Crossplat form technology reproducibility

The process of comparing gene expression information from a broad range of conditions has been widely adopted as an approach to functional discovery. As each individual laboratory may use different expression assaying plat forms to profile a common subset of genes, one obvious question arises as to how one may combine and compare expression measurements across diverse technologies. Assuming that such operations makes sense, the problem comes down to determining the reproducibility of cross-plat form measurements. Analyses by Kuo [113] suggest that for spotted cDNA probe microarrays and microarrays with Affymetrix-synthesized oligonucleotides circa 1999, the reproducibility may be poor. The mRNA expression measurements from oligonucleotide microarrays from Butte et al. [39] were paired with measurements from the cDNA microarrays from Ross et al. [154] by matching the corresponding cell lines and RNA transcript probes. When the oligonucleotide data (from the Avg Diff column) were compared to the normalized Cy5/Cy3 ratio (from the RAT2N column in the ScanAlyze output file) of the cDNA data, the overall linear correlation across all 162, 120 paired measurements was poor (r = .0326) as illustrated in figure 3.11. Corrections for GC content, cross-hybridization, and probe length all did not materially improve the correlation coefficient.

Figure 3.11: Overall correlation coefficients for each matching probe and probe set across two microarray technologies. Great variance is seen in reproducibility of measurements across probes and probe sets, with some probes showing correlation coefficients near 1.0, and some even showing a negative correlation (i.e., a gene which is reported as highly expressed in one technology has an opposite expression report in the other technology).

Figure 3.11: Overall correlation coefficients for each matching probe and probe set across two microarray technologies. Great variance is seen in reproducibility of measurements across probes and probe sets, with some probes showing correlation coefficients near 1.0, and some even showing a negative correlation (i.e., a gene which is reported as highly expressed in one technology has an opposite expression report in the other technology).

It does not seem likely that the reported measurements from these data sets can be normalized or transformed into a single standardized unit of gene expression. Moreover, even the qualitative correlation was poor. For instance, high measurements in the oligochip did not correspond well with high measurements in the cDNA chip, and vice versa. At the very least, the results of this study suggest that greater (pre)caution should be taken in making cross-plat form comparisons. While complementing existing assay technologies, new techniques for expression profiling such as optically coded beads and ink-jet arrays (see section 7.1) will add to the challenge of comparing and combining expression data across different plat forms, and are likely to require similar efforts for normalization and standardization.

Figure 3.12: Pooling total RNA extracts for replicate experiments. Two strategies for making replicated measurements from samples, pooled and not pooled. N may not necessarily equal n. 3.2.4 Pooling sample probes and PCR for replicate experiments

A typical microarray hybridization experiment today requires on the order of 10 1Ag of total RNA to generate the labeled sample probe compound for each array (spotted cDNA array, see [167]; oligochip, see [122]). Depending upon the experimental design, namely, how one defines a replicate experiment, it might not always be possible to obtain enough total RNA from a system, e.g., specific tissue from a test animal, to prepare sample probes or targets for one or more (as repeat measurements) chip experiments. For example, suppose that a biologist wishes to obtain the gene expression profile of the mouse heart during embryonic day 12 (E12) and finds that the total RNA extracted from one embryonic mouse heart is not sufficient for generating enough labeled target for even a single microarray experiment. In this situation, an alternative solution is to pool together the total RNA extract from several, say n many, different E12 mouse hearts to create a sample probe or target compound which is then split into , N < n, many targets for hybridizing on N separate chips, and hence resulting in N replicate experimental data.

When these tissue samples are drawn from n embryos that do not share a common parent, the pooling procedure will average out infra-species biological variations between the mice. This is independent of and in contrast to the subsequent splitting of one pooled sample into individual targets for N separate microarray assays and takes into consideration measurement error that inevitably arises in any empirical experiment.

Regarding the use of PCR to amplify these and any RNA extracts, the reader should note that beyond a small number of amplification cycles, a heterogeneous RNA mixture that has been amplified via PCR may not retain the same specific RNA percentagewise composition as in the original mixture [17]. Furthermore, one should also note that experimental data generated by re-using a target that has undergone a previous hybridization should not generally be considered a replicate assay because the biochemical composition of the target may have changed irreversibly after a hybridization procedure, due to both the protocol and time differences. Typically, a target needs to be incubated on a microarray for 12 hours or more during hybridization.

Was this article helpful?

0 0

Post a comment