Angiosperms employ an astonishing variety of visual and olfactory floral signals that are generally thought to evolve under natural selection. Those morphological and chemical traits can form highly correlated sets of traits. It is not always clear which of these are used by pollinators as primary targets of selection and which would be indirectly selected by being linked to those primary targets. Quantitative genetics tools for predicting multiple traits response to selection have been developed since long and have advanced our understanding of evolution of genetically correlated traits in various biological systems. We use these tools to predict the evolutionary trajectories of floral traits and understand the selection pressures acting on them. We used data from an artificial selection and a pollinator (bumblebee, hoverfly) evolution experiment with fast cycling Brassica rapa plants to predict evolutionary changes of 12 floral volatiles and 4 morphological floral traits in response to selection. Using the observed selection gradients and the genetic variance-covariance matrix (G-matrix) of the traits, we showed that the observed responses of most floral traits including volatiles were predicted in the right direction in both artificial- and bumblebee-selection experiment. Genetic covariance had a mix of constraining and facilitating effects on evolutionary responses. We further revealed that G-matrices also evolved in the selection processes. Overall, our integrative study shows that floral signals, especially volatiles, evolve under selection in a mostly predictable way, at least during short term evolution. Evolutionary constraints stemming from genetic covariance affected traits evolutionary trajectories and thus it is important to include genetic covariance for predicting the evolutionary changes of a comprehensive suite of traits. Other processes such as resource limitation and selfing also need to be considered for a better understanding of floral trait evolution.

中文翻译：

---

油菜的花信号在人工和传粉媒介选择下以可预测的方式演化

被子植物利用了令人惊讶的视觉和嗅觉花信号，这些信号通常被认为是在自然选择下进化的。这些形态和化学特征可以形成高度相关的特征组。目前并不总是清楚其中哪些被传粉者用作选择的主要目标，哪些将通过与这些主要目标联系而间接选择。用于预测多种性状对选择的反应的定量遗传学工具已经开发了很长时间，并增进了我们对各种生物系统中遗传相关性状进化的理解。我们使用这些工具来预测花性状的进化轨迹，并了解作用在它们身上的选择压力。我们使用来自快速循环芸苔属植物的人工选择和传粉昆虫（熊蜂、食蚜蝇）进化实验的数据来预测 12 种花挥发物和 4 种花形态性状响应选择的进化变化。利用观察到的选择梯度和性状的遗传方差-协方差矩阵（G 矩阵），我们表明，在人工选择实验和大黄蜂选择实验中，大多数花卉性状（包括挥发性成分）的观察到的反应都以正确的方向预测。遗传协方差对进化反应具有限制和促进作用。我们进一步揭示了 G 矩阵也在选择过程中进化。总体而言，我们的综合研究表明，花香信号，尤其是挥发性成分，在选择下以大多数可预测的方式进化，至少在短期进化过程中是这样。遗传协方差产生的进化约束影响性状的进化轨迹，因此包含遗传协方差对于预测一系列性状的进化变化非常重要。为了更好地理解花性状的进化，还需要考虑其他过程，例如资源限制和自交。