What do you know? Using a flakey light as a control for a flakey light led me on a merry dance. Here are some data (all mean ± standard deviation, for 20 frames each):

Light Source           Trial 1         Trial 2

IR diode array:       130 ± 3.9       167 ± 2.3

Compact fluoro:     151 ± 0.7         148 ± 1.4

100W tungs/halo: 163 ± 0.8       162 ± 0.5

40W incandesc:     145 ± 0.9       154 ± 0.9

The trials were independent and the measurements were from a rectangle placed on the image that had an area of 10,000 pixels squared, reflecting about 2% of the image area. The mean values are not important, only that they are similar to one another. The key datum is the standard deviation. Note how the IR diode array has the largest deviations. This matches with the perceptible flickering when watching sequences from that source but not so much from the others. I made similar 20-frame image sequences on the microscope too, using its light, and the deviation was around 5, high like the diode array. I did not bring those data home so it is not included on the table above.

I can accept standard deviations of around 0.5. Happily, the microscope has a hatch at its base (which remember is its side in my horizontal set-up) and by opening the hatch and removing the bulb, I can shine light from any other source through the optics and onto the sample. In fact, the third light source in the table above, the 100W tungsten halogen, is connected to a long fiber optic cable, and so the temporary solution is to use that light. A more permanent solution will be to repair the electronics in the microscope’s base, it is probably the socket that has corroded contacts. However, with Xiaoli counting down the days until she leaves I cannot yet send the base of the ‘scope out for a lengthy sojourn in the repair shop.

OK, so this sounds good: problem traced to defective electronics powering the microscope light and solution found — bypassing the light. With this, Xiaoli and I were excited to make some movies with 60 sec time intervals. Here is the principal component analysis for the first one:

The blue line, principal component number 1, fluctuates, and its period looks similar to the one I posted a few weeks ago, that was obtained with 30-sec frame intervals. But look at that huge rise in the line over the first 20% or so of the graph. Ah yes. This root grew faster and faster and faster over that period before stabilizing. Don’t know why. Maybe the temperature in the microscope room was higher than in the growth chamber? Maybe it was the music Xiaoli was listening to while she collected the images? I am kidding – I have no idea whether she listens to music, it is just an example of some random factor. Whatever caused it, makes the results a little less than beautiful.

For the next one, Xiaoli made sure the temperature in the ‘scope room was about the same as in the growth chamber and the plants were allowed to equilibrate on the microscope for 45 min or so before the image sequence was started. Here is the result:

No spurious change of root growth rate, and fluctuations in the first principal component showing a fast period. Good. But, to my chagrin, the velocity profile for three or four of the time points failed; that is, there was clear nonsense at each end of the profile. By nonsense, I mean values that are physiologically impossible. And, for these time points, the intensity was stable. Hmmm. I was able to get ‘normal looking’ profiles by adjusting one of the parameters used by the software to calculate the profile. This is hardly optimal but since I needed to make this adjustment on only a few of the time points, I did it anyway to have a look at the results.

Then, I tried the next root, and every time point failed (well, the first 10 failed, I gave up on doing more of them after that). The little tweak I gave the software (for the root in figure 2) made no difference at all. These failed velocity profiles look exactly like the ones that failed that led me to discover the light intensity noise. Now that I have stable light, I was dismayed, to put it mildly, to find clearly failed velocity profiles. Ugh.

After ruling out consuming large quantities of vodka and barbiturates, I considered scientific possibilities. What is different about these roots compared to ones I worked with in UK? The most salient thing is growth rate. The roots here are growing about 25% faster than they did across the pond. And the root that worked fine (Fig 1) was growing more slowly than the one that had a few failed points (Fig. 2). Why should that matter? Well, it means that, in the given (60 sec) time interval, the roots in Mass enlarge more than their UK counterparts. The velocity profile generating software (named StripFlow) first registers the two fames – it essentially centers them with respect to each other and then measures the local changes. If the amount of growth is too much, it might not be able to center the images properly.

Hmmm. Sounds plausible – but how can I tell? I realized that in all of these failed profiles, it was always the ends where the failure occurred. About 20% of the profile at each end was bad, occasionally at just one end, but never in the middle. So I took a time point from the really bad root and cropped the images to remove 20% of the end of the root, and ran it like that. Hey presto! A perfect velocity profile. I think this implicates the registration. If instead, there were some anomaly in the images (as for example caused by fluctuating gray levels caused by fluctuating light) then cutting off one end of the root should make no difference to the bad stuff at the other end. Conversely the fact that cutting off one end solved the glitches completely supports the idea that the large amount of growth over the 60 cannot be registered over the whole root image, but can be registered over a shorter length of root.

Now, what I can do, based on that claim, is to crop the whole sequence as described above, not just the first time point; and by doing so, all of the profiles should be fine. We will see. Ironically, by making the crop the total root length in the profiles will match the UK profiles rather well, so that could be useful. Frankly, if the third root with 60 sec frame intervals appears to have short-period fluctuations in principal component 1, then I’ll be entirely happy for the rest to move to 30 sec frame intervals. There is no “logical” reason why the frame interval should matter and seeing three examples where it doesn’t will be good enough for me.

Such faulty registration also explains something that always bothered me about the dodgy-light explanation – sequences made with 30 sec frame intervals worked well. Given how fast the light fluctuates, this really made no sense. Had I infinite amount of time, I would go back to some of those original sequences taken with bad fluctuations and see whether they can be fixed by cropping off the back end of the root. I suspect so.

But this whole rigmarole is salutary. The dodgy light hid my attention from the more important symptom, overly rapid growth. This illustrates nicely (or perhaps nastily) the difficulty about troubleshooting when two things go wrong at once. I am sure that the dodgy light cannot be good but it seems that the rapid growth was really bad.