Sánchez-Tójar et al. (2020) raised three main issues with our
meta-analysis study (Yin et al. 2019): (1) the study did not fully
account for non-independent, (2) it had limited coverage, and (3) it
lacked transparency and the peer review process was ineffective. Here, I
will focus this reply on these issues.
First, I appreciate Sánchez-Tójar et al. for their efforts in improving
the data quality and reanalysing the data with models fully accounting
for non-independence. However, I feel these efforts further demonstrate
the robustness of our previous conclusions (Fig. 1). The re-analysis
results show that all approaches generate similar estimates, and that
the difference between “significant” and “non-significant” results
only depends on different evolutionary assumptions rather than the data.
As previously indicated in our paper, we did not account for
phylogenetic non-independence because our data include many
non-homologous traits that do not share a common ancestor. Furthermore,
our data include both animals and plants, and in such a large phylogeny,
the non-independence caused by sharing a common ancestor is probably
very weak due to independent selection in different lineages (Fig. S1).
The comment ignored the effect of selection by adopting the Brownian
motion model (Felsenstein 1985), which overestimated the
non-independence and substantially decreased the independent sample size
and statistical power (Figs. S2 and S3). Although reducing the influence
of phylogenetic non-independence or adopting different evolutionary
models can generate significant results (Fig. 1), I do not consider
these approaches necessary, as the critical assumption of sharing a
common ancestor is still violated.
Second, I agree with Sánchez-Tójar et al. that our study has limited
coverage. Such a limitation is more likely associated with restricting
the subject areas (~78% reduction in searching records)
than with missing keywords (~30% reduction). However, I
disagree with the comment to focus the meta-analysis on F2 and F3
generations or to exclude the effects of parental condition transfer, as
they will both greatly reduce the number of collected studies. One
significant aspect of our study is the facilitation of acquiring
comprehensive data by proposing a consensus conceptual framework of
transgenerational effects (please see the discussion below). We,
therefore, hope that our study and this reply can be an incentive as
well as a reference for future more comprehensive meta-analysis in this
field.
In particular, I disagree with the comment that our study lacks
transparency or that the review is ineffective. Both the data and code
were submitted in our first submission and completely open to the three
reviewers, and their comments were critical but constructive (please
refer to the 69-page reply letter in the supporting file). As a
consequence, most issues raised by this comment, including those in the
supplementary information, have been identified and checked in the
revision. More importantly, for controversial issues, e.g., expected
fitness relationships (Supplementary Information 2.7 in Sánchez-Tójar et
al. 2020), they even suggested using sensitivity analysis to evaluate
the effect of our conductance. I emphasize that the decision of our
manuscript is not established on the ignorance of these limitations, but
the judgement by reviewers that the significance overwhelms the
limitations and fits the criteria of Ecology Letters.
Finally, I feel that the comment underestimates the significance of our
study. The real purpose of this meta-analysis is NOT to
demonstrate a significantly positive overall effect. Such a positive
effect has been anticipated by the previous meta-analysis, which
revealed positive but non-significant estimates (Uller et al.2013; Radersma et al. 2018), possibly limited by the number of
collected studies. One significant point of our study is that it raises
a consensus conceptual framework to enable data collection from
different environments, generations and types of designs, significantly
improving the comprehensiveness of data. Furthermore, by analysing the
heterogeneity underlining the comprehensive data, our study obtains the
most detailed picture of transgenerational effects, which shows when and
for which taxa such an effect is beneficial. The picture helps to weigh
empirical evidence with theories and instructs future studies and is
thus the real jury of our study. Intriguingly, our study also shows that
transgenerational effects are found across the tree of life, a pattern
consistent with previous studies (Salinas et al. 2013; Ulleret al. 2013). This pattern suggests that these effects either
have a deep evolutionary origin or have evolved repetitively.
To resolve the detailed scenarios that retain or drive the evolution of
transgenerational effects, I fully agree with the comment that more
comprehensive meta-analysis is necessary. Together, I hope that these
discussions will promote open science practices, methodological
development and the maturation of meta-analysis studies, which serve to
bridge empirical evidence with theories and enlighten further research.