Berry and pip form covariation in wild and domesticated grapevines: eco- evo-devo implications and archaeobotanical perspectives

1 ISEM, Univ Montpellier, CNRS, EPHE, IRD, Montpellier, France. Équipe « Dynamique de la biodiversité, anthropo-écologie », Place Eugène Bataillon CC065 34095 Montpellier Cedex 5, France. 2 Laurentian Forestry Centre, Natural Resources Canada, 1055 Du P.E.P.S. Street, P.O. Box 10380, Québec, Quebec, G1V 4C7 3 INRA, UMR 1334 AGAP, Equipe Diversité, Adaptation et Amélioration de la Vigne, F34060, Montpellier, France. & these authors contributed equally to this work @ corresponding author: bonhomme.vincent@gmail.com +33 (0)4 67 14 41 60

1. If grapevine phenotypic changes occurred during domestication and diversification processes are quite well known, particularly seed morphology used in archaeobiological studies, the functional causes and consequences behind these variations are still poorly understood. This study clarifies many aspects of size and shape (co)variation between pip and berry in both wild and domesticated Vitis vinifera. 2. The covariation of berry size, number of seeds per berry ('piposity'), pip size and pip shape were explored on 49 grapevine accessions sampled among Euro-Mediterranean traditional cultivars and wild grapevines. Consequences for archaeobotanical studies are discussed through the prediction of berry dimensions from archaeological pips from a Southern France Roman site. 3. For wild grapevine, the higher the piposity, the bigger the berry and the longer the pip.
For wild and domesticated grapevine, the longer is the pip, the more it has a « domesticated » shape. Such covariations allow to infer the berry dimensions in archaeological context where only the pips are recovered.

Introduction
Grapevine (Vitis vinifera L.) is the most important fruit species in the modern world (FAO, 2017) and acquired a central economical and cultural role since Classical times, particularly in the Mediterranean area (Brun, 2003;McGovern, 2007). Grapevine has long been cultivated for its fruits, mostly used to be transformed into wine but also consumed fresh or dried. Its wild progenitor, Vitis vinifera subsp. sylvestris, has likely been first domesticated in the South Caucasian area (Imazio et al., 2013), which provided the oldest wine making evidence (McGovern et al., 2017) dated to early Neolithic period (~8000 BP). The existence of other domestication centres has also been argued (Grassi et al., 2003;Arroyo-García et al., 2006).
Since the early times of domestication, grapevines varieties (or cépages) of Vitis vinifera subsp. vinifera have been selected, cultivated and propagated such as, today, several thousand varieties are recognized, and traditionally identified, by ampelography, the phenotypic description of grape varieties (Galet, 1988). Wild and domesticated grapes differ regarding their reproductive biology -wild plants being dioecious and cross-pollinated while most of the cultivars are hermaphrodite and self-pollinated, and an array of phenotypic changes with domesticated grape having bigger bunches and berries, higher diversity in their shape and skin colour, higher sugar content (Olmo, 1995;This et al., 2006) .
Tracing back in time and space the origins and spread of these changes is a big challenge.
Archaeobotanical material provides unique snapshots of ongoing domestication/diversification processes. Notwithstanding such material is often limited and frequently degraded, the quantitative and rigorous morphological description of preserved parts (mostly seeds), brought major insights into our understanding of the intertwined relationships between humans and domesticated plants in general (Zohary & Hopf, 1993; Morphometrics on archaeological pips have first been criticized for the overlapping of seed shapes and the differential deformation of wild versus domesticated pips shape that could make the shape of domesticated pips more similar to wild pips (Smith & Jones, 1990).
Experimental charring has allowed to validate resilience of identification even after distortions induced by charring (Ucchesu et al., 2016;Bouby et al., 2018). On the other hand, archaeological grapevine material is also found in waterlogged contexts and such conservation conditions possibly entail other, hardly testable, biases on shape.
It is well known that pips from wild and domesticate grapes differ in their form (that is size plus shape): wild grapes produce roundish pips with short stalks and cultivated varieties more elongated pips, with longer stalks (Levadoux, 1956). Such inter-(sub)specific, and also infra-specific, variations of form are now well known on archaeological grapevine pips (Terral et al., 2010;Bouby et al., 2013;Pagnoux et al., 2015). However, the functional causes and consequences (if any) behind these variations in seed form are still poorly understood. If size, shape and colour of berries are phenotypic traits that have been selected by humans, seed shape was likely not a direct target of selective pressures. The form of pips may possibly be affected by: the berry size; the number of pips per berry, that implies sharing space and embryonic resources; the growing environment, including cultivation practices; the status of the plant (wild or domesticated); the grapevine variety for Vitis vinifera subsp. vinifera; and developmental stochasticity. Previous works suggested that pip size and the number of pips per berry are positively correlated to berry size (Negrul, 1960;Galet, 1988;Houel et al., 2013). It was also evidenced that this correlation is stronger for wild grapevines than for domesticated varieties (Bouby et al., 2013). Additionally, Houel and colleagues (2013) advocate that although seed number, seed weight and berry flesh weight are correlated at the intragenotypic level in domesticated grapevine, there is no global correlation at the intergenotypic level.
A better understanding of the process of domestication relying on archaeological seeds goes hand in hand with a better understanding of the relationships between seed and berry forms. It is most likely that one of the first objective when cultivating grapevines was to obtain more numerous and bigger berries, as they are key yield factors for both wine and table production. It is therefore necessary to find out to what extent the shape changes observed in archaeological seeds imply changes in the size of berries. It is also important to know how this relation could be affected by simple cultivation of wild individuals and by domestication. populations cultivated "true" wild grapevines or "weakly" domesticated forms (Bouby et al., 2013;Pagnoux et al., 2015).
This paper aims at exploring how the form of berries and the pips they contain covariate.
A modern dataset composed of domesticated and wild grapevines allowed to compare patterns of covariation between these two Vitis vinifera compartments. To that respect, this paper considerably extents the previous work by Bouby and colleagues (2013) since it: i) enlarges the modern reference sampling, both in the cultivated and wild compartments; ii) tests the possible side effect of the number of pips per berry; iii) includes more berry measurements; iv) compares such length measurements with a geometric morphometric approach, namely outline analyses using elliptical Fourier transforms; v) tests the feasibility of archaeobotanical application.
This paper is divided into four parts: i) how does size (co)vary between pips and berries, and depending of the number of seeds?; ii) how does shape (co)vary between pips and berries and depending piposity?; iii) how much pip shape depends on berry size, number of seeds per berry, status, accession, and which practical consequences for archaeobotanical studies? iv) can we infer the dimensions of the (vanished) berry dimensions from the (recovered archaeological) pips?

Nomenclature
Hereafter, status designs compartment (domesticated or wild); accession designs the variety or cultivar (or cépage) for cultivated grapevines and the individual for wild grapevines; synecdochically, a domesticated/wild pip/berry refer to the accession they were collected from; usage designs the typical destination of cultivated grapevine (wine or table); cultivation designs for wild grapevines whether they were cultivated in collection or collected in natura; form is used when shape and size are used in combination; "piposity" is short for "given a pip, the number of pips in the berry where it was sampled". Archaeological material comes from wells of the Roman farm of Sauvian -La Lesse (US3022, US3063, US3171 and US3183) which has been extensively described elsewhere (Figueiral et al., 2015). The waterlogged conditions of the two wells allowed very good preservation of numerous uncharred pips (N=205) while dating of the archaeological layers was provided by numerous chronologically characteristic pottery and coin remains (2025( -1725. This material provided an archaeobotanical application.

Traditional measurements
On modern material, the berry diameter (berry Diameter ), height (berry Height ) and mass (berry Mass ) were obtained before dissection (Table 1). Mass was not available for 9 accessions; these accessions were removed for further analyses when mass was involved. Then the number of pips (hereafter "piposity") was recorded and, if not unique, one pip was haphazardly chosen.
In our dataset, a single berry from the variety "Kravi tzitzi" was found with 5 pips and the corresponding pip was discarded from further analyses. The dataset eventually consisted in 1469 pips (48×30 + 1×29) and was thus almost perfectly balanced between accessions.
Then, on each modern and archaeological pip, five length measurements were manually (Table 1, Figure 1) recorded using ImageJ by the same operator (L. Bouby): total length (pip Length ), length of stalk (pip LengthStalk ), position of the chalaza (pip PositionChalaza ), breadth (pip Breadth ) and thickness (pip Thickness ). All pips have been photographed in dorsal and lateral views by the same operator (T. Pastor) using an Olympus SZ-ET stereomicroscope and an Olympus DP 12 camera.

Statistical environment
All statistical analyses were performed using the R 3. measurements were log-transformed to focus on relative changes and minimize size differences (Bouby et al., 2013); the mass was cubic-root transformed for the same reason.
As preliminary analyses on modern material, differences between average piposity were tested using generalized linear model with Poisson error; differences in their distributions were tested using two-sided Fisher's exact tests on count data ( Figure 2).

Testing the covariation between pip and berry size in relation to the number of pips
On modern material, three sets of differences in pips and berries measurements were tested within and between piposity levels using Wilcoxon rank tests (not affected by log). First, differences were tested between: (1) domesticated and wild accessions and reported on Bivariate comparisons were then explored between the domesticated and wild accessions (piposity not considered), and tested using an analysis of covariance. When the "status" factor was found significant, separate regressions were then tested and, if significant, the adjusted r 2 was obtained and reported on Figure B (SI).

Testing the covariation between pip and berry shapes in relation to the number of pips
Outlines coordinates were obtained from photographs, centred, scaled, aligned and normalized for their first point before elliptical Fourier transforms (EFT). These preliminary steps removed positional, size, rotation and phasing differences between outlines and EFT could then be used without numerical normalization (Claude, 2008). EFT were performed on the two views separately, and the number of harmonics was chosen to gather 99% of the total harmonic power (8 for both views). Then 64 coefficients were obtained for each pip (2 views × 8 harmonics per view × 4 coefficients per harmonic).
To explore the overall variability of shape, a principal component analysis (PCA) was calculated on the full, raw matrix of Fourier coefficients ( Figure 4). The first two PCs were used as synthetic shape variables (See Results). To test the effect of piposity and pip dimension on pip shape, the same approach than for length measurements was used with PC1 and PC2 as the response variables. To test the relation between shape and pip length (only  Figure 5a). Then Wilcoxon rank tests were used to test shape differences between and within piposity levels ( Figure 5b).
To visualize shape differences between extreme piposity levels (1 and 4), and within subsets of interest, mean shapes for the dorsal and lateral views were calculated on the matrix of coefficients ( Figure 6). These differences were quantified with the mean absolute difference (MD) between each sets of Fourier coefficients. To make these differences meaningful, they were divided by the mean difference of Fourier coefficients between cultivated and wild accessions with all piposity levels pooled. For each subset, MD was thus calculated as (| coefficients subset, 1-pip -coefficients subset, 4-pips |)/(| coefficients domesticated, all pips -coefficients wild, all pips |). For example, a MD equals to 0 would indicate no difference between pips with a piposity of 1 and those with a piposity of 4; a MD greater than unit would indicate more differences relatively to differences that exist between cultivated and wild individuals ( Figure 6).

Pip shape and size in relation to status, accession and piposity
To evaluate the respective contribution of berry dimensions, accession and piposity onto pip shape, a multivariate analysis of variance was used according to the following model: all Fourier coefficients ~ berry Height + accessions + piposity within accession. Since it was highly correlated to other berry measurements, only berry Height was used to describe berry dimensions.
The contribution of each variable is the ratio of its sum of squares over the total sum of squares (thus including residuals). Again, this is tested on the different subsets of interest were also evaluated to compare their performance to classify the pips to their correct status and their original accession (Figure 8).
Given a combination of (status, accession) × (sizes+shape, sizes, shape), a leave-one-out cross-validation was used to assess status and accession classification accuracies. The principle of this method is to calculate a LDA on all-but-one pips of the sample and, then, to predict the classification of the omitted pip. This procedure is repeated for each pip of the sample. Classification accuracy was evaluated on all pips, to mirror the typical archaeobotanical context where the piposity cannot be known but also, on the four piposity levels ( Figure 8). To cope with unbalanced group structures, we also calculated a baseline for each subset that estimates the mean and maximum accuracy one can obtain by chance, using 10000 permutations (Evin et al., 2013). If the accuracy observed is higher than the maximum value obtained using permutations, then the LDA can be considered to perform better than random, with an estimated alpha below 10 -4 .

Predict the dimensions of the (vanished) berry dimensions from the (recovered archaeological) pips
To test whether berry dimensions can be inferred from the pip dimensions, separate multivariate regressions were calculated, on the reference material, for berry height and diameter (using the five length measurements on pips). For the sake of replicability, we decided to only use length measurements, that are straightforward to obtain from pips (whether modern or archaeological), which is not the case for Fourier coefficients. The difference between domesticated and wild grapevines regressions was first tested using an analysis of covariance: two regressions (one for cultivated, one for wild) were obtained for the berry height and two others for its diameter ( Figure 9). These four regressions were fitted using stepwise regression with backward elimination based on the AIC, and started with full models (berry Height /berry Diameter for wild/domesticated ~ pip Length + pip LengthStalk + pip PositionChalaza + pip Breadth + pip Thickness , all logged). Then, archaeological pips were classified into domesticated or wild using an LDA trained on the dimensions of modern pips. Pips with a posterior probability< 0.8 were filtered out. Finally, the berry height and diameter of this archaeological material were inferred using the corresponding models ( Figure 10).

Preliminary analyses on modern material
The average piposity, the number of pips per berry, is equivalent between domesticated and wild accessions (

Covariation between pip and berry size in relation to the number of pips
The average piposity, the number of pips per berry, was slightly lower for cultivated grapevines yet not different ( Figure A). Some pip dimensions were more contrasted between wild and cultivated grapevines than others (pip LengthStalk > pip PositionChalaza > pip Length > pip Thickness > pip Breadth ).
Differences were even more marked on berries, domesticated grapevines having larger berries than their wild counterparts (Wilcoxon one-tailed rank tests: all P ≤10 -16 ).
Whatever the piposity, the domesticated grapevines have the highest lengths (with four exceptions concerning the median, for 3-pips pip Breadth and for 4-pips pip Length , pip Breadth and pip Thickness but, differences were non-significant - Figure 3a). Pip LengthStalk and pip PositionChalaza are found different for all piposity levels. For pip Length , pip Breadth and pip Thickness are less contrasted between wild and domesticated grapevines, particularly when piposity is high. Overall, the highest the piposity, the lower the contrast between domesticated and wild. Indeed, for 1-pip berries, all lengths comparisons but pip Thickness were found different but, for 4-pips berries that is only true for pip LengthStalk and pip PositionChalaza (Figure 3a). Interestingly, the pip dimensions of wild grapevines increase faster with the piposity than their domesticated counterpart decrease.
Indeed, for 1-pip berries, all lengths comparisons but pip Thickness were found different but, for 4-pips berries that is only true for pip LengthStalk and pip PositionChalaza (Figure 3a).
In wild grapevines, larger berries have more and larger pips (Figure 3a). Berries with 2 pips have higher berry diameter, height and mass than those with a single pip (no exception, all P<10 -9 ; similarly for 3 versus 2 (for berry Mass , P was above the chosen alpha but with   (Figure 3b). With increasing piposity, table varieties tend to have bigger berries which does not seem to be the case for wine varieties, but differences between low (1+2) and high (3+4) piposity levels were not significant. For pips, the only significant differences between low and high piposities were found for wine varieties and for pip Breadth and pip Thickness (P<10 -16 ).
Wild in cultivation vs. wild in natura. Similarly, wild grapevine pips and berries appear to be bigger when in cultivation than their counterparts growing in natura (again some P above the chosen alpha yet the highest P was for 4-pips with P=0.005 - Figure 3c). Besides these global differences, trends of all measured variables are similar along increasing piposity. The berry mass ratios, relatively to the berry mass of wild collected in their habitat, were on average, 6.4 for wine varieties, 15.6 for table ones and 1.8 for cultivated wild,.
Bivariate comparisons ( Figure B, SI) indicate positive correlations between all pairwise dimensions (no exception among significant regressions). The total pip Length appears to be the most consistent variable, between domesticated and wild grapevines: indeed, only the correlation with the pip LengthStalk show a significant interaction. Inversely, the correlations implying the length of stalk always show a significant interaction and thus the need for separate regressions. For pips dimensions, the best correlations were found between pip Length and pip PositionChalaza (adj. r 2 =0.787) among those with non-significant interactions and between pip LengthStalk and pip PositionChalaza (adj. r 2 wild =0.615, adj. r 2 domesticated =0.717) among those with significant interactions. Compared to pips dimensions, correlations between berry dimensions were much better (all significant and adj. r 2 =0.864) and the three possible interactions were all significant.

Covariation between pip and berry shape in relation to the number of pips
The PCA obtained on the combined views shows that the first two PCs ( Figure 5) gathered 69% of the total shape variation, and higher rank components clearly levelling off further used as two synthetic shape variables. Convex hulls illustrate that shape differences between wild and domesticated grapevines are mostly captured on PC1. Nevertheless, scores on both PC1 (Wilcoxon rank tests, P<10 -16 ) and PC2 (P<10 -16 ) were different. PC1 is a synthetic shape variable that summarizes how prominent is the stalk and how round is the pip.
PC2 represents the circularity, a more global length/width ratio of pips, for the two views.
Average shapes reconstructed from the raw matrices of Fourier coefficients ( Figure 6) illustrate these results. The mean absolute difference (MD) confirms that larger changes between extreme piposities are observed within wild grapevines (particularly for cultivated ones) and reveals that most of these changes affect the dorsal side of the pips.

Pip shape and size in relation to status, accession and piposity; consequences for archaeobotanical inference
The respective contributions of berry height, accession and piposity on the shape of pips ( Figure 7) show that the accession is, overall and by far, the most correlated factor affecting pip shape. Among the different subsets, the accession factor has a higher impact on domesticated than on wild, and on cultivated wild accessions than on those collected in natura. By contrast, its contributions for wine and table domesticated varieties were similar. Here again, piposity and berry height both affect the pip shape of wild accessions but have a limited (piposity) and very low (berry height) contribution for domesticated accessions. Classification accuracies were compared using different training data and on different subsets (Figure 8). When different piposity levels were pooled, mirroring archaeobotanical admixtures, discrimination was very good at the status level ( Figure 8a). As expected, size + shape performed better (95%), than shape (93.7%) and size (92.5%) alone. When these models were evaluated on piposity subsets, they all have an accuracy above 91%, except for 4-pips berries. In all cases, accuracies were much higher than what could be obtained by chance alone. Accuracies were lower at the accession level ( Figure 8b) and when piposity levels were pooled size + shape (89.8%) outperformed shape alone (81.3%) and size alone (46.3%). The same model ranking was observed on piposity subset, except for 4-pips berries here again. Overall, accuracies were nevertheless much better than chance alone.

Application to the archaeological pips from the Roman site of Sauvian -La Lesse (Hérault, France): can we predict the dimensions of the (vanished) berry dimensions from the (recovered) pips?
On modern material, we used the size of pips to infer berry heights and diameters. Both regressions show a significant interaction of the status (berry Diameter : df=1, F=8075, P<10 -16 ; berry Diameter : df=1, F=7124, P<10 -16 ), and two regressions for berry diameter and two others for its height were obtained (Figure 9a, b). All were significant (all P<10 -16 ) yet the adjusted r 2 were quite low (berry Diameter adj. r 2 wild =0.583, adj. r 2 domesticated =0.468; berry Height adj. r 2 wild =0.609, r 2 domesticated =0.485). Final models all used pip Length + pip LengthStalk + pip Thickness + pip Thickness predictors, except berry Height for wild that used pip PositionChalaza "instead" of pip LengthStalk (Table 2). On unlogged berry diameter and height, the relative deviations were obtained (Figure 9a, b).
Mean relative deviation per accession for berry Diamater ranged from -13.8% to +10.5% for wild, and from -23.2% to +18.1% for domesticated; for berry Height they ranged from -13.3% to +13.7% for wild, and from -36.5% to +28.7% for domesticated. Then, these four models were applied on the archaeobotanical material after being classified at the wild/domesticated level using LDA. 46 pips (22%) were classified with a posterior probability < 0.8 and were filtered out. Among the remaining pips, 113 (71%) were classified as domesticated, 46 (29%) as wild.
When compared to their modern analogues (Figure 10), the length of "domesticated pips" were closer to those of wine varieties than

Discussion
This study contributes to understand Vitis vinifera phenotypic changes under domestication by clarifying many aspects of size and shape (co)variation between pip and berry in both wild and domesticated vines. It helps disentangle the interplay of the number of seeds per berry, berry dimensions, domestication, varietal diversity and cultivation practices onto the pip shape. We discuss implications for Vitis vinifera eco-evo-devo and perspectives for archaeobotanical studies, where size and shape are largely used to discriminate seeds from wild and cultivated vines, and for which a possible application is proposed.

Patterns of covariation between the form of the pip, the form of the berry and the piposity
With four ovules per ovary, the theoretical maximum number of pips per berry is four, yet one (over almost 1500) was observed with five. Such abnormal piposities have already been reported (T. Lacombe, pers. comm.). Most berries collected had two pips, and more than 70% of them had only one or two pips. This is in accordance with previous publication (Houel et al., 2013). There were no differences either between domesticated or wild (Figure 2), or between cultivated wild individuals and those collected in their habitat. One could have expected, on average, higher piposity, for domesticated (since they have hermaphroditic flowers), and more generally for cultivated accessions (since the pollen rain is expected to be lower, or even limiting, in the natural habitat).
Overall, and this is no surprise, wild pips and berries are smaller than their domesticated counterparts. Similarly pips and berries of wine varieties are smaller than those of table varieties, as pips and berries of wild vines collected in their habitat are smaller than those from cultivated wild individuals. This study clarifies the possibly confounding effect of piposity reported by previous studies (Negrul, 1960;Olmo, 1995;Bouby et al., 2013;Houel et al., 2013). Among vertebrate dispersed plants, the reward (the fruit pulp mass) associate with a given seed mass is commensurate with work required to move it, and expected to scale relatively (Edwards, 2006). For wild grapevine, and vertebrate dispersed plant species in 14 405 410 415 420 general, fruit and pip dimensions/masses is furthermore constrained by their dispersers and by a general trade-off between seed size/number (Leishman, 2001).
On the piposity side, and again for all but wine varieties, the higher the piposity the longer the pip and the bigger the berry in which they develop (Figure 3). For these groups, it seems that more numerous pips are not limited by space or nutrients but rather contribute the development of bigger berries. The berry development is well known (Dokoozlian, 2000;Ristic & Iland, 2005) and can be divided into two phases of enlargement. The first, prior to anthesis, is a period of rapid berry growth mostly due to cell division. After anthesis, berry growth is largely due to cell enlargement. Coombe (1973) reported that 17 doublings of cell number occurred in the ovaries before anthesis, compared to only two after anthesis. Ojeda et al. (1999) suggested that seed growth may also increases cell mitosis in the developing berry.
As suggested by Houel et al. (2013) auxins, cytokinins and gibberelins, upregulated shortly after fertilisation in grapevine ovaries, are likely to trigger fruit growth by cell expansion.
The absence of positive (or even negative) correlations between piposity, pip and berry dimensions for wine varieties remains unclear. Perhaps, the regulation, if any, is at the bunch or stock scale and, whether it has been selected (for example to concentrate sugars, aromas and flavours) or it is a by-product of another trait under selection for accessions destined to wine production. Since table varieties are bigger than wine varieties, the berry dimensions of the latter cannot be argued to have reached a developmental limit.
Finally, bivariate correlations concerning berry dimensions and mass are the best observed. This indicate robust allometries in berry sizes and mass, in other words that berry largely remain ellipsoid in shape, independently of their dimensions.

Morphometrics and domestication draw the curtain into grapevine eco-evo-devo
Domestication releases "natural" constraints by actively selecting for bigger seeds or fruits, for grapevine and for many cultivated plants in general (Fuller et al., 2014;Bonhomme et al., 2015), or by taking advantage of this trade-off with cultivation practices such as pruning. This may explain why cultivated wild accessions, have bigger berries for high piposities: the number of grapes is reduced, leading to bigger ones. Cultivation also releases most of the constraints of the wild, competition for water, taller plants competition for light, selfsupporting and climbing costs, etc.
Such evidence of plastic and canalized phenotypic expression may be fuel for further eco-evo-devo studies. The latter brings a conceptual and experimental framework that relies on environmentally mediated regulatory systems to better understand ecological and  (Sultan, 2007). Here, the norm of reaction of the pip size and shape, along increasing piposity and berry dimensions, is clearly different at the three investigated levels: between wild and domesticated, between wine and table, between cultivated wild and those collected in their habitat.
As seen from an eco-evo-devo perspective, domestication results in a change of phenotypic patterns that are selected or at least desirable. These phenotypic outputs arise from genetic background and signalling networks and that are rooted in phylogenetic history of the wild progenitor, constrained by physics, and modulated by environmental inputs. The aim of eco-evo-devo is to uncover the rules that underlie the interactions between an organism's environment, genes, and development and to incorporate these rules into evolutionary theory (Abouheif et al., 2014).

Consequences for archaeological inference
Taken independently or in pairwise comparisons, some pip lengths differences between wild and domesticated appear more "robust" to increasing piposities than others, notably the length of stalk (pip LengthStalk ) and the position of the chalaza (pip PositionChalaza - Figure 3a, Figure A, SI).
Overall shape variability, as captured by the first two principal components, also tend to distinguish the wild and domesticated grapevines as already observed elsewhere (Terral et al., 2010;Terral & Bouby, 2013). Yet the proper way to discriminate is using LDA and will be discussed below, the wild/domesticated differences are largely collinear with the largest shape variability components when an admixture of wild and domesticate pips is considered.
The piposity being positively correlated to pip length of wild accessions, the same pattern is observed: piposity is correlated with shape changes for wild accessions ( Figure 5).
Most of these changes affect the dorsal side of wild pips, that are much larger when piposity is high. This is illustrated and quantified in the Figure 6 where mean absolute difference indicate that there is ~70% as much difference, between 1-and 4-pips for wild accessions, than those observed between average wild and domesticated. Differences in extreme piposities are even larger than unit, that is larger than this "domestication gap", for the cultivated wild. This does not answer the question whether past vineyards cultivated "true" wild grapevines or "weakly" domesticated forms (Bouby et al., 2013;Pagnoux et al., 2015) but it points out a possible mechanism underlying this confusion.
Pip shape being largely used in archaeobotany, it is crucial to point out which factors contribute to its variability, or at least covary with it. In this study, the main factor associated to pip differences was, by far, the accession and it was even more important for domesticated accessions than for wild ones (Figure 7). Relatively to accession, berry height and piposity poorly contributed to observed differences. This first confirms the usefulness of shape and its robustness to identify varieties (or morphotypes that are shape archetypes). It may also indicate that domestication, by creating (or revealing) agrobiodiversity, chiefly in traits of interest, may also favour pip shape diversification whether genes controlling it may be linked to those directly selected or the product of drift, that is non-adaptive.
Here we show the reliability of classification, independently of piposity. Indeed, classification accuracies at the status level were all good (Figure 8), even when the models trained on the pip admixture where evaluated on piposity subsets. Shape was nonetheless superior to size alone but, when considered jointly, classification were improved. Whenever possible, size should thus be included along morphometric coefficients and used jointly in classification models. Accuracies at the accession levels evidenced even more clearly the latter conclusions: the superiority of shape over dimensions, and the benefit of using them jointly when it comes to infra-specific classification.
Shape is overall more robust than sizes when models were evaluated on piposity subsets.
The only exception, for both status and accession levels, were obtained on the 4-pips subset. This is likely due to the reported different piposity structure due to this 4-pips subset. This illustrates one limit of our experimental design, intended to reflect real-world admixtures rather than designing one where all levels of all studied factors would have been perfectly controlled and, primarily, balanced. That being said, such bias in the piposity structure is very unlikely to affect archaeobotanical identification when it includes shape, either at the (sub)specific level between wild and domesticated, or at the infraspecific level when one wants to identify accessions or morphotypes.

An application on archaeological material: inferring berry size from pip
Berry is home to the most selected traits, from the beginnings of domestication to varietal breeding and diversification times. Unfortunately, its dimensions cannot be quantified directly on archaeological material. The latter is indeed limited to pips, mostly charred or waterlogged; fleshy parts are usually absent, or too degraded to allow any morphometric approach. relative deviations from actual dimensions were nevertheless centered on zero and overall in the ±25% range.
Archaeological material from Sauvian -La Lesse presents an admixture of wild and domesticated type (Figueiral et al., 2015). Berry dimensions inferred from pips of this site are intermediate between the wild growing in their habitat and those cultivated ( Figure 10). This may suggest that past vineyards cultivated wild individuals. The berry dimensions inferred for domesticated varieties were closer to wine varieties than to table ones. This is congruent with the unequivocal wine production at this period and in this region (Figueiral et al., 2010(Figueiral et al., , 2015.

Conclusion
In our view, the main finding of this systematic exploration of berry and pip form covariation is that the more "pipose" is a wild grapevine, the bigger is the berry and the longer is the pip.
For both wild and domesticated, the longer is the pip, the more its shape looks like "domesticated". Further studies will clarify the contribution of cultivation practices contribution on pip shape, largely used in archaeobotanical studies to better understand viticulture history. These findings on grapevine pave the way for dedicated studies to shed light on genetic, functional and evolutionary changes that occurred in Vitis vinifera between the seed, its reproductive unit, and the berry, its dispersal reward and main target of its domestication and amelioration. More largely, domesticated taxa appear as model of choice to better understand this interplay for other organisms.  Table 1 Accessions used in this study. Dimensions are reported with mean±sd and given in mm, except for berry mass expressed in g.