Culture conditions, and strain conditioning
Stocks of 10 T. thermophila strains (Table 1) were maintained in an axenic (bacteria-free) culture medium (2% proteose peptone, Difco no. 3; 0.2 % yeast Difco; demineralized water; all autoclaved after mixing) (see also ), and new cultures were made every three to five weeks by transferring 500 μl of an old stock to a 50 ml tube filled with 40 ml of new culture medium. All work was done under sterile conditions under a flow-hood. Culture tubes were kept at 27°C at a 12 h light/12 h dark cycle in an incubator with sun-glow bulbs emitting light corresponding precisely to natural sunlight.
Data collection using digital imaging
A procedure based on image analysis was developed to collect data from digital pictures taken under the microscope. For each sample to be measured, aliquots of 10 μl were taken by pipette after vortexing for 10 seconds at medium speed to homogenize the cell suspension. Each aliquot was then placed in a chamber of a Plexiglas cell count slide (KOVA Glasstic Slide 10 no grids from Hycor Biomedical Inc., California, USA, reference 87146). One grey level picture was taken per chamber using a digital camera (Nikon Coolpix 4500; settings: focal length at 8 mm, fixed exposure at 1/60 s, f/6, manual focus fixed to infinity with focus obtained manually using the microscope) mounted via an eyepiece (Olympus Coolpix digital coupler) on a microscope (Olympus CX-41; settings: 10 × oculars, 4x/013PhL planar lens, condensator placed on phase 2, maximum light); an example of such a picture is shown in Figure 6. Conversion of measurement units on pictures (pixels) into real units (μm and μm2) were obtained from a picture made with the same setup of a similar cell count slide with a calibrated grid of known size and volume (KOVA Glasstic Slide 10 with counting grids, reference 87144).
Pictures were analyzed using the public domain image-analysis program Scion Image  with a macro that we developed to automate the treatment of large numbers of pictures. The macro first converted each grey level pixel into black or white, with the threshold set at 200 as it proved to optimize discrimination of cells from dust and scratches. Then the "Analyze Particles" function was used to export to a text file a list of parameters for each particle present on the picture. Particles smaller than 35 or bigger than 500 pixels were discarded because these sizes are far outside normal sizes for T. thermophila cells  and hence represented dust. For each cell, the following variables were measured: size, color density (mean, mode and std; using the grey level picture), position (X and Y coordinates), perimeter, and three variables from the best fitting ellipse: major axis, minor axis and angle (angle between the major axis and a line parallel to the x-axis of the image). A shape variable was computed as the ratio major/minor axis for each cell; a perfect circle had shape = 1 and shape increased for more elongated cells. The density of cells per picture was obtained from the number of particles identified.
To ensure that results were comparable between all the pictures, the method was designed using fixed parameters only, both for taking the pictures (microscope and camera parameters) and analyzing them (Scion Image macro parameters) with settings described above. Extensive fine-tuning of each step was performed to define an optimal method, and results of picture analyses were manually checked on several hundreds of cells to verify the validity of measures recorded. Artefacts (missed cells, or dust considered as cells) were less than 5% of the results and distributed randomly among pictures; therefore they did not bias the results even if they did increase the random noise in the data.
For each of our 10 T. thermophila strains (Table 1), we studied (1) growth from low density in presence of nutrients, (2) survival in a medium devoid of nutrients (starvation), (3) dispersal in presence of nutrients, and (4) single cell colonization capacity in presence of nutrients. Experiments were started from cell cultures inoculated one week before. 10 pictures of the initial tube were taken and density of cells was measured so that experiments could be started with a precise cell density by diluting the solution accordingly. All experimental setups were autoclaved before experiments and conserved in the incubator under the same conditions as strain stocks (see above).
Growth from low density in presence of nutrients
Growth rate and carrying capacity in nutrient-rich conditions were examined by placing approximately 50 000 cells in a 12 ml polypropylene culture tube (Greiner Bio-One), adding nutrient medium till a total volume of 5 ml to reach a starting cell density of 10 000 cells/ml. Three replicate tubes were made per strain. Pictures were taken at six different times (from 0 to 192 h).
From the raw cell measurements at the replicate level, we estimated five major variables describing cell population growth: carrying capacity K and growth rate r, and final (i.e. at carrying capacity) cell density, size and shape (Table 2). Growth was estimated by fitting the following logistic growth function to successive densities Dt over time t (t > 0): +++++++++++++++++++++
with D0 being the initial density at time 0, and two parameters to be estimated: growth rate r and carrying capacity K. The fitting was achieved through least-squares non linear programming using PROC NLP in SAS . Statistical difference between replicates and strains were tested by comparing a model with equality constraints of K and r parameters between replicates of each strain, a model with equality constrains of K and r parameters between strains, and a model where parameters K and r were free to vary between strains and replicates; model selection was achieved using the AIC criterion [94-96].
Survival in a medium devoid of nutrients (starvation)
To remove all traces of nutrients from cell cultures, the contents of five tubes per strain were concentrated and washed with water (similar to ). This was done by centrifugating tubes at 2000 r.p.m. for three minutes, after which the supernatant (approximately 35 ml) was aspirated using a vacuum pump. The pellet, containing the cells, was then resuspended by adding 35 ml of autoclaved demineralized water and then tubes were vortexed gently for 5 seconds to detach any cells sticking to tube walls. This process was repeated four times for each tube. After the fifth centrifugation, the cells were left in a volume of 5 ml (i.e. no water was added) but still vortexed gently. Hence, this washing process had the effect of diluting the nutrient solution originally present in the tube 4096-fold, effectively leaving the cells nutrient-deprived. The five tubes of concentrated cells per strain were then pooled into a single initial tube. A volume containing approximately 200 000 cells was placed in a 12 ml polystyrene culture tube and autoclaved demineralized water was added to increase the total volume to 2 ml. Three replicate tubes were set up for each strain. For each replicate, pictures were taken at 14 time steps up to 408 h (every four hours during the first 24 h, this delay increasing to 24 h and 48 h in the course of the experiment).
From the raw cell measurements at the replicate level, we estimated five major variables describing the changes in cell density and shape: survival as a density sum over time, mean and variance of maximal cell elongation, elongation persistence as a sum of cell elongation over time, and the frequency of putative disperser morphs after 8 h, time when the maximal elongation was observed (Table 2; see also Figure 6 and Additional file 1).
Additional file 1. Fast, directional swimming behavior of dispersal morphs compared to other cells. This video clip shows the rapid, directional swimming behavior of T. thermophila dispersal morphs compared to other cells.
Format: WMV Size: 2.6MB Download file
Dispersal in presence of nutrients
Dispersal rate was measured using a two patch system setup consisting of two tubes (1.5 ml polypropylene test tubes, Greiner Bio-One) connected by a small horizontal pipe (silicone Tygon tubing, 6 mm diameter, 17 mm long) inserted through a hole drilled on the side of each tube. For each setup, a cell suspension volume corresponding to 300 000 cells of a given strain was taken from a culture tube (the initial tube) and introduced in one patch (the start tube), by the use of a pipette. By pipette we then filled up the target tube with cell-free nutrient medium, filling up in this way also the connecting tube and the remaining free space in the start tube. Both start and target tube contained 1 ml in the filled-up state. The setup was then incubated at 24°C. Pilot experiments using an ink solution instead of a cell suspension had verified that filling up the setup and placing it in the incubator did not induce perturbations that could displace cells from the start patch.
After 17 h we estimated the number of cells that had swum from the start to the target patch. This was done by closing both ends of the connecting tube with clamps and then pipetting off the solution in both start and target tubes and taking pictures for the two tubes. The experiment was performed over several days in a semi-block design due to constraints on manipulation time. A total of six replicates were done for each strain, in the manner of three replicates per strain for two out of four blocks (dates).
From the raw cell measurements, we estimated three major variables at the replicate level: initial cell shape, dispersal rate and cell elongation (Table 2).
Single cell colonization capacity in presence of nutrients
To assess colonization capacity, individual cells were isolated from a one-week old cell culture tube using a handheld 10 μl pipetteman and a binocular dissection microscope. Each cell was placed separately in a 12 ml test tube with nutrient medium (the new patch to be colonized), and 10 replicates (10 different tubes) were created per strain. The entire set-up was replicated one week later, leading to a total of 20 single-cell tube experiments per strain. In the second experiment, however, some tubes became infected and were discarded. All tubes were incubated at standard temperatures in the incubator for eight days, after which the success of single cells to colonize the new patches by surviving and undergoing multiple cell divisions was determined.
To determine colonization success we took out a 10 μl aliquot from each experimental tube, after gentle vortexing, placed it on a slide and counted the number of cells directly under binoculars. We then assigned a 1 or a 0 to each experimental tube according to whether any cells were found or not (presence of at least one cell versus none), and calculated (over replicate tubes) the probability of successful colonization for each strain.
The basic data for studying life-history trait covariation between strains were the estimates at the replicate level of 14 parameters (Table 2). Covariation between these parameters was studied through Spearman's correlation  and principal component analysis , and differences between strains by generalized linear models  and discriminant analyses , all implemented using SAS software.
Because replicates of a given strain were not linked between experiments, we used a permutation at the replicate level procedure to correlate parameters from different experiments. This procedure prevented discarding the information on variation between replicates of a given strain. The replicates of a given strain were randomly associated across experiments 1000 times, and a correlation was computed for each random association. The mean Spearman's correlation (r) was reported as covariation measure between the two parameters studied. Statistical significance of this correlation (showing that it differed significantly from zero), however, was based on the probability of obtaining the observed proportion of significant correlations by chance (in the 1000 simulations). The distribution of this proportion of significant correlations under the null hypothesis of no correlation was also obtained by resampling, with observed values randomized for replicate and strain, breaking any existing correlation. In total, 1000 such random associations were done and the proportion of significant correlations computed. This procedure was repeated another 1000 times, and the P-value for the test was computed as the proportion of random associations with a proportion of significant correlations greater than or equal to the one observed.