Early terrestrial plants colonizing land probably relied on arbuscular mycorrhizal (AM) associations to meet their nutrient needs (Smith & Read, 2008; but see Bidartondo et al., 2011). Despite occasional diversifications towards other mycorrhizal association strategies (Hoeksema, 2010; Feijen et al., 2018), the AM symbiosis shows a remarkable persistence over evolutionary time (Brundrett & Tedersoo, 2018). Not all plants, however, benefit equally from associating with mycorrhiza (Wilson & Hartnett, 1998), and it remains unclear why some plant species for which we often observe negative responses (such as Bromus inermis, Poa pratensis and Koeleria pyramidata in Wilson & Hartnett, 1998) to mycorrhiza continue to associate with AM fungi. An obvious shortcoming of many experimental studies using arbuscular mycorrhiza is that for logistical reasons they are carried out over relatively short periods. In the long run, the fitness of an organism is an expression of the geometric, and not the arithmetic, mean of payoffs realized over generations (Sæther & Engen, 2015), which means that expectations from short-term studies might be biased and overstate benefits (or lacks of benefits) from AM fungi (Supporting Information Fig. S1a; Notes S1). Here, we explore the possibility that, over evolutionary time, AM fungi benefit plant hosts by enabling them to survive unfavourable events at the expense of a relatively lower fitness (compared to a noncolonized state) under less stressful periods, in what is known as ‘evolutionary bet-hedging’.

A well-described case of evolutionary bet-hedging across biological systems is the evolution of dormancy in plant seeds (e.g. Evans, 2005; Childs et al., 2010): natural selection has resulted in the ability of seeds to germinate over several years, potentially to account for environmental stochasticity. There have been, however, observations of evolutionary bet-hedging across many organisms spanning the tree of life, such as viruses (to the degree that they can be classified as living organisms; Maslov & Sneppen, 2015), bacteria (Beaumont et al., 2009), fungi (Levy et al., 2012) and vertebrates (McAllan et al., 2012). Lekberg & Koide (2014) proposed that AM-associating plants engage in associations with less beneficial AM fungi as part of a bet-hedging strategy and possibly AM fungi do the same in relation to partner choices (Babikova et al., 2013; Veresoglou & Rillig, 2014). Field et al. (2015) formulated a very similar hypothesis in relation to symbiotic partner choices in early terrestrial plants. Bet-hedging (but also evolutionary bet-hedging) in AM associations could also arise, however, if upon colonization plants can better tolerate transient environmental stress such as droughts (e.g. Augé, 2001). To the best of our knowledge, the possibility of evolutionary bet-hedging by plants forming symbioses with AM fungi has not been addressed.

We propose three different scenarios that could give rise to an evolutionary bet-hedging strategy in AM-associating plants (Fig. 1). The most intuitive scenario (Scenario A) describes strong positive mycorrhizal responses during adverse years which offset likely negative growth benefits over favourable years (Fig. 1). The scenario shares expectations with nonevolutionary forms of bet-hedging (describing bet-hedging that occurs at timescales of a single generation such as how mycorrhiza can promote plant fitness under adverse soil conditions but simultaneously suppress it in areas of high fertility; Lekberg & Koide, 2014) and thus shares possible mechanisms with them. One possible mechanism is better protection from pathogens (Veresoglou & Rillig, 2012), which could slow down adaptations to low phosphorus (P) availability, giving rise to a tradeoff between pathogen protection with mycorrhiza and low-P tolerance without mycorrhiza (Laliberté et al., 2015). A further mechanism is improved tolerance to extreme weather conditions (e.g. Auge, 2001). Here, an evolutionary bet-hedging strategy could also arise if AM fungi reduce relative fitness differences across hosts (as shown in Veresoglou et al., 2018) under different weather conditions. This may lead to plants that are disadvantaged by weather conditions exhibiting reduced fitness losses relative to plants favoured by the weather conditions.

In Scenario B, climatic conditions do not necessarily modify average growth effects of mycorrhiza, but AM fungi stabilize plant fitness in time (i.e. reduce temporal variability, which is often assayed in experimental studies as the coefficient of variation of a metric of fitness in time). In the longer term (i.e. when we encounter over 10 generations), fitness depends on the geometric mean (and not the arithmetic mean; Notes S1) of growth effects in time, implying that organisms that on average experience a reduction in fitness gains, can still have greater fitness if they experience lower temporal variability (Fig. S1; see Methods S1 for a reproducible example). There have been several recent studies addressing how mycorrhiza alters the temporal variability of plant fitness (in most cases in the form of biomass production) and most studies report that AM fungi reduce temporal variability (and thus support the idea; e.g. G. Yang et al., 2014; X. Yang et al., 2021; but see Veresoglou et al., 2020; Table 1).

Scenario C presents a special case of Scenario B and specifically describes fitness benefits in the form of a reduction in temporal variability exclusively under adverse conditions. A means by which plants could experience such a reduction in temporal variability under adverse conditions is if they can recover faster from environmental perturbations (i.e. have a higher resilience; Veresoglou et al., 2020; Jia et al., 2021b).

Remarkably, some of the expectations (those in Scenario A and Scenario B) of evolutionary bet-hedging are routinely captured in the existing literature because of the propensity of mycorrhizal ecologists to extensively invest in metaanalyses, using log-response ratios as effect sizes. Let us assume that w_i presents the relative fitness of a plant species that does not form mycorrhizas over generation i. Counterparts of the plant species that associate with mycorrhiza have a multiplicative relative fitness increment of λ_i (i.e. the inverse of an AM fungal response ratio – RR). The fitness of the plant species forming mycorrhiza, R_AM, compared to that of its nonmycorrhizal counterparts, R_NM, after n generations will be:

(Eqn 1)

which following a log transformation of both sides becomes:

(Eqn 2)

It follows that a necessary and sufficient condition (in Scenario A and Scenario B) for AM fungi to benefit plants is that the average log response ratios of plant fitness in response to AM fungi be above zero. This is an expectation that has been routinely tested (even though mostly via procedures using weighting techniques) in numerous mycorrhizal metaanalyses (e.g. Treseder, 2004; Hoeksema et al., 2010). Metaanalytical approaches also capture many of the abstractions (and thus biases) of the experimental procedures that are routinely used in mycorrhizal ecology, such as the unrealistic growth settings with nutrient-deficient sand, soil mixtures used for brief growth assays (Hoeksema et al., 2010), and the use of plant biomass production as a good proxy of fitness (Younginger et al., 2017). It would nevertheless be useful to further explore the degree to which we could take advantage of such metaanalyses to explore evolutionary bet-hedging as well as to develop approaches to discriminate between the two underlying mycorrhizal effects (i.e. growth stimulation and reduced temporal variability) on plant growth. Despite some preliminary studies providing evidence that points in this direction (e.g. Veresoglou et al., 2020; Jia et al., 2021b), it is not yet clear whether AM fungi additionally contribute to a higher resilience (Scenario C) in the systems where they occur, and this now represents a pressing topic in mycorrhizal ecology (e.g. Yang et al., 2018).

Revisiting existing syntheses could probably quantify the variability of growth responses to mycorrhiza over iterative trials but, because it is difficult to reconstruct environmental conditions in the field, it probably cannot answer the question of whether eventually plants profit from an evolutionary bet-hedging. Finding appropriate settings to test the hypothesis of evolutionary bet-hedging is challenging. A promising avenue in palaeoecology is to reconstruct past distribution ranges of plants and assess how variable they have been over time (e.g. Gavin et al., 2014): if biomass of AM-associating plants varies less with time than across non-AM-associating plants, this could be an indication of increased evolutionary fitness, which can then be compared with respective benefits from short-term experiments. Alternatively, it might be easy to use a space-for-time substitution approach (Johnson & Miyanishi, 2008): for example, by monitoring the growth of plants over a range of settings, even outside their distribution range and assess whether the benefits (but also the respective temporal variability) gained from associating with AM fungi are systematically greater for any particular type of settings.

We developed the idea that associations with AM fungi could persist even if, for some hosts, such associations do not result in intermediate, short-term (i.e. in a single generation timespan) fitness gains. We can envisage two ways through which studying bet-hedging in AM systems has relevance to other disciplines. First, given that stability of food yield is an essential constituent of food security (Schmidhuber & Tubiello, 2007), it is worth exploring whether managing land to support arbuscular mycorrhiza promotes consistency in delivering ecosystem services. A key part of sustainable agricultural intensification is to improve management of soil biodiversity (e.g. Tilman et al., 2011) and to this end it is important to explore any possible ways that arbuscular mycorrhiza could contribute (Rillig et al., 2016). Second, arbuscular mycorrhiza could serve as a model system in exploring bet-hedging across other symbiotic systems. AM systems present some desirable features such as ubiquity in nature (Smith & Read, 2008) and a relative ease of assaying fitness benefits (at least in the form of pragmatic proxies) to plants through biomass production. Using AM associations as a model system could streamline the study of bet-hedging across mutualisms, reveal parallels to comparable systems that possibly experience bet-hedging, such as orchids (Shefferson et al., 2003), and uncover the degree to which bet-hedging differs between symbiotic and nonsymbiotic systems because of coevolution (Hoeksema, 2010).

早期陆生植物在陆地上定居可能依赖于丛枝菌根 (AM) 组合来满足其养分需求（Smith & Read，2008 年；但参见 Bidartondo等人，2011 年）。尽管其他菌根关联策略偶尔会多样化（Hoeksema，2010；Feijen等，2018），但 AM 共生在进化时间中表现出显着的持久性（Brundrett & Tedersoo，2018）。然而，并非所有植物都能从与菌根的结合中获益（Wilson & Hartnett, 1998），而且我们还不清楚为什么我们经常观察到一些负面反应的植物物种（例如无芒雀麦）, Poa pratensis和Koeleria pyramidata in Wilson & Hartnett, 1998 ) 到菌根继续与 AM 真菌相关联。许多使用丛枝菌根的实验研究的一个明显缺点是，出于逻辑原因，它们在相对较短的时间内进行。从长远来看，生物体的适应度是几代人实现的收益的几何平均值而非算术平均值的表达（Sæther & Engen, 2015），这意味着短期研究的预期可能存在偏见，并夸大了 AM 真菌的益处（或缺乏益处）（支持信息图 S1a；注释 S1）。在这里，我们探索了这样一种可能性，即随着进化时间的推移，AM 真菌使植物宿主能够在不利的事件中生存，但代价是在压力较小的时期（与非定殖状态相比）相对较低的适应度（与非定殖状态相比），在所谓的“进化对冲”。

一个很好描述的跨生物系统的进化对冲案例是植物种子休眠的进化（例如 Evans，2005；Childs等，2010）：自然选择导致种子能够在几年内发芽，可能考虑环境随机性。然而，已经观察到跨越生命之树的许多生物体的进化对冲，例如病毒（在某种程度上可以将它们归类为活生物体；Maslov & Sneppen, 2015）、细菌（Beaumont et al . , 2009 )、真菌 (Levy et al ., 2012 ) 和脊椎动物 (McAllan et al .,., 2012 年）。Lekberg & Koide ( 2014 ) 提出，作为对冲策略的一部分，AM 相关植物与不太有益的 AM 真菌进行关联，并且可能 AM 真菌在伙伴选择方面也有同样的作用 (Babikova et al ., 2013 ; Veresoglou &里利格，2014 年）。菲尔德等人。( 2015 ) 提出了一个与早期陆生植物共生伙伴选择非常相似的假设。然而，如果在定植后植物能够更好地耐受诸如干旱等短暂的环境压力，那么 AM 关联中的对冲（也包括进化对冲）也可能出现（例如 Augé，2001）。据我们所知，植物与 AM 真菌形成共生体的进化对冲的可能性尚未得到解决。

我们提出了三种不同的情景，它们可能会在 AM 相关植物中产生进化的对冲策略（图 1）。最直观的情景（情景 A）描述了不利年份的强烈阳性菌根反应，抵消了有利年份可能产生的负面生长收益（图 1）。该情景与非进化形式的对冲（描述在单代时间尺度上发生的对冲，例如菌根如何在不利的土壤条件下促进植物适应性但同时在高肥力地区抑制它；Lekberg 和 Koide，2014），因此与他们分享可能的机制。一种可能的机制是更好地保护免受病原体的侵害（Veresoglou & Rillig，2012)，这可能会减慢对低磷 (P) 可用性的适应，从而在使用菌根的病原体保护和没有菌根的低磷耐受性之间进行权衡（Laliberté等人，2015 年）。另一个机制是提高对极端天气条件的耐受性（例如 Auge，2001 年）。在这里，如果 AM 真菌在不同的天气条件下减少宿主之间的相对适应度差异（如 Veresoglou等人，2018 年所示），也可能出现进化的对冲策略。这可能导致受天气条件不利的植物相对于受天气条件有利的植物表现出减少的适应性损失。

在场景 B, 气候条件不一定会改变菌根的平均生长效应，但 AM 真菌会及时稳定植物适应度（即减少时间变异性，这通常在实验研究中作为适应度及时度量的变异系数进行测定）。从长远来看（即当我们遇到超过 10 代时），适应度取决于时间增长效应的几何平均值（而不是算术平均值；注释 S1），这意味着平均而言，适应度增益降低的生物体可以如果他们经历较低的时间变异性，仍然有更大的适应度（图 S1；参见方法 S1 以获取可重复的示例）。等人，2014 年；X.杨等，2021；但参见 Veresoglou等人，2020 年；表格1）。

场景 C呈现了场景 B的一个特例，并专门以在不利条件下以减少时间变异性的形式具体描述了健身益处。植物可以在不利条件下经历这种时间变异性降低的一种方法是，它们能否更快地从环境扰动中恢复（即具有更高的恢复力；Veresoglou等人，2020 年；Jia等人，2021b年）。

值得注意的是，由于菌根生态学家倾向于广泛投资于元分析，使用对数响应比作为效应大小，进化对冲的一些预期（情景 A和情景 B中的那些）经常在现有文献中得到体现。让我们假设w _i表示在第i代不形成菌根的植物物种的相对适应度。与菌根相关的植物物种的对应物具有λ _i的相乘相对适应度增量（即 AM 真菌响应比的倒数 – RR）。形成菌根的植物物种的适应度，R _AM，与其非菌根对应物R _NM相比，在n代之后将是：

（方程式 1）

在双方的对数变换之后变为：

（方程式 2）

由此可见，AM 真菌对植物有益的充分必要条件（在情景 A和情景 B中）是植物适应度对 AM 真菌的平均对数响应比大于零。这是在许多菌根荟萃分析（例如 Treseder， 2004 年；Hoeksema等人，2010 年）中常规测试的预期（尽管主要通过使用加权技术的程序）。元分析方法还捕获了菌根生态学中常规使用的实验程序的许多抽象（以及由此产生的偏差），例如营养缺乏的沙子的不切实际的生长环境，用于简短生长测定的土壤混合物（Hoeksema等人）., 2010 )，以及使用植物生物质生产作为适应度的良好代表 (Younginger et al ., 2017 )。尽管如此，进一步探索我们可以在多大程度上利用这些元分析来探索进化对冲以及开发区分两种潜在菌根效应（即生长刺激和减少时间变异性）对植物的方法将是有用的生长。尽管一些初步研究提供了指向这个方向的证据（例如 Veresoglou等人，2020；Jia等人，2021b），但尚不清楚 AM 真菌是否还有助于提高弹性（情景 C）在它们发生的系统中，这现在代表了菌根生态学中的一个紧迫主题（例如 Yang等人，2018 年）。

重新审视现有的合成可能可以量化迭代试验中菌根生长反应的可变性，但由于难以重建该领域的环境条件，它可能无法回答最终植物是否从进化对冲中获利的问题。寻找合适的设置来检验进化对冲的假设是具有挑战性的。古生态学的一个有前途的途径是重建植物过去的分布范围并评估它们随时间的变化情况（例如 Gavin等人，2014)：如果 AM 相关植物的生物量随时间的变化小于非 AM 相关植物的生物量，这可能表明进化适应性增加，然后可以将其与短期实验的各自益处进行比较。或者，使用空间换时间替代方法可能更容易（Johnson & Miyanishi，2008 年）：例如，通过监测植物在一系列环境中的生长，甚至在其分布范围之外，并评估收益是否（对于任何特定类型的设置，从与 AM 真菌相关联中获得的相应时间变异性也系统性地更大。

我们提出了这样的想法，即与 AM 真菌的关联可以持续存在，即使对于某些宿主而言，这种关联不会导致中期、短期（即在单代时间跨度内）的适应度增益。我们可以设想两种方式，通过这些方式研究 AM 系统中的赌注对冲与其他学科相关。首先，鉴于粮食产量的稳定性是粮食安全的重要组成部分（Schmidhuber 和 Tubiello，2007 年），管理土地以支持丛枝菌根是否能促进提供生态系统服务的一致性是值得探索的。可持续农业集约化的一个关键部分是改善土壤生物多样性的管理（例如 Tilman等，2011）为此，探索丛枝菌根可能发挥作用的任何可能方式非常重要（Rillig等人，2016 年）。其次，丛枝菌根可以作为一个模型系统来探索其他共生系统的对冲。AM 系统具有一些理想的特征，例如自然界中的普遍性（Smith & Read，2008 年）以及通过生物质生产测定植物的适应性益处（至少以实用代理的形式）相对容易。使用 AM 关联作为模型系统可以简化互惠互利对冲的研究，揭示与可能经历对冲的可比系统的相似之处，例如兰花 (Shefferson et al ., 2003)，并揭示共生系统和非共生系统之间的对冲因共同进化而不同的程度（Hoeksema，2010 年）。