A galling insect activates plant reproductive programs during gall development

Many insect species have acquired the ability to redirect plant development to form unique organs called galls, which provide these insects with unique, enhanced food and protection from enemies and the elements. Many galls resemble flowers or fruits, suggesting that elements of reproductive development may be involved. We tested this hypothesis using RNA sequencing to quantify the transcriptional responses of wild grapevine (Vitis riparia) leaves to a galling parasite, phylloxera (Daktulosphaira vitifoliae). If development of reproductive structures is part of gall formation, we expected to find significantly elevated expression of genes involved in flower and/or fruit development in developing galls as opposed to ungalled leaves. We found that reproductive gene ontology categories were significantly enriched in developing galls, and that expression of many candidate genes involved in floral development were significantly increased, particularly in later gall stages. The patterns of gene expression found in galls suggest that phylloxera exploits vascular cambium to provide meristematic tissue and redirects leaf development towards formation of carpels. The phylloxera leaf gall appears to be phenotypically and transcriptionally similar to the carpel, due to the parasite hijacking underlying genetic machinery in the host plant.

Many insect species have acquired the ability to redirect plant development to form unique organs called galls, which provide these insects with unique, enhanced food and protection from enemies and the elements. Many galls resemble flowers or fruits, suggesting that elements of reproductive development may be involved. We tested this hypothesis using RNA sequencing to quantify the transcriptional responses of wild grapevine (Vitis riparia) leaves to a galling parasite, phylloxera (Daktulosphaira vitifoliae). If development of reproductive structures is part of gall formation, we expected to find significantly elevated expression of genes involved in flower and/or fruit development in developing galls as opposed to ungalled leaves. We found that reproductive gene ontology categories were significantly enriched in developing galls, and that expression of many candidate genes involved in floral development were significantly increased, particularly in later gall stages. The patterns of gene expression found in galls suggest that phylloxera exploits vascular cambium to provide meristematic tissue and redirects leaf development towards formation of carpels. The phylloxera leaf gall appears to be phenotypically and transcriptionally similar to the carpel, due to the parasite hijacking underlying genetic machinery in the host plant.
Plant galls are unique organs formed in response to a parasite, which may be a virus, fungus, bacterium, nematode, or arthropod 1 . Among insects, the ability to elicit galls has evolved independently many times in six orders 1,2 . Insect galls are extremely diverse phenotypically, ranging from cell hypertrophy and hyperplasia of a single tissue to development of highly organized and complex organs comprising several tissue types with specialized functions [3][4][5] . Galls can develop on any plant tissue, but the great majority occur on leaves 2 . Leaf galls have been classified along a complexity gradient from simple folds and thickenings without tissue differentiation to curls, "pouches", and "covered" galls completely enclosing the insect inside a differentiated organ 1 . These more complex galls are highly divergent from the plant tissues on which they develop and share the production of a specialized "nutritive tissue" on which the insect feeds 1 . They may be as highly organized and complex as normal plant organs 1 . All insect galls are elicited via chemical signaling, although exactly how this is accomplished is poorly understood 1 .
Darwin was fascinated by insect galls, and wrote about insect galls that bear a striking resemblance to specific flowers or fruits, including conifer cones and peaches 6 . Indeed, many remarkable flower-and fruit-like traits are seen in galls formed by many insect families and orders on many plant species (Fig. 1). Darwin also noted the similarity between some galls and fruits in the number, complexity and arrangement of internal tissues 6 . These tissues include a nutritive layer rich in carbohydrates and proteins for the insect 1 much as nucellus or endosperm provide nutrition to plant embryos. A sclerotized capsule often protects the insect or plant embryo, and the surrounding cortex and epidermis can contain defensive chemistry 3,7 . Gall and fruit growth, development, and functions are under the direction of chemical signals from the non-host (hormones, in the case of the embryo) and are encoded by a set of transcriptionally co-regulated genes 8,9 . It is common for galling insects to infest and modify the development of flowers and fruits. They may displace the plant embryo and direct otherwise normal development 10 , develop in a manner similar to flowers or fruits 1,11,12 , or even revert to flowers if the insect dies 13 .   These observations have led to the hypothesis that insects eliciting complex galls recruit portions of the host plant's reproductive program to produce these necessary characteristics 1,[12][13][14][15] . We examined the hypothesis that a galling insect, grape phylloxera (Daktulosphaira vitifoliae Fitch 1855) co-opts portions of flower and/or fruit transcriptional pathways to produce a fairly complex leaf gall on grapevine (Vitis riparia) leaves. Assessing the degree to which a developing gall's transcriptome diverges from that of the vegetative leaf tissue on which it develops and specifically, the degree to which the gall's transcriptome is reproductive, should indicate to what degree the insect hijacks the underlying reproductive developmental programs in the host plant. We employed RNA sequencing (RNAseq) to characterize the transcriptome of this gall and the leaves on which it develops, sampling at four developmental intervals (Fig. 2). We confirmed that the insect reprograms leaf cell transcriptomes to direct gall development 9 . We asked whether genes typical of reproductive development from the decision to flower through meristem establishment and floral organ formation 16 , were significantly enriched among genes differentially expressed in the gall compared to the leaf. Results confirmed that phylloxera gall development engages portions, but not all, of the floral developmental programs in grapevine.

Results
Gall and ungalled leaf transcriptomes diverge significantly as the gall develops. We extracted RNA from phylloxera leaf galls on Vitis riparia at four intervals as they developed (Fig. 2). Aligning reads to the Vitis vinifera genome (Version 12 × ; Phytozome Version 7, Joint Genome Institute) allowed us to identify 26,346 grape transcripts expressed in either gall or leaf or both. Of these, 11,049 were differentially expressed ( > 1.5-fold, P < 0.01) at least once in galls compared with ungalled leaves (Fig. 3). Because the Vitis genome is not yet fully functionally annotated, we integrated Vitis transcripts with Arabidopsis thaliana TAIR v.9 functional annotations. This process produced 11,049 differentially-expressed transcripts we could potentially use for functional evaluation. The number of transcripts expressed differentially (DEGs) in galls compared with leaves increased dramatically as the gall developed, from 1,763 in stage 1 to 8,318 in stage 4 (Fig. 3).
The functional makeup of transcripts in developing galls was distinct from that in age-matched ungalled leaves. We assorted transcripts that were significantly up-or down-regulated in galls compared with leaves into gene ontology (GO) categories using the PANTHER classification system 17 . Significantly enriched GO categories related to reproduction were present throughout gall development but their number increased dramatically in later gall development stages (Table 1). Reproductive GO categories were enriched among both upregulated and downregulated DEGs throughout gall development, although more frequently among upregulated DEGs (Table 1).
Flowering pathways. Normal flowering is initiated at the shoot apical meristem (SAM) in response to environmental cues and endogenous signals via several major pathways [18][19][20][21] (Fig. 4). These may include the photoperiod, light quality/intensity, vernalization, gibberellin, and autonomous pathways. These flowering pathways are largely conserved among herbaceous plant species like Arabidopsis but can vary somewhat in woody plants 22 .
In grapevine, ambient temperature, light intensity, age and gibberellin (GA) are the primary influences on initiating flowering 22 . There is little evidence of photoperiod or vernalization impacts on flowering in grapevine 22 . We identified differentially-expressed genes from these pathways in our dataset by searching gene ontology categories We identified 162 Arabidopsis orthologs of known flowering related genes among 237 Vitis loci expressed in phylloxera galls and leaves via RNAseq (Supplementary Data S1). Of these, 123 putative genes (184 loci) were differentially expressed in galls. We identified the best-supported function of each DEG ortholog using information curated by TAIR 23 , UNIPROT 24 , and FLOR-ID 25 . We then used this information to infer each DEG's likely impact on flowering as expressed (up-or down-regulated) in the galls. This examination of the functions of the The particular way in which flowering transition is regulated in grapevine 22 led us to focus on orthologs related to ambient temperature/light intensity and GA signaling. Two orthologs that may be related to ambient temperature-or light-regulated flowering were differentially expressed in either the first or second gall stage. Both are normally flowering repressors. One was ZEITLUPE (ZTL), a flowering suppressor involved in light intensity and photoperiod signaling 26 ; it was downregulated. The other was PHOTOPERIOD-INDEPENDENT EARLY FLOWERING 1 (PIE1), which normally suppresses FLOWERING LOCUS C (FLC) expression to promote flowering 26 ; it too was downregulated. The number of orthologs potentially involved in ambient light or temperature signaling increased through gall development (Supplementary Data S1). However, each of the DEGs in this category acts by increasing the expression of FT, SOC1 or GI 26 . The expression of each of these was suppressed or unchanged in galls so that they could not promote flowering there (Fig. 4).
In grapevine, flowering is triggered by an absence or decline in GA signaling 27 and we observed differences in GA signaling and metabolism as galls developed. Sixteen DEGs (24 loci) from canonical flowering pathways involved in GA biosynthesis or signaling were upregulated in galls, while 26 genes (33 loci) were downregulated in galls (Supplementary Data S1). Upregulated biosynthesis DEGs included GIBBERELLIN 20-OXIDASE 1 and 2 (GA20OX1, GA20OX2), GIBBERELLIN  Since these DEGs may be flowering promoters or repressors, we tallied their likely impact on flowering as expressed. Sixty-two GA-related DEGs would promote flowering in Arabidopsis, while 53 DEGs in this list would delay or repress flowering in Arabidopsis as expressed in galls. Those numbers reverse if it is indeed true that GA signaling delays or prevents flowering in grapevine 22,27 . Thus, gene expression patterns provide little conclusive evidence about the role of gibberellins in gall development.

Flowering integrators.
The key to initiating flower development is activating the floral meristem initiator LFY 28,29 . In normal flowering, these regulatory pathways converge on a small set of floral pathway integrator genes 30,31 . In Arabidopsis and other plants, expression of one or more of these integrators must be increased or decreased to allow the floral meristem identity gene LFY to initiate flowering 32 . The key integrators are CONSTANS (CO), SUPPRESSOR OF OVEREXPRESSION OF CO 1 (SOC1), AGAMOUS 24 (AG24), FLOWERING LOCUS T (FT), and FLC. CO and FLC integrate signals from the various pathways and promote or inhibit flowering via their impact on expression of FT and SOC1/AGL24, which in turn activate LFY [33][34][35] (Fig. 4). The influence of the GA pathway is mediated by SOC1 and the GAMYB transcription factor MYB33 36,37 .   Four of the flowering integrators we found were upregulated in galls while 11 were downregulated or unchanged in galls (  Sources of meristem. Flowering is normally initiated at the apical meristem in response to the signaling pathways and integrators described above. Since gall development is a form of de novo organogenesis, it presumably requires stem cells as a starting point 39,40 . While some galls appear on apical buds, phylloxera galls, and many others, form on leaves or stems. In leaves, the only meristem is (pro)cambium from which vascular tissue is derived.

GO Term Search Term
To determine whether cambial meristem might provide a foundation for gall development, we examined the expression of genes in GO categories specific or related to vascular cambium formation and activity: GO:0010067 procambium histogenesis, GO:0010065 primary meristem tissue development, GO:0010305 leaf vascular pattern formation, GO:0001944 vascular development, GO:0010087 phloem or xylem histogenesis, and GO:001005 xylem and phloem pattern formation. We found 96 orthologous loci from these categories differentially expressed in galls; expression of 44 genes (54 loci) was significantly greater in galls than leaves (  (Fig. 5). Also activated in galls were three DEGS involved in auxin synthesis, TRYPTOPHAN AMINOTRANSFERASE 1 (TAA1) and TRYPTOPHAN AMINOTRANSFERASE RELATED 2 (TAR2), plus YUCCA6 (YUC6), which controls the formation of vascular tissues as well as floral organs in Arabidopsis 49 (Fig. 5). Cambium-related cytokinin signaling in galls was suggested by elevated expression of CYTOKININ RESPONSE 1 (CRE1) (Fig. 5). However, cytokinin activators LONELY GUY 1 and 3 (LOG1, LOG3) were downregulated in late stage galls ( Fig. 5; Supplementary Data S3).
The divergence in expression of cambium-related genes in gall and leaf as they developed exhibited several different temporal patterns (Fig. 6). Expression of many genes declined in both leaves and galls as they aged, but less rapidly in galls, producing statistically significant differences by gall stage 4 (Fig. 6). In a second pattern, gall values showed little or no decline with development and more or less exceeded leaf values over the entire course of development (Fig. 6). A third pattern involved gall values that declined more precipitously than values in leaves (Fig. 6).

Vegetative-to-reproductive transition. LFY is the master initiator of floral meristem development and
indicator of the vegetative-to-reproductive meristem transition 29,49 . VFL, the grape homolog of AtLFY, functions similarly in the grapevine flowering transition 50 . Having established that the canonical flowering pathways appear unlikely to trigger elements of flower development in galls, we examined expression of flowering triggers by identifying DEGs in our gall data set found in the GO category vegetative to reproductive phase transition of meristem (GO:0010228). We found altered expression of 76 genes (93 loci) from that GO category in developing galls ( Fig. 3; Supplementary Data S4). Thirty-nine genes (43 loci) would promote the transition to flowering in Arabidopsis if they were expressed as they were in galls, while 30 DEGs (42 loci) would repress it.
While all 76 genes were differentially expressed in the gall stages 3 and 4, several genes involved in the vegetative-to-reproductive transition were also expressed in the earliest stages (Supplementary Data S4). These included AGL6, PROTEIN ARGININE METHYLTRANSFERASE 10 (PRMT10), one locus of COL9, and GA20OX1, all of which were upregulated in the youngest galls. Genes that were downregulated early include SUPPRESSOR OF PHYA-105 1 (SPA1) and DNAJ HOMOLOGUE 3 (J3) (Supplementary Data S4). However, these and many other DEGs in this category act as part of, or together with, one or more canonical flowering pathways and depend for their influence on the flowering integrators we found inactive or downregulated (Fig. 4).
The same vegetative-to-reproductive transition DEG set included meristem transition triggers not affiliated with the canonical pathways and their integrators. Krizek 51 and Yamaguchi et al. 29 have described an auxin-responsive pathway in Arabidopsis leading to flowering, dependent on AINTEGUMENTA (ANT), AIL6 and LFY (Fig. 4). They showed that ANT and AIL6 expression is elevated in response to auxin, and that they in turn activate LFY to initiate flowering. Auxin sources include polar transport involving PIN1, as well as synthesis by members of the YUCCA (YUC) family 51 ; expression of both was elevated in developing galls (Supplementary Data S4). Krizek 51 implicated auxin response factors ARF3 and ARF4 in this signaling network. We found elevated expression of ARF2, ARF3, ARF4 and ARF6 orthologs in developing galls (Supplementary Data S4). In Arabidopsis, ARF4 is a target of LFY 52 and regulates polarity 53 , ARF6 regulates gynoecium maturation, and ARF2 and ARF3 are involved in carpel and ovule development 54,55 . All of the elements of auxin-triggered transition to flowering were activated in developing galls (Fig. 4).
An age-based pathway to flowering transition was also active in developing galls (Fig. 4). Plants must mature over some period of time before they become competent to flower 56 . Grapevine generally requires 3-6 years before it can reproduce 57 . As plants age, the expression of micro RNA miRNA156 decreases. miRNA156 suppresses expression of the transcription factor SQUAMOSA PROMOTER BINDING-LIKE 9 (SPL9), which is a promoter of LFY expression. As miRNA156 activity decreases, SPL9 expression increases, and eventually increased LFY expression triggers flowering, independent of the canonical flowering pathways. While we could not assess miRNA abundance or activity using our methods, the expression of SPL9 increased significantly in galls as they aged; this increase could promote the flowering process in galls.
Some gall DEGs found in GO:0010228 influence the flowering transition via pathways or genes that were not found to be activated in galls.  While LFY is the master regulator and indicator of floral meristem development, it also triggers the transcription of key components of flower organ determination through its interaction with AGAMOUS (AG) 59 . We found that LFY expression was significantly elevated in gall stage 4 (Table 1; Fig. 4; Supplementary Data S5), whereas its target AG, which terminates meristem activity so that floral organogenesis can proceed 59   downregulated in galls. It interacts with LFY but is not necessary for flower formation 61 . Upregulated REBELOTE (RBL) contributes to floral meristem termination so as to prevent the formation of supernumerary flowers or floral organs 62 . The activation of LFY and AG in developing galls should set the stage for flower organ development.
To determine how carpel development and related genes might be involved in gall formation, we examined the expression of all unique genes from ontology category gynoecium development (GO:0048467), augmented with a list developed by Reyes-Olalde 63  Once the vegetative-to-reproductive transition has been achieved, AG interacts with floral homeotic genes to regulate floral organ development in Arabidopsis and other species (Fig. 4) 67 . Proteins encoded by a small number of homeotic genes interact in a combinatorial way to determine each of the major floral organs: sepals, petals, stamens, and carpel 67 . The homeotic genes required to produce these structures have been classified A, B, C, D, or E 67 . We found no differential expression of orthologous homeotic genes from class-A or -B ( Fig. 4; Supplementary Data S5). However, orthologs of the class-C carpel identity genes AG 68 and SHATTERPROOF 1 (SHP1) 69,70 were strongly upregulated in gall stages 3 and 4 compared with leaves (Supplementary Data S5). Class-C proteins interact with class-E proteins to direct development of the floral organs 71 (Fig. 4). In Arabidopsis, the major class-E genes comprise the SEPALLATA family 72 . The combination of AG and SEPx is required to produce a carpel 68 . SEPALLATA 1 (SEP1) was strongly upregulated in galls (Supplementary Data S5). The protein encoded by AGL6, which was strongly upregulated throughout gall development, also fulfills SEPx functions in some plant species 73 . All the elements necessary for flower development, from activated LFY through AG expression to elevated transcripts for SEP1 and AGL6 are present in phylloxera galls.

Discussion
We found that gall and leaf transcriptomes differ at the earliest point in gall development, and diverge increasingly as galls and leaves develop. The transcription of many grape genes orthologous or homologous to genes responsible for triggering flowering and regulating flower development in Arabidopsis is altered in phylloxera leaf galls. The general pattern is that expression of these genes, many of which have little or no role in the development of the leaf on which the gall grows, is up-or down-regulated in ways that could lead to flowering and eventual fruiting. Expression of many floral repressors were found to be downregulated, while promoters were upregulated. The frequency of differentially-expressed flowering genes increased dramatically as the gall developed and the leaf matured.
The number of genes differentially expressed during gall development was rather large: 11,049 were differentially expressed (> 1.5-fold, P < 0.01) at least once in galls compared with ungalled leaves. There are likely several reasons for this. First, flowers and fruits are complex organs whose development can involve differential expression of many genes. For example, Ramos et al. (2014) 74  Flowering requires a transition from a vegetative state to the reproductive meristematic state. This transition is elicited by the influence of environmental or hormonal signals on a few key floral integrator genes, which in turn increase the expression of the master regulator LFY to establish a floral meristem and promote flower development 32 . Indeed, ectopic LFY expression is sufficient to produce flowers in the absence of repressors 80 .
Overexpressing these genes as well as other flowering genes in Arabidopsis can produce ectopic flower development, particularly carpels, or gall-like morphological changes 49,71,81 . LFY expression was significantly elevated in late stage galls. We therefore consider the expression of LFY and its targets a key step if gall development involves aspects of flower development, and we so investigated all of the ways in which LFY expression could be elicited.
Flowering pathways. We first asked whether phylloxera could be exploiting the canonical pathways that culminate in activating LFY to trigger flowering (Fig. 4). We found that the differences between galls and leaves in the expression of the many grape orthologs of Arabidopsis genes in those pathways were mixed. Some pathway orthologs were expressed in ways that would prevent their impact on LFY while expression of others could promote LFY expression. Expression of genes in the gibberellin pathway was consistent with GA's role in normal flower promotion in many plant species. For example, orthologs of many GA biosynthesis and response genes were upregulated in late gall stages while catabolic genes were downregulated ( Fig. 4; Supplementary Data S1). However, GA signaling suppresses flowering in grapevine 27 so positive GA signaling could prevent flower development as part of gall development. The impact of GA signaling on flower development may depend on signaling by the GAMYB transcription factor MYB33 36,37,82 . MYB33 was downregulated in late stage galls, in principal blocking GA signaling. Since gibberellin's influence on flowering switches from negative to positive during flower development 83 , a more detailed study of the timing of GA signaling will be needed to determine its role in gall development.
Overall, we did not find convincing evidence that gall elicitation or development depends on the canonical flowering pathways as they normally function in flowering.

Flowering integrators.
Signaling by all the canonical flowering elicitation pathways converges on a few integrating genes 18,58 . These integrators in turn elevate LFY expression to bring about the meristem transition to flowering and flower development 49 . The only integrator gene expressed in galls in a way that would influence LFY expression was the floral repressor FLC. However, FLC's impact on flowering comes about when it is downregulated and its repression of FT and SOC1/AGL24 is stopped. Expression of FT and SOC1/AGL24 was unchanged or decreased in galls as compared with leaves. This fact alone would appear to rule out most or all canonical environmental signaling pathways as gall elicitors (Fig. 4).
Vegetative-to-reproductive transition. Indicators of a meristematic transition from vegetative to reproductive state were conspicuous in galls. Genes involved in floral meristem identity and/or maintenance were upregulated in gall stages 3 and/or 4, including LFY, AG, FUL, CAL, and UNUSUAL FLORAL ORGANS (UFO) (Fig. 3). One exception was AP1, which was unchanged. AP1 expression is not associated with flowering in grapevine 84 . AG has a dual role in floral meristem identity early and meristem termination plus organ differentiation later 85 . TFL and relatives, which repress floral meristem formation, were unchanged or downregulated in late gall developmental stages.
We found evidence suggesting that localized auxin signaling could play a role in the vegetative-to-reproductive transition and gall development. Auxin signaling mediated by auxin response factors (ARFs) and acting via expression of ANT and AIL6 can elevate LFY expression and lead to flowering transition and flower development in Arabidopsis 29 . All of the orthologs in this short pathway were significantly elevated in late gall stages, as were other auxin-responsive signaling and biosynthetic genes (Fig. 4). Phylloxera could initiate flowering processes via local elevation of auxin concentrations or signaling.
An age-related flowering pathway could also be involved in gall development. Like most woody plants, juvenile grapevines require a maturation period of several years before becoming reproductively competent. During this time, the expression of microRNA miR156 declines, and its suppression of SPL9 decreases. Increasing expression of SPL9 then provokes LFY expression to trigger flowering 56 . SPL9 expression was significantly elevated (about 2-fold) in late stage galls, but our methods provided no evidence concerning miR156 expression. Medina et al. 86 found that miR159 played a role development of galls elicited by the root-knot nematode Meloidogyne incognita. The potential role of microRNAs in insect gall development warrants further attention.

Sources of meristem.
As in normal flowering and organogenesis, gall development requires undifferentiated stem cells. Normal flowering is initiated at the SAM in response to hormonal and/or environmental cues. There is no SAM in plant leaves, but we found evidence that vascular cambial meristem remains active in galls long after it declines in the leaf and so is a possible source of stem cells for exploitation by the insect (Fig. 5). Phylloxera galls (and many others) are always associated with leaf veins and may obtain undifferentiated cells there from which to develop a novel organ. Expression of the key cambial activation genes, CLE44, PXY, and WOX4 41,42 , was significantly elevated in galls compared with leaves as leaves and galls aged (Figs 5 and 6). We found elevated expression of genes associated with hormonal signaling normally involved in cambium activation, and reduced expression of genes that terminate cambium activity and promote vascular differentiation well after leaves and their vasculature were mature (Fig. 5). While activated cambium could reflect increased vascular development, phylloxera galls do not exhibit increased vascularization 87 . The gall transcriptome is consistent with phylloxera manipulating vascular cambium to provide stems cells for organ development.
Expression of the CLE, PXY, and WOX cambial activation pathway is also key to development of the gall-like structures elicited by root-knot and cyst nematodes and nodulation by Rhizobium in legumes 88 , suggesting that phylloxera and other parasites have converged on altering developmental regulation of vascular stem cells during gall elicitation. We are aware of no studies of CLE peptide production in phylloxera or other insects, as has been shown to be important for root-galling nematodes 88 .
Flower development. We found transcriptional indications of flower development, including organ determination, in the phylloxera galls. Many orthologs of genes that positively regulate flower development were upregulated. Differential expression of grape orthologs of canonical Arabidopsis "ABCE" model homeotic genes that determine floral organ identities was significant for class-C genes. Class-C AG protein normally associates with class-E SEP proteins to determine carpel identity 89 , and expression of Vitis AG, SHP and SEP1 orthologs was strongly upregulated in galls (Fig. 4). We also found enhanced expression of the grape ortholog of AGL6, a close relative that plays a SEP role in several other species 73,90 and has an ancestral role in carpel identity 85  genes, was also elevated in galls. The phylloxera gall resembles a carpel more than any other floral organ transcriptionally, anatomically and functionally. The view that galls are convergent on carpels or fruits is supported by diverse observations from other studies. At least one gall's nutritive layer includes proteins normally found only in the seed 92 . Gall development and growth of the nutritive layer depend on chemical cues from the insect 93 , much as embryos direct development of surrounding tissues hormonally 13 . Interestingly, rolled leaf edges are considered ancestral elements in the evolution of both insect galls and in the origin of the carpel 94 .
The absence of evidence for signaling from the canonical flowering pathways led us to examine other means by which flowering can be elicited. There are many paths to flowering, some of which are independent of environmentally-cued pathways. All of the known pathways generally culminate in hormone-regulated gene expression. Most of the major plant hormones have been found to play some role in flowering, and their interaction during flowering and flower development is complex. Our results and these observations suggest that direct provision or manipulation of phytohormones is the most plausible means of gall elicitation, although we cannot rule out the injection of CLE peptides or small RNAs, which has not been described in insects.
The idea that galling insects somehow manipulate plant hormones to accomplish their ends is very old, and accumulation of various hormones in galls has been reported frequently 95 . LFY responds to both GA and auxin 96,97 . Manipulating signaling by one or more of these hormones would seem a likely way for galling insects to trigger flowering programs in producing a gall. Our results, which found elevated expression of auxin-responsive genes and auxin transport genes in the galls, suggest an important role for auxin in phylloxera gall formation. The requirement for local auxin accumulation to prompt organ development, including flowers, is well established 51,96,97 . On the other hand, our results suggest that gibberellin signaling may be suppressed in developing galls, which could stimulate reproductive development at gall sites in grapevine 27 . Definitive resolution of hormone signaling in gall development will require an integration of detailed chemical and transcriptional analyses.

Limitations to this study. Our conclusions are based on the assumption that similarity between computed
Arabidopsis and Vitis protein sequences suitably indicates similarity in function for a given gene in both Vitis and Arabidopsis. While we are confident in the assignment of orthologs between the two species, this assumption about functional similarity is no doubt more valid for some genes than for others due to expansion of some gene families in Vitis and sub-or neo-functionalization. Network level rewiring may have altered activator and repressor roles of transcriptional regulators in Vitis compared to Arabidopsis 98 . Thus, even when gene families are of similar size, there is no guarantee of a one-to-one function concordance. For example, the key floral meristem gene LFY is expressed in a wider range of situations in grapevine than is the Arabidopsis LFY 50 . While floral meristem indicator AP1 is key to the development of flowering competence in Arabidopsis, that is not true in Vitis, where its impact is restricted to tendrils 84,99 . On the other hand, the functions and expression of many of the reproductive genes we identified in galls, such as AG, SHP, the hormone signaling elements, the pathway integrators, and others are highly conserved among plant species and exhibit the same or similar expression patterns in grapevine 18 .
We also did not identify putative orthologs for all genes using the current methods. While we might find more matches through broader searches, we are missing very few important flowering genes, and none that would significantly change our conclusions. Our conclusions are based entirely on transcriptional data, and ignore post-transcriptional and other regulatory mechanisms. In particular, our methods did not allow an assessment of the impact of small RNAs, which are important regulators of many reproductive genes including those we studied 31 .
Our conclusions are also based on statistically significant differences in the numbers of RNA transcripts between gall and leaf tissues. Very few genes were present in one tissue and not the other, despite the fact that many are involved in flower development. It is important to remember that flowers are modified leaves, evolutionarily 100 . Phylloxera galls are not flowers or fruits, but their transcriptomes show greater commitment to flowering than do ungalled leaf tissues; they are neither flowers nor leaves, but are unique organs incorporating traits of both.
In summary, we have shown that phenotypic similarities between galls and fruits extend to their transcriptomes. The likely reason for this is that the plant embryo and galling insect have similar requirements for success and manipulate plant development similarly to achieve similar goals. Both need the conditions provided by an expanded carpel. The patterns we obtained support the hypothesis that the phylloxera leaf gall -and probably other similar galls -is developmentally and transcriptionally convergent on floral organs, particularly the carpel.
Methods study system. Grape phylloxera (Daktulosphaira vitifoliae Fitch 1855) is an aphid relative, native to North America, that feeds on leaves and roots of certain Vitis species. It elicits complex galls on abaxial leaf surfaces, and causes swelling on roots when feeding there. Its life history and gall development have been described by Sterling 101 (Fig. 2). Females emerge from eggs in the spring and feed from the upper surfaces of the youngest leaves, sucking contents from parenchyma cells beneath them. Within 24-48 hours a disk-shaped depression forms under the feeding insect. Cell division and expansion are altered at the disc margins and soon a circular ridge or wall surrounds the feeding insect. Within 48-72 hours the abaxial depression containing the insect deepens due to differential cell division and expansion and the adaxial wall closes over her, leaving a narrow opening protected by dense trichomes. Two tissue layers several cells thick underlie the depression, an inner layer that is densely cytoplasmic and an outer layer that contains larger vacuoles, enlarged nuclei and nucleoli, and cytoplasmic globules. These 'secretory' characteristics spread to other cell layers, becoming a thick 'nutritive zone' . Development of a complete gall takes 4-5 days, at which point the insect has matured and begins producing eggs. "Crawlers" hatch from eggs in the gall, exit through the abaxial opening, and proceed to feed and form galls on younger leaves. Gall development stops if the insect is removed before this final stage. tissue sampling. Galled (Fig. 2) and dissected on ice; midribs were removed from ungalled control leaves. Because the two earliest gall stages developed on the same leaves, there were only three control leaf size classes matched to the four gall stages.
To obtain enough RNA, samples were pooled from three individual vines, producing three biological replicates for each of the four gall developmental stages and three control leaf sizes (i.e., originating from twelve independent grapevines for galled tissue, and nine for control leaves). All tissues were immediately frozen in liquid nitrogen and stored at −80 °C. Illumina read processing and expression quantification. A custom Perl script was used to parse the libraries and remove barcode sequences resulting in approximately 40.9 million reads per biological replicate for the ungalled library and 40.3 million reads per biological replicate for the galled library. NextGENe V2.3.3.1 (SoftGenetics, LLC., State College, PA, USA) was used to quality filter the fastq data, remove reads with a median quality score of less than 22, trim reads at positions that had three consecutive bases with a quality score of less than 20, and remove any trimmed reads with a total length less than 40 bp. The reads were aligned against Vitis vinifera V2 genome (DOE-JGI; ftp://ftp.jgi-psf.org/pub/compgen/phytozome/v9.0/Vvinifera/), thus eliminating transcripts originating from phylloxera insects. Gene expression was quantified using TopHat/Cufflinks software 102 . Differential expression between galled and ungalled leaf tissue was analyzed for each mapping, using two discrete probability distribution based methods, DESeq and edgeR (https://bioconductor.org) and the annotated Vitis vinifera V2 genome (DOE-JGI; ftp://ftp.jgi-psf.org/pub/compgen/phytozome/v9.0/Vvinifera/). Read counts and RPKM values (reads per kilobase per million) were calculated for each library. An RPKM cutoff of 0.1 per gene model was applied for comparing expression values. Functional analyses were limited to genes with a differential expression significance < 0.05 and > 1.5-fold difference. Fold-change between galls and their respective ungalled control leaves was calculated for each gene by subtracting the base-2 logarithm of the RPKM value of galls to the base-2 logarithm of the RPKM value of ungalled control leaves.

RNA
Genome-wide syntenic analyses were performed to identify Arabidopsis thaliana -Vitis vinifera orthologs using CoGe (http://genomevolution.org/CoGe/). In addition, Arabidopsis -Vitis orthologs were identified using reciprocal BLASTp analyses (protein databases) with a 0.00001 p-value cutoff resulting in the annotation of ~86.7 % of all coding sequences in the Vitis vinifera V2 genome.
Gene Ontology (GO) enrichment analyses were performed for each of the gall and leaf gene expression sets using the PANTHER classification system 17 . Statistical significance for enrichment scores was set at < 0.005.  Table S1). PCR reaction parameters were optimized with qPCR using a MJ Research Opticon2 PCR thermal cycler (Bio-Rad, Hercules, CA, USA), with iQSYBR Green Supermix (Bio-Rad #170-8882; Bio-Rad, Hercules, CA, USA). Droplet digital PCR (ddPCR) reactions were performed on the Bio-Rad QX100 ddPCR System using using QX200TM ddPCR ™ EvaGreen Supermix (Bio-Rad #1864034; Bio-Rad, Hercules, CA, USA). Primer sequences, cDNA dilution and volume, and annealing temperature for each gene tested by ddPCR are presented in Supplementary Table S1. Six biological replicates per each of seven tissue types were used for ddPCR analysis.

Validation of
We used two abundantly expressed neutral genes whose expression was uniform across all gall and leaf samples, orthologs of AtDEC and AtDNAJ, as internal controls to normalize the amount of starting RNA used for RT-PCR for all samples (N = 6 biological replicates per developmental stage for both galls and ungalled control leaves). Normalized gene copies for each gene were calculated by dividing their absolute gene copies by the average gene copies of the two neutral genes. Fold-change between galls and their respective ungalled control leaves was calculated for each gene by subtracting the base-2 logarithm of the normalized gene copy of galls to the base-2 logarithm of the normalized gene copy of ungalled control leaves followed by one-way ANOVA. ddPCR results for these genes were consistent with results obtained via RNAseq (Fig. 7).

Data Availability
RNAseq data that were generated for this study are available at NCBI Gene Expression Omnibus (https://www. ncbi.nlm.nih.gov/geo/) under study accession GSE118569. The authors declare that all other data supporting the findings of this study are available within the article and its Supplementary Information files, or are available from the authors upon request.

Figure 7.
Concordance of expression differences for selected genes between galls and leaves found via ddPCR and RNAseq. (A) Normalized sequence concentrations obtained using ddPCR. Asterisks identify gall/leaf comparisons that are > 1.5-fold difference. (B,C) Respectively, ddPCR and RNAseq differences for the same genes as presented in (A). Values are log 2 (-fold differences). Genes in orange were upregulated in galls, genes in blue were downregulated, and expression of genes in white was unchanged.