Why do I get young ages (like uncalibated BLs) for node dating in MrBayes?

Question

[17th Level Botanist]

I am new to divergence time dating in MrBayes 3.2, though I have used BEAST often. I am analyzing some data for which I prefer to use MrBayes for reasons not applicable to this question.

I ran an analysis with node constraints and calibrations as follows:

Begin mrbayes;

[Hard constraints here; Identical to those listed in prset topologypr below. Not provided here in Biostars for space considerations.]

lset nst=6 rates=gamma ngammacat=4;

prset topologypr = constraints(Ingroup,AesculusHandeliodendron,AesculusGenus,sectCalothyrsus,AsianClade,node1,node2,node3,sectsParryanaPaviaMacrothyrsusHippocastanum,sectsParryanaPaviaMacrothyrsus,sectPavia,node4,node5,sectAesculus);                                                                                                                                                                                                                
prset clockratepr = lognormal(-5,0.6); [mean 0.00001, SD 0.005; centered around small numbers with room for increasing]
prset clockvarpr = tk02; [use with default settings]
prset sampleprob = 0.01;

prset treeagepr = offsetlognormal(72.1,73.53,1.25);
calibrate AesculusGenus = offsetlognormal(66.0,66.1,0.8);
calibrate sectAesculus = uniform(1.75,22.5);

prset nodeagepr = calibrated;

mcmc ngen=20000000 nruns=2 nchains=4 samplefreq=1000 checkpoint=yes checkfreq=1000 temp=0.2 mcmcdiagn=no;

;
end;

I can confirm from a log of what is echoed to screen in MrBayes that these priors were set as intended:

  Defining constraint called 'sectAesculus' [others defined above; not included here for space]
  Setting Nst to 6
  Setting Rates to Gamma
  Setting Ngammacat to 4
  Successfully set likelihood model parameters
  Setting Topologypr to Constraints
  Successfully set prior model parameters
  Setting Clockratepr to Lognormal(-5.000000,0.600000)
  Successfully set prior model parameters
  Setting Clockvarpr to TK02
  Successfully set prior model parameters
  Setting Sampleprob to 0.01000
  Successfully set prior model parameters
  Setting Treeagepr to Offsetlognormal(72.10,73.53,1.25)
  Successfully set prior model parameters
  Setting age of constraint node 'AesculusGenus' to Offsetlognormal(66.00,66.10,0.80)
  Setting age of constraint node 'sectAesculus' to Uniform(1.75,22.50)
  Setting Nodeagepr to Calibrated
  Successfully set prior model parameters
  Setting number of generations to 20000000
  [...and so on]

However, my results in no way reflect these node age calibrations and look more like uncalibrated branch lengths. See here for a screen shot of the MCC Tree displayed in FigTree with node ages on branches.

This is not a case of extremely strong signal in the data leading to results outside of the prior bounds as I do get the expected old ages using these calibration priors in BEAST.

What have I done wrong?

Thanks in advance!

Answer 1

[17th Level Botanist]

I will answer my own question and feel a little silly in the process. The issue was that I actually had to summarize the trees following the analysis using sumt. This generates the BLs in units time on a consensus tree. I find this a little objectionable as I should be able to derive the BLs in units time myself easily from the output. I'm sure it's possible, but not made apparent.