Identifying fluorescently labeled single molecules in image stacks using machine learning

Automatic classification and segmentation of single-molecule fluorescence time traces with deep learning

Traces from single-molecule fluorescence microscopy (SMFM) experiments exhibit photophysical artifacts that typically necessitate human expert screening, which is time-consuming and introduces potential for user-dependent expectation bias. Here, we use deep learning to develop a rapid, automatic SMFM trace selector, termed AutoSiM, that improves the sensitivity and specificity of an assay for a DNA point mutation based on single-molecule recognition through equilibrium Poisson sampling (SiMREPS). The improved performance of AutoSiM is based on accepting both more true positives and fewer false positives than the conventional approach of hidden Markov modeling (HMM) followed by hard thresholding. As a second application, the selector is used for automated screening of single-molecule Förster resonance energy transfer (smFRET) data to identify high-quality traces for further analysis, and achieves ~90% concordance with manual selection while requiring less processing time. Finally, we show that AutoSiM can be adapted readily to novel datasets, requiring only modest Transfer Learning.

An automated workflow for quantifying RNA transcripts in individual cells in large data-sets

Advanced molecular probing techniques such as single molecule fluorescence in situ hybridization (smFISH) or RNAscope can be used to assess the quantity and spatial location of mRNA transcripts within cells. Quantifying mRNA expression in large image sets usually involves automated counting of fluorescent spots. Though conventional spot counting algorithms may suffice, they often lack high-throughput capacity and accuracy in cases of crowded signal or excessive noise. Automatic identification of cells and processing of many images is still a challenge. We have developed a method to perform automatic cell boundary identification while providing quantitative data about mRNA transcript levels across many images. Comparisons of mRNA transcript levels identified by the method highly correlate to qPCR measurements of mRNA expression in Drosophila genotypes with different levels of Rhodopsin 1 transcript. We also introduce a graphical user interface to facilitate analysis of large data sets. We expect these methods to translate to model systems where automated image processing can be harnessed to obtain single-cell data.

The described method:

•

Provides relative intensity measurements that scale directly with the number of labeled transcript probes within individual cells.

•

Allows quantitative assessment of single molecule data from images with crowded signal and moderate signal to noise ratios.

Mutagenesis of GATA motifs controlling the endoderm regulator elt-2 reveals distinct dominant and secondary cis-regulatory elements

Cis-regulatory elements (CREs) are crucial links in developmental gene regulatory networks, but in many cases, it can be difficult to discern whether similar CREs are functionally equivalent. We found that despite similar conservation and binding capability to upstream activators, different GATA cis-regulatory motifs within the promoter of the C. elegans endoderm regulator elt-2 play distinctive roles in activating and modulating gene expression throughout development. We fused wild-type and mutant versions of the elt-2 promoter to a gfp reporter and inserted these constructs as single copies into the C. elegans genome. We then counted early embryonic gfp transcripts using single-molecule RNA FISH (smFISH) and quantified gut GFP fluorescence. We determined that a single primary dominant GATA motif located 527bp upstream of the elt-2 start codon was necessary for both embryonic activation and later maintenance of transcription, while nearby secondary GATA motifs played largely subtle roles in modulating postembryonic levels of elt-2. Mutation of the primary activating site increased low-level spatiotemporally ectopic stochastic transcription, indicating that this site acts repressively in non-endoderm cells. Our results reveal that CREs with similar GATA factor binding affinities in close proximity can play very divergent context-dependent roles in regulating the expression of a developmentally critical gene in vivo.

Simultaneous detection of nuclear and cytoplasmic RNA variants utilizing Stellaris® RNA fluorescence in situ hybridization in adherent cells

RNA fluorescence in situ hybridization (FISH) has long been an indispensable tool for the detection and localization of RNA and is increasingly becoming an important complement to other gene expression analysis methods. We detail a streamlined RNA FISH protocol for the simultaneous imaging of multiple RNA gene products and RNA variants in fixed mammalian cells. The technique utilizes fluorescently pre-labeled, short DNA oligonucleotides (20 nucleotides in length), pooled into sets of up to 48 individual probes. The overall binding of multiple oligonucleotides to the same RNA target results in punctate fluorescent signals representing individual RNA molecules without the need for enzymatic signal amplification. Visualization of these punctate signals, through the use of wide-field fluorescence microscopy, enables the quantification of single RNA transcripts. Additionally, by utilizing probe sets with spectrally distinct fluorophores, multiplex analysis of specific RNAs, or RNA variants, can be achieved. We focus on the detection of a cytoplasmic mRNA and a nuclear long noncoding RNA to illustrate the benefits of this method for cell-specific detection and subcellular localization. Post-processing of images and spot counting is briefly discussed to demonstrate the capabilities of this method for the statistical analysis of RNA molecule number per cell, which is information that can be utilized to determine overall gene expression levels and cell-to-cell gene expression variation.

MED GATA factors promote robust development of the C. elegans endoderm

The MED-1,2 GATA factors contribute to specification of E, the progenitor of the Caenorhabditis elegans endoderm, through the genes end-1 and end-3, and in parallel with the maternal factors SKN-1, POP-1 and PAL-1. END-1,3 activate elt-2 and elt-7 to initiate a program of intestinal development, which is maintained by positive autoregulation. Here, we advance the understanding of MED-1,2 in E specification. We find that expression of end-1 and end-3 is greatly reduced in med-1,2(-) embryos. We generated strains in which MED sites have been mutated in end-1 and end-3. Without MED input, gut specification relies primarily on POP-1 and PAL-1. 25% of embryos fail to make intestine, while those that do display abnormal numbers of gut cells due to a delayed and stochastic acquisition of intestine fate. Surviving adults exhibit phenotypes consistent with a primary defect in the intestine. Our results establish that MED-1,2 provide robustness to endoderm specification through end-1 and end-3, and reveal that gut differentiation may be more directly linked to specification than previously appreciated. The results argue against an "all-or-none" description of cell specification, and suggest that activation of tissue-specific master regulators, even when expression of these is maintained by positive autoregulation, does not guarantee proper function of differentiated cells.

Aro: a machine learning approach to identifying single molecules and estimating classification error in fluorescence microscopy images

Background

Recent techniques for tagging and visualizing single molecules in fixed or living organisms and cell lines have been revolutionizing our understanding of the spatial and temporal dynamics of fundamental biological processes. However, fluorescence microscopy images are often noisy, and it can be difficult to distinguish a fluorescently labeled single molecule from background speckle.

Results

We present a computational pipeline to distinguish the true signal of fluorescently labeled molecules from background fluorescence and noise. We test our technique using the challenging case of wide-field, epifluorescence microscope image stacks from single molecule fluorescence in situ experiments on nematode embryos where there can be substantial out-of-focus light and structured noise. The software recognizes and classifies individual mRNA spots by measuring several features of local intensity maxima and classifying them with a supervised random forest classifier. A key innovation of this software is that, by estimating the probability that each local maximum is a true spot in a statistically principled way, it makes it possible to estimate the error introduced by image classification. This can be used to assess the quality of the data and to estimate a confidence interval for the molecule count estimate, all of which are important for quantitative interpretations of the results of single-molecule experiments.

Conclusions

The software classifies spots in these images well, with >95% AUROC on realistic artificial data and outperforms other commonly used techniques on challenging real data. Its interval estimates provide a unique measure of the quality of an image and confidence in the classification.

Electronic supplementary material

The online version of this article (doi:10.1186/s12859-015-0534-z) contains supplementary material, which is available to authorized users.

MicroRNA-based single-gene circuits buffer protein synthesis rates against perturbations

Achieving precise control of mammalian transgene expression has remained a long-standing, and increasingly urgent, challenge in biomedical science. Despite much work, single-cell methods have consistently revealed that mammalian gene expression levels remain susceptible to fluctuations (noise) and external perturbations. Here, we show that precise control of protein synthesis can be realized using a single-gene microRNA (miRNA)-based feed-forward loop (sgFFL). This minimal autoregulatory gene circuit consists of an intronic miRNA that targets its own transcript. In response to a step-like increase in transcription rate, the network generated a transient protein expression pulse before returning to a lower steady state level, thus exhibiting adaptation. Critically, the steady state protein levels were independent of the size of the stimulus, demonstrating that this simple network architecture effectively buffered protein production against changes in transcription. The single-gene network architecture was also effective in buffering against transcriptional noise, leading to reduced cell-to-cell variability in protein synthesis. Noise was up to 5-fold lower for a sgFFL than for an unregulated control gene with equal mean protein levels. The noise buffering capability varied predictably with the strength of the miRNA-target interaction. Together, these results suggest that the sgFFL single-gene motif provides a general and broadly applicable platform for robust gene expression in synthetic and natural gene circuits.