use of up to 48 individually labeled 20 nucleotide-long oligonu-
cleotides that hybridize to their target RNA [7]. The accumulation
of a large number of probes on a single transcript produces a
diffraction limited signal that can easily be distinguished from
background [7]. However, smFISH gives the best results when
used on thin samples like single cells or tissue sections, and no
protocols were available for zebrafish yet. Furthermore, available
analysis pipelines for quantification of transcripts at cellular resolu-
tion in multicellular organisms relied on manual cell segmentation
which is very labor-intensive and time-consuming [8, 12–14].
We previously developed a protocol for smFISH on zebrafish
cryosections as well as an automated analysis pipeline for transcript
detection and cell segmentation (Fig.1)[15]. As input samples for
our smFISH protocol, we use cryosections of zebrafish embryos
embedded in OCT. Although sectioning at early zebrafish stages is
difficult because cells are large and embryos fragile, our protocol
generates high quality sections at a broad range of developmental
stages. We then adapted a protocol for Stellaris smFISH on tissue
sections [8] for use on these zebrafish sections (Fig.1a). With the
resulting protocol, even very low transcript levels can be detected at
high specificity and sensitivity. To increase image analysis speed and
throughput, we developed a freely available analysis pipeline for
automated transcript detection and semiautomated cell segmenta-
tion. This pipeline can be applied to sections of zebrafish and other
multicellular organisms (Fig.1b)[15]. The transcript analysis pipe-
line enables quantification of transcript levels, as well as quantifica-
tion of the number of active transcription sites (transcription foci),
and the number of transcripts per focus (Fig.1b 6 ). Nuclear seg-
mentation is integrated in the transcript analysis pipeline to be able
to assign transcripts to nuclei or cytoplasm. The membrane seg-
mentation pipeline we developed consists of three parts. First, a
random forest pipeline in KNIME is used to predict for each pixel
whether it is part of the membrane, a membrane intersection point
(vertex), or background (Fig.1b 2 ). Then, the PathFinder plugin in
Fiji uses these predictions to generate a cell mask (Fig.1b 3 ). Finally,
the Fiji Cell annotation tool can be used to correct small errors in
the segmentation, and to group cells according to cell type
(Fig.1b 4 ). The resulting cell mask can be used in combination
with the transcript analysis pipeline in Fiji to assign transcripts to
individual cells and nuclei (Fig.1b 6 ).
Here, we describe (1) a method for high-quality cryosectioning
of zebrafish embryos, (2) an optimized smFISH protocol for use on
zebrafish cryosections, and (3) a pipeline for (semi)automated tran-
script detection and cell segmentation that can be applied to
smFISH samples of any multicellular organism.144 L. Carine Stapel et al.