcription dynamics data weconcatenated individual cell tracks to generate time-series of around 1,000–5,000 frames for each cell line per imaging session. Concatenated trackswere log-transformed and normalized to give mean = 0 and SD = 1 (SIAppendix, Fig. S6 A and B), for comparison of time-series with differentoffsets and amplitudes. A MATLAB package was used to derive word histo-grams (https://cs.gmu.edu/∼jessica/sax.htm) before collating the bag-of-patternsfor individual time-series (27). Parameter sets were defined empirically usingact5, act6, and act8 data as a training set (SI Appendix, Fig. S6C) before applyingthese to promoter-switched datasets.

ACKNOWLEDGMENTS. We thank Rafael Rosengarten and Gad Shaulsky foraccess to data prior to publication. Work was supported by Wellcome TrustSenior Fellowship 202867/Z/16/Z (to J.R.C.) and Medical Research Council(MRC) funding (MC_U12266B) to the MRC LMCB University Unit at UCL.Imaging was carried out at the MRC LMCB Light Microscopy Facility. E.T. wassupported by a London Interdisciplinary Doctoral Training ProgrammeBBSRC studentship.

1. Zhang J (2003) Evolution by gene duplication: An update. Trends Ecol Evol 18:292–298.

2. Marzluff WF, Gongidi P, Woods KR, Jin J, Maltais LJ (2002) The human and mousereplication-dependent histone genes. Genomics 80:487–498.

3. Eickbush TH, Eickbush DG (2007) Finely orchestrated movements: Evolution of theribosomal RNA genes. Genetics 175:477–485.

4. Holmes WF, et al. (2005) Coordinate control and selective expression of the fullcomplement of replication-dependent histone H4 genes in normal and cancer cells.J Biol Chem 280:37400–37407.

5. Joseph JM, et al. (2008) The actinome of Dictyostelium discoideum in comparison toactins and actin-related proteins from other organisms. PLoS One 3:e2654.

6. Vandekerckhove J, Weber K (1980) Vegetative Dictyostelium cells containing 17 actingenes express a single major actin. Nature 284:475–477.

7. Parikh A, et al. (2010) Conserved developmental transcriptomes in evolutionarily di-vergent species. Genome Biol 11:R35.

8. Yates LD, Greaser ML (1983) Quantitative determination of myosin and actin in rabbitskeletal muscle. J Mol Biol 168:123–141.

9. Uyemura DG, Brown SS, Spudich JA (1978) Biochemical and structural characterizationof actin from Dictyostelium discoideum. J Biol Chem 253:9088–9096.

10. Romans P, Firtel RA (1985) Organization of the Dictyostelium discoideum actin multigenefamily. Flanking sequences show subfamily homologies and unusual dyad symmetries.J Mol Biol 183:311–326.

11. Hori R, Firtel RA (1994) Identification and characterization of multiple A/T-rich cis-acting elements that control expression from Dictyostelium actin promoters: TheDictyostelium actin upstream activating sequence confers growth phase expressionand has enhancer-like properties. Nucleic Acids Res 22:5099–5111.

12. Muramoto T, et al. (2012) Live imaging of nascent RNA dynamics reveals distinct typesof transcriptional pulse regulation. Proc Natl Acad Sci USA 109:7350–7355.

13. Rosengarten RD, et al. (2015) Leaps and lulls in the developmental transcriptome ofDictyostelium discoideum. BMC Genomics 16:294.

14. Bertrand E, et al. (1998) Localization of ASH1 mRNA particles in living yeast. Mol Cell2:437–445.

15. Symmons O, Raj A (2016) What’s luck got to do with it: Single cells, multiple fates, andbiological nondeterminism. Mol Cell 62:788–802.

16. Antolovi�c V, Miermont A, Corrigan AM, Chubb JR (2017) Generation of single-celltranscript variability by repression. Curr Biol 27:1811–1817.e3.

17. Corrigan AM, Tunnacliffe E, Cannon D, Chubb JR (2016) A continuum model oftranscriptional bursting. eLife 5:e13051.

18. Carey LB, van Dijk D, Sloot PMA, Kaandorp JA, Segal E (2013) Promoter sequencedetermines the relationship between expression level and noise. PLoS Biol 11:e1001528.

19. To T-L, Maheshri N (2010) Noise can induce bimodality in positive transcriptionalfeedback loops without bistability. Science 327:1142–1145.

20. Bar-Even A, et al. (2006) Noise in protein expression scales with natural proteinabundance. Nat Genet 38:636–643.

21. Mangiarotti G, Altruda F, Lodish HF (1981) Rates of synthesis and degradation of ri-bosomal ribonucleic acid during differentiation of Dictyostelium discoideum.Mol CellBiol 1:35–42.

22. Lenstra TL, Rodriguez J, Chen H, Larson DR (2016) Transcription dynamics in livingcells. Annu Rev Biophys 45:25–47.

23. Fukaya T, Lim B, Levine M (2016) Enhancer control of transcriptional bursting. Cell166:358–368.

24. Michalak P (2008) Coexpression, coregulation, and cofunctionality of neighboringgenes in eukaryotic genomes. Genomics 91:243–248.

25. Bartman CR, Hsu SC, Hsiung CC, Raj A, Blobel GA (2016) Enhancer regulation oftranscriptional bursting parameters revealed by forced chromatin looping. Mol Cell62:237–247.

26. Hocine S, Vera M, Zenklusen D, Singer RH (2015) Promoter-autonomous functioningin a controlled environment using single molecule FISH. Sci Rep 5:9934.

27. Lin J, Khade R, Li Y (2012) Rotation-invariant similarity in time series using bag-of-patterns representation. J Intell Inf Syst 39:287–315.

28. Hebenstreit D (2013) Are gene loops the cause of transcriptional noise? Trends Genet29:333–338.

29. Soares LM, et al. (2017) Determinants of histone H3K4 methylation patterns. Mol Cell68:773–785.e6.

30. Nozawa RS, et al. (2017) SAF-A regulates interphase chromosome structure througholigomerization with chromatin-associated RNAs. Cell 169:1214–1227.e18.

31. Rodríguez-Carballo E, et al. (2017) The HoxD cluster is a dynamic and resilient TADboundary controlling the segregation of antagonistic regulatory landscapes. GenesDev 31:2264–2281.

32. Shaner NC, et al. (2013) A bright monomeric green fluorescent protein derived fromBranchiostoma lanceolatum. Nat Methods 10:407–409.

33. Raj A, van den Bogaard P, Rifkin SA, van Oudenaarden A, Tyagi S (2008) Imagingindividual mRNA molecules using multiple singly labeled probes. Nat Methods 5:877–879.

34. Mueller F, et al. (2013) FISH-quant: Automatic counting of transcripts in 3D FISHimages. Nat Methods 10:277–278.

Tunnacliffe et al. PNAS | August 14, 2018 | vol. 115 | no. 33 | 8369

CELL

BIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

3, 2

020