rounds of registration between different samples. The scripts
necessary for these steps can be found in [4].- Select the target channels of a relatively large number of sam-
ples. This number depends on the stereotypicity of the animal
and the quality of the imaging, but a rule of thumb is above a
hundred. - Roughly orient each sample to the same position (e.g., anterior
side up and ventral side forward). - Measure the size of all the samples, and select one of them (e.g.,
one that has a median size) as the preliminary reference. - Perform a round of rigid registration (e.g., using the plugin
‘Rigid Registration’ in Fiji) to the preliminary reference, and
average over the aligned samples to create a “rigid reference.” - Perform a round of affine registration over the oriented sam-
ples to the rigid reference, and average the resulting stacks to
create an “affine reference.” - Perform a round of affine registration, followed by deformable
registration, over the oriented samples to the affine reference, and
average the resulting stacks to create a “deformable reference.” - At this point, it is advised to inspect the registered individuals
and remove from the oriented samples those that do not regis-
ter properly. - Repeatstep 6, registering to the deformable reference, and
average to create the “final reference.” We have not found
any significant improvements by repeating this step further,
but it is advisable to check for each case, as further iteration
rounds might improve the final reference.
3.1.3 Sample
Registration and Averaging
The next step requires the alignment of the individual samples to
the reference template. During the image registration process, the
intensity patterns of the target channel are compared with those of
the reference to calculate a geometrical transformation that can be
based on rigid and/or nonrigid registration methods (scaling,
translation, rotation, shearing and local deformations). The same
geometrical transformation is then applied to the second channel,
which overlays the expression signal onto the common reference.- For each sample for each gene, use the script “Affine_Defor-
mable_Registration.py” in Vergara et al. (2016) [5] to register
both channels. - Use “Average_registered.py” to average the signal channels
and create a canonical representation of the gene expression. - Detemine, for each gene, a binary threshold for each expression
pattern. This can be done manually (see[4]) or using a more
systematic image analysis procedure that adjust the threshold in
a gene and body region-dependent manner (see[5]).
114 Kaia Achim et al.