Experiments are rarely performed in isolation. Usually, several conditions are compared in parallel or sequential experiments. This experimental strategy also applies to time-dependent data, e.g. from timelapse imaging. So, naturally, after I published a ‘walk-through for plotting temporal data using R and ggplot2, I was immediately asked how to plot two (or more) sets of data in the same graph.

To respond to this request, I made a ‘walk-through’ for generating graphs that visualize temporal changes in signal from multiple experimental conditions. This tutorial will also feature some of the ‘tidy tools’ (footnote 1) that are designed to work on tidy data. The final figure, shown below, displays the changes in the activities of three (separately measured) Rho GTPases that occur after stimulating endothelial cells. For more background on the experiments, the reader is referred to the paper by Reinhard et al (2017).

Combining and plotting the data from different conditions

First we will read the experimental data from three different csv files (available here) and put them into three separate dataframes:

> df1 <- read.csv("Fig3_Rac_S1P.csv")
> df2 <- read.csv("Fig3_Rho_S1P.csv")
> df3 <- read.csv("Fig3_Cdc42_S1P.csv")

Each dataframe holds the information from a particular experimental condition. In this example, the dataframes contain the activities of three Rho GTPases, Rac, Rho or Cdc42, measured over time. Each dataframe has data from individual cells organized by columns. The values that were measured are a “Ratio” that is a result of a FRET imaging experiment. The first column is a record of the Time. Let’s see what a typical dataframe looks like:

> head(df1)

returns:

       Time    Cell.1    Cell.2    Cell.3    Cell.4    Cell.5 ...
1 0.0000000 1.0012160 1.0026460 1.0022090 0.9917870 0.9935569 ...
2 0.1666667 0.9994997 0.9928106 0.9997658 0.9975348 1.0018910 ...
3 0.3333333 0.9908362 0.9964057 0.9905094 0.9946743 0.9961497 ...
4 0.5000000 0.9991967 0.9972504 0.9972806 1.0074250 1.0060510 ...
5 0.6666667 1.0093450 1.0109910 1.0103590 1.0084080 1.0022130 ...
6 0.8333333 0.9941078 0.9940830 0.9990720 1.0181230 1.0110220 ...

To keep track of which data or values belong to which experimental condition (Rac, Rho or Cdc42) we assign a unique identifier to each of the dataframes. To achieve this, we add another column, named “id”, that identifies the condition:

> df1$id <- "Rac"
> df2$id <- "Rho"
> df3$id <- "Cdc42"

The next step is to merge the datasets into one dataframe. The function bind_rows() from the ‘dplyr’ package will do this (to learn how to load the package see footnote 1):

>df_merged <- bind_rows(df1,df2,df3)

Columns with identical names will be merged. For instance, there will be one column with “Time” and another column named “id” that indicates which experiment was performed. The remaining columns list the values that were obtained during the measurements for each of the cells.

The dataframes  from the different conditions have data from different numbers of cells. Therefore, there will be columns that are not completely filled with values. For instance, the dataframe df1 (Rac) has data from 32 cells, while df2 (Rho) has only data from 12 cells. As a result, the df_merged dataframe has a column “Cell.32” that has only data from the Rac experiment and not for Rho and Cdc42. These empty fields (or “missing values”) will be replaced in the dataframe by “NA”.

To do all of the above at once for all csv files that are present in a defined working directory, this R script can be used.

The merged dataframe, df_merged, is still in a wide, spreadsheet format. To convert it into a tidy format (as has been detailled here) we use the function gather() from ‘tidyr’. Since the “Time” and “id” column do not need to be changed, they are excluded from the gathering operation by the preceding minus sign:

>df_tidy <- gather(df_merged, Cell, Ratio, -id,-Time)

To check the result use:

> head(df_tidy)

which returns:

       Time  id   Cell     Ratio
1 0.0000000 Rac Cell.1 1.0012160
2 0.1666667 Rac Cell.1 0.9994997
3 0.3333333 Rac Cell.1 0.9908362
4 0.5000000 Rac Cell.1 0.9991967
5 0.6666667 Rac Cell.1 1.0093450
6 0.8333333 Rac Cell.1 0.9941078

Previously, we used a similar tidy dataframe to draw the lines from individual cells and color those according to the column “Cell”:

>ggplot(df_tidy, aes(x=Time, y=Ratio)) + geom_line(aes(color=Cell))

This results in:

This graph looks like a mess. The reason is that in the column “Cell”, the factor Cell.1 can be linked to the Rho, Rac and Cdc42 dataset.

To solve this, we need to define another column that has a unique identifier for each cell from every condition. One way to achieve this, is by combining the string in the “id” column with the string in the “Cell” column by using unite(). The new, combined string (connected with an underscore) is placed into a new column named “unique_id”:

>df_tidy <- unite(df_tidy, unique_id, c(id, Cell), sep="_", remove = FALSE)

Now, each sample can be grouped according to the “unique_id” to draw the line correctly:

>ggplot(df_tidy, aes(x=Time, y=Ratio)) +
  geom_line(aes(color=Cell, group=unique_id))

To obtain a plot that shows all the individual traces colored according to the experimental condition we can use:

>ggplot(df_tidy, aes(x=Time, y=Ratio)) + geom_line(aes(color=id, group=unique_id))

Generating a graph with data summaries

The next thing you may want to do is to calculate statistics. In the previous post it was explained how to calculate the mean value for each time point by using base R code.

Here, the same calculation will not give a meaningful result, since each time point has data from the three different conditions. To calculate the mean for each condition (defined by “id”) for each of the time points I will use the function group_by(). This function is part of the “tidyverse”, a set of tools that work well on tidy data (footnote 2).

Another feature of tidy tools is the “pipe” which is encoded by %>%. It basically moves the result from a function onto the next function. The advantage is that we can concatenate a number of functions that are still understandable by humans because the sequence of events can be read from left to right (footnote 3).

So, to calculate the mean value we first group the data by “Time” and “id”. Next, we calculate the number of points (n) and the average (mean) with the function summarise(). The pipe is used to perform the operations all at once. The result is assigned to a new dataframe, df_tidy_mean:

>df_tidy_mean <- df_tidy %>%
    group_by(Time, id) %>%
    summarise(n = n(), mean=mean(Ratio))

To show what the first rows of the dataframe looks like we use head():

> head(df_tidy_mean)

which yields:

# A tibble: 6 x 4
# Groups:   Time [2]
Time id        n   mean
<dbl> <chr> <int>  <dbl>
1 0     Cdc42    32 NA
2 0     Rac      32  0.998
3 0     Rho      32 NA
4 0.167 Cdc42    32 NA
5 0.167 Rac      32  0.999
6 0.167 Rho      32 NA

The data structure of the result is a “tibble“, which is a next-generation dataframe. In this walk-through, we use tibbles in the same way as dataframes and do not make a distinction.

The content of df_tidy_mean is not what we want, since each row has n=32 observations and the mean of de conditions “Cdc42” and “Rho” returns “NA”. We know that n=32 for Rho cannot be right, since there are only 12 cells cells measured for Rho. The n=32 for each of the conditions is due to the “NA” values in the columns that are counted as well. This also explains why we do not see a proper mean for “Cdc42” and “Rho”. To address this issue, we need to get rid of the missing values in the “Ratio” column. We can do this by adding another function to the previously used command that uses the filter() function to only select “Ratio” values that are not “NA”:

> df_tidy_mean <- df_tidy %>%
    filter(!is.na(Ratio)) %>%
    group_by(Time, id) %>%
    summarise(n = n(), mean=mean(Ratio))

Let’s see if that worked out the way we want:

> head(df_tidy_mean)

# A tibble: 6 x 4
# Groups:   Time [2]
Time id        n  mean
<dbl> <chr> <int> <dbl>
1 0     Cdc42    19 1.00
2 0     Rac      32 0.998
3 0     Rho      12 0.998
4 0.167 Cdc42    19 0.995
5 0.167 Rac      32 0.999
6 0.167 Rho      12 1.00

This gives the expected result. The condition “Rho” has n=12 and the condition “Rac” has n=32. These number of samples match with the data (csv files) that were supplied as input. We can continue to add some more useful statistics to the dataframe:

>df_tidy_mean <- df_tidy %>%
    filter(!is.na(Ratio)) %>%
    group_by(Time, id) %>%
    summarise(n = n(),
           mean = mean(Ratio),
         median = median(Ratio),
             sd = sd(Ratio))

Finally, we will use the function mutate() to calculate the inferential statistics, i.e. the standard error of the mean and the 95% confidence interval of the mean:

>df_tidy_mean <- df_tidy %>%
    filter(!is.na(Ratio)) %>%
    group_by(Time, id) %>%
    summarise(n = n(),
           mean = mean(Ratio),
         median = median(Ratio),
             sd = sd(Ratio)) %>%
    mutate(sem = sd / sqrt(n - 1),
      CI_lower = mean + qt((1-0.95)/2, n - 1) * sem,
      CI_upper = mean - qt((1-0.95)/2, n - 1) * sem)

Now, the df_tidy_mean dataframe has all the summary data that we need, for each condition (“id”) and for each time point.

> head(df_tidy_mean)

# A tibble: 6 x 9
# Groups:   Time [2]
Time id        n  mean median      sd      sem CI_lower CI_upper
<dbl> <chr> <int> <dbl>  <dbl>   <dbl>    <dbl>    <dbl>    <dbl>
1 0     Cdc42    19 1.00   1.00  0.00756 0.00178     0.999    1.01
2 0     Rac      32 0.998  0.998 0.00550 0.000988    0.996    1.000
3 0     Rho      12 0.998  0.997 0.00477 0.00144     0.995    1.00
4 0.167 Cdc42    19 0.995  0.997 0.00565 0.00133     0.993    0.998
5 0.167 Rac      32 0.999  0.999 0.00684 0.00123     0.997    1.00
6 0.167 Rho      12 1.00   1.00  0.00615 0.00186     0.999    1.01

That’s all we need to make a graphical summary of the data:

>ggplot(df_tidy_mean, aes(x=Time, y=mean, color = id)) +
  geom_line(aes(x=Time, y=mean, color=id)) +
  geom_ribbon(aes(ymin=CI_lower,ymax=CI_upper,fill=id),color="grey70",alpha=0.4)

Improving the presentation and annotation of the graph

Further styling of the figure can be done. Below, we get rid of the default ggplot2 theme, change the scale of the x-axis and add a vertical line at t=1.75 to indicate when the stimulation took place. This will generate the final figure that was shown above:

>ggplot(df_tidy_mean, aes(x=Time, y=mean, color = id))+
  geom_line(aes(x=Time, y=mean, color=id))+
  geom_ribbon(aes(ymin=CI_lower,ymax=CI_upper,fill=id),color="grey",alpha=0.4)+
  theme_light(base_size = 16) + xlim(0,10) + geom_vline(xintercept = 1.75)

Color

User defined colors can be added to the lines and ribbons by using the functions scale_color_manual() and scale_fill_manual() respectively. The list of color values can consist of colornames or hexadecimal RGB code:

>color_list <- c("#EE6677", "#228833", "#4477AA")

Or:

>color_list <- c("darkgoldenrod", "limegreen", "turquoise2")

When the colors are defined, they can be used for plotting:

>ggplot(df_tidy_mean, aes(x=Time, y=mean, color = id))+
  geom_line(aes(x=Time, y=mean, color=id), size=1)+
  geom_ribbon(aes(ymin=CI_lower,ymax=CI_upper,fill=id),color="grey",alpha=0.4) +
  theme_light(base_size = 16) + xlim(0,10) + geom_vline(xintercept = 1.75) +
  scale_fill_manual(values=color_list) + scale_color_manual(values=color_list)

Side-by-side

Thus far, we made a single graph for all the data. To see the data from the different conditions (“id”) in separate plots, we add “facet_grid(.~id)”:

>ggplot(df_tidy_mean, aes(x=Time, y=mean, color = id))+
  geom_line(aes(x=Time, y=mean, color=id), size=1)+
  geom_ribbon(aes(ymin=CI_lower,ymax=CI_upper,fill=id),color="grey",alpha=0.4) +  theme_light(base_size = 16) + xlim(0,10) + geom_vline(xintercept = 1.75) +
  scale_fill_manual(values=color_list) + scale_color_manual(values=color_list) +
  facet_grid(.~id)

The result is three plots next to each other (in a horizontal format). However, in this case it probably makes more sense to arrange the plot in a single column. This is achieved by adjusting the order between the brackets: facet_grid(id~.). Since a legend seems redundant, it is removed in the following example:

>ggplot(df_tidy_mean, aes(x=Time, y=mean, color = id))+
  geom_line(aes(x=Time, y=mean, color=id), size=1)+
  geom_ribbon(aes(ymin=CI_lower,ymax=CI_upper,fill=id),color="grey",alpha=0.4) +  theme_light(base_size = 16) + xlim(0,10) + geom_vline(xintercept = 1.75) +
  scale_fill_manual(values=color_list) + scale_color_manual(values=color_list) +
  facet_grid(id~.)+ theme(legend.position="none")

Final words

The use of transparent layers for the different conditions allows to combine the data in a single graph. The resulting plot provides a clear view of the relation between the activities. Alternatively, the plots can be readily shown as a column in which the time axes are aligned (similar to the original figure 3 from Reinhard et al., 2017). I hope that providing this ‘walk-through’ that shows how to build a graph from multiple datasets will be a good starting point to generate plots of your own data with R/ggplot2.

Acknowledgments: Thanks to Eike Mahlandt and Max Grönloh for testing and debugging the code.

Footnotes

Footnote 1: The R packages that are necessary for this walk-through are: ‘dplyr’, ‘tidyr’, ‘ggplot2’ and ‘magrittr’. To load a package (after downloading and installing it) use require(), for example:

>require('dplyr')

The aforementioned packages are part of the ‘tidyverse’ and can be installed all at once.

Footnote 2: I usually mix basic R functions with functions from tidytools. This is mostly for ‘historical’ reasons; I first learned a bit of base R before I learned about the tidyverse. In my experience, some of the base R code is easier to remember and therefore my default code is base R. On the other hand, the tidy tools allow to string a number of functions together (using the pipe; %>%) which condenses the code, while maintaining readability.

Footnote 3: There are many places where the application of tidy tools is explained in more detail, for example here and here.

(9 votes)

Loading...

BBSRC funded postdoc position in the laboratory of Natalia Sanchez-Soriano (https://sanchezlab.wordpress.com), to study the cell biology of neuronal ageing and the underlying mechanisms.  The aim of the project is to understand the harmful changes that neurons undergo at the subcellular level during ageing, and unravel the cascade of events that cause them, with a focus on intracellular degradation systems and the upstream regulatory pathways. Ideally applicants should be trained in neuro- and/or in vivo cell biology, and imaging, and have experience with Drosophila.

The post is available from 1st of September 2018 until 31st of August 2021.
For full details and to apply online, please visit: https://recruit.liverpool.ac.uk
Job Ref: 009401, closing Date: 28 June 2018.

(No Ratings Yet)

Loading...

The Maehr lab (http://maehrlab.net/) is looking for a team member in the area of developmental immunology and human pluripotent stem cell-based disease models.

Envisioned projects will utilize pluripotent stem cell differentiation approaches, together with assays of thymus and T cell development, to decipher the molecular underpinnings of human immune syndromes such as thymus dysfunction-based autoimmunity and immunodeficiency. In addition to stem cell differentiation and immunological assays, the team will apply technologies such as functional genomins (e.g. CRIPSR) and single cell RNA-sequencing as well as computational approaches.

Candidates should possess a Ph.D. and have a strong background in immunology and/or developmental biology. Experience with immunological assays, stem cell differentiation, bioengineering and/or next generation sequencing assays is desirable. Excellent communication, writing, and collaboration skills are essential.

Interested candidates should email a cover letter, CV, and 3 references to Dr. René Maehr, Associate Professor of Molecular Medicine, UMass Medical School (rene.maehr@umassmed.edu).

(No Ratings Yet)

Loading...

An emerging trend in developmental biology focuses on the role of cell adhesion in modulating tissue morphogenesis. Spectacular advances have shed light on how modulation of adhesion between equivalent cells orchestrates the acquisition of forms. However, how interactions at complex interfaces between distinct neighboring cell types influence tissue growth remains to be elucidated. In a recent paper, staring mostly at a flat muscle that spreads subcutaneously and its partner motor neurons, I investigated how neuromuscular morphogenesis involves complementary activities of the Fat1 Cadherin in three interdependent cell types – myogenic cells, motor neurons, and the connective tissue through which muscles spreads.

Tissue-specific activities of the Fat1 Cadherin cooperate to control neuromuscular morphogenesis Françoise Helmbacher, PLoS Biology, May 16, 2018 | doi: 10.1371/journal.pbio.2004734 | PMID : 29768404

The story started as I was transitioning from being a post-doc in the lab of Rüdiger Klein in Munich, to establishing my own group in Marseilles. At the time, I was interested in events of communication between motor neuron (MN) populations, whereby a pool of MNs influences the specification of neighbor MNs. I was looking for molecules mediating such function, and Fat1 originally came up as one of my candidate genes. Although Fat1 met many of my criteria, it was initially considered a low priority candidate for the specific question I was asking (which I won’t discuss here). However, there were also good reasons for deciding to work on Fat1, aside from the main project of my freshly established lab: (1) Fat1 was expressed in my favorite MN population (Figure 1), a group of neurons recognizable by the transcription factor they expressed, which innervates a fascinating muscle, the Cutaneous Maximus (CM), (2) there were mouse mutants available (which by chance happened to be already in my former lab), and (3) through pioneer experiments with these mutants, I had gathered preliminary data indicating that Fat1 disruption resulted in defective specification of the CM motor pool and in defective axonal arborization within the target muscle – the phenotype I was expecting.

Whatever the underlying gene function, starting a project with a robust in vivo phenotype guaranteed that an interesting story would emerge from it, even though this was not my main project. For this reason, I kept this as my own side project. I had no hypothesis, but it felt like starting a classical forward genetics approach, with the advantage that I already knew the gene identity. This is also when the field of planar cell polarity (PCP) started blooming, mostly through work in Drosophila. The Drosophila Fat Cadherin, initially regarded as a putative tumor suppressor, had been attributed the mysterious function of propagating information on cell polarity within the plane of epithelia across long distances. I was fascinated by these studies, and curious to ask if MNs could propagate some polarity information to their neighbors too. Well, the story took a completely different path!

The two first key results that I had obtained were that 1) axons of the motor neuron pool innervating my favorite muscle, the CM, were clearly shorter in Fat1-deficient embryos, when all other nerves appeared unaffected; and 2) expression of one of the markers specifically characterizing this MN population was drastically reduced. Given the fact that at that time, I had selected Fat1 because of its selective expression in subsets of MNs encompassing the CM-innervating pool (marked by Etv4 expression, Figure 1B), these data were compatible with the simplistic and straight forward hypothesis of a cell autonomous activity of Fat1 in MNs. Thus, it looked as if this project would be devoted to finding out whether specification or axonal growth was affected first, and to tackle the mechanism, keeping PCP principles in mind, possibly using a combination of genetics, in vitro, and in vitro approaches.

–

Prior to jumping to the conclusion that Fat1 indeed controlled MN specification and axon patterning cell autonomously, it was essential to determine whether the target muscle, the CM, had developed normally in Fat1 deficient embryos. Any muscle defect might indeed represent a potential cause – or consequence – of these MN phenotypes, or might even indicate an independent phenotype. With my team, we therefore performed the experiment that changed the course of this story. We visualized muscle development by whole mount in situ hybridization to detect MyoD RNA in myogenic progenitors. To my surprise, this experiment turned out not to yield the anticipated negative result. Instead, Fat1-deficient embryos exhibited a series of very robust muscle defects affecting morphogenesis of subsets of muscles in the face and shoulder, phenotypes that we described in our PLoS Genetics paper in 2013 [1].

When we first saw these muscle phenotypes, there was no reason a priori to dissociate our findings on the altered development of a selective group of muscles and associated defects in the corresponding motor neuron population. However, as I was reflecting on the possible significance of these results in terms of health and biomedical implications, and as, in these early days, I was also in a phase of heavy grant writing, I wondered whether these phenotypes could possibly correspond to symptoms of a rare human pathology that would not have been assigned yet any genetic culprit. I thus started exploring the human genome to see where the human FAT1 gene was located, and discovered the OMIM database (Online Mendelian Inheritance in Man). It didn’t take long to realize that the 4q35 region where FAT1 was localized was also known for being subject to chromosomal abnormalities associated with a human muscular dystrophy called Facioscapulohumeral dystrophy (FSHD) [2]. I was really amazed to learn that the muscle wasting symptoms in this disease affected a highly regionalized group of muscles, with a topography reminiscent of what we had seen in Fat1-deficient embryos. I was also intrigued by the fact that additional non-muscular symptoms of this disease, auditory and vascular retinal abnormalities, were emerging as hallmarks of defective planar cell polarity.

At that time, this was a completely unexpected finding. First, the FAT1 gene was relatively far from the true FSHD locus (where changes in the D4Z4 macrosatellite repeat array were considered to be the primary event linked to FSHD). This distance (~3Mb) was sufficiently large to guarantee unambiguous genetic mapping of a disease locus, but was small enough for a genetic linkage, and for FAT1 regulation to be affected secondarily to chromatin architecture changes occurring as a result of the FSHD abnormalities (although the concept of topologically associated domains (TADs) was not as popular then as it has become). Even though a possible deregulation/contribution of neighboring genes had been considered, FAT1 had largely been ignored or dismissed. Nothing in the literature at that time indicated any likely involvement of Fat1 in muscle biology, whereas Fat1 disruption in mice had instead been reported to result in severe kidney abnormalities [3], the latter not being part of the panoply of FSHD symptoms.

This is when I decided to radically change the topic of the lab. I therefore sought to identify potential collaborators who would help approaching the human disease link. I was lucky to turn to Nicolas Levy, who headed a department of medical genetics at La Timone Hospital in Marseilles, and was really enthusiastic about this new idea. With his help, we were later joined by Marc Bartoli, who started a group on translational myology in Marseilles, and by Julie Dumonceaux (now at UCL, London). To make a long story short, this fruitful collaboration has led to date to 3 papers exploring the possible link between FAT1 dysfunction and FSHD [1, 4, 5]. First, my team provided compelling data indicating that disrupting Fat1 in mice caused muscular and non-muscular phenotypes with an FSHD-like topography [1]. Second, we detected lowered FAT1 RNA and protein levels in affected muscles from FSHD patients with classical diagnosis, at fetal [1] and adult [5] stages. Third, we identified FSHD-like patients carrying pathogenic FAT1 variants, in absence of the traditional FSHD-causal abnormalities [1, 4]. These FAT1 variants included single nucleotide variants affecting RNA splicing or amino-acid structure [4], but also copy number variants deleting a putative cis-regulatory element in the FAT1 locus [1]. Such deletions had the potential to cause tissue-specific depletion of FAT1 expression, thus offering a logical explanation for why some phenotypes resulting from Fat1 deficiency in mice were not part of the clinical picture of FSHD symptoms in human patients.

Because the FAT1/FSHD link was so unanticipated and provocative, the road to our first publication turned into a long and emotionally agitated roller coaster ride. As a result, this paper was submitted multiple times, and went through several rounds of revisions for several journals, until it proudly landed in PLoS Genetics in 2013 [1]. As in the meantime we had produced mice with a conditional Fat1 allele, we were able to show in this first study that Fat1 deletion in the myogenic lineage reproduced in part the aberrant dispersion of myogenic cells in the limb observed in constitutive knockouts [1]. I already knew however that this only gave a partial account of the complexity of Fat1 functions. I just figured that it was unnecessarily challenging to simultaneously report the identification of a potential new genetic actor in a human pathology (even when only considered a modifier gene) and the complexity of its biology. This deserved to be studied as a developmental question per se, independently of its relevance for this disease, which somehow is still a controversial issue, despite our genetic evidence obtained with the Levy/Bartoli groups.

To approach the complexity of the Fat1-driven control of neuromuscular morphogenetic, the new study I have just published [6] gives a central position to one emblematic couple: the Cutaneous Maximus (CM) muscle, and its cognate MN pool. This MN pool can be recognized through its expression of the transcription factor Etv4, which is essential to specify the cell body position and dendritic architecture of these MNs [7]. The CM muscle emerges from the bulk of forelimb levels myogenic progenitors that undergo long range migration. Unlike other limb muscles, the CM muscle progenitors avoid entering the limb bud. Instead, they undergo a sharp caudal turn and start their subcutaneous progression. Throughout this migration phase, CM progenitors secrete Glial-cell-derived-neurotrophic-factor (GDNF), previously known for its neurotrophic activity and for its roles in motor axon guidance and in CM motor pool specification (as it induces Etv4 expression) [8]. Thus, by way of GDNF production, there is an intimate link between muscle progression and partner MN development. Among the muscles vulnerable to Fat1 deficiency, the CM muscle was also the easiest to approach quantitatively, by measuring the orientation of individual progenitor cells in our first study [1], and with simple area measurements in the new study [6]. Such an analysis illustrates how Fat1 disruption robustly interferes with expansion of the area occupied by Gdnf-expressing myogenic progenitors, by differentiating muscle fibers (Figure 2), as well as by axons of the cognate MNs.

–

Given my initial selection of Fat1 for its expression in the pool of MNs innervating the CM muscle, this was a region of non-myogenic Fat1 expression of obvious relevance to the phenotype. I was really excited by the possibility that Fat1 expression in the CM pool of MNs might represent a key driver of CM subcutaneous morphogenesis. Quantifications of morphometric parameters confirmed that MN-specific Fat1 ablation interfered with axonal growth and specification of the CM-innervating MNs. In addition, this also non-autonomously slows the subcutaneous progression of CM myogenic progenitors, hence expansion of the GDNF producing domain, without altering the rate of myogenic differentiation. However, this role turned out to be way more modest than what I initially anticipated, with an effect size considerably smaller than the phenotypes of constitutive mutants. This implied that another site of Fat1 activity had to be making a major contribution to neuromuscular morphogenesis.

An unanticipated third site of Fat1 expression that turned out to play a major role is the loose connective tissue (CT) that surrounds muscles. During their collective progression, myogenic progenitors and motor axons explore such connective tissues and select their migration trajectory. The CM progenitors in particular encounter an increasing gradient of Fat1 expression in this tissue. Furthermore, Fat1 expression appears fully preserved in mouse mutants lacking migratory muscles (Met or Pax3 mutants, in which myoblast migration is abrogated), indicating that the main site of peripheral Fat1 expression represented non-myogenic cells. To establish the contribution of CT-Fat1 to muscle morphogenesis, Fat1 activity was ablated in the lateral plate derived mesenchyme, in a domain (driven by Prx1-cre) including the limb and flank connective tissues encompassing a large part of the territory through which the CM migrates. This led to severe non-cell autonomous disruption not only of the progression of CM progenitors and subsequent myofiber elongation, but also of motor axon elongation across the same territory, and of acquisition of CM fate characteristics. Similar results were obtained with a Cre line allowing stage-controlled Tamoxifen-inducible Fat1 excision in Pdgfrα-expressing connective tissue at the time of myoblast migration, with a lesser effect size matching the lower excision efficiency. These data identify connective tissue as a tissue type in which Fat1 activity is mandatory for neuromuscular morphogenesis. They imply that Fat1 signaling activity in CT cells is necessary for them to emit proper signals promoting collective myoblast migration, myogenic differentiation, motor axon growth, and MN specification, with the exact mechanisms underlying these cell interactions remaining to be explored.

–

In conclusion, looking back on how the story started, the loop has been closed by the identification of gene functions that underlie the phenotype which initially attracted my attention. Although my intuition that this had to be connected to the neuronal expression of this atypical Cadherin was not completely wrong after all, conditional mutagenesis nevertheless revealed that this function was modest in regard of the unexpected predominant activity in a cell type – connective tissue – that I had initially not considered. I have learned many lessons along the way of this unusual story. Above all, I learned not to fall in love with a favorite hypothesis but instead to be open to what the facts tell us: when the results of an experiment are the opposite of what you expect, this pushes you to think outside the box, and challenges you to remain unbiased in conceiving experimental plans and analyzing data. It is now time to switch scale, and start exploring what happens at the connective tissue muscle interface from a cell biologist’s point of view.

References

1. Caruso, N., et al., Deregulation of the protocadherin gene FAT1 alters muscle shapes: implications for the pathogenesis of facioscapulohumeral dystrophy. PLoS Genet, 2013. 9(6): p. e1003550.
2. DeSimone, A.M., et al., Facioscapulohumeral Muscular Dystrophy. Compr Physiol, 2017. 7(4): p. 1229-1279.
3. Ciani, L., et al., Mice lacking the giant protocadherin mFAT1 exhibit renal slit junction abnormalities and a partially penetrant cyclopia and anophthalmia phenotype. Mol Cell Biol, 2003. 23(10): p. 3575-82.
4. Puppo, F., et al., Identification of variants in the 4q35 gene FAT1 in patients with a facioscapulohumeral dystrophy-like phenotype. Hum Mutat, 2015. 36(4): p. 443-53.
5. Mariot, V., et al., Correlation between low FAT1 expression and early affected muscle in facioscapulohumeral muscular dystrophy. Ann Neurol, 2015. 78(3): p. 387-400.
6. Helmbacher, F., Tissue-specific activities of the Fat1 cadherin cooperate to control neuromuscular morphogenesis. PLoS Biol, 2018. 16(5): p. e2004734.
7. Livet, J., et al., ETS gene Pea3 controls the central position and terminal arborization of specific motor neuron pools. Neuron, 2002. 35(5): p. 877-92.
8. Haase, G., et al., GDNF acts through PEA3 to regulate cell body positioning and muscle innervation of specific motor neuron pools. Neuron, 2002. 35(5): p. 893-905.

(2 votes)

Loading...

We recently published a manuscript in Cell that describes a method to image transcription factor concentration dynamics in real time, in living embryos, using a nanobody-based protein tag that we call the “LlamaTag.” We were particularly excited about these investigations because this new technology overcomes a major technical obstacle to understanding how gene-expression dynamics are regulated in development, which has held back the field for decades.

From Snapshots of Dead Embryos to Movies

Remember when you were first starting to study developmental biology? In one of your lab sections, you probably participated in that classic teaching activity where you watched the cephalic furrow form in Drosophila embryos. Maybe you were one of those students struggling to move embryos with an eyelash glued to a toothpick, and the whole time you were thinking, “I need to hurry up—this thing is gastrulating while I’m messing around here!” There’s a strong contrast between watching development unfold before your eyes and scrutinizing snapshots of FISH data from dead, fixed embryos. This contrast hinders more than budding scientific excitement: how can we truly understand dynamic processes by analyzing static data?

I was particularly struck by this conundrum as a postdoctoral researcher as my mind kept replaying a video of a gastrulating embryo. How could I bridge this gap so that developmental biologists could accurately model—and ultimately construct—dynamic systems as complex as embryos? At the simplest level, we would need to know the input and the output of each genetic circuit as it changes throughout development in order to deduce the logic gates at play. I tackled the output end first: as a postdoc, I developed a technique to quantify transcription in flies at the single-cell level by tagging a gene’s nascent mRNA molecules with fluorescent proteins. How could we accomplish the same feat with the DNA-protein interactions that constitute the input to these genetic circuits? I decided that solving this challenge would be a major driver of the research in my own lab. Jacques Bothma was equally excited about this challenge and decided to join the project as a postdoctoral fellow.

Typically, one would quantify the concentration of a transcription factor by fusing it to a fluorescent protein. However, in flies, worms, fish, and frogs, these fluorescent take more than 40 minutes before they mature and become fluorescent. This delay is actually a major problem: the activators and repressors that drive development often exist for less than 10 minutes before they are degraded. By the time the fluorescent proteins actually became fluorescent, the action they were supposed to report on is already over!

Jacques had a great idea: instead of relying on the synthesis of fluorescent proteins, why not use the localization of already matured fluorescent proteins to report gene expression? Our Cell paper describes the engineering and implementation of this idea with LlamaTags, a technology that enables quantitation of the dynamics of transcription-factor activity without limits from the slow maturation of traditional fluorescent fusion proteins. Instead of these fusions, we employed nanobodies, which are small, highly specific, single-domain antibodies that are raised in llamas (hence LlamaTags). We fused a transcription factor of interest (we started with Hunchback) to a nanobody raised against eGFP, and expressed the construct from the endogenous locus for that transcription factor. Importantly, the embryo was engineered to contain maternally deposited eGFP, which means that eGFP is already mature before the transcription factor of interest is expressed (no waiting around for the fusion to mature!). When the transcription factor-nanobody is translated, it binds cytoplasmic eGFP within seconds, yielding a quantitative increase in nuclear fluorescence when the transcription factor moves to the nucleus to perform its regulatory function. LlamaTags therefore deliver a direct readout of the instantaneous transcription-factor concentration in a given nucleus.

Through a variety of experiments, we showed that LlamaTags serve as specific and faithful reporters of the endogenous concentration dynamics of transcription factors during development, thus capturing the input pertinent to these circuits. So, we finally had the two pieces needed to solve the puzzle of measuring input-output functions. LlamaTags made it possible to measure the input concentration of transcription factors in individual nuclei, while MS2 revealed the transcriptional activity of specific genes to these input levels.

By fusing a LlamaTag to Fushi-Tarazu (Ftz), we obtained measurements of its nuclear concentration during development, from which we extracted the in vivo degradation rate of the construct. We were pleased when our data revealed rapid fluctuations in Ftz concentration—consistent with previous reports of protein bursts that likely arise from stochastic fluctuations in mRNA concentrations. We nailed down this relationship, and uncovered exciting evidence of inter-nuclear communication within the embryo. This coupling could be a major driver of the sharp boundaries that dictate development of the fly embryo. However, our greatest satisfaction came from simultaneously visualizing input transcription-factor activity and output transcription, at the single-cell level, in real time, as stripe 2 of eve was laid down in live embryos:

A New “Microscope” for the Central Dogma in Development

Our recent work establishes a powerful pair of technologies—labeling transcription with MS2 and labeling DNA-protein interactions with LlamaTags—that together constitute a “microscope” for visualizing and interrogating the activity of genetic circuits as they function. We envision that LlamaTags can be applied to quantitatively measure the flow of information along regulatory networks in any multicellular organism that is amenable to transgenic control and live imaging. LlamaTags literally light up the central dogma, in real time, as development unfolds. Importantly, we can do more than map input to output: we can quantitate the dynamics of these connections over time. That’s the difference between trying to create a stop-motion movie where you need a new actor for each frame (a dead embryo), and actually observing development unfold in real time.

Physical Biology of Living Embryos

Like many of you, we believe that in order to construct a system, we should first understand it in quantitative detail. We view this as a call for reaching a “predictive understanding” of development, through which we can calculate developmental outcomes from knowledge of the concentrations of input transcription factors and the DNA regulatory sequence. To reach this predictive understanding, we believe that a powerful dialogue is necessary in which theoretical models make predictions that are subsequently tested experimentally, with measurements fed forward into the model to generate a new cycle of experiments. Our work in Cell is a crucial early step toward enabling the biophysical dissection of developmental programs by making it possible to measure the very same input-output functions predicted by our theoretical models.

(6 votes)

Loading...

Applications are now open for this year’s Gene Regulatory Networks for Development which will be at The Marine Biological Laboratory in Woods Hole, USA from October 14- 27.  The application deadline is July 20.  The course is for graduate students, postdoctoral researchers, and faculty members and it focuses on using experimental data and computational modeling to analyze gene regulatory networks that are key to development in animals and plants. 

This unique course is an intense and always interesting experience and has drawn rave reviews in all of its previous incarnations. Students will meet with experts in the field for an in-depth treatment of experimental and computational approaches to GRN science. Through lectures, highly interactive discussions, and group projects we will explore the GRN concept and how it can be applied to solve developmental mechanisms in various systems and contexts. Topics include structural and functional properties of networks, GRN evolution, cis-regulatory logic, experimental analysis of GRNs, examples of solved GRNs in a variety of developmental contexts, and the computational analysis of network behaviour by continuous and discrete modelling approaches. 

Travel fellowships are available.

For more information about the course, go to www.mbl.edu

The 2018 GRN course faculty:

Scott Barolo, University of Michigan

James Briscoe, The Francis Crick Institute, London

Fernando Casares, Andalusian Center for Developmental Biology, Spain

Ken Cho, University of California, Irvine

Doug Erwin, Smithsonian Institution

Robb Krumlauf, Stowers Institute

Bill Longabaugh, Institute for Systems Biology, Seattle

Lee Niswander, University of Colorado, Denver

Isabelle Peter, Caltech

John Reinitz, University of Chicago

Ellen Rothenberg, Caltech

Trevor Siggers, Boston University

(No Ratings Yet)

Loading...

The Karam Teixeira lab (https://karamteixeira.wixsite.com/website) at the University of Cambridge (Department of Genetics) is looking to recruit a Research Assistant to support a range of projects being carried out in our research group. We are primarily interested in germline development (stem cell regulation and genome defense mechanisms) and we use the Drosophila ovary as a model system (see Sanchez et al, Cell Stem Cell 2016; Teixeira et al., Nature 2017).

We are looking for enthusiastic and proactive candidates with expertise in standard molecular biology techniques (including genotyping, cloning, and RT-qPCR). Prior experience in fly genetics would be an advantage – although training can be provided where necessary. Candidates should have a B.Sc. or M.Sc. degree in a relevant biological subject and should possess good communication skills and the ability to work effectively as part of a team.

• How to apply:

To apply, please follow the link: http://www.jobs.cam.ac.uk/job/17837/

The position start date is flexible. Application deadline: July 13^th, 2018.

For an informal discussion about this position, please contact Dr. Felipe Karam Teixeira (fk319@cam.ac.uk; https://www.gen.cam.ac.uk/research-groups/karam-teixeira/).

(No Ratings Yet)

Loading...

Post-doctoral positions are available in Dr. Samantha Brugmann’s lab to study vertebrate craniofacial development and disease, with an emphasis on the role of the primary cilium in these processes. For information about specific research areas see http://www.cincinnatichildrens.org/research/divisions/p/plastic/labs/brugmann/default/.

Qualifications: Applicants should possess a Ph.D. in a relevant field, such as Biology, Biochemistry, Genetics or another related discipline and be highly motivated, independent and organized. Successful applicants will have a record of communicating research results via publications and/or professional presentations, and be willing and able to participate in collaborative, interdisciplinary research projects. Experience in developmental biology, cell and molecular biology and avian/murine model systems is desirable. Preference will be given to applicants with a proven record in craniofacial research.

Please submit your application to Dr. Brugmann with the following information: A cover letter, a statement of research interests, and a CV with contact details for 3 referees.

Contact: Samantha Brugmann, PhD    Email Address: Samantha.Brugmann@cchmc.org

Cincinnati Children’s Research Foundation

Cincinnati Children’s Hospital Medical Center (CCHMC) is a premier pediatric research institution with over 900 diverse and productive faculty members. Here, researchers work collaboratively across specialties and divisions to address some of the biggest challenges we face today in improving child health. A strong network of research support services and facilities, along with institutional commitment to research, push our team of faculty, postdocs and support staff to explore the boundaries of what is possible, leading to significant breakthroughs. We are driven by our mission to improve child health and transform the delivery of care through fully integrated, globally recognized research, education and innovation.

Post-doctoral research fellows at Cincinnati Children’s are valued for their unique interests and strengths, and are supported by our institution’s strong programming for post-docs through the Office of Postdoctoral Affairs and the Office of Academic Affairs and Career Development. Mentoring, support for international students and an emphasis on crafting high-quality grant proposals are only a few of the features that set our program apart. Cincinnati Children’s is a respected part of the broader, and very vibrant, Cincinnati community. With a thriving arts scene, numerous festivals celebrating music and food, a passionate fan following for our college and professional sports teams, and a variety of opportunities for outdoor activities, our region is truly a great place to work and live.

For further information about working at CCHMC or living in Cincinnati, please contact the CCHMC Postdoc recruiter, Uma Sivaprasad, PhD, at research@cchmc.org

To apply online go to: http://www.cincinnatichildrens.org/careers/apply/default.htm and search for job (requisition) number 95516 or 95517.

Cincinnati Children’s Hospital Medical Center is an Affirmative Action/Equal Opportunity Institution

(No Ratings Yet)

Loading...

Applications are invited from highly motivated individuals dedicated to peruse a PhD and who are interested in nerve/tumour interactions and nervous system development. 

PROJECT DESCRIPTION

During tumour progression nerves and tumours interact resulting in tumour cells using nerves as a metastatic route out of the organ, through a process called perineural invasion (PNI). While this is robustly documented at a histological level, the molecular mechanisms underling this, remain poorly understood. You will join an interdisciplinary collaborative team focusing on the signalling between nerves and tumours, leading to provocation of nerve plasticity (such as growth and remodelling) and tumour metastasis via nerves. You will engage a systematic strategy to identify these mechanisms by (i) examining how nerves form/grow/remodel and migrate normally during embryo development and (ii) pathologically in tumour models with an initial focus on pancreatic ductal adenocarcinoma (PDAC). The formal PhD qualification you will obtain is a PhD in medical science with a focus on cell biology.

LOCATION

The laboratory is located at IMB, Umeå University, Sweden. IMB is an interdisciplinary department, which focuses on questions in basic and medical sciences and provides an interactive modern environment with good core facilities within the wider university. The working ‘day to day’ language in the laboratory is English. The position is for 4 years and is fully funded.

APPLICATION

Apply at the following online system by 24^thAugust 2018 (reference code AN 2.2.1-1186-18): https://umu.mynetworkglobal.com/en/what:job/jobID:215647/

Further application information, including application qualifications, requirements and merits can be found at that webpage.

INFORMAL ENQUIRIES

Informal enquiries may be directed to Dr. S. I. Wilson (sara.wilson@umu.se).

Laboratory webpage: www.imb.umu.se/english/research/research-groups/wilson-laboratory/?languageId=1

We look forward to your application!

(No Ratings Yet)

Loading...

Following our latest data release, the DMDD website (dmdd.org.uk) now contains detailed phenotype data for nearly 700 embryos from 82 different knockout mouse lines. Highlights include the identification of limb defects and cysts in Col4a2 knockouts and replication of the major features of Meckel syndrome in B9d2 knockouts.

We have begun to add immunohistochemistry image data for the brain and spinal cord of some embryos at E18.5. These images give further information about lines in which the embryos appeared morphologically normal at E14.5, but were still not viable. We have also added viability data for every line at both E9.5 and E14.5.

Together with the placental phenotype data that we hold for more than 100 knockout lines, the DMDD website is a rich resource for those investigating the effect of gene mutations on embryo development, and may provide clues about the genetic basis of rare diseases.

LIMB DEFECTS SEEN IN Col4a2 KNOCKOUTS

In humans, COL4A2 mutations have been linked to porencephaly, a rare disorder with phenotypes that include the development of intracranial cysts. In the latest DMDD data, Col4a2 knockouts have a variety of nervous system disorders in line with porencephaly. However, all four embryos also show abnormal autopod morphology and cysts between the nasal septum and the oral cavity, as well as other morphological defects.

B9d2 KNOCKOUTS MODEL MECKEL SYNDROME

In humans, mutations of the gene B9D2 have been linked to Meckel syndrome, a severe disorder caused by dysfunction of the primary cilia during the early stages of embryogenesis. Meckel syndrome is characterised by multiple kidney cysts, occipital encephalocele (where a portion of the brain protrudes through an opening in the skull) and polydactyly, but it also commonly affects the brain and spinal cord, eyes, heart, lungs and bones.

B9d2 knockout mouse embryos included in our latest data release show the major features of Meckel syndrome, including polydactyly and defects in the brain, peripheral nervous system, heart and vascular system. They also display situs defects, where the left-right asymmetry of the body did not develop as expected. The image below shows a B9d2 knockout embryo with left pulmonary isomerism and symmetric branching of the principle bronchi from the trachea.

NEURAL IMAGE DATA NOW AVAILABLE

In around 20% of embryonic lethal lines, embryos appear morphologically normal at E14.5 but still go on to die before or shortly after birth. To understand more about why these embryos were not viable, DMDD colleagues Professor Corinne Houart and Dr Ihssane Bouybayoune at Kings College London analysed the lines at E18.5 – when embryo development is almost complete. They used immunohistochemistry to identify abnormalities in the brain and spinal cord that could not be picked up in our standard, whole-embryo morphological analyses. This data is now available for the line Trappc9, and additional lines will be added in future data releases.

Neural images are available as 20-micron sections through the brain and spinal cord, and the images from different embryos can be compared side by side using the stack viewer. A separate Nissl stain was used to highlight neural death and these images can also be explored online.

A FULL LIST OF NEW DATA IN THE LATEST RELEASE

• HREM image data added for Auts2, Col4a2, Ehbp1l1, Frrs1, Gtpbp3, Nr0b1, Pitx1, Pthlh
• HREM phenotype data added for Auts2, B9d2, Col4a2, Morc2a, Zmynd11

(No Ratings Yet)

Loading...