Determining the subcellular location of new proteins from microscope images using local features.

MOTIVATION: Evaluation of previous systems for automated determination of subcellular location from microscope images has been done using datasets in which each location class consisted of multiple images of the same representative protein. Here, we frame a more challenging and useful problem where previously unseen proteins are to be classified. RESULTS: Using CD-tagging, we generated two new image datasets for evaluation of this problem, which contain several different proteins for each location class. Evaluation of previous methods on these new datasets showed that it is much harder to train a classifier that generalizes across different proteins than one that simply recognizes a protein it was trained on. We therefore developed and evaluated additional approaches, incorporating novel modifications of local features techniques. These extended the notion of local features to exploit both the protein image and any reference markers that were imaged in parallel. With these, we obtained a large accuracy improvement in our new datasets over existing methods. Additionally, these features help achieve classification improvements for other previously studied datasets. AVAILABILITY: The datasets are available for download at http://murphylab.web.cmu.edu/data/. The software was written in Python and C++ and is available under an open-source license at http://murphylab.web.cmu.edu/software/. The code is split into a library, which can be easily reused for other data and a small driver script for reproducing all results presented here. A step-by-step tutorial on applying the methods to new datasets is also available at that address. CONTACT: murphy@cmu.edu SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

A variable light domain fluorogen activating protein homodimerizes to activate dimethylindole red.

Novel fluorescent tools such as green fluorescent protein analogues and fluorogen activating proteins (FAPs) are useful in biological imaging for tracking protein dynamics in real time with a low fluorescence background. FAPs are single-chain variable fragments (scFvs) selected from a yeast surface display library that produce fluorescence upon binding a specific dye or fluorogen that is normally not fluorescent when present in solution. FAPs generally consist of human immunoglobulin variable heavy (V(H)) and variable light (V(L)) domains covalently attached via a glycine- and serine-rich linker. Previously, we determined that the yeast surface clone, V(H)-V(L) M8, could bind and activate the fluorogen dimethylindole red (DIR) but that the fluorogen activation properties were localized to the M8V(L) domain. We report here that both nuclear magnetic resonance and X-ray diffraction methods indicate the M8V(L) forms noncovalent, antiparallel homodimers that are the fluorogen activating species. The M8V(L) homodimers activate DIR by restriction of internal rotation of the bound dye. These structural results, together with directed evolution experiments with both V(H)-V(L) M8 and M8V(L), led us to rationally design tandem, covalent homodimers of M8V(L) domains joined by a flexible linker that have a high affinity for DIR and good quantum yields.

Blue fluorescent dye-protein complexes based on fluorogenic cyanine dyes and single chain antibody fragments.

Fluoromodules are complexes formed upon the noncovalent binding of a fluorogenic dye to its cognate biomolecular partner, which significantly enhances the fluorescence quantum yield of the dye. Previously, several single-chain, variable fragment (scFv) antibodies were selected from a yeast cell surface-displayed library that activated fluorescence from a family of unsymmetrical cyanine dyes covering much of the visible and near-IR spectrum. The current work expands our repertoire of genetically encodable scFv-dye pairs by selecting and characterizing a group of scFvs that activate fluorogenic violet-absorbing, blue-fluorescing cyanine dyes, based on oxazole and thiazole heterocycles. The dye binds to both yeast cell surface-displayed and soluble scFvs with low nanomolar K(d) values. These dye-protein fluoromodules exhibit high quantum yields, approaching unity for the brightest system. The promiscuity of these scFvs with other fluorogenic cyanine dyes was also examined. Fluorescence microscopy demonstrates that the yeast cell surface-displayed scFvs can be used for multicolor imaging. The prevalence of 405 nm lasers on confocal imaging and flow cytometry systems make these new reagents potentially valuable for cell biological studies.

Enhanced photostability of genetically encodable fluoromodules based on fluorogenic cyanine dyes and a promiscuous protein partner.

Fluoromodules are discrete complexes of biomolecules and fluorogenic dyes. Binding of the dyes to their cognate biomolecule partners results in enhanced dye fluorescence. We exploited a previously reported promiscuous binding interaction between a single-chain, variable fragment antibody protein and a family of cyanine dyes to create new protein-dye fluoromodules that exhibit enhanced photostability while retaining high affinity protein-dye binding. Modifications to the dye structure included electron-withdrawing groups that provide resistance to photo-oxidative damage. Low nanomolar equilibrium dissociation constants were found for the new dyes. Fluorescence microscopy illustrates how yeast can be surface-labeled with three different colors based on a single protein and appropriately chosen dyes.

STED nanoscopy in living cells using Fluorogen Activating Proteins.

We demonstrate the effectiveness of a genetically encoded Malachite Green (MG) binding fluorogen activating protein (FAP) for live cell stimulated emission depletion nanoscopy (STED). Both extracellular and intracellular FAPs were tested in living cells using fluorogens with either membrane expressed FAP or as an intracellular FAP-actin fusion. Structures with FWHM of 110-122nm were observed. Depletion data however suggests a resolution of 70nm with the given instrument.

A rainbow of fluoromodules: a promiscuous scFv protein binds to and activates a diverse set of fluorogenic cyanine dyes.

Combined magnetic and fluorescence cell sorting were used to select Fluorogen Activating Proteins (FAPs) from a yeast surface-displayed library for binding to the fluorogenic cyanine dye Dimethyl Indole Red (DIR). Several FAPs were selected that bind to the dye with low nanomolar Kd values and enhance fluorescence more than 100-fold. One of these FAPs also exhibits considerable promiscuity, binding with high affinity to several other fluorogenic cyanine dyes with emission wavelengths covering most of the visible and near-IR regions of the spectrum. This significantly expands the number and wavelength range of scFv-based fluoromodules.

A cell surface display fluorescent biosensor for measuring MMP14 activity in real-time.

Despite numerous recent advances in imaging technologies, one continuing challenge for cell biologists and microscopists is the visualization and measurement of endogenous proteins as they function within living cells. Achieving this goal will provide a tool that investigators can use to associate cellular outcomes with the behavior and activity of many well-studied target proteins. Here, we describe the development of a plasmid-based fluorescent biosensor engineered to measure the location and activity of matrix metalloprotease-14 (MMP14). The biosensor design uses fluorogen-activating protein technology coupled with a MMP14-selective protease sequence to generate a binary, "switch-on" fluorescence reporter capable of measuring MMP14 location, activity, and temporal dynamics. The MMP14-fluorogen activating protein biosensor approach is applicable to both short and long-term imaging modalities and contains an adaptable module that can be used to study many membrane-bound proteases. This MMP14 biosensor promises to serve as a tool for the advancement of a broad range of investigations targeting MMP14 activity during cell migration in health and disease.

Directed evolution of a fluorogen-activating single chain antibody for function and enhanced brightness in the cytoplasm.

Directed evolution is an exceptionally powerful tool that uses random mutant library generation and screening techniques to engineer or optimize functions of proteins. One class of proteins for which this process is particularly effective is antibodies, where properties such as antigen specificity and affinity can be selected to yield molecules with improved efficacy as molecular labels or in potential therapeutics. Typical antibody structure includes disulfide bonds that are required for stability and proper folding of the domains. However, these bonds are unable to form in the reducing environment of the cytoplasm, stymieing the effectiveness of optimized antibodies in many research applications. We have removed disulfide-forming cysteine residues in a single chain antibody fluorogen-activating protein (FAP), HL4, and employed directed evolution to select a derivative that is capable of activity in the cytoplasm. A subsequent round of directed evolution was targeted at increasing the overall brightness of the fluoromodule (FAP-fluorogen complex). Ultimately, this approach produced a novel FAP that exhibits strong activation of its cognate fluorogen in the reducing environment of the cytoplasm, significantly expanding the range of applications for which fluoromodule technology can be utilized.

scFv-based fluorogen activating proteins and variable domain inhibitors as fluorescent biosensor platforms.

Single chain antibodies (scFvs) are engineered proteins composed of IgG variable heavy (V(H)) and variable light (V(L)) domains tethered together by a flexible peptide linker. We have characterized the individual V(H) or V(L) domain activities of several scFvs isolated from a yeast surface-display library for their ability to bind environmentally sensitive fluorogenic dyes causing them to fluoresce. For many of the scFvs, both V(H) and V(L) domains are required for dye binding and fluorescence. The analysis of other scFvs, however, revealed that either the V(H) or the V(L) domain alone is sufficient to cause the fluorogenic dye activation. Furthermore, the inactive complementary domains in the original scFvs either contribute nothing to, or actually inhibit the activity of these active single domains. We have explored the interactions between active variable domains and inactive complementary domains by extensive variable domain swapping through in vitro gene manipulations to create hybrid scFvs. In this study, we demonstrate that significant alteration of the fluorogenic dye activation by the active V(H) or V(L) domains can occur by partnering with different V(H) or V(L) complementary domains in the scFv format. Hybrid scFvs can be generated that have fluorogen-activating domains that are completely inhibited by interactions with other domains. Such hybrid scFvs are excellent platforms for the development of several types of genetically encoded, fluorescence-generating biosensors.

Cell cycle dependence of protein subcellular location inferred from static, asynchronous images.

Protein subcellular location is one of the most important determinants of protein function during cellular processes. Changes in protein behavior during the cell cycle are expected to be involved in cellular reprogramming during disease and development, and there is therefore a critical need to understand cell-cycle dependent variation in protein localization which may be related to aberrant pathway activity. With this goal, it would be useful to have an automated method that can be applied on a proteomic scale to identify candidate proteins showing cell-cycle dependent variation of location. Fluorescence microscopy, and especially automated, high-throughput microscopy, can provide images for tens of thousands of fluorescently-tagged proteins for this purpose. Previous work on analysis of cell cycle variation has traditionally relied on obtaining time-series images over an entire cell cycle; these methods are not applicable to the single time point images that are much easier to obtain on a large scale. Hence a method that can infer cell cycle-dependence of proteins from asynchronous, static cell images would be preferable. In this work, we demonstrate such a method that can associate protein pattern variation in static images with cell cycle progression. We additionally show that a one-dimensional parameterization of cell cycle progression and protein feature pattern is sufficient to infer association between localization and cell cycle.

Fluorogen-activating single-chain antibodies for imaging cell surface proteins.

Imaging of live cells has been revolutionized by genetically encoded fluorescent probes, most famously green and other fluorescent proteins, but also peptide tags that bind exogenous fluorophores. We report here the development of protein reporters that generate fluorescence from otherwise dark molecules (fluorogens). Eight unique fluorogen activating proteins (FAPs) have been isolated by screening a library of human single-chain antibodies (scFvs) using derivatives of thiazole orange and malachite green. When displayed on yeast or mammalian cell surfaces, these FAPs bind fluorogens with nanomolar affinity, increasing green or red fluorescence thousands-fold to brightness levels typical of fluorescent proteins. Spectral variation can be generated by combining different FAPs and fluorogen derivatives. Visualization of FAPs on the cell surface or within the secretory apparatus of mammalian cells can be achieved by choosing membrane permeant or impermeant fluorogens. The FAP technique is extensible to a wide variety of nonfluorescent dyes.

The spectrum of Trp- mutants isolated as 5-fluoroanthranilate-resistant clones in Saccharomyces bayanus, S. mikatae and S. paradoxus.

5-Fluoroanthranilic acid (FAA)-resistant mutants were selected in homothallic diploids of three Saccharomyces species, taking care to isolate mutants of independent origin. Mutations were assigned to complementation groups by interspecific complementation with S. cerevisiae tester strains. In all three species, trp3, trp4 and trp5 mutants were recovered. trp1 mutants were also recovered if the selection was imposed on a haploid strain. Thus, FAA selection may be more generally applicable than was previously described.

Large-scale automated analysis of location patterns in randomly tagged 3T3 cells.

Location proteomics is concerned with the systematic analysis of the subcellular location of proteins. In order to perform high-resolution, high-throughput analysis of all protein location patterns, automated methods are needed. Here we describe the use of such methods on a large collection of images obtained by automated microscopy to perform high-throughput analysis of endogenous proteins randomly-tagged with a fluorescent protein in NIH 3T3 cells. Cluster analysis was performed to identify the statistically significant location patterns in these images. This allowed us to assign a location pattern to each tagged protein without specifying what patterns are possible. To choose the best feature set for this clustering, we have used a novel method that determines which features do not artificially discriminate between control wells on different plates and uses Stepwise Discriminant Analysis (SDA) to determine which features do discriminate as much as possible among the randomly-tagged wells. Combining this feature set with consensus clustering methods resulted in 35 clusters among the first 188 clones we obtained. This approach represents a powerful automated solution to the problem of identifying subcellular locations on a proteome-wide basis for many different cell types.

Complete nucleotide sequence of a P2 family lysogenic bacteriophage, varphiMhaA1-PHL101, from Mannheimia haemolytica serotype A1.

The 34,525 nucleotide sequence of a double-stranded DNA bacteriophage (phiMhaA1-PHL101) from Mannheimia haemolytica serotype A1 has been determined. The phage encodes 50 open reading frames. Twenty-three of the proteins are similar to proteins of the P2 family of phages. Other protein sequences are most similar to possible prophage sequences from the draft genome of Histophilus somni 2336. Fourteen open reading frames encode proteins with no known homolog. The P2 orthologues are collinear in phiMhaA1-PHL101, with the exception of the phage tail protein gene T, which maps in a unique location between the S and V genes. The phage ORFs can be arranged into 17 possible transcriptional units and many of the genes are predicted to be translationally coupled. Southern blot analysis revealed phiMhaA1-PHL101 sequences in other A1 isolates as well as in serotype A5, A6, A9, and A12 strains of M. haemolytica, but not in the related organisms, Mannheimia glucosida or Pasteurella trehalosi.