Comparing mRNA levels using in situ hybridization of a target gene and co-stain

Abstract

In situ hybridization is an important technique for measuring the spatial expression patterns of mRNA in cells, tissues, and whole animals. However, mRNA levels cannot be compared across experiments using typical protocols. Here we present a semi-quantitative method to compare mRNA levels of a gene across multiple samples. This method yields an estimate of the error in the measurement to allow statistical comparison. Our method uses a typical in situ hybridization protocol to stain for a target gene and an internal standard, which we refer to as a co-stain. As a proof of concept, we apply this method to multiple lines of transgenic Drosophila embryos, harboring constructs that express reporter genes to different levels. We generated this test set by mutating enhancer sequences to contain different numbers of binding sites for Zelda, a transcriptional activator. We demonstrate that using a co-stain with in situ hybridization is an effective method to compare mRNA levels across samples. This method requires only minor modifications to existing in situ hybridization protocols and uses straightforward analysis techniques. This strategy can be broadly applied to detect quantitative, spatially resolved changes in mRNA levels.

Keywords: Drosophila melanogaster; In situ hybridization; Transcriptional activator; Vielfaltig; Zelda; mRNA levels.

Figure 1. A test set of transgenic reporters drives lacZ to different levels

(A) To compare mRNA levels between different transgenic reporter lines, we stain embryos from these lines for the gene of interest, lacZ, and a co-stain, hkb, both in red, and DNA in green. Using a image processing pipeline, we can segment the image into cells. We then normalize the lacZ signal using the hkb co-stain and analyze the gene expression pattern, shown here as a line trace along the anterior-posterior axis of the embryo. (B-C) To test our ability to detect changes in mRNA levels, our goal was to create a set of transgenic reporters that drive lacZ to different levels. We selected a set of enhancers that contain binding sites for Zelda, a transcriptional activator. We identified predicted Zelda sites in each enhancer and mutated strong (Δstrong), weak (Δweak) or all (Δall) Zelda sites. In the diagrams, light pink circles are weak Zelda sites, and red circles are strong Zelda sites. For each enhancer, we created a transgenic D. melanogaster line with the enhancer driving lacZ. We show cartoons of the expression patterns of the wild-type enhancers in blastoderm-stage embryos. In the cartoons, anterior is left, posterior is right, dorsal up and ventral down. (D) qPCR results from the zen lines verify that deleting Zelda binding sites alters mRNA level. Results are plotted as fold change relative to the Δall lines. The error bars represent the standard error of the mean estimated from two biological replicates.

Figure 2. The expression patterns of huckebein and our target genes do not overlap

(A) On the top, we show a maximum intensity projection of an embryo at the beginning of the blastoderm stage of development stained for hkb expression (red) and DNA (green). Using an image processing pipeline to computationally align many embryos, we created a gene expression atlas that includes hkb expression [8]. The pose of the embryo is the same as in Fig. 1B. hkb is expressed in both the anterior and posterior of the embryo. (B) We show the expression pattern of hkb for half the embryo over the 6 time points in the atlas, using a rectangular projection and a heat map for expression levels. (C) We show the mean (dotted line) and 95% quantile values (solid line) of hkb expression in the anterior 10% (purple) and posterior 10% (orange) of the embryo. We chose the posterior region as our normalization domain because of the relative stability of hkb levels in time points 2 and 3.

Figure 3. Quantile measurements capture representative levels of expression

Here we show the distribution of hkb and lacZ expression values in each cell for three sample embryos from the zen Δall line. Note that the scale of expression differs across these plots, supporting the need for normalization to calibrate measured expression values. When plotted for all cells (left), the peak in the distribution near zero represents cells expressing neither hkb nor lacZ, and the width and location correlates with the degree of experimental noise in our measurements. When plotted for the posterior (middle column) and middle domains (right column), the distributions are also asymmetric, supporting the use of quantiles to characterize the levels of hkb and lacZ expression consistently across transgenic lines (orange lines). Note that the quantiles select a similar part of the distribution in every embryo, while the mean would be influenced by the varying shapes of the distributions.

Figure 4. Huckebein and lacZ levels are correlated within a genotype

In the first column (A, C, E), we compare the expression levels of hkb (the 95% quantile of expression levels in the posterior of the embryo) to lacZ (the 99% quantile of the expression levels in the middle of the embryo). The points are colored according to reporter construct, and we fit a line to the results from each reporter (colored lines). In the second column (B, D, F), we plot the standardized residuals, which are the deviations of each data point from the fitted line. These plots indicate that the linear fit is appropriate because the residual scatter is roughly equally spread between values above and below zero without any particular pattern.

Figure 5. Spatial expression patterns can be normalized by two methods

The expression patterns of zen (A and B), sog (C and D) and gt enhancers (E and F), and their variants are shown as line traces through the embryo. For zen and sog, the trace is around the circumference of the embryo; dorsal is in the middle, ventral is on the edges. For gt, the trace is along the lateral side of the embryo; anterior is left, posterior is right. In the gt plots, the anterior and posterior peaks are from the hkb co-stain. The average expression level is shown as a solid line, standard errors of the mean are shown as shaded areas in the corresponding color. In A, C, and E, measurements were normalized to the hkb levels directly. In B, D, and F, measurements were normalized to both lacZ and hkb levels, and we subtracted the background (see Results).