We designed a novel bimodal readout system that can be used to measure reporter protein fluorescence in vivo. We opted to use a microplate reader to measure fluorescence to allow high throughput measurement. For all fluorescence measurements, we used living protoplasts in transfection regeneration medium (Hohe et al., 2004). Initial tests showed that a minimum number of 2,000 protoplasts are necessary to get a signal above background fluorescence (Supplementary Figure 1). The constructs used for transient transfection contained the promoter:GFP fusion and additionally a normalization cassette consisting of the 2x 35S promoter, the mCherry gene sequence and a 35S terminator (Figure 1). We selected the 2x 35S promoter because it is widely used in the P. patens community and shows a medium gene expression in P. patens, which makes it a good candidate for comparisons with new promoters. The fluorescence from this second reporter was used to normalize the GFP fluorescence values, which show high between-experiment variation due to different transformation efficiencies (Supplementary Figure 2 and Supplementary Table 1). By integrating 2xp35S:mCherry into the same plasmid as the GFP fusion of interest, we also account for uptake of multiple plasmids during transformation. To test the feasibility of our plate reader measurement system, we performed protoplastation of protonemal tissue from stable mutant lines that express GFP or mCherry (Perroud et al., 2011), mixed the protoplasts and found that we can measure both signals in parallel from the same well. Additionally, we prepared a dilution series and found a linear relationship between number of protoplasts and fluorescence intensity for both reporters (Supplementary Figure 3).

We calculated the codon frequencies for all v1.6 genes of P. patens and grouped them by expression strength based on a broad range of transcriptome microarray experiments (Hiss et al., 2014). The top 233 highly expressed genes (0.9%) are enriched in the Gene Ontology (GO) terms translation, gene expression, and protein synthesis (Supplementary Figure 4). By comparing the codon usage of these highly expressed genes with the codon usage of all genes, we find a significant codon usage bias (Fisher’s Exact Test, FDR adjusted p < 0.05) for nine amino acids, namely Glu, Phe, Tyr, Cys, His, Gln, Ile, Asn, and Lys (Figure 2A and Supplementary File 1). These codons are preferred not only in P. patens highly expressed genes but also in other organisms [Figure 2B; Arabidospsis thaliana (Wright et al., 2004; Morton and Wright, 2007), Saccharomyces cerevisiae (dos Reis and Wernisch, 2009), Schizosaccharomyces pombe (Hiraoka et al., 2009), and Homo sapiens (van Hemert and Berkhout, 1995)]. Using microarray data (Hiss et al., 2014), we can confidently detect the first biased codon usage (Lys) already within the top 1,967 highly expressed genes (7.3% of all genes measured by the array), and the codon usage bias for all nine amino acids can be detected based on the top 291 highly expressed genes (1.1%, Supplementary File 2). Thus, by using the biased codon usage as an indicator, we detect around 7% of the P. patens genes as highly expressed. These genes also show high expression in microarray-based transcriptome studies [Supplementary Figure 5, (Hiss et al., 2014; Ortiz-Ramirez et al., 2015)]. For the highly expressed genes, we detect a GC bias toward higher GC content and a lower effective number of codons (ENC) as compared to the overall gene set (Supplementary Figure 6).

To determine whether the codon bias we have seen in highly expressed genes in P. patens can be used to enhance protein expression, we prepared GFP variants with adapted codon usage. Starting from the smGFP (Davis and Vierstra, 1998), we changed all triplets either to the one preferred in highly expressed genes, or to the one not preferred. This led to 39 changes in the coding sequence for “GFPhigh” and 58 changes for “GFPlow,” respectively (Supplementary Figure 7 and Supplementary Table 2). The GC content of the smGFP was 43.7% as compared to 49.1% (GFPhigh) and 35.6% (GFPlow). In both adapted GFP versions, the resulting amino acid sequence was not changed.

All three GFP versions were combined with different promoters 2xp35S, pactin1, and plhcsr1 (1,956 bp upstream region of the Physcomitrella patens lhcsr1 gene) and the mCherry normalization cassette. These constructs were transfected transiently into P. patens protoplasts and the fluorescence signal of GFP and mCherry measured in vivo with a microplate reader system.

The measurements were normalized and the 2x 35S promoter in combination with the smGFP signal set to one. The rice actin1 promoter:GFPhigh fusion is at 1.0 ± 0.6-fold in our system and therefore not significantly different from the 2x 35S promoter with the same GFP version (p = 0.15, Student’s t-test). We see a shift in signal intensity for the 2x 35S promoter constructs with different GFP versions, with the GFPhigh giving the strongest signal at 1.8 ± 0.5-fold and the GFPlow the weakest at 0.4 ± 0.3-fold. Different signal intensities can also be seen with the different plhcsr1:GFP combinations starting with GFPlow at 0.9 ± 0.8-fold to smGFP at 1.7 ± 0.3-fold up to GFPhigh at 5.7 ± 0.4-fold (Figure 3).

We selected a strong and constitutively expressed gene from our transcriptome microarray data which cover several developmental stages and perturbations (Hiss et al., 2014). To find a suitable endogenous promoter we calculated the coefficient of variation (c. v.) of the expression values of all genes represented on the Combimatrix microarray (Wolf et al., 2010) and selected genes with high expression values and low coefficient of variation, leading to candidate genes with a high expression across many developmental stages and perturbations. Additionally we filtered for genes whose 3 kbp upstream region does not overlap with another gene or transposable element (TE). The 3 kbp gene-free upstream region was chosen in order to prevent the regulatory sequences of the gene of interest to fall into the sequence of another gene, and in order not to select atypically short genes (Zimmer et al., 2013). The 3kbp upstream region further should not contain TEs because they are known to be silenced via methylation of the corresponding genomic region (Zemach et al., 2010; Widiez et al., 2014). We chose the cab protein LHCSR1 (Phypa_169593, Pp1s213_80V6.1, Pp3c9_3440V3.1) since the promoters of cab genes of plants have successfully been used as strong endogenous and exogenous promoters before (Abe et al., 2008). We used about 2,000 bp upstream of the coding sequence (Chr09:1,975,316…1,977,272) as putative promoter for cloning and subsequent transfection (expression profile of the lhcsr1 gene, see Supplementary Figure 8). The lhcsr1 gene does not show the bias in codon usage we generally observe in the top highly expressed genes (Supplementary Table 3). The constructs containing the full plhcsr1 promoter show higher signals than the corresponding 2x p35S constructs (Figure 3). We prepared eight shortened promoter sequences by restriction digestion of the original 1,956 bp sequence (Figure 4, constructs A–H). All promoter versions were fused to GFPhigh:35S-terminator and the mCherry normalization cassette, transfected transiently into moss protoplasts and the GFP and mCherry signal measured in vivo with the microplate reader system.

Based on restriction enzyme sites, we divided the 1,956 bp promoter sequence into four regions (Figure 4). Region I contains the 202 bp 5′-UTR and 84 bp of the upstream region. Region II consists of a 391-bp fragment, Region III of a 520-bp fragment, and Region IV of the 759 bp at the 5′ end of the chosen promoter sequence. The removal of the regions IV and III without modifying the rest of the promoter does not strongly affect promoter activity, and the remaining fragments display an activity between 121 and 129% compared to the full length (1,956 bp) promoter (Figure 4, constructs A, B, and C). None of the other partial deletions abolish totally the promoter activity, most of the regions I and II deletions display 12–20% total plhcsr1 value (Figure 4, constructs D–H). The only exception is the construct D that does not contain the activating region II, but still retains 35% promoter activity (Figure 4, construct D). The region IV appears to contain a domain able to activate transcription once it is fused to the 5′-UTR. This is especially interesting in contrast to construct E which has longer fragments for regions I and IV but a lower signal intensity.

We searched the potential promoter sequence for motifs using Signal Scan and the plant cis-acting regulatory DNA elements database [PLACE, (Higo et al., 1999)] as well as the Plant cis-acting regulatory element (PlantCARE) database. PLACE finds several known cis-acting elements that are associated with light-regulated genes, e.g., GT1-sites (Green et al., 1988) or I-boxes (Giuliano et al., 1988), whereas many TATA boxes are found in region IV, further away from the ATG but not in the 5′-UTR or the regions II and III (Figure 5A). PlantCARE predicts a TATA box core promoter element for both lhcsr genes within the 5′-UTR at around -130 bp of the translation start (Figure 5B). The transcription start site (TSS) data implemented in the Plant Promoter database (Hieno et al., 2014) support this TATA box based on 5′ end sequencing of the lhcsr1 gene with a TSS found 102 bases upstream of the start codon (Supplementary Figure 9). This TATA box is further supported by a possible TSS present in 5′ Cap-capture sequencing data available at the P. patens CoGe representation (Figure 6).

Physcomitrella patens Gransden (Rensing et al., 2008) was cultivated on solidified [1% (w/v) agar] mineral medium, also known as modified (Reski and Abel, 1985) Knop’s (1868) medium, on 9-cm petri dishes enclosed by laboratory film at 22°C with a 16-h-light/8-h-dark regime under 70 μmol m^-2 s^-1 white light (long-day conditions).

To assess codon usage bias, the coding sequences of v1.2 gene models on the Combimatrix microarray experiments from Hiss et al. (2014) that show expression values above the detection limit were analyzed (26,856 genes). Genes were sorted according to their expression level and genes with a normalized fluorescence intensity above 200,000 (567 genes, 2.1%) were termed strongly expressed genes whereas a value of 450,000 (233 genes, 0.87%) put them into the group of highly expressed genes. The codon frequencies for each group were calculated with R and afterward a Fisher’s Exact Test was used for each of the 64 codons to find significant changes in codon frequencies between the groups. To account for the high number of statistical tests, multiple testing adjusted p-values were calculated with the R function p.adjust and the significance level was set at adjusted p < 0.05.

The assessment at which expression level a codon usage bias can be detected was based on the coding sequences of v1.6 gene models with data based on the v1.2 Combimatrix microarray experiments. Custom software used to calculate codon usage and adjusted p-values can be found in the Github repository https://github.com/kullrich/bio-scripts/tree/master/codonusage.

Plasmids for transient transfection were assembled by Golden Gate Cloning using the SapI enzyme. A modified pAM-PAT vector (accession number: AY436765) was used for cloning of 2x p35S constructs. The multiple cloning site (MCS) was removed, and together with a chloramphenicol resistance gene and a ccdb kill cassette, two SapI restriction sites were introduced to create the vector pAM-PAT-SapI. To accommodate the insertion of other promoters, the pAM-PAT-SapI vector was modified by restriction with XhoI and SalI, removing the 2x 35S promoter. For cloning reactions promoter and reporter sequences were amplified with primers containing the SapI restriction site with matching overhang. SapI-cut vector, amplified promoter and amplified reporter were incubated with SapI and T4 ligase in ligase buffer and after 1 h incubation at room temperature transformed into TOP10 E. coli cells. The mCherry gene was amplified from a modified p123 vector (kindly provided by Michael Bölker) with SapI restriction sites as overhang (mCherry-SapI_fwd, mCherry-SapI_rev) and inserted into the pAM-PAT vector behind the 2x p35S sequence. The 2xp35S:mCherry fusion was amplified with NotI restriction sites as overhang (NotI-p35SS_fwd, NotI-mCherry) and ligated into the pAM-PAT vector via the NotI site. Although insertion was not directed by specific restriction sites the clones obtained always contained the 2x p35S:mCherry in the same direction as the promoter:GFP insert. To test whether a read-through from the lhcsr1 promoter could lead to increased expression of mCherry, we removed the 2x 35S promoter in front of the mCherry. We could not detect a mCherry signal above background when using these constructs (Supplementary Table 6). Plasmid DNA was extracted by the Bibdo protocol (Birnboim and Doly, 1979) or with the NucleoBond Xtra Midi Kit (Macherey-Nagel, Germany).

Modified GFP versions were synthetized by Geneart (Germany) and delivered in the pMA-T vector, inserted via the SfiI restriction site. The GFP versions were amplified with primers containing SapI restriction sites as overhang (GFP_high-for_SAP, GFP_high_rev, GFP_low_for, GFP_high_rev). During primer design, the second codon of the smGFP was changed from serine (S) to valine (V). To test whether this has an effect on expression, we compared the GFP/mCherry ratio of the 2x p35S:S-V_smGFP to a correct 2x p35S:smGFP. The plate reader measurements did not show a difference between the S and V versions of the smGFP (Supplementary Figure 10). 1,956 bp of genomic sequence upstream of the coding sequence for the cab protein (Phypa_169593) were amplified by PCR with primers containing the SapI restriction site as overhang (p169593_for, p169593_rev). Shortened versions of the lhcsr1 gene promoter sequence were created by restriction digest and blunt/compatible end ligation with EcoRV + SalI (A), EcoRV + BclI (B), DraII (C), StuI (D), BclI + BglII (E), EcoRV + BglII (F), and SalI + XhoI (G). The EcoRV + SalI construct without 5′-UTR (H) was amplified from the full length construct with SapI restriction sites as overhang (sh_pCAB_for, sh_pCAB_rev). The rice actin1 (McElroy et al., 1990) promoter sequence was amplified by PCR from the PIG-AN vector (Schallenberg-Rudinger et al., 2017) with primers containing the SapI restriction site as overhang (pActin_Sap_for, pActin_Sap_rev).

Cloning success was tested by selection on ampicillin and either by colony PCR or test digestion. Positive candidate plasmids were sent for Sanger sequencing to GATC (Konstanz, Germany) or Macrogen (Amsterdam, Netherlands). The primers used for plasmid construction are shown in Supplementary Table 6.

Transfection protocol was adapted from (Hohe et al., 2004). Regularly disrupted protonemal tissue in a 200 mL liquid culture, pH 5.8 was adjusted to 60 mg/L dry weight and transferred to 200 mL liquid Knop medium pH 4.5. After 5–6 days, the culture was harvested by sieving (100 μm sieve). Protonemal tissue was equilibrated in 12 mL 0.51M Mannitol (pH 5.6–5.8) for 30 min, 4 mL Driselase solution (4%) was added and incubated for 1–2 h on a slowly tumbling shaker. The protoplast solution was sieved first on a 100 μm, then on a 50 μm sieve and afterward centrifuged 10 min at 50 g. Supernatant was removed, protoplasts resuspended in 10 mL 0.51 M Mannitol and centrifuged 10 min at 50 g. Supernatant was removed, protoplasts resuspended in 10 mL 0.51 M Mannitol and protoplast number counted on a Fuchs-Rosenthal counting chamber. Protoplast suspension was centrifuged for 10 min at 50 g, supernatant removed and a concentration of 1.2 × 10⁶ protoplasts per mL adjusted with MMM medium [MMM medium, 0.51 M Mannitol, 15 mM MgCl₂, 0.1% w/v 2-(N-morpholino)ethanesulfonic acid, pH 5.6]. For transfection, 100 μL DNA in 0.1 M Ca(NO₃)₂, 250 μL protoplast suspension, and 350 μL PEG solution (40% PEG 400 in MMM medium) were mixed during a 30 min incubation time. To slowly dilute the transfection solution, first 1 mL of MMM medium is added and mixed, next 2, 3, 4, and 5 mL are added and mixed every 5 min. Suspension was centrifuged for 10 min at 50 g and protoplasts resuspended in regeneration medium (0.28 M glucose and 0.24 M mannitol in Knop medium, pH 5.8). For transient transfections, circular plasmid was used. DNA amounts used for transient transfections were 10–50 μg. After transfection, protoplasts were left to regenerate in 1 mL of regeneration medium.

Fluorescence intensity measurements were performed in a FLUOstar microplate reader (BMG Labtech, Germany) with transfected P. patens protoplasts. Sample volumes of 100 μL (up to 30,000 protoplasts) were placed into black 96-well microplates with transparent bottom (Greiner Bio-one, Austria). The samples were detected using the bottom optic, orbital averaging with 2 mm diameter and 15 flashes per well. For GFP and mCherry fluorescence, emission filters at 485 and 584 nm as well as excitation filters at 520 and 620 nm were used with 10 nm bandpass width. Regeneration medium was used as the blank value and non-transfected protoplasts as background control. Blank values were subtracted and the ratio of GFP and mCherry signals was calculated for each well to normalize for the transfection efficiency. Measurements were done 6–7 days after transfection since time course experiments found the highest fluorescence signal intensity after this time period (Supplementary Figure 11).

The GO bias analyses used Fisher’s Exact Test to calculate p-values. Multiple testing corrected (Benjamini and Hochberg, 1995) q-values were calculated in R with the function p.adjust (R Development Core Team, 2008). Word cloud visualizations were created using the online tool wordle¹. Word size is proportional to the -log10(q-value) and over-represented GO terms were colored dark green if q ≤ 0.0001 and light green if q > 0.0001. Under-represented GO terms were colored dark red if q ≤ 0.0001 and light red if q > 0.0001.