Introduction: FISH is a powerful technique to detect specific genomic sequences within an individual cell. Differences in the genetic makeup of AC and SqCC of the lung have been documented. A robust genetic based test applicable to sections from FFPE for the classification of AC vs SqCC tissue could have clinical utility as this classification can determine treatment selection. The optimal translation of array based genetic alterations data (in the form of copy number alterations) into a select set of FISH probes is nontrivial. One must consider the type (gain or loss), the extent and fidelity of the alteration (length in base pairs of the alteration common across the data set being considered which will affect FISH probe signal strength). Also one must consider effects the sectioning process will have on the actual FISH measurements; thick sections – more complete nuclei (better genetic loci fidelity) but more nuclear overlap (more difficult recognition of individual nuclei), thin sections – less nuclei overlap between cells but less complete nuclei. We use an in silico simulation to model the effect sectioning will have on the robustness and efficacy of a FISH based classifier for the genetic based differentiation of AC from SqCC tissue.

Methods: High resolution array comparative genomic hybridization (aCGH) tiling array data was used to identify recurrent copy number alterations that could be used to differentiate between 169 AC and 92 SqCC cases. An in silico simulation of the sectioning process on the detection of these alterations was programmed in MATLAB. This simulation modeled the physical distribution of the DNA sequence, chromosome by chromosome within individual modeled nuclei. For each of the 261 cases chromosomal loci gains were randomly inserted into the cell's DNA irrespective of chromosome and chromosomal loci deletions were assumed to occur in only one of that specific chromosome type. Gains and losses were assumed to linearly affect the size of chromosomes and cell nuclei. Cell nuclei sectioning was simulated 300 times for each case (10 different nuclear rotations for each of ten different chromosomal placement simulations within the nucleus for three different section thicknesses of 4um, 7um and 10 um). The number of spots that would be detected by FISH for each of these 300 simulations for each case for all 24,817 loci measured by the aCGH was saved for later classification accuracy analysis.

Results: The accuracy of the in silico model was checked against expected out comes; 1) such as the average loss of FISH spots as a function of the loss of nuclear volume to sectioning for loci with 1, 2 or 3 expected copies and 2) the concurrent loss of adjacent loci, or the loss of all loci within a chromosome with the loss of that chromosome. These measures were validated across at least 10K simulations. From the simulations we computed graphs of the number of loci retained as a function of the maximum radius of the nuclear material left within the section for each cell (an observable feature under microscopy). Further we calculated what percentage of nuclei within the section would have a maximum radius larger than 50, 60, 70, 80 or 90% of the pre-section nuclear radius for each case. This was done the 3 section thickness. Finally we used the simulation data to see the effect of sectioning on case classification, the number of cells which must be measured per case for optimal classification and the selection of the optimal genomic loci to differentiate the two types of tissue.

This content is only available via PDF.