The main difficulties in the analysis of animal language have to date been predominantly methodological in nature. Addressing this perennial problem, the elaborated experimental paradigm presented here has been applied to ants, and can be extended to other social species of animals that have the need to memorize and relay complex "messages".
Accordingly, the method opens exciting new dimensions in the study of natural communications in the wild. Grand Eagle Retail is the ideal place for all your shopping needs! With fast shipping, low prices, friendly service and over 1,, in stock items - you're bound to find what you want, at a price you'll love! Please view eBay estimated delivery times at the top of the listing.
We are unable to deliver faster than stated. NOTE: We are unable to offer combined shipping for multiple items purchased. This is because our items are shipped from different locations. Please contact Customer Services and request "Return Authorisation" before you send your item back to us. Unauthorised returns will not be accepted. Returns must be postmarked within 4 business days of authorisation and must be in resellable condition.
Our results are consistent with this interpretation, because M. Our findings that nurse-upregulated genes are more rapidly evolving and less conserved among social insect lineages relative to forager-upregulated genes suggest that nurse traits have been a major focus of evolutionary innovation between social insect lineages.
This result seems surprising given that foragers of different lineages experience diverse environments outside the nest compared to the relatively constant within-nest environment experienced by nurses and could be expected to experience more diverse selective pressures. One explanation is that the physiological mechanisms associated with metabolically costly foraging activities and older adult life M. Nursing behavior, occurring during very early adult life, may involve more diverse physiological and developmental processes, and nursing itself may also involve more diverse behaviors and physiological processes, such as food processing and the synthesis of glandular secretions that are fed to larvae.
Perhaps the relatively more complex genetic architecture less tightly connected, involving more modules, and diverse processes has served as less of a constraint and facilitated more evolutionary change for nurse-related genes. If so, we predict that nurse-specific functions and functions for early adult life may be generally more evolutionarily labile as well as more physiologically and behaviorally labile within and across lineages than forager-specific functions.
Note that this prediction is opposite the typical expectation that genes acting early in development have more pleiotropic effects and are thus especially constrained Roux and Robinson-Rechavi, ; Piasecka et al. We identified two discrete sets of genes with distinct genetic architecture associated with age-based division of labor. The majority of forager-upregulated genes were contained within a single gene module module 14; Figure 1—figure supplement 3 that was significantly positively associated with age.
Another module with expression negatively associated with age contained the largest number of nurse genes, but nurse genes were also broadly spread out across a number of other modules with complex expression patterns across age and behavioral groups. Interestingly, the modules differed in the proportion of constituent genes which had identifiable S.
That said, the modules were enriched for various gene ontology terms, providing some insight into their putative functional importance Supplementary file 4. By explicitly studying regulatory architecture and inferring modules of tightly connected genes in other species as well as M. Building on the genetic toolkit conceptual framework, it will be possible to ask to what degree diverse lineages repeatedly use the same modules, and importantly approaches already exist for quantifying module overlap in the absence of functional information Oldham et al.
Similarly, after finding non-significant overlap in lists of genes associated with queen- and worker-caste development in paper wasps and honey bees, Berens et al. Focus on co-expressed modules may actually improve the feasibility of inferring the function of co-expressed genes based on observed expression patterns together with standard functional information inferred from the subset of conserved annotated genes with identifiable orthologs from model systems. It will also be possible to determine the relative contribution of conserved vs taxonomically restricted genes to co-expression modules.
Two replicate M. Each colony was established from a separate source colony, which came from a stock of approximately 40 colonies that have been repeatedly mixed across generations so that they are genetically similar. Every 3 days, newly eclosed callow workers, which were inferred to be approximately 0—1 days old, were collected from 8—10 stock colonies. These callow workers were briefly anesthetized with CO 2 and individually paint marked on the gaster with a unique age cohort color dot using a Sharpie extra fine oil based paint pen, and then were added to each of the observation colonies.
Five uniquely marked age cohorts were thus added to the colonies on days 1, 4, 7, 10, and 13 of the study. Nestmate recognition is at most weak and transient in M. We also set up a camera to automatically take images of the nest areas of each colony once every 20 min for the entire period of the study, although we do not further discuss these images. Previous literature indicates that M. We ran the study for 1 month, expecting to capture the major age-based transitions in worker behavior e.
In practice, such late transitions are difficult to detect as sample size necessarily declines as increasing numbers of workers die. A behavioral scan of each colony was completed once each day for the duration of the month-long study by recording the instantaneous behavior and location observed for every visible paint-marked worker. We recorded 30 distinct behaviors, but only 15 were observed more than 15 total times during the study period Supplementary file 1.
We defined an individual as foraging if it was observed on a food or water source or actually carrying food i. Each experimental colony contained four identifiable locations that were redefined prior to each behavioral scan: brood area, brood periphery, remaining nest area, and foraging area. The brood area was defined as the central area within the nest containing all brood and queens Edwards, The nest periphery was defined as the region directly adjacent to the brood area, where workers were dense in aggregation but not in contact with any of the brood.
The nest area was defined as the sparsely occupied remainder of the space within the nest, not including the brood area and nest periphery. The foraging area included all areas outside of the nest. Analyses of behavioral data were conducted in R www. In addition, for each of the two replicate observation colonies, we collected and pooled 20 non-paint marked workers in the act of the following five behaviors: nursing larvae, grooming larvae, engaged in trophallaxis with other workers, foraging for protein collecting egg, mealworm, or liver , and foraging for carbohydrates collecting honey solution.
RNA was extracted from pools of worker samples of known age or observed behavior using Qiagen RNeasy kits with standard protocols. RNA sequencing was conducted at the University of Arizona Genetics Core on an Illumina HiSeq with bp paired ends reads, with six samples multiplexed per lane, randomly distributed across four lanes. DNA from a single haploid male ng was used to prepare a TruSeq library, which was sequenced in multiplex on an Illumina HiSeq , yielding 70,, million bp read pairs.
We then chose the assembly with the longest N50 as the reference for transcriptome assembly. The transcriptome was mapped to the reference using Tophat 2, and assembled into transcripts using Cufflinks 2. Gene expression data were obtained by re-mapping the transcript reads to the extracted transcripts using RSEM and calculating the expected counts at the gene level Li and Dewey, When multiple isoforms of a single locus were found, only the longest transcript was used for gene annotation. Transcript counts were filtered by abundance, removing those with less than 1 fragment per kilobase mapped FPKM in more than half of the libraries Mortazavi et al.
Differential gene expression analysis was carried out in edgeR, using a GLM fit to the count data and identifying differentially expressed genes using planned linear contrasts Robinson et al. In order to infer co-expression modules and gain an insight into network structure of gene interactions, we performed a weighted gene co-expression network analysis WGCNA on the count data Langfelder and Horvath, WGCNA was conducted on the entire transcript set, after filtering out the low-abundance transcripts. This analysis relies on patterns of gene co-expression, but has been shown to reconstruct protein—protein interaction networks with reasonable accuracy Zhao et al.
We used total connectivity as a measure of gene interaction strength, because it is not as sensitive to module assignments, and most likely reflects the overall selective pressures acting on the gene, beyond those imposed by its role in age polyethism. As with most gene expression analysis, WGCNA provides better estimates for highly abundant genes, and in particular for genes showing variation in their expression levels.
Consequently, low-abundance and invariant genes will show lower connectivity. Fire ant S. This parameterization allowed for high specificity, though at the cost of sensitivity, since paralogs were ignored Chen et al. These results were used to predict the M. Using the list of differentially expressed genes in foragers vs nest workers in the fire ant Manfredini et al. We repeated the analysis above using honey bee A. Finally, to study the main and interaction effects of connectivity, expression, and behavioral category on evolutionary rate, we used a linear model log transformed rate as the dependent variable, log transformed connectivity and expression as continuous predictors, and behavioral category as a categorical predictor.
Statistical analysis was performed with R. In the case of differential gene expression, data analyses were corrected for multiple comparisons using the Benjamini-Hochberg FDR procedure Benjamini and Hochberg, Most of the workflow and output is shown in Supplementary file 2 , with the corresponding R script shown in Source code 1. All behavioral and gene expression data, including a MySQL database for the gene expression data have been deposited to Dryad, doi An edited version of the letter sent to the authors after peer review is shown, indicating the substantive concerns or comments; minor concerns are not usually shown.
Reviewers have the opportunity to discuss the decision before the letter is sent see review process. Similarly, the author response typically shows only responses to the major concerns raised by the reviewers. Your article has been favorably evaluated by Diethard Tautz Senior editor , a Reviewing Editor, and three reviewers. The Reviewing editor and the reviewers discussed their comments before we reached this decision, and the Reviewing editor has assembled the following comments to help you prepare a revised submission.
While all three reviewers are enthusiastic about the study, several important concerns were voiced by all of them. All reviewers pointed out that this paper does not provide the necessary statistical details to be able to assess the quality of the work. The respective comments are combined below and it will be necessary to address them before a final decision about acceptance can be reached.
This is certainly interesting and worthwhile. However, the overlap is marginally significant fifth paragraph of the Results section. So the question is, what percent overlap exactly would constitute support for the toolkit hypothesis? The authors are not alone in having to deal with this issue, and so I don't want to lay this completely at their feet, but it would be nice if they stated explicitly what exactly would constitute evidence, or lack thereof, for the toolkit hypothesis somewhere before they get to the results.
Related to this issue, I wanted more detail about the analyses comparing expression patterns between taxa. Did the authors just ask if the same genes were differentially expressed between behavioral types in the two taxa to be viewed as 'consistent' with the toolkit model? Or did a particular gene have to be differentially expressed in the same direction in both taxa? The latter, I think, although this wasn't clear.
And did a gene need to be significantly differentially expressed to be counted? Or was it sufficient that it show the same directionality in expression, regardless of significance? Directionality without significance is as important as significance, given that the studies in the taxa had different power, used different methods, etc.
These are not trivial issues and they may affect the outcome and interpretation of the results. I urge the authors to look into this more closely. They refer to genes related to social behavior throughout the manuscript. Or morphological changes related to other functions, such as exposure to exterior environments? They do not state they used whole ants or just ant heads for transcriptome profiling, which is highly crucial for interpreting the results especially given that some other studies have used the animal head.
Four categories are compared and 2 of these are said to have different expression. What is the reference here; are the other two categories grooming and trophallaxis not different from these two? Or perhaps I am missing something? This can be partly followed in the Supplementary Material, but should be referred in the main text, e.
The foragers and nurses were most different as they represent the youngest and oldest. I would have stated this explicitly. I am also concerned somewhat about the PCA in the Supplementary Material: There seems to be two groups emerging, but this is likely technical I would guess sample processing dates. It might be difficult to control for this, but if possible, could improve DE analysis significantly. What was the p-value cutoff?
How did the authors control for multiple-testing? This can also be followed in the Supplementary Material, but should be referred in the main text. Thus, without additional analysis I think one cannot decide on the existence of distinct classes. The authors could consider applying some other test; e. In the fifth and sixth paragraphs of the Results section: In the gene expression conservation analysis, we are given no information how many genes are used in the comparisons i. If the numbers are low, they could instead check the effect size of orthologous genes identified as DE for honeybee, for example.
Was the honeybee data generated by Manfredini et al.
If not, the authors should state that. Most importantly, if the honeybee data was generated from the brain as done by quite a few studies and the data in this study from the whole body, this could also be a reason for finding limited overlap. In comparisons with the Fisher's exact test, it would be useful to state what the background is non-DE genes, genes up-regulated in the other category, or both?
In the seventh paragraph of the Results section: Connectivity—this could be more explicitly defined, such as emphasizing that the prediction comes from transcription data correlations e. I think the authors could also discuss potential biases here. One would want to make sure that these factors are not influential on the reported result.
Especially so, as lower conservation among up-regulated genes is one of the paper's main points. But no information is given regarding the magnitude of the effect. The authors could also plot expression versus connectivity.
In the sixth paragraph of the Results section: There is little discussion on the GO analysis. Does the UV response pathway have to do with sudden exposure to the sun? At least would one not expect to see the same pathway up-regulated in foragers of other taxa?
Please indicate the p-value cutoffs for the GO analysis. This is also found in the Supplementary Material, but should be in the main text or Methods. What is the estimated genome size? What was the average coverage per sample for the genomic and transcriptomic data? How did they evaluate this? A table with summary statistics would be very useful. How was the paralog problem dealt with, particularly with respect to the molecular evolution analyses? Similarly, for the network analyses: Were these co-expression networks calculated only on significant transcripts or on all transcripts?
Does that mean all transcripts show tightly-correlated expression levels? Finally, why didn't the authors include Polistes in their comparative analyses? There are at least 2 studies on Polistes , both of which are already cited in this manuscript. This seems like it would be another independent data point worth discussing. The statistical analysis for this project, as for most other bioinformatics projects, is quite complex, and we chose to present the complete analytical pipeline and data as a supplement to make it completely reproducible.
We may have relied too much on our supplement containing the complete analysis, omitting details from the main text. In this revision we spent a lot of time addressing statistical questions and included the relevant details in the main text. We also added several additional analyses. We agree completely. Similarly, a recent book chapter by Adam Wilkins points out that the concept of a developmental genetic toolkit is widely accepted, but it is also very unclear what actually makes up the genetic toolkit, so that the hypothesis is very difficult to actually test.
Because these two hypotheses are entrenched in the literature—and the idea of a genetic toolkit underlying social behavior is frequently cited as the major finding of sociogenomic research—we describe them in the Introduction to motivate our work. In the Discussion, we explain how our results fit with these two hypotheses, but then we move on to our main, and we believe, most exciting point: the genes we identified as being associated with ant division of labor show different patterns of connectivity and evolutionary conservation, and thus the genetic architecture of ant division of labor includes both highly connected and conserved genes as well as more loosely-connected and evolutionarily labile genes.
Following this suggestion we conducted the suggested analysis using the fire ant data. When we look at just the direction of expression, or the fold-change in expression across the genome, we find results very similar to those obtained when we only consider differentially expressed genes.
Whereas we previously found a 3. So, the results of the two analyses are consistent. This significant but seemingly low overlap is also consistent with previous studies interpreted as supporting the genetic toolkit hypothesis. For example, Woodard et al. Considering only genes with identifiable homologs between the two study species and A.
We agree that these are very important points. We used whole ant bodies and we emphasize this point more in the revised text. We also agree that the differentially expressed genes we observed may be related to a wide range of physiological and behavioral processes, and not just social signal production and response.
It demonstrates:. Day, E. It is just as impossible that there should exist a human brain without a mind, as a mind without a brain, and to every normal or pathological change in the mental activity, there corresponds a normal or pathological change of the neurocymic activity of the brain, i. A competition hierarchy among boreal ants: impact on resource partitioning and community structure. But what exists is surely in itself sufficiently interesting and important. In practice, such late transitions are difficult to detect as sample size necessarily declines as increasing numbers of workers die.
The underlying physiological processes are particularly intermeshed because by definition, age-based division of labor in social insects means that the individuals are aging as they transition from nursing to foraging tasks. We have clarified these issues in the revised text. There were four behavioral categories: foragers, nurses, and workers engaged in trophallaxis and grooming. We wished to see which of the categories were transcriptionally distinct from the others, and made comparisons of the focal category vs all of the others.
This section was indeed unclear and we re-wrote it with extra detail. All the samples were processed simultaneously and in the same facility. Most importantly, the separation between the clusters does not correlate with any biological signal detected in the data. Finally, worker samples on the first day of eclosion age 0 are an outlier on this plot.
As these workers are just beginning their transition to adult life, with numerous physiological changes, such as cuticle hardening, it is logical to expect that they would be separate from the others. If we discount these points, there is far less evidence for the existence of separate groups in the scatter. We followed the standard practice of 0. We now specify this in the new Statistical Analysis section of the Methods.
We were interested in testing whether our temporal data could be classified into two groups, corresponding to nurses and foragers, and, if so, where would the transition point be. We also include a simple cluster diagram as Figure 1—figure supplement 1 , to show that the there is indeed a breakpoint before day We have also clarified the logic behind this analysis in the text.
We added the numbers of genes, which were indeed low, and a parallel analysis using effect sizes log fold expression change for fire ants, which produced the same result. Unfortunately, for the honey bee study, only the list of differentially expressed genes is publicly available, so that we could not perform a similar analysis for the honey bee data set.
The unavailability of previous data sets is one reason that we have strived to make our full analysis available and completely transparent, as well as making all of the data available. The honeybee data was indeed generated from the brain, and this is certainly a plausible reason for the low observed overlap. We have added this very important point to our discussion, qualifying our conclusions in this light.
For our measure of connectivity, we used the total connectivity of a gene, which is less sensitive to how modules are defined, and most likely reflect the overall role of the gene, beyond the modules we detect in our data set. Although the authors of the WGCNA package suggest that the method can run on normalized count data, in the course of considering potential biases, we found the gene lengths varied somewhat among behavioral categories. We then re-ran the analysis using FPKM data, which are length-standardized.
This analysis also captured major network effects, but had a much better fit to the data. In particular, we had to use a smaller soft thresholding level before an approximately scale-free topology of the network was observed a WGCNA requirement. We were also able to detect many more modules at the same cutoff levels, suggesting greater network resolution. The most obvious remaining bias from this sort of analysis is that genes with low expression and low variability will not be detected as differentially expressed. Indeed, this effect can clearly be seen in Figure 2B , with the average expression level of non-differentially expressed genes is lower.
However, contrary to what you would expect if the pattern was driven by this bias, the nurse-upregulated genes show lower connectivity than non-differentially expressed genes Figure 2A. We also explicitly control for expression level by including expression level in our GLM analyses. We have added each of these suggested plots and have also added plots showing the effects of connectivity and expression on whether genes had identifiable fire ant and honey bee orthologs Figure 3. We have added further discussion of the GO analysis and include the full set of GO terms enriched for both the forager- and nurse-upregulated genes as well as each module separately.
As we mention in response to Reviewer 1, a genome project was not the goal of this study. That being said, as the quality of the reference genome assembly is important to this study, we now include quality control statistics in the Results, such as coverage statistics and CEGMA, as requested. Unfortunately, there is no independently estimated size of the M. As with all studies involving gene orthology, particularly with-non models species, there is no genome-wide gold standard that allows the performance of a method to be evaluated.
Because the data set is based on transcriptomic data, there will necessarily be false negatives associated with poorly expressed transcripts. That being said, our choice of reciprocal best hit is well justified, based on comparisons of various available methods, as it has high specificity, though at the cost of sensitivity, as it ignores paralogous relationships for genes that have been duplicated in one of the lineages Chen et al.
The Author of this new volume on ant communication demonstrates that information theory is a valuable tool for studying the natural communication of animals. Zh. Reznikova, Studying Animal Languages Without Translation: An Insight from Ants, DOI /_1. 1. Communication, Speech and.
Another important consideration is that we used an independently assembled genome to estimate evolutionary rates.