Antisense Deletion Library: Strategy, Library Construction, and Validation
(A) Abrogation of gene-specific asRNA transcription using seamless gene tagging. A cassette containing one full sfGFP and a ΔN-sfGFP fragment and two I-SceI sites flanking a counter selectable marker (URA3) is inserted using PCR targeting at the 3′ end of the ORF in a strain also containing a Gal1-I-SceI endonuclease cassette. Upon expression of I-SceI on galactose-containing medium, the inserted cassette is cleaved and recombination between the sfGFP repeats occurs with high efficiency.
(B) Three different cassettes, containing only sfGFP, or sfGFP and either PHO5_T or PHO5_T:scr were used to generate for each gene three different strains.
(C) RNA and protein expression analysis for the IME4 gene using strains constructed with the method described in (A) and (B). Top left: expression data from tiling arrays of the IME4 genomic region in haploid cells for the Watson (top) and Crick (bottom) strands are shown. Plots show normalized signal intensities from three hybridizations (YPD1, YPD2, and YPD3) using cells grown in rich medium (YPD). Transcript boundaries are depicted in red, and the darker blue color indicates a higher hybridization signal. Top right: IME4 RNA and asRNA abundances measured by qRT-PCR using strand specific primers for the sfGFP sequence. The values were normalized to the ACT1 gene. Bottom: western blotting and quantification of the signal of Ime4-sfGFP (Ime4). Pgk1 was used as a loading control. One representative blot is shown. Quantifications are based on three replicates, normalized using Pgk1 as a reference. Error bars indicate SD.
(D) Three categories of genes were chosen for tagging based on the data from , as indicated. The transcript start site (TSS) area of the sense gene is shown in purple.

High-Throughput Quantitative Microscopy Identifies Antisense-Regulated Genes
(A) Cells were imaged, and whole-cell fluorescence intensities were measured using co-cultured non-fluorescent cells (identified by differential labeling with an mCherry reporter) for background subtraction and fluorescence normalization on a per-well basis (). Example cells for IME4 in YPAD are shown. Gray, non-fluorescent cells; green, sfGFP-expressing sample cells.
(B) Venn diagram showing the number of antisense-regulated genes under the different growth conditions. Numbers in parentheses indicate the total number of genes with sfGFP levels above background (for PHO5_T and/or PHO5_T:scr). In total, 41 genes were found to be antisense regulated under at least one condition.
(C) A volcano plot of p values versus log2-fold changes: ratio of sfGFP levels between PHO5_T:scr (with antisense) and PHO5_T (no antisense) strains. Each dot represents one gene in one growth condition. Red indicates genes that met our criteria for being regulated by antisense (see text).
(D) Log2-fold changes between PHO5_T:scr and PHO5_T for each growth condition. Selected examples are labeled. Short black bars indicate the median. Colored circles indicate genes whose sfGFP intensities could not be reliably distinguished from background for the PHO5_T:scr construct. To be able to visualize these genes in the diagram, a small offset for the PHO5_T:scr expression value was introduced if necessary (see for details).
(E) Box plots of sfGFP intensities (sfGFP_norm, normalized to the fluorescence of the control cells without sfGFP) are shown for genes found to exhibit condition-specific regulation by antisense. Only the conditions differing in their regulatory strength are shown. Numbers indicate the log2-fold change of PHO5_T:scr over PHO5_T. Red numbers indicate that the gene was found to be significantly regulated in that condition and fill colors represent the construct (PHO5_T or PHO5_T:scr). Populations of both biological replicates were pooled in the boxplots for simplicity.

Correlation of Regulation by Antisense with Antisense and Protein Levels
(A) The effect of antisense on protein levels (log2-fold change of sfGFP levels between PHO5_T:scr and PHO5_T) versus antisense transcript levels. Data obtained for all conditions from fluorescence microscopy and from tiling arrays are shown. These were obtained using the same strain background and the same growth conditions. Blue circles represent genes not found to be regulated by antisense. Red and green circles indicate genes with regulation by antisense with the latter being subject to an antisense-dependent on/off switch (see and main text). Spearman’s correlation coefficient ρ is indicated in the plot.
(B) As in (A), but with sfGFP intensities of the PHO5_T:scr construct plotted on the x axis.

TSS Overlap Correlates with Regulation by Antisense Transcription
(A) Genes that were tested for antisense regulation in at least one condition were divided into three groups that differ in the degree of antisense overlap with the sense (see main text). Next, the number of genes that were regulated by antisense in at least one condition was determined. The frequency of genes regulated in at least one condition is significantly higher when the antisense is annotated to overlap the TSS than in the group without annotated TSS overlap or the group where no antisense has been annotated (Fisher’s exact test, p < 0.05).
(B) For every gene found to be regulated in at least one condition the maximum significant log2-fold-change between PHO5_T:scr and PHO5_T was determined across growth conditions. This does not include genes with an “all or nothing” behavior (see main text).
(C) Genes were aligned at their TSS (left) or stop codons (right). The aligned NET-seq traces are shown using group-wise smoothed medians (). Colors indicate the extent of SUT overlap with the sense as annotated in . Solid lines indicate antisense-regulated genes, whereas dashed lines indicate non-regulated genes.

Features Correlating with Regulation by Antisense
(A and B) H3K4me2 and H3K4me3 densities in antisense-regulated (red) versus non-regulated (green) genes. All genes were aligned at their stop codons. Lines and ribbons indicate bootstrapping-based mean and 95% confidence interval estimates, respectively.
(C) The number of Nrd1 binding sites as determined previously () was determined for all transcripts in our library and the number of binding sites per gene is shown, grouped by whether the gene was found to be regulated by antisense in at least one condition or not.
(D) Transcript lengths of genes grouped by whether they were regulated by antisense in at least one condition or not. The groups are significantly different (Wilcoxon’s rank sum test, p < 0.05).

Impact of Antisense Suppression on Gene Expression Noise
(A) Log₁₀[CV (%)²] versus log₁₀(sfGFP) of 66 genes permit robust linear fitting (solid black line) and the calculation of noise levels corrected for differences in abundance (CV_res; one example is indicated). Different fits are possible, but the result (shown in B) is the same in all cases (see also D and . PHO5_T and PHO5_T:scr strains are indicated by triangles and squares, respectively. Genes regulated by antisense are shown in orange, whereas non-regulated genes are shown in green.
(B) Distribution of the gene-wise differences of CV_res values of PHO5_T:scr minus PHO5_T constructs from (A), grouped depending on whether the genes were found to be antisense regulated under the growth condition used for FACS (SC) or not. The difference between the groups is significant (Wilcoxon’s rank sum test, p < 0.05).