Supplementary Material From Singer Lab Publications:

Materials and Methods

Probe Preparation

Three to five 50-mer DNA probes were synthesized for each target transcript. Sequences were designed using OLIGO 4.0 (Molecular Biology Insights) and verified for specificity using our own JAVA-coded BLAST client that queries the publicly available GenBank library at the NCBI. Amine-modified thymidine bases (five per probe) were used as substrates for amidation using commercially available succinimidyl ester conjugates of fluorescent dyes. In practice, we have used commonly available cyanine dyes as well as fluorescein and rhodamine derivatives. For experiments shown here the fluorophores were Cy3, Cy3.5, Cy5 (Amersham), Oregon Green 488 and Alexa Fluor 488 (Molecular Probes) (Table S1, column 2).

Probe Sequences

Three to five oligomer DNA probes were synthesized for each transcript detected:
[note:Tthe sequences for beta and gamma actin published on the Science website are transposed!]
[The sequences below are correct.]

Early Growth Response Protein 1 (EGR-1, accession NM_001964):
1 ATCATCTCCTCCAGCTTAGGGTAGTTGTCCATGGTGGGCGAGTGAGGAAA
2 TTCAATTGTCCTGGGAGAAAAGGTTGCTGTCATGTCCGAAAGCCCTGTGG
3 CCTCTCCTATGGCCATCTCCTCCTCCTGTCCTTTAAGTCTCTTGTGTTTC
4 CGATTTTCTAGCATTGAAGGGAGCAAGGGGTCAGGCATATGATGGAGATG

ß-actin (NM_001101)
1 GTGAACTTTGGGGGATGCTCGCTCCAACCGACTGCTGTCACCTTCACCGT
2 TCCTTAGAGAGAAGTGGGGTGGCTTTTAGGATGGCAAGGGACTTCCTGTA
3 CTTTTATTCAACTGGTCTCAAGTCAGTGTACAGGTAAGCCCTGGCTGCCT

g-actin (NM_001614)
1 TGTCTTCCACAAGCACGTTCTTACCTTAGCCACGAAGTGACCAAGCCACA
2 TTCTTGCCAGCCTCTAGAGAAATCCCAGAACACTCAGCCCTGACACGTTA
3 TTACGGCAGCACTTTTATTTTTCCTTACACAATGACGTGTTGCTGGGGCC

c-myc (NM_002467)
1 TCGTAGTCGAGGTCATAGTTCCTGTTGGTGAAGCTAACGTTGAGGGGCAT
2 CCACATACAGTCCTGGATGATGATGTTTTTGATGAAGGTCTCGTCGTCCG
3 TGACCTTTTGCCAGGAGCCTGCCTCTTTTCCACAGAAACAACATCGATTT
4 CTGGTGCATTTTCGGTTGTTGCTGATCTGTCTCAGGACTCTGACACTGTC
5 GGCCTTTTCATTGTTTTCCAACTCCGGGATCTGGTCACGCAGGGCAAAAA

c-jun (NM_002228)
1 GGCGTTGAGGGCATCGTCATAGAAGGTCGTTTCCATCTTTGCAGTCATAG
2 TCGGCCAGGTTCAGGGTCATGCTCTGTTTCAGGATCTTGGGGTTACTGTA
3 GTTTTCACTTTTTCCTCCAGCCGGGCGATTCTCTCCAGCTTCCTTTTTCG
4 CGTGGTTCATGACTTTCTGTTTAAGCTGTGCCACCTGTTCCCTGAGCATG

CyclinD1 (NM_001758)
1 CTACGCGCGGATGGTTTCCACTTCGCAGCACAGGAGCTGGTGTTCCATA
2 TTGACAGGAAGCGGTCCAGGTAGTTCATGGCCAGCGGGAAGACCTCCTCTTCGCACTA
3 CTTCTTAGAGGCCACGAACATGCAAGTGGCCCCCAGCAGCTGCAGGCGGCTCTA
4 ATCTTGAGCTTGTTCACCAGGAGCAGCTCCATTTGCAGCAGCTCCTCGGGCCGTC

Interleukin-8 (IL-8, NM_000584)
1 ACAGTGAGATGGTTCCTTCCGGTGGTTTCTTCCTGGCTCTTGTCCTAGAA
2 CTAAGTTCTTTAGCACTCCTTGGCAAAACTGCACCTTCACACAGAGCTGC
3 TTCTGTGTTGGCGCAGTGTGGTCCACTCTCAATCACTCTCAGTTCTTTGA
4 GAATTCTCAGCCCTCTTCAAAAACTTCTCCACAACCCTCTGCACCCAGTT

Myeloid Cell Leukemia 1 (MCL1, NM_021960)
1 AATCCTGCCCCAGTTTGTTACGCCGTCGCTGAAAACATGGATCATCACTC
2 AACCCATCCCAGCCTCTTTGTTTAACTAGCCAGTCCCGTTTTGTCCTTAC
3 AGCAGCACATTCCTGATGCCACCTTCTAGGTCCTCTACATGGAAGAACTC
4 TCCACCCTACCATCTTCACTAAATCTAAAAGTCCTCCTCCATAGCTTCCC

Transforming Growth Factor ßImmediate Early Gene (TIEG, NM_005655)
1 TTGGCAGTATCTGAGAGTGACTTGAAGTGTACAGTAGATGGCGCTGGTGC
2 GGCATCAGCTGTATGACGAATCACACTTGTTGCCTGAGCTTTGGGGAGTT
3 GGCATCAGCTGTATGACGAATCACACTTGTTGCCTGAGCTTTGGGGAGTT
4 TGCGTCCTCGTGTGGGCCTTCAGATGGGAACTTTTAAAGTATGTCTTGCC

Dual Specificity Kinase 1 / MAP Kinase-Phosphatase (DUSP, NM_004417)
1 TGTCCTCCACAGGGATGCTCTTGTACTGGTAGTGACCCTCAAAATGGTTG
2 ATTAGTCCTCATAAGGTAAGCAAGGCAGATGGTGGCTGACCGGGAAATGC
3 GCTGAAGTTGGGAGAGATGATGCTTCGCCTCTGCTTCACAAACTCAAAGG
4 TAAATAAGGACCAGCCCTCTCGAGCCCCTCCCAGAGTTATTGCATTTCTC

c-fos (NM_005252)
1 CCATGCTGGAGAAGGAGTCTGCGGGTGAGTGGTAGTAAGAGAGGCTATCC
2 GGATGAACTCTAGTTTTTCCTTCTCCTTCAGCAGGTTGGCAATCTCGGTC
3 GATGCTGGGAACAGGAAGTCATCAAAGGGCTCGGTCTTCAGCTCCATGCT
4 TAGGTGAAGACGAAGGAAGACGTGTAAGCAGTGCAGCTGGGAGTACAGGT

Cysteine-Rich Angiogenic Factor (Cyr61, NM_001554)
1 CCATTCCAAAAACAGGGATCCGCTTCAGTGAGCTGCCTTTTCCAACTGCA
2 TTTCACAAGGCGGCACTCAGGGTTGTCATTGGTAACTCGTGTGGAGATAC
3 CGCAGTACTTGGGCCGGTATTTCTTCACACTCAAACATCCAGCGTAAGTA
4 CCCCGCCCATTTTCTCCATGATTCTGATTCTGACACTCTTCTCCCTTGTT

Fos-related Antigen 1 (Fra-1, NM_005438)
1 CTCCTCTTCCTCCGGGCTGATCTGTTCACAAGGCCTTCGACGTACCCCTG
2 ATCTCTCGCTGCAGCCCAGATTTCTCATCTTCCAGTTTGTCAGTCTCCGC
3 TGAAGACCAGGCTGGGGGTGAAAGGAGTTAGGGAGGGTGTGGTCATGAGT
4 AGGATGGGTCTCCGCTGCTGCTGCTACTCTTGCGATGAGCTGAGGCACAA

Cell Culture and Hybridization

DLD-1 cells (ATCC CCL-221) and normal human fibroblasts (ATCC CRL-2091) were grown under standard conditions and plated onto cover slips. To augment and synchronize transcription, cells were serum-starved for 24 hours and serum-stimulated (10% FBS) for 30 min in the presence of cycloheximide (10ug/ml). Cytoplasm was extracted with 0.5% Triton X-100 to increase probe penetration and decrease cytoplasmic fluorescence. The nuclei were fixed in 4% paraformaldehyde for 15 min at room temperature, washed in PBSM (phosphate buffered saline with 5mM MgCl2), and stored in 70% ethanol at 4C. For hybridization, cells were rehydrated into PBSM and equilibrated in 50% formamide / 2X SSC for 10 min at room temperature. Hybridization was performed for 3 hours in 50% formamide / 2X SSC / 10% BSA at 37C. The concentrations of probes labeled in each color for each gene were determined empirically. Cover slips were then incubated twice in 50% formamide / 2X SSC for 20 min at 37C. Next, the samples were processed through decreasing concentrations of SSC, and finally returned to PBSM for DAPI counterstaining. Slides were mounted in 90% glycerol in PBS with phenylenediamine.

Imaging

Three-dimensional stacks of images were acquired using AX70 and BX51 microscopes (Olympus) with a PZ54 E piezoelectric translator (Physik Instrumente) and CH-350(502) and CoolSNAP-HQ Charge-Coupled Device (CCD) cameras (Roper Scientific), using IPLab software version 3.07 (Scanalytics). We used PlanApo 60x, 1.4 NA objectives (Olympus) and HiQ band pass filters for DAPI (#31000), FITC (#41001), Cy3 (#SP-102v1), Cy3.5 (#SP-103v1) and Cy5 (#41008) (Chroma Technology).

Image Processing

Transcription site detection algorithms and nuclear segmentation were developed in our own software package, using the Java Development Kit 1.3.0 and the Java Advanced Imaging Extension 1.0.2 (Sun Microsystems). CCD-captured image volumes for an arbitrary number of fluorescent colors are read and compiled into a single data image. Then, feature extraction by convolution filtering is performed to highlight areas of punctate signaling using kernels devised and tested empirically. Areas of peak intensity are chosen by processing the entire image in all color bands. These regions are sorted by barcode, or the combination of signals present, based on an adjustable threshold. Finally, the identity of the signal is interpreted based on a color coding scheme. A minimum number of color signals can be imposed in order to better distinguish between true signals from multiple fluorophores and noise from the assay. The data is exported to a spreadsheet program for analysis.
A second step of analysis involves determining the co-localization of the detected signals inside individual nuclei. The nuclear counterstain (DAPI image) is binarized using a selectable threshold and contiguous areas are “flood-filled”. Bodies that are not totally surrounded by background are ruled out as they are only partially imaged and cannot be completely analyzed. Then, the set of punctate signals is superimposed on the nuclear regions. Signal outside any body can be ignored to reduce further processing. Signals within bodies are scored for number (or simply for absence and presence) for each detectable species in the color code. Statistics are compiled for the entire field of view as well as for each body in a form exportable for further analysis.

Supporting Text

Number of Barcodes

Simple combinations of colors, or barcodes, are used to encode unique transcript identities. For c total colors, Bc, the number of barcodes, is given by:

As the number of fluorophores increases, the number of detectable transcripts increases exponentially. Examples are given below (Table S1, row 1).
To increase the number of detectable transcripts beyond the capability provided by using color alone we used a ratio-labeling scheme. This approach has been demonstrated previously for detection of DNA targets by in situ hybridization (1, 2). The high fidelity output of our computerized algorithms allows the precise discrimination of gradations of color intensity values for RNA FISH. Our preliminary studies have shown that computer-analyzed transcription site profiling can allow the discrimination of at least two gradations of color intensity values in the detection of sites. For c total colors, Rc, the number of ratio-labeled barcodes, is defined by the following formula given Ni discernable gradations of intensity, Ci possible sets of ratios for a single color combination, and Bi binary color combinations, for i colors:

Note that a minimum of two colors is always used. Example values are shown below (Table S1, rows 3-6).
Increases in the number of detectable targets can also be accomplished by separating existing fluorochromes that have similar emission spectra using computerized algorithms for spectral decomposition (3), or through the use of additional fluorophores. Using these means and the resolution afforded by digital imaging, a large cohort of genes could be assayed, which will provide a more comprehensive understanding of transcription.

Supporting Tables

Table S1. Numbers of barcodes. Using a given number of colors, the resulting quantity of distinct combinations that can be used for barcodes is determined by the formula given above in supporting online text (row 1). In order to increase fidelity of the assay, barcodes are limited to a minimum of two colors (row 2). In order to increase the number of detectable species, we apply a ratio labeling protocol yielding barcode numbers as above in supporting online text (rows 3-6). These predictions include only separation of two levels of intensity (Ni=2). The numbers are far greater given a higher number of discernable intensity thresholds.

Table S2. Pair wise correlations of genes expressed in stimulated DLD-1 cells. Single cell gene expression correlations are analyzed by scoring the number of nuclei positive for any two given genes, the number positive for one, and the number of double-negatives to determine an odds ratio. These are the only statistically significant, positively-correlated gene pairs for the ten genes and 45 possible pairs, as determined to 99% confidence.

Supporting References and Notes

1. P. M. Nederlof, S. van der Flier, J. Vrolijk, H. J. Tanke, A. K. Raap, Cytometry 13, 8 (1992).
2. J. G. Dauwerse et al., Hum. Mol. Genet. 1, 8 (1992).
3. E. Schrock et al., Science 273, 5274 (1996).