<p>UCYN-A enriched proteins were selected from a protein quantification dataset [26]. Multiple sequence alignment was performed on these proteins using COBALT [15] with default parameters, followed by alignment cleanup using CIAlign [16] with the remove-insertions option. Alignments exhibiting at least 60% coverage in the 880-1010 region were selected for further analysis.</p>
<p>The Gblocks [castresana] software was employed to crop the selected alignments, with a minimum of 104 sequences for both conserved and flank positions, a maximum of 15 contiguous nonconserved positions, a minimum block length of 10, and allowing all gap positions. MEME [18] was subsequently run on the cropped alignments using protein mode, with 10 motifs requested and a negative dataset, employing the differential enrichment objective function.</p>
<p>Motif analysis involved computing the positions of each motif's occurrences relative to motif #1, which was present in 184 out of 204 sequences. The distribution of these relative positions was visualized, along with the different combinations of motifs and their relative positions within these combinations.</p>
<p>The relationship between the mature domain and the motifs in the corresponding transit peptide was analyzed using embeddings computed with the prot_t5_xl_uniref50 [19] model, obtained from HuggingFace Model Hub. Logistic regression, decision tree, random forest, and support vector machine classifiers, as implemented by scikit-learn [20] using default parameters, were trained to predict motif combinations based on protein embeddings and amino acid contents of the mature domain. The performance of classifiers was tested using 5-fold cross-validation. The statistical significance of classifier accuracy was evaluated using permutation tests [21], as implemented by scikit-learn’s permutation_test_score, with a significance threshold of p <0.05.</p>
<p>The design of fluorescent uTP constructs was guided by a previously trained logistic regression classifier, which predicted the most probable motif combinations for each sequence. To more accurately mimic the native protein structure, the spatial arrangement of motifs was carefully considered. In native proteins, motifs are typically separated by short intervening sequences rather than occurring in direct succession. To replicate this natural spacing in the constructs, gaps between motifs were incorporated. These gaps were populated with amino acids randomly selected based on the frequency distribution observed in the corresponding native protein sequences.</p>
<p>For structural analysis, AlphaFold3 [22] was used to predict structures for each UCYN-A imported protein with default parameters. The resulting structures were analyzed using Biopython's PDB module [23]. A subset of each structure with strong C-terminal sequence similarity was selected for further analysis. These structures were aligned using PyMOL's [schroedinger] cealign command, and the alignments were used to create a consensus structure by averaging the position of the aligned residues from each structure.</p>
<p>To evaluate the predicted fluorescent uTP constructs, their structure was predicted using AlphaFold3 and aligned onto the consensus structure found earlier using PyMOL’s cealign command.</p>
<p>Protein localization was predicted for each UCYN-A enriched protein using MuLocDeep [25]. The attention weights as a function of sequence position were visualized, and the total C-terminal and N-terminal attention weights were compared across different localizations.</p>
<p>Using raw RNA-seq data from [27], a transcriptome assembly was constructed using DRAP [35] with the oases assembler and default parameters. The completeness of the assembly was tested with BUSCO [33]. Proteins were predicted from the transcriptome assembly with TransDecoder [34]. Ortholog analysis was performed using OrthoFinder [36].</p>
</div>
<!-- 3 -->
<divclass="h"id="three">
<divclass="h1">Plasmid Design and Construction</div>
<p>Plasmids were designed to express fluorescent proteins fused to various targeting sequences in Saccharomyces cerevisiae and Chlamydomonas reinhardtii. These sequences included uTP1/2 (UCYN-A transit peptide, identified through bioinformatics analysis), MTS1 (mitochondrial targeting sequence 1) [13], and CTS (chloroplast targeting sequence) [14]. The uTP1/2 sequence was inserted at the C-terminal end of the fluorescent proteins, while MTS1 and CTS were incorporated at the N-terminal end. For uTP1/2 constructs, a His6-tag was incorporated between the transit peptide and the fluorescent protein to facilitate future purification experiments.</p>
<p>Two existing vectors with fluorescent proteins were used as backbones for this study. For S. cerevisiae, we utilized pUDE1311 [11], which contains mNeonGreen and a URA3 marker for selection (full genotype: ConLS-ScHHF2p-ymNeonGreen-ScENO1t-ConR1-URA3-2µ-AmpR). This vector was kindly provided by Marcel Vieira Lara of TU Delft. For C. reinhardtii, we used pOpt2 [9, 10], containing mVenus and a ble marker for selection. This plasmid, kindly provided by Friedrich Kleiner from the MEA lab at TU Delft, also features specific introns in the mVenus coding sequence to prevent gene silencing. Both plasmids contain an ampicillin resistance gene for selection in E. coli. The complete sequences of both unmodified vectors and plasmids with all inserts have been made publicly available [link].</p>
<p>All plasmids, inserts, and primers were designed in SnapGene (www.snapgene.com). The inserts containing the transit peptides were synthesized by GenScript (Piscataway, NJ, USA). Gibson Assembly was employed for plasmid construction. Primers with specific overhangs were designed for PCR amplification of inserts and linearization of plasmid backbones at both the C-terminal and N-terminal regions, for inserting uTP1/2 and MTS1/CTS respectively.</p>
<p>PCR amplification was performed using KOD polymerase. To eliminate leftover circular plasmid from the backbone amplification process, DpnI digestion was utilized. Gibson Assembly was then carried out to combine the linearized plasmid backbones with the uTP, MTS1, and CTS inserts using the Gibson Assembly® Cloning Kit by New England Biolabs [1], following the manufacturer’s protocol [2]. pUC19 and a DNA fragment provided by the kit were used as positive controls.</p>
</div>
<!-- 4 -->
<divclass="h"id="four">
<divclass="h1"><i>Escherichia coli</i> Transformation and Verification</div>
<p>5-alpha competent E. coli cells from New England Biolabs Gibson Assembly® Cloning Kit were transformed with the assembled plasmids and the positive control using the manufacturer’s protocol for chemical transformation [3].</p>
<p>Transformed cells were plated on LB (lysogeny broth) agar containing ampicillin (50 µg/mL). Colony PCR targeting the inserts was conducted on the resulting E. coli colonies to verify correct insert incorporation.</p>
<p>Plasmids were isolated from selected and verified E. coli colonies using the PureYield™ Plasmid Miniprep System (Promega, Madison, WI, USA), following the manufacturer's protocol [miniprep_protocol] with slight modifications. Specifically, the elution step was performed using pre-warmed nuclease-free water (37°C) and the incubation time was extended to 10 minutes to increase yield. Plasmid purity and concentration were subsequently assessed using a NanoDrop™ Microvolume Spectrophotometer.</p>
<p>S. cerevisiae CEN.PK113-5D (full genome available at [12]), kindly provided by Marcel Vieira Lara of TU Delft, was cultured on YPD (yeast extract peptone dextrose) media. This particular strain carries a mutation in the URA3 gene, rendering it auxotrophic for uracil. This auxotrophy was used as a selective marker in subsequent transformation experiments.</p>
<p>S. cerevisiae was chemically transformed with plasmids encoding the fluorescent protein mNeonGreen tagged with either the mitochondrial transit peptide (MTS1) or UCYN-A transit peptide (uTP2), alongside a control without transit peptide (mNeonGreen only) and negative control (no plasmid) (see Protocols). Transformed cells were plated on uracil-deficient media to select successful transformants.</p>
<p>Colony PCR (see Protocols) was performed on the formed E. coli colonies, targeting the inserts to confirm the presence of the transit peptide sequences. Positive colonies were inoculated in liquid selective media for plasmid amplification and isolation.</p>
<p>Verification of isolated plasmids was performed via whole-plasmid sequencing done by Plasmidsaurus Inc. (Pasadena, CA, USA) using Oxford Nanopore Technologies long-read sequencing. The results were interpreted in SnapGene software (www.snapgene.com) by aligning the consensus sequence generated from sequencing with the expected plasmid.</p>
<p>S. cerevisiae colonies that passed the screening were inoculated in liquid selective medium and examined under a fluorescent spinning disk confocal microscope (Nikon Eclipse Ti-2, Crest X-light V3, 488-nm laser, FI emission filter) using a 100x oil-immersion objective to investigate sub-cellular localization of expressed fluorescent proteins.</p>
<p>The UVM4 strain of C. reinhardtii [31] was selected for transformation due to its higher transformation efficiency compared to more common strains. The strain was kindly provided by Dr. Ralph Bock from the Max Planck Institute of Molecular Plant Physiology. Zeocin was used for selection as this strain carries no resistance to it and our plasmids contained the ble gene conferring resistance to successful transformants.</p>
<p>Transformation of C. reinhardtii is more efficient with small linear DNA than with larger and circular DNA [9]. Therefore, before transformation, plasmid constructs were digested with KpnI and XbaI restriction enzymes. Then, we attempted to excise the bands with the expected length (... bp) for the TP-mVenus-ble expression cassette from the gel, followed by column purification. However, the DNA yield after purification was very low. Therefore, we then used an ethanol precipitation method (see protocols) to purify the DNA. As a result, the purified DNA solution contained the TP-mVenus-ble expression cassette as well as the vector backbone.</p>
<p>For the transformation of C.reinhardtii cells, a protocol adopted from [30] and [37] was used (see Protocols). C. reinhardtii was transformed using electroporation with an exponential decay pulse. The digested plasmids were used containing fluorescent protein constructs with an antibiotic marker (mVenus-ble) tagged with either the mitochondrial transit peptide, chloroplast transit peptide, or UCYN-A transit peptide (uTP2), alongside a control without transit peptide (mVenus-ble only) and a negative control (no plasmid) (see Protocols). Transformed cells were plated on TAP agar media containing zeocin to select successful transformants.</p>
<p>A protocol for fusing E. coli into S. cerevisiae using polyethylene glycol (PEG) was adapted from [6]. E. coli NCM3722 expressing fluorescent reporter PlsB-msGFP2 (plasmid carrying GFP available here: [plsb-mgfp2]), kindly provided by Jaïrus Beije from TU Delft, was used to track fusion success under microscopy.</p>
<p>The E. coli cells were grown to OD600 ~0.8, harvested, and resuspended in bacterial resuspension buffer. Yeast cells were converted to spheroplasts using zymolyase treatment. The E. coli suspension was mixed with sorbitol and immediately added to the spheroplast suspension. After incubation, the mixture was combined with PEG buffer. Following centrifugation, the pellet was resuspended in YPDS (yeast extract peptone dextrose sorbitol) medium and incubated.</p>
<p>Control samples were also prepared: S. cerevisiae treated with PEG but without E. coli and S. cerevisiae with E. coli but without PEG treatment.</p>
<p>All samples were imaged using fluorescent scanning confocal microscopy (Nikon confocal A1R, excitation 488-nm@16mW, emission filter 525/50) to evaluate fusion events. Cells were segmented using the transmitted light channel and instances of bright fluorescent spots within cells were evaluated. Image analysis was performed in ImageJ.</p>
