Recurrent rewiring and emergence of RNAregulatory networksDaniel Wilinskia, Natascha Butera, Andrew D. Klockob, Christopher P. Lapointea, Eric U. Selkerb, Audrey P. Gaschc,and Marvin Wickensa,1

aDepartment of Biochemistry, University of Wisconsin–Madison, Madison, WI 53706; bInstitute of Molecular Biology, University of Oregon, Eugene, OR97403; and cLaboratory of Genetics, University of Wisconsin–Madison, Madison, WI 53706

Edited by Robert H. Singer, Albert Einstein College of Medicine, Bronx, NY, and approved February 14, 2017 (received for review October 26, 2016)

Alterations in regulatory networks contribute to evolutionary change.Transcriptional networks are reconfigured by changes in the bindingspecificity of transcription factors and their cognate sites. The evolutionof RNA–protein regulatory networks is far less understood. The PUF(Pumilio and FBF) family of RNA regulatory proteins controls the trans-lation, stability, and movements of hundreds of mRNAs in a singlespecies. We probe the evolution of PUF–RNA networks by direct iden-tification of themRNAs bound to PUF proteins in budding and filamen-tous fungi and by computational analyses of orthologous RNAs from62 fungal species. Our findings reveal that PUF proteins gain and losemRNAs with related and emergent biological functions during evolu-tion. We demonstrate at least two independent rewiring events forPUF3 orthologs, independent but convergent evolution of PUF4/5 bind-ing specificity and the rewiring of the PUF4/5 regulons in differentfungal lineages. These findings demonstrate plasticity in RNA regula-tory networks and suggest ways in which their rewiring occurs.

RNA regulation | PUF proteins | evolution | 3′UTR elements

Coordinated regulation of genes and mRNAs pervades bi-ology. Cells simplify the challenge by coregulating groups of

functionally related genes via the same regulatory proteins. Inthis way, cells establish “networks” consisting of a site-specificregulatory protein, the downstream targets it controls, and therecognition sequences in DNA or RNA that link each target toits regulator. The discrimination of specific sequences in DNAand RNA is a cornerstone of biological regulation in living systems,as these binding sites determine which genes will be controlled.Transcriptional networks display a remarkable capacity to evolve,

as evidenced by comparative genomic studies. Networks can changeat the individual gene level, in which specific transcriptional targetsare gained or lost via changes in transcription factor binding siteswithin genomic DNA (1–4). Over extended periods of evolutionarytime, the entire set of targets of a regulatory protein can change, untilthis regulator effectively controls a distinct set of genes. Alternatively,coregulated genes can fall under the control of different regulators. Inthis case, each individual target acquires DNA sites that are recog-nized by the alternate regulator. Ribosomal protein genes in fungi area striking example: Their coregulation has been maintained for over abillion years of fungal evolution despite regulation by completelydifferent systems in different species (5–7). This type of wholesalerewiring requires concerted evolution across the suite of targets. Al-though the mechanisms through which transcriptional networksevolve are unclear, it has been proposed to involve a period of “re-dundant” regulation, during which the same group of targets istemporarily controlled by two different regulatory systems, followedby loss of the ancestral connections (6, 8, 9). Rewiring often correlateswith duplication and divergence of the responsible regulator (10–13).The evolution of RNA regulatory networks is less well understood.

A single RNA-binding protein (RBP) often controls groups of RNAswith common biological functions (14–17) and so governs their sta-bility, translation, and localization. RBPs in the same protein familyexhibit similar though not identical RNA-binding specificities (18,19). Differences in family members generate divergence in theirregulatory networks. 3′UTRs, like introns, are particularly favorabletargets for network divergence, as they are relatively unconstrained in

length and sequence. Alternative polyadenylation events, in which asingle pre-mRNA species can give rise to two or more mRNAs thateither contain or lack regulatory sites, provide a mechanism of net-work divergence unique to mRNAs (20–22). Other molecularmechanisms for the evolution of binding elements and mRNA–pro-tein regulatory networks exist but are largely opaque.Here we focus on the evolution of RNA–protein networks. We

concentrate on the PUF (Pumilio and FBF) family of RBPs, becausethey are exemplary regulators of cytoplasmic mRNAs and provide astrong foundation based on genetics, biochemistry, and structuralbiology. PUF proteins generally bind single-stranded RNA elementsin the 3′UTR to trigger mRNA repression, activation, or decay andcan influence localization as well (23–27). PUF family members arepresent in all eukaryotes and maintain stem cells and promotememory formation in metazoa (28, 29). RNA targets for the ca-nonical Saccharomyces cerevisiae PUF proteins have been identifiedby immunoprecipitation methods, including RIP-chip, HITS-CLIP,and RNA tagging (14, 30–32). Each PUF protein binds a group offunctionally related mRNAs, forming distinct networks that are eachenriched for specific cellular processes. For example, of the476 tagged RNAs bound to S. cerevisiae Puf3, 191 encode proteinsthat contribute to mitochondrial organization (32).PUF proteins bind different RNAs due to variations in a con-

served protein scaffold. The proteins consist of eight largely helicalrepeats (PUF repeats). Three amino acids within each repeat canpotentially contact a single RNA base (23, 33). The sequence of the5′-end of most PUF binding elements is UGUR (where R is U orC), and the 3′-end is AU or UA. Differences in RNA bindingspecificity among PUF proteins are due to changes in the curva-ture of a conserved, largely helical PUF protein scaffold, whichimposes different limitations on the RNAs that can bind with high

Significance

Regulatory networks change during evolution. A protein thatcontrols many genes in one species may control a different set ofgenes in another. We examined how mRNA networks evolve,focusing on the PUF (Pumilio and FBF) family of RNA-bindingproteins. These govern stability and translation of hundreds ofmRNAs and enable coordinate regulation of discrete biologicaloutcomes. To understand how RNA networks evolve, we usedknowledge of the RNA specificity of each PUF protein to predictits mRNA targets and directly identified mRNAs bound to eachprotein in divergent fungi via biochemical methods. We findnetworks controlled by one protein switch during evolution tobe controlled by another and that proteins with different spec-ificities can share, gain, or lose batteries of mRNAs.

Author contributions: D.W. and M.W. designed research; D.W., N.B., and A.D.K. per-formed research; D.W., A.D.K., C.P.L., and E.U.S. contributed new reagents/analytic tools;D.W., N.B., A.D.K., C.P.L., E.U.S., A.P.G., and M.W. analyzed data; and D.W., A.D.K., E.U.S.,A.P.G., and M.W. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. Email: wickens@biochem.wisc.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617777114/-/DCSupplemental.

E2816–E2825 | PNAS | Published online March 20, 2017 www.pnas.org/cgi/doi/10.1073/pnas.1617777114

affinity (30). RNA nucleotides in the binding element contort toaccommodate the protein scaffold (30, 34, 35). For example, in S.cerevisiae, Puf3 has high specificity for an 8-nucleotide (nt) PUFelement, Puf4 instead recognizes a related 9-nt element, andPuf5 binds most tightly to 9-nt and 10-nt elements (14, 30, 34, 36).Each of their binding elements begins with UGUR and ends withUA. Their specificities diverge through the presence or absence ofinternal bases in the RNA binding element. The identities of thesebases generally are unimportant.Elegant computational studies of mRNAs across fungal species

support a model in which the Puf3 regulon, which controls nuclear-encoded mitochondrial transcripts in S. cerevisiae, was rewired duringevolution (37). Bioinformatics revealed that many of these mRNAspossessed 9-nt elements in filamentous fungi and 8-nt elements inS. cerevisiae (37, 38). However, PUF–RNA networks have not beenidentified experimentally in any fungal species besides S. cerevisiae.To probe the evolution of RNA regulatory networks at a molecular

level, we directly identified RNAs bound to multiple PUF proteins inbudding and filamentous fungi. Our studies revealed multiple in-dependent rewiring events in the Puf3, Puf4, and Puf5 networksduring fungal evolution.We show that the binding specificities of PUForthologs from the filamentous fungi Neurospora crassa and Asper-gillus nidulans are similar to one another as well as to their S. cerevisiaecounterparts. We find strong evidence for at least two independentrewiring events for Puf3 orthologs and demonstrate the independentbut convergent evolution of Puf4/5 binding specificities and therewiring of the Puf4/5 networks in different fungal lineages. Ourfindings demonstrate the plasticity of RNA regulatory networks, howthese networks emerge, and ways in which they have been rewired.

ResultsPattern of PUF Binding Elements Suggests Rewiring. We identifiedmatches to the known binding elements of S. cerevisiae PUF proteinsin the 3′UTRs of orthologous RNAs in 62 species of fungi to in-vestigate the evolution of binding sites across orthologous gene sets

(Fig. 1A; full Phylogenetic tree in Fig. S1). The phylum represents atleast 400 million years of evolution and includes species with strongfoundations in genomics and functional information (39). We focusedon species from two subphyla of Phylum Ascomycota: Saccha-romycotina (budding yeasts) and Pezizomycotina (filamentous fungi)(40) (Fig. 1A). We considered in our analysis only mRNAs with aclear ortholog in at least 50% of species—that is, mRNAs that en-code orthologus proteins—as defined previously (41, 42).We identified matches to PUF binding elements in the 3′UTRs of

orthologous mRNAs, comparing against sequence motifs that cor-respond to the major binding elements of the canonical S. cerevisiaePUF proteins (14). Because the S. cerevisiae PUF proteins binddifferent lengths of sequence elements, each bearing a UGU at oneend and a UA or AU at the other (see the Introduction), wesearched separately for matches to the 8-, 9-, and 10-nt PUF posi-tion–weight matrices (PWMs) defined by previous biochemicalanalysis of S. cerevisiae Puf3 (Sc_Puf3), Puf4 (Sc_Puf4), and Puf5(Sc_Puf5) (14, 30, 34, 36). For each sequence in each 3′UTR, wedetermined the log-likelihood ratio of the PWM model comparedwith a background genomic mononucleotide model (Fig. 1B). Weassigned to each 3′UTR the best log-likelihood ratio identified foreach PWM. For each PWM comparison, we then clustered groupsof orthologous genes based on the assigned log-likelihood score (Fig.1 C–E; gene ontology (GO) enrichments for each cluster are foundin Table S1).Striking patterns of conservation and evolution of PUF binding-

sequence enrichment emerged across groups of orthologous genes.However, we detected no clear clusters for which enrichment of asingle PUF PWM was conserved in all fungi. Many clusters exhibitfunctional relatedness. A few appear to lack it [e.g., cluster 2 in theEurotiomycetes (Aspergilli)] (Fig. 1C).In Saccharomycotina, the 8-nt cluster 1 and the 9-nt cluster 5 have

GO term enrichments for nuclear-encoded “mitochondrion” (Pvalue 1.4 E-96) and “ribosome biogenesis” (P value 4.0E-21), re-spectively, as noted previously (14, 37, 38). Strikingly, the enrichment

A B

C

UGU

ED

UGUUGAC

UGAC

UGAC

UGAC

UGAC

UAGAC

UGC

UGUUGAC

UGAC

UGAC

UGAC

UAGAC

UGC

UGAC

UGAC

UGAC

UAGAC

UGC

Saccharomycotina“Budding yeast”

Pezizomycotina“Filamentous fungi”

Taphrinomycotina“Fission yeast”

8 nt sequence element

9 nt sequence element

10 nt sequence element

Other Fungi

8 nt sequence element 10 nt sequence element9 nt sequence element

Aspergillus nidulans

Neurospora crassa

Schizosaccharomyces pombe

Candida albicansSaccharomyces cerevisiae

Allomyces macrogynus

Batrachochytrium dendrobatidis

Pneumocystis jirovecii

Cluster 5ribosome biogenesis4.04E-21

Cluster 4mitochondrion1.10E-15

Cluster 3mitochondrion3.08E-51ribosomal protein4.09E-09

Cluster 1mitochondrion1.39E-96

Cluster 2no enrichment

Cluster 6mitochondrion3.12E-28

Loglikelihood

No data-5 5

Fig. 1. Conservation of putative PUF binding elementsin Asycomocota fungi. (A) Phylogenetic tree of selectedfungi (see full tree in Fig. S1). Colored branches repre-sent each subphylum. Pink, Saccharomycotina (buddingyeasts); orange, Pezizomycontina (filamentous fungi);blue, Taphrinomycontina (fission yeast), black, otherfungi. (B) Representation of 8-, 9-, and 10-nt PWMmodels used to parameterize the log-likelihood func-tion. Height of base represents probability of a base ateach position in the binding element. (C–E) k-meansclustering of orthologous transcripts based on log-likelihood scores for putative PUF binding elementsin 3′UTRs for species in A, as shown in the key. Eachrow plots the log-likelihood PWM-match scores for anorthologous 3′UTR, and each column represents aspecies (with the phylogenetic tree shown above).Clustering was done independently for each heat map.Only clusters with a GO term P value lower thanE-9 were highlighted (except cluster 2). Shown are thelog-likelihood scores based on the 8-nt binding ele-ments (C) found in one or more of 4425 orthologoustranscripts. (D) The 9-nt binding elements found in4,898 transcripts. (E) The 10-nt binding elements foundin 4,423 transcripts.

Wilinski et al. PNAS | Published online March 20, 2017 | E2817

BIOCH

EMISTR

YPN

ASPL

US

for similar-length PUF binding elements was not conserved acrossother fungi: The enrichment for 8-nt-long sequences in cluster 1 wasonly seen for budding yeasts, whereas enrichment for 9-nt elements

in clusters 5 were observed for sublineages in filamentous fungi.Clusters of conservation restricted to the Pezizomycotina include9-nt clusters 3 and 4 and 10-nt cluster 6 (Fig. 1D and E). All three of

A

B CS. cerevisiae

A. nidulans

N. crassa

80

120

β-ga

l (n

orma

lized

%)

Binding element length

0204060

100

140160180

N. crassaA. nidulansACA 8 nt 10 nt9 nt ACA 8 nt 10 nt9 nt

156

9

22

1165

27

D

bits

bits

bits

1 2 3 4 5 6 7 8 9 10-2 -1

S. cerevisiae

N. crassa

A. nidulans

C. albicansbits

2

0

1

2

0

1

2

0

1

2

0

1

UGUACU

UA

GUAUAC

GT

UGUACU

UGAUA

CAUU

AUCA

G

AUC

ACUU

CTGU

U

GACUA AUU

AG

AUC

ACUU

CT

A

GUU

GAC

UA

UGAUU

A

BdenSpun

Amac

Tmes

Cgat

Cneo

Ccin

Pchr

Mlar

Pgra

Mglo

SreiUmay

SoctSjap

Spom

Pjir

Bcin

Sscl

Tree

Foxy

Fver

Fgra

MorySmac

Ntet Ndis

NcraTmar Tsti Aory Ater Afum Anid

AcarAnig

Afla

AacuCimm

CposClus

YlipAgos

KlacScas

algC

Sklu

Scer

Smik

Spar

Sbay

Lelo

Calb

Ctro

Cpar

Dhan

Cgui

RdelMcirPbla

1

2

34

8nt 9nt 10nt

Binding element length

Abbreviated species name

Number of PUF genes

0 1 2 3

1

2

34

log(p-value-1)

Species

PUF3

PUF4/5

Taph

rinom

ycot

ina Saccharom

ycotina

Pezizomycotina

0

100

+

Fig. 2. Evolution of PUF–mitochondrial RNA network. (A) Enrichment of PUF binding elements in transcripts from the mitochondrial cluster across Asco-mycota fungi. Abbreviations used for fungal species are provided in Fig. S5. Ring 1 represents binding element enrichment orthologs in each species. Theinner ring is the phylogenic species tree. Black arrows mark the species used in C. (B) MEME-derived PWMs identified in 3′UTRs of mitochondrial clustertranscripts in each species. (C) Overlap of genes in mitochondrial cluster containing a putative PUF binding element for three species. (D) Relative lumi-nescence values as a proxy for binding affinity for each PUF protein binding to each PUF site as assayed by yeast-three hybrid assays (48).

E2818 | www.pnas.org/cgi/doi/10.1073/pnas.1617777114 Wilinski et al.

these clusters were enriched for mitochondrial function, but notably,none was enriched for the 8-nt PUF sequences known to mediatetheir control in S. cerevisiae. Interestingly, 9-nt cluster 3 was enrichedfor genes for both mitochondrion (3.1E-51) and ribosomal protein(4.1E-9). (A complete list of GO enrichments for each cluster is inTable S1.)

Predicted Switch Between Budding and Filamentous Fungi. The309 transcripts in cluster 1 (Fig. 1C) are highly enriched for 8-ntelements, consistent with their binding to Sc_PUF3 (14, 32). Werefer to these transcripts as the “mitochondrial cluster.” In fila-mentous fungi, other clusters are enriched for mitochondria-relatedfunctions but contain transcripts with putative 9-nt or 10-nt bindingelements (Fig. 1 D and E, clusters 3, 4, and 6). Our data are con-sistent with bioinformatic analyses that suggested a “rewiring” eventfor this group (37, 38). They further suggest widespread evolution ofenriched sequences among transcripts in the mitochondrial cluster.We determined the enrichment of each length of putative bindingelement for transcripts in the mitochondrial cluster across 62 fungi(Fig. 2A, ring 1). All of the budding yeasts exhibit significant en-richment of 8-nt sequences in their 3′UTRs but not of 9-nt and10-nt elements. Conversely, none of the filamentous fungi havesignificant enrichment for the 8-nt site. Instead, the filamentousfungi exhibit enrichment for putative binding elements 9 or 10 ntlong (Fig. 2A, ring 1). No other clade showed a significant enrich-ment of putative binding elements in these orthologs. These resultswere supported by unbiased motif discovery in the groups’ 3′UTRs(43), which identified PWMs matching the 9- and 10-nt bindingelements seen in A. nidulans and N. crassa transcripts orthologousto the mitochondrial cluster (Fig. 2B). Most of the filamentous fungitranscripts orthologous to the mitochondrial cluster have a PUFbinding element, as is the case in S. cerevisiae (Fig. 2C). Thus, pu-tative PUF binding elements in the mitochondrial cluster extend tofilamentous fungi but have diverged so that they are of a differentlength, including 10-nt elements not previously detected.

Evolution of Putative Binding Elements Despite Conservation of PUFBinding Specificities. Two models could explain the differences inbinding site lengths: Transcripts in the mitochondrial cluster couldbe regulated by the same PUF ortholog whose binding specificitychanged during evolution; alternatively, the sequence of thebinding elements could have changed, while the protein specificitydid not. To address this, we investigated the specificities of PUFproteins between species. All proteins containing a Pumilio do-main for all 62 species were aligned, and then a protein tree wasgenerated by maximum likelihood (Fig. S2) (44–46).Nearly all species have a single ortholog of Puf3 (Fig. 2A, ring 3).

In contrast, Saccharomycotina and Schizosaccharomycetes have twoparalogs that belong to a Puf4/Puf5 clade, whereas Pezizomycotinaand other fungi have only one copy of the Puf4/5 protein (Fig. 2A,ring 4). This suggests there was one ancestral gene, which we referto as PUF4/5, that was duplicated in the Saccharomycotina (37) andthen again in the Schizosaccharomycetes. This is particularly in-triguing because S. cerevisiae Puf4 and S. cerevisiae Puf5 have dif-ferent binding specificities, suggesting that their binding preferencesdiverged after duplication (14, 30).To distinguish between changes in protein specificity or

binding site sequence, we assayed the RNA binding specificitiesof A. nidulans PUF4/5 (An_PUF4/5; AN10071) and N. crassaPUF-4/5 (Nc_PUF4/5; NCU16560.1) using a series of RNAswith binding elements ranging in length from 8 to 10 nt, using thethree-hybrid system (47). In this assay, levels of LacZ expressionare directly related to binding affinity (48). A. nidulans PUF4/5 bound best to sequences matching the 9-nt PUF binding ele-ment, similar to the known binding preference of Sc_Puf4, butalso bound weakly to a 10-nt element (Fig. 2D). In contrast,N. crassa PUF-4/5 (Nc_PUF-4/5) bound well to both 9- and10-nt-long binding elements (Fig. 2D). The broader specificity ofNc_PUF-4/5 echoes the expanded binding specificity of Sc_Puf5(30). Together, these data suggest that the ancestral PUF4/5 protein could bind both 9-nt and 10-nt elements and that the

known specificities for the S. cerevisiae proteins emerged afterthe PUF4/5 duplication (see Discussion).The binding specificities of the filamentous fungi PUF4/

5 orthologs and the enrichment for putative 9- and 10-nt ele-ments in the orthologs of the mitochondrial cluster suggest thatthere was a “regulator switch,” in which A. nidulans and N. crassaunderwent a change in the protein that mediates control oforthologous mitochondrial RNAs. In this hypothesis, mRNAsthat possess 8-nt elements in S. cerevisiae, which are recognizedby Sc_Puf3, instead possess 9- or 10-nt elements in Neurosporaand Aspergillus and are likely bound by PUF4/5 in those species.

In Vivo Targets of S. cerevisiae Puf3 and N. crassa PUF4/5 Demonstratea Regulatory Switch. To test the regulator switch hypothesis, weidentified mRNAs bound in vivo to Sc_Puf3, N. crassa PUF3(Nc_PUF3), and Nc_PUF4/5. We cross-linked RNA–proteincomplexes in vivo with UV irradiation of intact cells, immuno-purified these protein–RNA complexes, and identified the RNAsby high throughput sequencing (“HITS-CLIP”) (49). We recentlyanalyzed Sc_Puf5 using the same methods and included those datain the meta-analysis (30). The S. cerevisiae PUF genes were TAP-tagged and the N. crassa PUF genes were FLAG-tagged; all geneswere integrated by homologous recombination at the endogenouschromosomal locus under control of the natural promoter. Thethree biological replicates for each protein displayed excellentreproducibility (Fig. 3 A–C).We identified 500, 835, and 746 significant CLIP peaks for

Sc_Puf3, Nc_Puf3, and Nc_PUF4/5, respectively. These arisefrom 7.6, 8.4, and 14.3 million uniquely assignable reads (forcomplete mapping statistics, see Table S2). Data quality inHITS-CLIP is often assessed by the presence of binding elementsin peaks of read enrichment (50–52). In our data, 90%, 68%, and99% of Sc_Puf3, Nc_PUF3, and Nc_PUF4/5 binding peaks, re-spectively, contain one or more sequences matching the PWMsderived for each set of targets, suggesting our sequencing dataidentified genuine PUF protein targets.The peaks, predominantly found in the 3′UTRs of mRNAs,

represent 465, 803, and 686 transcripts bound by Sc_Puf3,Nc_PUF3,and Nc_PUF4/5, respectively (Fig. S3 and Dataset S1). Represen-tative CLIP peaks are shown in Fig. 3 D–G for COX17 orthologs, awell-characterized target of Sc_Puf3 (Fig. 3D) that is not efficientlybound by Sc_Puf5 (Fig. 3E). Strikingly, the Neurospora COX17ortholog was bound by Nc_PUF5 but not Nc_PUF3 (Fig. 3 F andG). In all cases, binding peaks were centered over the known bindingelements in the S. cerevisiae COX17 mRNA or the two predictedelements in the N. crassa ortholog. This trend was seen for theprofiles of the five loci with the largest binding signal in the N. crassaand S. cerevisiae Puf3 and PUF4/5 proteins (Fig. S4 A–C).To assess whether targets were shared between Sc_Puf3 and

N. crassa PUFs, we determined their overlap across those species(Fig. 3H). The overlap in transcripts bound by Sc_Puf3 andNc_PUF4/5 was much more significant (6.1E-41, hypergeometricdistribution test) than the overlap between Sc_Puf3 and Nc_PUF3(5.0E-3, hypergeometric distribution test) (Fig. 3H): Orthologs of42% of Sc_Puf3 targets were bound by Nc_PUF-4/5, whereas only17% were bound by Nc_PUF3. We conclude that Sc_Puf3 andNc_PUF4/5 proteins bind a similar set of RNAs, even more thando the two PUF3 orthologs.

PUF Binding Specificities Across Species. To determine the in vivo se-quence preference of each PUF protein, we identified PWMs basedon overrepresented sequences in targets from the HITS-CLIP dataset(43). The binding preferences of the N. crassa PUFs were strikinglysimilar to those of their orthologous proteins in S. cerevisiae (Fig. 4 Aand C). Sc_Puf3-bound sites contained sequences matching theknown 8-nt binding element comprised of a 5′ UGU and a 3′ AUA(Fig. 4A). In addition to the core PUF binding element, ∼50% of thesequences associated with Sc_Puf3 contain a cytosine at the minus2 position, which enhances binding in vitro and biological activity invivo (32, 36). Nc_PUF3-bound regions contained 8-nt sequencessimilar to the known Sc_Puf3 binding element, but the PWM lacked

Wilinski et al. PNAS | Published online March 20, 2017 | E2819

BIOCH

EMISTR

YPN

ASPL

US

the –2C enrichment seen in S. cerevisiae (Fig. 4B) (14, 32, 36). ThePWM identified for targets of Sc_Puf5 contains the canonical 5′UGUA with apparent degeneracy in the 3′-end; we previouslyshowed that the single matrix includes discrete 8–12-nt-long bindingelements that differ in spacer length (30). When training a singlePWM on Nc_PUF4/5 targets, we identify the same type of apparentdegeneracy (Fig. 4D) that can be explained as a combination of 9-nt-and 10-nt-long elements (Fig. 4E). The PWMs derived from in vivotargets agree well with our bioinformatic predictions of sequenceenrichment (Fig. 2B) and the binding specificities measured ex vivo(Fig. 2D). We conclude that the in vivo binding preferences ofPUF3 orthologs is highly conserved and that the binding preferenceof the single Nc_PUF4/5 is similar to that of Sc_Puf5.

Evolution of PUF Target Networks in N. crassa. To determine thefunctional enrichment for the targets of each PUF protein, weperformed GO analyses. As expected, the Sc_Puf3 targets exhibiteda strong enrichment for nuclear-encoded mitochondrial proteins (Pvalue 1E-103) and translation factors (P value 6.5E-37) (Fig. 4F andDataset S2). The targets of Nc_PUF4/5 were dramatically enrichedfor those encoding nuclear-encoded mitochondrial proteins (P value1E-25); Nc_PUF3 targets also were enriched for this annotation,although much more modestly (P value 1E-5) (Fig. 4F and DatasetS2). In agreement with these enrichments, we found that 50% (155/309) of orthologous transcripts in the mitochondrial cluster (Fig. 1C)

were bound by Sc_Puf3 in S. cerevisiae, whereas 44% (137/309) werebound by Nc_PUF4/5 and 13% (39/309) by Nc_PUF3 in N. crassa(Fig. 4G). Interestingly, of the 39 targets of Nc_PUF3 in the mito-chondrial cluster, 71% were also bound by Nc_PUF4/5 (Fig. 4G).These transcripts often contained both 8-nt- and 9/10-nt-long puta-tive PUF binding elements. Two examples of cotargets of Nc_PUF-3 and Nc_PUF-4/5 are transcripts CBP3 and NCU07386, whichharbor both binding elements, one under each respective peak (Fig.4H). Thus, a subset of mitochondrial-related transcripts are boundby both PUF4/5 and PUF3 through their respective recognition el-ements. Nc_PUF4/5 targets were enriched for other functional an-notations beyond mitochondrial transcripts. The group of targets wasalso enriched for those encoding cytosolic ribosomal proteins (Pvalue 3E-16), which is intriguing, as previously identified Sc_Puf4targets are enriched for transcripts encoding ribosome biogenesisproteins (14) (see Discussion).

Evidence for Multiple Rewiring Events and the Emergence of NewNetworks. Our data suggest that orthologous transcripts haveevolved different lengths of PUF binding elements in their 3′UTRsand hence are regulated by different PUF proteins. To further in-vestigate this, we assembled a combined set of all orthologoustranscripts bound by Sc_Puf3, Nc_PUF3, and Nc_PUF4/5, orSc_Puf4 (identified in a previous study by a different method) (14).We clustered this combined set of transcripts based on putative

A S. cerevisiae PUF3

8 10 12 14 16 18 20 8

10

12

14

16

18

810

1214

16

Replicate 1Replica

te 2Repli

cate

3

Replicate 1Replica

te 2Repli

cate

3

Replicate 1Replica

te 2Repli

cate

3

N. crassa PUF3

6 8 10 12 14 16 18 20 6 8101214161820

6 81012141618

N. crassa PUF5

8 10 12 14 16 18 8101214161820

810

1214

16

N. crassaPUF4/5

N. crassaPUF3 S. cerevisiae

PUF3

193

207

186

112

21

73

41

H

cox17

339

09

UGUACAACUA10

UGUAUAAUA

E

UGUAUAUA88 88

UGUAAAUA

4,179

0cox17

CB

GF

UGUAUAUA8 8

UGUAAAUA

4,179

0cox17

RPMRPM

RPMRPM

S. cerevisiaePUF3

S. cerevisiaePUF5

N. crassaPUF3

cox17

339

09

UGUACAACUA10

UGUAUAAUA

N. crassaPUF4/5

D

ρ ≥ 0.65r ≥ 0.77

ρ ≥ 0.64r ≥ 0.99

ρ ≥ 0.79r ≥ 0.67

Fig. 3. HITS-CLIP data. (A–C) Reproducibility of biological replicates. Height of CLIP peaks were log-transformed and then plotted on each axis. Data werecolored based on the precipitated protein (Sc_PUF3, blue; Nc_PUF3, red; Nc_PUF4/5, green). Spearman’s correlation coefficients (ρ), associated P values (P), andPearson’s correlation coefficient (r) are indicated in A (ρ = 0.69, 0.84, 0.64; P = 0, 0, 0; r = 0.77, 0.96, 0.81; n = 500), B (ρ = 0.64, 0.85, 0.76; P = 0, 0, 0; r = 0.99, 0.99,0.99; n = 835), and C (ρ = 0.85, 0.92, 0.79; P = 0, 0, 0; r = 0.71, 0.97, 0.67; n = 746). (D–G) Examples of binding peaks (read depth) for COX17, a canonical target ofSc_PUF3. YLL009C (COX17), NCU0058 (al-2), and NCU02530 (cox17) are depicted, with ORFs annotated in blue and the likely binding elements displayed beneath.(H) Overlapping CLIP targets for each protein. Only genes with orthologs in S. cerevisiae and N. crassa are included. RPM, reads per million mapped reads.

E2820 | www.pnas.org/cgi/doi/10.1073/pnas.1617777114 Wilinski et al.

PUF binding elements identified in 3′UTRs of each species’orthologs across 62 fungal species (Fig. 1A). This analysis captureddifferential enrichment of putative PUF elements in the ortholo-gous mRNAs, all of which are targets identified in our CLIP studies.Several striking patterns emerged that implicate additional PUF

regulatory rewiring and the emergence of networks. As reportedabove, transcripts encoding mitochondrial proteins (Fig. 5, cluster A)are enriched for 8-nt elements in S. cerevisiae and, to some extent inN. crassa, but are strongly enriched for 9-nt and 10-nt elements infilamentous fungi. Interestingly, many transcripts linked to mito-chondrial envelope and organization (Fig. 5, cluster B) were specifi-cally depleted of PUF binding elements in the Neurospora clade. Incontrast, cluster C (Fig. 5) was bound by Nc_PUF3 and enriched forthe 8-nt PUF element—strikingly, this group was enriched for tran-scripts encoding hydrolases and cellulose-binding proteins, functionsthat may be important for the optimal growth of this filamentous

fungus. This cluster may represent the emergence of a new networkin that species. Cluster D reveals ribosomal protein mRNAs with9-nt elements inNeurospora but not in S. cerevisiae. Cluster E revealsmRNAs with 9-nt elements in S. cerevisiae (and recognized by itsPuf4 and Puf5 proteins), which again may imply the emergence of anew network during evolution. Finally, we observe an enrichment ofPUF targets in Neurospora that are responsive to light (Nc_PUF3:233, P value 1.7E-13, Nc_PUF4/5: 257, P value 1.1E-22; hyper-geometric distribution test). Thus, this regulatory system is also im-pacted by PUF regulation.To visualize the overlaps, we depicted the connection between

every orthologous target and its cognate binding protein(s) (Fig.6). Each target of the four proteins (Sc_Puf3, Sc_Puf5, Nc_PUF3,or Nc_PUF4/5) is represented as a node (small circle). Many ofthe transcripts with mitochondrial functions are bound bySc_Puf3 and Nc_PUF4/5 in the respective species, whereas others

UGUGAG

ACU

TAG

CGUA

CGAU

CGUA

C

GUAUA

GAGCU

GCA

AGUC

CGA

CAU

UGA

UUUGUAC

UUCG

UG

UGU

GA A

AAAA

AU

G G GCC

AC

CCC

PUF3 PUF5

S. cerevisiae

N. crassa

99% (739 peaks out of 746)68% (570 peaks out of 835)

90% (451 peaks out of 500)

A C

B D

G

AUC

CAU

AGUU

GA

CUA

UA

AU

G

UAbit

s

2

0

1

G

bits

2

0

1

G

bits

2

0

1

G

bits

2

0

1

bits

2

0

1

bits

2

0

1

G

bits

2

0

1

1 2 3 4 5 6 7 8 9 10-2 1 2 3 4 5 6 7 8 9 10-2-1 -1

1 2 3 4 5 6 7 8 9 10-2 1 2 3 4 5 6 7 8 9 10-2-1

1 2 3 4 5 6 7 8 9 10-2 -1

-1

E

71% (741 genes out of 1,043)

1E0 1E-8 1E-16 1E-24 1E-32 1E-40 1E-108

S. cerevisiae PUF3N. crassa PUF3S. cerevisiae PUF5

p-value-1

F

U

UGUTGA

GACU

CAUG

CGUA

CGAU

CGAU

CAGU

UGUGAACU

TA

G

CGUA

C

UAUGUG

AACU

TA

G

CGUA

C

UAUAU

33% 245/739)

26%(193/739)

38%(282/739)

N. crassaPUF4/5

G

N. crassaPUF3

S. cerevisiaePUF3

N. crassaPUF4/5 u = 117

PUF3 PUF4/5

1396

0

UGUACAAUUA1010 88

UGUAGAUACBP3CBP3

146

HNc_PUF3Nc_PUF4/5

1800

0

120

0

UGUACGAUA99 88

UGUCUAUANCU07386

Nc_PUF3Nc_PUF4/5

RPMRPM

RPM

N. crassa PUF4/5

44

25

84 235

47

Fig. 4. In vivo binding elements of PUF proteins. MEME-derived PWMs from CLIP data. (A) S. cerevisiae PUF3, (B) N. crassa PUF3, (C) S. cerevisiae PUF5 (30), (D)N. crassa PUF4/5. (D) The previously defined Par-CLIP ofHomo sapiens PUF3 ortholog, PUM2, defined by PhyloGibbs (52, 59). (E) Deconvolution ofNc_PUF4/5 compositebinding element (D) into binding elements of 9 nt and 10 nt in length. (F) Mitochondrion GO term enrichment for all mRNAs targets of each PUF protein. (G) Venndiagram depicting the overlap between PUF targets from the mitochondrial cluster. (H) Nc_PUF3 and Nc_PUF4/5 interaction peaks identify separate binding elementsfor each PUF protein. One biological replicate for Nc_PUF3 and Nc_PUF4/5 are plotted in red and brown, respectively. Likely PUF binding elements are highlighted.RPM, reads per million mapped reads.

Wilinski et al. PNAS | Published online March 20, 2017 | E2821

BIOCH

EMISTR

YPN

ASPL

US

are uniquely associated with one protein or another (Fig. 6A). Thefact that many connections are unique, but with targets that en-code proteins in the same biological process, implies that PUFregulation of this biological function is conserved, whereas thespecific mRNAs that are regulated diverged (Fig. 6 B–D).

The Emergence of New Sites Underlies Rewiring. Two evolutionaryroutes could produce an apparent switch from 8-nt PUF3 regulationto 9/10-nt PUF4/5 regulation. An element of one length could changeinto a different length simply by gain or loss of nucleotides within theelement; such changes are particularly facile, as the identity of basesin the central region of the site is unimportant in binding (see theIntroduction). Alternatively, a new binding element could ariseelsewhere in the same 3′UTR before the original motif degenerates.We find a preponderance of evidence for the latter model.Nc_PUF4/5 targets in the mitochondrial cluster are significantly enriched fortranscripts with multiple binding elements per mRNAUTR (averageof 2.04 elements per 3′ UTR) compared with all Nc_PUF4/5 targets(average of 1.67) (P value 6.5E-4, two-tailed t test). Furthermore,transcripts within a species can be bound by both proteins, as pre-

viously detailed for the cotargeting of CBP3 and NCU07386 byNc_PUF-3 and Nc_PUF-4/5 (Fig. 4H). Although element-lengthswitching likely exists, the presence of targets with two sites in asingle 3′UTR suggests a transient intermediate may often exist inwhich both proteins act on the same mRNAs.

DiscussionOur work provides compelling experimental support for exten-sive rewiring of PUF-dependent regulatory networks in fungi andinsights into how that occurred. Computational analyses promptedmodels of the rewiring of the nuclear-encoded mitochondrial net-work (5, 37, 38). Our biochemical data provide molecular evidencefor such a regulator switch, reveal multiple rewiring events, andsupport fluiditiy in the emergence of new PUF–RNA networks.The most parsimonious model for the evolution of the mito-

chondrial networks posits that mitochondrial transcripts (along withcytosolic ribosomal protein mRNAs and other transcripts) wereregulated by both PUF3 and PUF4/5 in the common ancestor (Fig.7). The relative prominence of PUF3 versus PUF4/5 control of mi-tochondrial transcripts evolved in the modern-day lineages, with

Cluster

8nt 9nt 10nt

A

mitochondrion organization4.19E-101mitochondrial translation4.70E-62mitochondrion 1.3E-25

B

mitochondrion organization4.17E-09mitochondrial envelope 1.3E-4

E

ribonucleoprotein complexbiogenesis 3.34E-10ribosome biogenesis3.90E-09ncRNA metabolic process1.9E-2

C

Nonehydrolase1.0E-3fCBD (Fungal-type cellulose-binding domain)6.7E-2

DNoneribosomal protein 3.4E-2

Sc_PUF4Sc_PUF3 Nc_PUF3

Nc_PUF4/5Log likelihood No data-5 5

Fig. 5. Evolution of PUF elements in PUF-boundtranscripts. (A–E) k-means clusters. Orthologs of tran-scripts bound by Sc_PUF3, Sc_PUF4 (14), Nc_PUF3, orNc_PUF4/5 were aligned and clustered based on thelog-likelihood match of 3′UTR sequences to the 8-nt,9-nt, and 10-nt PUF PWMs (Materials and Methods).Gray indicates a lack of ortholog in that species. Tar-gets bound by each protein are annotated by coloredbars to the left of the figure. Enriched GO terms foreach cluster are shown to the right (S. cerevisiae inblack and N. crassa in tan).

E2822 | www.pnas.org/cgi/doi/10.1073/pnas.1617777114 Wilinski et al.

PUF3 playing the predominant role in Sacharamycotina and PUF4/5 the major role in filamentous fungi. Orthologous PUF3 proteinshave stringent specificity for 8-nt-long elements in virtually every or-ganism tested to date, including mammalian PUM (14, 30, 52, 53).We show that, in contrast, Nc_PUF-4/5 and An_PUF4/5 can bind9 nt, and even 10 nt in the case of Nc_PUF4/5 long binding elements(Fig. 2D), as does Sc_Puf5 (30). Together, this suggests that the an-cestral mitochondrial transcripts were regulated by PUF3 and PUF4/5 bound to their 8-nt and 9/10-nt 3′ UTR elements, respectively.Several events produced specialization of the PUF networks

along the branch leading to budding yeasts. Our phylogeneticanalysis indicates that PUF4/5 duplicated after the split of buddingand filamentous fungi and then diverged in binding preference;broad binding specificity was likely ancestral (37). We propose thatSc_Puf5 retained the ancestral specificity whereas Sc_Puf4 special-ized to bind a 9-nt element. These changes likely involved alter-ations in the topology of the PUF scaffold (30). As protein binding

specificity evolved, so too did the elements of the downstream tar-gets: Mitochondrial transcripts bound by Sc_Puf3 evolved toward 8-ntelements, whereas transcripts bound by Sc_Puf4 and Sc_Puf5 evolved9-nt or maintained mixed-length PUF sequences, respectively. It isintriguing that Sc_Puf4 targets are enriched for ribosome-biogenesistranscripts, which are functionally related to the ribosomal proteintranscripts bound by the orthologous Nc_PUF4/5 in N. crassa. Wepropose that the ancestral PUF4/5 protein regulated both mito-chondrial and cytosolic ribosome-biogenesis functions but that thiscoregulation was lost in budding yeasts when the two networks spe-cialized toward PUF3- and PUF4-dependent control, respectively.This may have facilitated a further decoupling of mitochondrial andcytosolic ribosomal protein gene expression in respiro-fermentativeyeasts, as previously proposed to have occurred through upstreamtranscription factor rewiring (54, 55). The acquisition of sugar-responsive PUF3 regulation could have provided selective pressureto evolve and maintain the PUF3 network in respiro-fermentative

mitochondriaother

light responsiveother

ribosomeother

hydrolaseother

Sc_PUF3 Sc_PUF5

Nc_PUF3 Nc_PUF4/5

A

D

B

C

Sc_PUF3 Sc_PUF5

Nc_PUF3 Nc_PUF4/5

Sc_PUF3 Sc_PUF5

Nc_PUF3 Nc_PUF4/5

Sc_PUF3 Sc_PUF5

Nc_PUF3 Nc_PUF4/5

Fig. 6. Functional relatedness of all mRNAs bound by PUF3 and PUF4/5 proteins of S. cerevisiae and N. crassa. All orthologous targets of PUF proteins definedby CLIP are plotted as nodes (gray balls). Edges (lines) link the PUF to the RNAs it binds. Targets are colored based on their annotated function: (A) mito-chondria (purple), (B) ribosome (red), (C) hydrolase (black), and (D) light responsive (blue).

Wilinski et al. PNAS | Published online March 20, 2017 | E2823

BIOCH

EMISTR

YPN

ASPL

US

yeasts. Sc_Puf3 is extensively phosphorylated when S. cerevisiae isdepleted of glucose and switches from a repressor to an activator ofmitochondrial mRNAs (24). Acquisition of regulated activation viaphosphorylation may have committed the budding yeast lineage tousing Puf3 to control mitochondrial RNAs.A different story emerges for the evolutionary trajectory in the

filamentous fungi. Our experimental data demonstrate PUF rewiringof the mitochondrial targets. Orthologs of Sc_Puf3-bound targetsdisplay strong enrichment for 9-nt or 10-nt sites in filamentousspecies. We show here that these orthologs of Sc_Puf3 targets arebound by Nc_PUF4/5 in N. crassa and to a lesser extent byNc_PUF3. We find one set of transcripts that is strongly enriched for9-nt PUF elements in the Neurospora species and, to a lesser extent,in the budding yeasts; however, orthologs in the Aspergilli areenriched primarily for 8-nt PUF elements (Fig. 5, cluster D, and Fig.7, purple box). This observation suggests a second hand-off of targetspotentially regulated by PUF3 in Aspergilli. Given the enrichment inthis group for ribosome-related functions and for 9-nt elements inbudding-yeast transcripts, we propose that a subset of these mRNAsare regulated by Puf4 in S. cerevisiae (see Fig. 5, cluster DSc_PUF4 binding profiles). This model suggests a complex interplaybetween three different PUF regulators in three branches of fungiand implicates a second independent rewiring of PUF3 regulation.Our data also demonstrate an additional, independent evo-

lution of PUF4/5 binding specificity in Aspergilli. Whereas thesingle PUF4/5 protein in N. crassa binds 9-nt and 10-nt se-quences equally well, An_PUF4/5 strongly prefers the 9-nt-longelement (Fig. 2D). Concordantly, the mitochondrial transcriptsfrom Neurospora species are nearly equally enriched for 9-nt and10-nt elements, whereas those from Aspergilli are significantly biasedtoward 9-nt elements (Fig. 2A, arrows). Together, these data in-dicate that Aspergilli have evolved both in terms of PUF4/5 speci-ficity and in the 3′UTR sequences found in hundreds of itstranscripts. It also confirms that there have been two independentorigins of PUF4/5 binding preference: once after the split of As-pergilli andNeurospora species and again after PUF4/5 duplication inthe Saccharomycotina lineage. It is plausible that the determinantsfor 9-nt versus 10-nt site preference in the PUF4/5 proteins may berelatively simple to change; modest changes in the curvature of thePUF scaffold may be sufficient (30, 35, 56). The related bindingspecificities of PUF proteins and the relaxed binding preferences ofthe ancestral PUF4/5 protein likely both contributed.Regulatory redundancy is a recurring theme in transcription-

factor network rewiring. Ancestral and newly derived regulators areproposed to cofunction until ancestral regulation is lost, completing

the “handoff” of targets. In most examples, that redundancy isprovided by the co-occurrence of distinct transcription factor bind-ing sites (6, 8, 57). This model of multiple elements likely pertains torewiring of the PUF regulons. For example, many mitochondrialtranscripts bound by Nc_PUF3 are also bound by Nc_PUF4/5—inthese cases, binding appears to occur through 8-nt or 9/10-nt ele-ments, respectively. The co-occurrence of multiple elements mayhave facilitated evolution toward predominantly PUF3-dependentregulation in the budding yeasts versus primarily PUF4/5-dependentregulation in the filamentous fungi, simply through loss of one classof binding site. Once initiated, this situation could have amplifiedthe selective pressure for other functionally related targets to ac-quire 8-nt or 9/10-nt elements in the respective lineages to maintaincoregulation under the new regulatory system.PUF proteins bind a set of related binding elements—this, com-

bined with the relaxed specificity of PUF4/5, could provide an al-ternate model of redundant regulatory systems. PUF elements in thetarget transcripts could evolve by de novo creation of new bindingsequences and loss of ancestral sites from elsewhere in the 3′UTRs.Alternatively, an existing PUF element could rapidly switch into adistinct-length PUF sequence through simpler changes. Insertion ordeletion of bases into the PUF-element spacer region could instantlychange PUF element length and thus change which PUF proteinbinds that element. Alternatively, addition of a simple UA di-nucleotide downstream of an existing 8-nt element (e.g., UGUAN-NUA) could switch specificity from PUF3 to PUF4/5.The recurrent and large-scale rewiring of PUF networks across

fungal species may reflect the simplicity with which specificities can bechanged via simple alterations of the PUF scaffold and its bindingelements. It is intriguing that many of the rewiring events we reporthere are linked to groups of transcripts whose functional relationshipsare associated with species-specific differences in niche or environ-mental responsiveness. 3′UTR-based networks such as these mayevolve particularly rapidly, given the rapid divergence of 3′UTRs insequence and length. This may enable rewiring of transcript modulesto different upstream regulatory systems that respond to novel signalsas species evolve. It will be of interest to determine which features ofPUF network evolution are idiosyncratic and which are shared withother families of mRNA regulatory proteins.

Materials and MethodsDetailed methods and descriptions of yeast three-hybrid assay, HITS-CLIP, N. crassamanipulations, and accession numbers can be found in SI Materials and Methods.Ascomycota fungi were chosen based on previously determined orthology (41).Phylogenetic tree shown in Fig. 1A and Fig. S1was calculated using ribosomal RNA

Ance

stral

State

Puf4/5Puf3

8nt 9/10nt

S. ce

revis

iae

Puf5Puf3

8nt 9nt

Puf4

10/11/12nt

PUF-4/5PUF-3

8nt 9/10nt

MtTargets

RiboTargets

Mt RiboMt Ribo Mt Ribo

GeneDuplication

Puf4Puf5

N. cr

assa

PUF4/5PUF3

8nt 9nt

A. n

idulan

s

10nt9nt

Light-Responsive

TargetsDashed boxes represent

acquisition (or loss in other species)of different targets

Withoutfunctional enrichment

Withoutfunctional enrichment

Fig. 7. Evolutionary model of PUF network rewir-ing. Colored boxes represent orthologous sets ofmRNA targets. Solid arrows indicate PUF binding ofthe denoted binding element enriched in the targetmRNA UTRs; dashed arrows represent weak bindingor secondary modes of regulation (see Discussion fordetails). Dashed colored boxes represent the acqui-sition or loss of transcripts from the denoted reg-ulon. Mt, mitochondrial; Ribo, ribosomal.

E2824 | www.pnas.org/cgi/doi/10.1073/pnas.1617777114 Wilinski et al.

sequences (58), based on the maximum likelihood model executed in thePhyML program (45). Sequences of 300 bases downstream of the translationtermination codon were obtained from organism-specific databases. Each3′UTR sequence was probed for matches to PMWs representing PUF bindingelements. The sequence with the high log-likelihood score was filtered toretain orthologous transcripts in which >50% of species had an ortholog. Theremaining genes were then k-means clustered with k = 13. (We chose13 clusters based on the “elbow” method, which considers the percentage of

variance as a function of the number of clusters.) Additional descriptions ofbioinformatic analyses are provided in SI Materials and Methods.

ACKNOWLEDGMENTS. We thank the M.W. and Kimble laboratories, andElena Sorokin, in particular, for helpful discussions of this work. We alsothank Laura Vanderploeg of the UW Biochemistry Media Lab for help withfigures. This work was supported by NIH Grants R01 GM035690 and R01GM093061 (to E.U.S.), F32 GM097821 (to A.D.K.), R01 GM083989 (to A.P.G.),and R01 GM50942 (to M.W.).

1. Doniger SW, Fay JC (2007) Frequent gain and loss of functional transcription factorbinding sites. PLOS Comput Biol 3(5):e99.

2. Wohlbach DJ, Thompson DA, Gasch AP, Regev A (2009) From elements to modules:Regulatory evolution in Ascomycota fungi. Curr Opin Genet Dev 19(6):571–578.

3. Villar D, Flicek P, Odom DT (2014) Evolution of transcription factor binding in meta-zoans—Mechanisms and functional implications. Nat Rev Genet 15(4):221–233.

4. Whiteway M, Tebung WA, Choudhury BI, Rodríguez-Ortiz R (2015) Metabolic regulation inmodel ascomycetes–Adjusting similar genomes to different lifestyles. Trends Genet 31(8):445–453.

5. Gasch AP, et al. (2004) Conservation and evolution of cis-regulatory systems in asco-mycete fungi. PLoS Biol 2(12):e398.

6. Tanay A, Regev A, Shamir R (2005) Conservation and evolvability in regulatory networks:The evolution of ribosomal regulation in yeast. Proc Natl Acad Sci USA 102(20):7203–7208.

7. Hogues H, et al. (2008) Transcription factor substitution during the evolution offungal ribosome regulation. Mol Cell 29(5):552–562.

8. Martchenko M, Levitin A, Hogues H, Nantel A, Whiteway M (2007) Transcriptionalrewiring of fungal galactose-metabolism circuitry. Curr Biol 17(12):1007–1013.

9. Tsong AE, Tuch BB, Li H, Johnson AD (2006) Evolution of alternative transcriptionalcircuits with identical logic. Nature 443(7110):415–420.

10. Reece-Hoyes JS, et al. (2013) Extensive rewiring and complex evolutionary dynamics ina C. elegans multiparameter transcription factor network. Mol Cell 51(1):116–127.

11. Baker CR, Hanson-Smith V, Johnson AD (2013) Following gene duplication, paraloginterference constrains transcriptional circuit evolution. Science 342(6154):104–108.

12. Conant GC, Wolfe KH (2008) Turning a hobby into a job: How duplicated genes findnew functions. Nat Rev Genet 9(12):938–950.

13. Chen S, Zhang YE, Long M (2010) New genes in Drosophila quickly become essential.Science 330(6011):1682–1685.

14. Gerber AP, Herschlag D, Brown PO (2004) Extensive association of functionally and cyto-topically related mRNAs with Puf family RNA-binding proteins in yeast. PLoS Biol 2(3):E79.

15. Keene JD (2007) RNA regulons: Coordination of post-transcriptional events. Nat RevGenet 8(7):533–543.

16. Mukhopadhyay R, Jia J, Arif A, Ray PS, Fox PL (2009) The GAIT system: A gatekeeperof inflammatory gene expression. Trends Biochem Sci 34(7):324–331.

17. Ule J, et al. (2003) CLIP identifies Nova-regulated RNA networks in the brain. Science302(5648):1212–1215.

18. Ray D, et al. (2013) A compendium of RNA-binding motifs for decoding gene regu-lation. Nature 499(7457):172–177.

19. Campbell ZT, et al. (2012) Cooperativity in RNA-protein interactions: Global analysisof RNA binding specificity. Cell Reports 1(5):570–581.

20. Mayr C (2016) Evolution andbiological roles of alternative 3’UTRs. Trends Cell Biol 26(3):227–237.21. Mangone M, et al. (2010) The landscape of C. elegans 3’UTRs. Science 329(5990):432–435.22. Jan CH, Friedman RC, Ruby JG, Bartel DP (2011) Formation, regulation and evolution

of Caenorhabditis elegans 3’UTRs. Nature 469(7328):97–101.23. Wickens M, Bernstein DS, Kimble J, Parker R (2002) A PUF family portrait: 3’UTR

regulation as a way of life. Trends Genet 18(3):150–157.24. Lee C-D, Tu BP (2015) Glucose-regulated phosphorylation of the PUF protein

Puf3 regulates the translational fate of its bound mRNAs and association with RNAgranules. Cell Reports 11(10):1638–1650.

25. Spassov DS, Jurecic R (2003) The PUF family of RNA-binding proteins: Does evolu-tionarily conserved structure equal conserved function? IUBMB Life 55(7):359–366.

26. Quenault T, Lithgow T, Traven A (2011) PUF proteins: Repression, activation andmRNA localization. Trends Cell Biol 21(2):104–112.

27. Miller MA, Olivas WM (2011) Roles of Puf proteins in mRNA degradation and trans-lation. Wiley Interdiscip Rev RNA 2(4):471–492.

28. Kershner AM, Kimble J (2010) Genome-wide analysis of mRNA targets for Caenorhabditiselegans FBF, a conserved stem cell regulator. Proc Natl Acad Sci USA 107(8):3936–3941.

29. Dubnau J, et al. (2003) The staufen/pumilio pathway is involved in Drosophila long-term memory. Curr Biol 13(4):286–296.

30. Wilinski D, et al. (2015) RNA regulatory networks diversified through curvature of thePUF protein scaffold. Nat Commun 6:8213.

31. Freeberg MA, et al. (2013) Pervasive and dynamic protein binding sites of the mRNAtranscriptome in Saccharomyces cerevisiae. Genome Biol 14(2):R13.

32. Lapointe CP, Wilinski D, Saunders HAJ, Wickens M (2015) Protein-RNA networks re-vealed through covalent RNA marks. Nat Methods 12(12):1163–1170.

33. Wang X, McLachlan J, Zamore PD, Hall TMT (2002) Modular recognition of RNA by ahuman pumilio-homology domain. Cell 110(4):501–512.

34. Miller MT, Higgin JJ, Hall TMT (2008) Basis of altered RNA-binding specificity by PUFproteins revealed by crystal structures of yeast Puf4p. Nat Struct Mol Biol 15(4):397–402.

35. Opperman L, Hook B, DeFino M, Bernstein DS, Wickens M (2005) A single spacer nucleotidedetermines the specificities of twomRNA regulatory proteins.Nat StructMol Biol 12(11):945–951.

36. ZhuD, Stumpf CR, Krahn JM,WickensM, Hall TM (2009) A 5′ cytosine binding pocket in Puf3pspecifies regulation of mitochondrial mRNAs. Proc Natl Acad Sci USA 106(48):20192–20197.

37. Hogan GJ, Brown PO, Herschlag D (2015) Evolutionary conservation and diversificationof Puf RNA binding proteins and their mRNA targets. PLoS Biol 13(11):e1002307.

38. Jiang H, Guo X, Xu L, Gu Z (2012) Rewiring of posttranscriptional RNA regulons: Puf4pin fungi as an example. Mol Biol Evol 29(9):2169–2176.

39. Taylor JW, Berbee ML (2006) Dating divergences in the Fungal Tree of Life: Reviewand new analyses. Mycologia 98(6):838–849.

40. Sugiyama J, Hosaka K, Suh S-O (2006) Early diverging Ascomycota: Phylogenetic di-vergence and related evolutionary enigmas. Mycologia 98(6):996–1005.

41. Wapinski I, Pfeffer A, Friedman N, Regev A (2007) Natural history and evolutionaryprinciples of gene duplication in fungi. Nature 449(7158):54–61.

42. Li L, Stoeckert CJ, Jr, Roos DS (2003) OrthoMCL: Identification of ortholog groups foreukaryotic genomes. Genome Res 13(9):2178–2189.

43. Bailey TL, Elkan C (1994) Fitting a mixture model by expectation maximization todiscover motifs in biopolymers. Proc Int Conf Intell Syst Mol Biol 2:28–36.

44. Mitchell A, et al. (2015) The InterPro protein families database: The classificationresource after 15 years. Nucleic Acids Res 43(Database issue):D213–D221.

45. Guindon S, et al. (2010) New algorithms and methods to estimate maximum-likelihoodphylogenies: Assessing the performance of PhyML 3.0. Syst Biol 59(3):307–321.

46. Edgar RC (2004) MUSCLE: Multiple sequence alignment with high accuracy and highthroughput. Nucleic Acids Res 32(5):1792–1797.

47. SenGupta DJ, et al. (1996) A three-hybrid system to detect RNA-protein interactions invivo. Proc Natl Acad Sci USA 93(16):8496–8501.

48. Hook B, Bernstein D, Zhang B, Wickens M (2005) RNA-protein interactions in the yeast three-hybrid system: Affinity, sensitivity, and enhanced library screening. RNA 11(2):227–233.

49. Licatalosi DD, et al. (2008) HITS-CLIP yields genome-wide insights into brain alterna-tive RNA processing. Nature 456(7221):464–469.

50. Zhang C, Darnell RB (2011) Mapping in vivo protein-RNA interactions at single-nucleotide resolution from HITS-CLIP data. Nat Biotechnol 29(7):607–614.

51. König J, et al. (2010) iCLIP reveals the function of hnRNP particles in splicing at in-dividual nucleotide resolution. Nat Struct Mol Biol 17(7):909–915.

52. Hafner M, et al. (2010) Transcriptome-wide identification of RNA-binding protein andmicroRNA target sites by PAR-CLIP. Cell 141(1):129–141.

53. Galgano A, et al. (2008) Comparative analysis of mRNA targets for human PUF-family proteinssuggests extensive interaction with the miRNA regulatory system. PLoS One 3(9):e3164.

54. Ihmels J, et al. (2005) Rewiring of the yeast transcriptional network through theevolution of motif usage. Science 309(5736):938–940.

55. Thompson DA, et al. (2013) Correction: Evolutionary principles of modular generegulation in yeasts. eLife 2:e01114.

56. Wang Y, Opperman L, Wickens M, Hall TMT (2009) Structural basis for specific recognition ofmultiplemRNA targets by a PUF regulatory protein. ProcNatl Acad Sci USA 106(48):20186–20191.

57. Lavoie H, et al. (2010) Evolutionary tinkering with conserved components of a tran-scriptional regulatory network. PLoS Biol 8(3):e1000329.

58. Quast C, et al. (2013) The SILVA ribosomal RNA gene database project: Improved dataprocessing and web-based tools. Nucleic Acids Res 41(Database issue):D590–D596.

59. Siddharthan R, Siggia ED, van Nimwegen E (2005) PhyloGibbs: A Gibbs sampling motiffinder that incorporates phylogeny. PLOS Comput Biol 1(7):e67.

60. Saldanha AJ (2004) Java Treeview–Extensible visualization of microarray data.Bioinformatics 20(17):3246–3248.

61. Dennis G, Jr, et al. (2003) DAVID: Database for Annotation, Visualization, and In-tegrated Discovery. Genome Biol 4(5):3.

62. Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: A practical andpowerful approach to multiple testing. J R Stat Soc B 57(1):289–300.

63. Koh YY, Wickens M (2014) Determining the RNA specificity and targets of RNA-binding proteins using a three-hybrid system. Methods Enzymol 539:163–181.

64. Langmead B, Salzberg SL (2012) Fast gapped-read alignment with Bowtie 2. NatMethods 9(4):357–359.

65. Uren PJ, et al. (2012) Site identification in high-throughput RNA-protein interactiondata. Bioinformatics 28(23):3013–3020.

66. Honda S, Selker EU (2009) Tools for fungal proteomics: Multifunctional neurospora vectorsfor gene replacement, protein expression and protein purification. Genetics 182(1):11–23.

Wilinski et al. PNAS | Published online March 20, 2017 | E2825

BIOCH

EMISTR

YPN

ASPL

US