Poly A + RNA polya 3' XXXXX. SMART CDS primer First-strand synthesis & tailing by SMARTScribe RT. SMARTer II A Oligonucleotide. Single step 5' XXXXX

Transcription

1 5' 5' XXXXX SMARTer II A Oligonucleotide XXXXX 5' 5' XXXXX Poly A + RNA polya 3' SMART CDS primer First-strand synthesis & tailing y SMARTScrie RT polya Template switching and extension y SMARTScrie RT Single step 5' XXXXX XXXXX polya Amplify cdna y PCR Doule-stranded cdna Shearing (Covaris) End repair Base addition Tagmentation (Tn5) Adapter ligation PCR amplification Sequence lirary Supplementary Figure 1. The Smart-Seq protocol Protocol for cdna amplification and lirary preparation from few or single cells. Cells are first lysed in reverse transcription compatile uffer and the reaction is initiated with oligodt containing primer. First strand synthesis is completed with the addtion of a few untemplated C nucleotides followed y template switching and the incorporation of SMARTer IIA oligonucleotide. Full-length cdnas are amplified using PCR to otain a few nanograms of DNA. Fragmentation and adapter introduction was performed using either acoustic shearing or transposase tagmentation ased procedures to generate sequencing liraries.

2 c -1 k 1-2 k 2-3 k read coverage 1% 8% 6% 4% 2% % 1% 8% 6% 4% 2% % 1% 8% 6% 4% 2% % 1% 8% 6% 4% 2% % 1% 8% 6% 4% 2% % 1% 8% 6% 4% 2% % 1% 8% 6% 4% 2% % d e f 3-4 k 4-5 k 5-7 k g read coverage read coverage 1% 8% 6% 4% 2% % k h 9-15 k oocytes (Smart-Seq, n=3) oocytes, 2-cell and 4-cell lastomeres (Ref. 7, n=24) ES cells, ICM, epilast (Ref. 8, n=34) i j k oocyte, Smart-Seq oocyte, Smart-Seq oocyte, Smart-Seq oocyte, Tang et al. 29 oocyte, Tang et al. 29 5, 1, 1% detection % RPKM, log1 std dev, log1 scale oocytes (Smart-Seq) oocytes (Ref. 7) RPKM, log1 Supplementary Figure 2. Comparison of Smart-Seq and Tang et al. data (29, 21) (a-h) Read coverage across transcripts for Smart-Seq oocytes, oocytes and lastomeres from Ref. 7 and cells from Ref. 8. Transcripts has een separated according to their lengths as indicated over the plots. The gray vertical line indicate the point from where the coverage is expected to drop ased on the span of transcript lengths included in each figure. (i-k) Comparisons of sensitivity etween mouse oocytes generated with Smart-Seq (green) and from Ref. 7 (red), using 5 million reads per cell. (i) The numer of genes detected in each oocyte (result not sensitive to read cutoff for detection). (j) The percentage of genes detected in pairs of oocytes, as a function of the gene expression level (max over iologial replicates). Error ars, s.e.m. (n 7). (k) Variation in expression level estimation etween pairs of oocytes as a function of the mean expression. Error ars, s.e.m (n 7).

3 mbrain Smart-Seq ' end ES cells (Tang21) 56 k 57,85 k 57,86 k 57,87 k 57,88 k 57,89 k 57,9 k POLR2B [ - 77] Supplementary Figure 3. Full-transcript coverage and reconstruction (a) Smart-Seq reads mapping to a 14-k region of the RNA polymerase II alpha gene locus containing 27 of the 28 exons. The read coverage is displayed for Smart-Seq data from diluted amounts of mouse rain RNA (green) and for previously pulished single-cell data from mouse emryonic stem (ES) cells (red) from Tang et al. 21. () Example of a fully reconstructed transcript (human POLR2B) from Smart-Seq reads from an individual T24 cancer cell line cell. Reads were visualized using the Integrated Genome Viewer (IGV, Broad Institute).

4 Numer of genes detected c 15, 12, 9, 6, 3, 8 cells (n~13,4 genes) 1 cells (n~11,9 genes) 5 cells (n~11,2 genes) millions Numer of LNCaP cells (Smart-Seq) d Gene detection (%) 1% 8% 6% 4% 2% non-ampl. Smart-Seq (PE) 1ng 1pg 1pg % Gene expression (RPKM), log 1 Smart-Seq (Tn5) 1ng 1ng 1pg 1pg 1%.71 Gene detection (%) 8% 6% 4% 2% within cell lines full depth 2M reads 1M reads 5M reads 1M reads Spearman correlation % Gene expression (RPKM), log 1 Read depth (million uniquely mapped reads) Supplementary Figure 4. Sensitivity and variation in Smart-Seq data (a) The numer of detected genes in individual cells or when comining data from two or more cells as indicated on the x-axis. For comparison, the last data point show the numer of genes detected in standard from millions of cells. () Gene sensitivity presented as the mean percentage of genes detected in replicates, inned according to expression levels. We performed all pair-wise comparisons within groups of replicates anre report the mean and 95% confidence interval. Comparing sequence liraries constructed using either transposase tagmentation (Tn5) or shearing followed y ligation (PE) from indicated low amounts of starting materials. (c) Gene detection sensitivity for cancer cells as in (Fig. 2) ut using different sequence depths. (d) Spearman correlation (rho) in gene expression levels etween different cells of the same origin, as a function of the sequence depth.

5 Refseq transcripts 2, 18, 16, 14, 12, 1, 8, Numer of PCR cycles NA Total RNA starting amount (ng) 1 Read coverage (%) ' end Transcript (%) 3' end Smart-Seq 1ng, 18 cycles 1pg, 18 cycles 1ng, 12 cycles Supplementary Figure 5. Transcript detection and coverage with varying PCR cycles (a) The numer of Refseq transcripts detected as a function of PCR cycles (left) and starting amount of total RNA (right). A RPKM threshold of.1 was used for detection. Colors separate independent dilution series. The error ars show standard error. () Mean read coverage over transcripts. We compared read coverage for Smart-Seq data generated from 1ng or 1pg of total RNA using either 12 or 18 cycles of PCR. No reproducile trends of PCR cycles on read coverage were oserved. Errors ars represent standard deviation.

6 1..9 Pearson correlation ng B 1pg B 1pg U dilution (Br UHRR PE) Brain 1ng 1pg 1pg UHRR 1ng 1pg 1pg dilution (Brain, UHRR Tn5) Brain vs. UHRR: 1ng 1pg 1pg PC3 LNCaP T24 lncap vs. PC3 T24 vs. PC3 T24 vs. lncap human cancer cell line cells skmel uacc PMCTC hes PM vs: skmel uacc CTChES human cells and caner cell line cells CTC vs: skmel uacc PMhES c d mbrain (1ng, rep 2) RPKM, log spearman r=.97 pearson r= mbrain (5pg, rep 2) RPKM,log spearman r=.81 pearson r= T24 cell (cell 2) RPKM, log spearman r=.78 pearson r= mbrain (1ng, rep 1) RPKM, log 2 mbrain (5pg, rep 1) RPKM, log 2 T24 cell (cell 1) RPKM, log 2 Supplementary Figure 6. Correlations in gene expression levels etween technical and iological replicates. (a) Correlations over gene expression levels in dilution- or single cell replicates were shown as oxplots (lack). Analyses of different types of cells or diluted RNA of different origins were shown in red. Correlations were computed from log2 transformed RPKM gene expression levels. (-d) Representative scatter plots of gene expression levels from liraries generated from independent technical replicates of 1ng total RNA (), or 5pg total RNA (c), or liraries generated from independent T24 cancer cell line cells (d). These scatter plots are representative of gene expression levels variations found and the results from all pairwise comparisons are summarized in Fig. 2c,d.

7 Smart-Seq (PE) Smart-Seq (Tn5) Smart-Seq (PE) Smart-Seq (Tn5) Correlation with transcript G+C fraction Correlation with transcript length LNCaP UHR hbrain mbrain LNCaP UHR hbrain mbrain Supplementary Figure 7. Analyses of GC and length iases in Smart-Seq and data (a) Pearson correlation coefficients etween log transformed expression values and the transcript GC content for pre-amplified (Smart-Seq) and standard data. () Pearson correlation coefficients etween log transformed expression values and log transcript length for pre-amplified (Smart-Seq) and standard. As the true iological GC and length iases are not known, there is no ovious est correlation. We found no systematic difference etween Smart-Seq data generated with shearing and ligation (PE) or Tn5-mediated tagmentation (Tn5) or standard samples, ut a tendency towards lower GC content correlations for Smart-Seq data.

8 Smart-Seq ESC ESC rain 1 pg 1+ pg CTC ±.5 ±.3 ±.8 ±.16 standard rain melanoma ±.4 ±.3 ±.9 ± ±.2 ±.2 ±.2 ±.6 lymph node leukocytes ±.4 ±.2 ±.7 ± ±.4 ±.3 ±.6 ±.15 Supplementary Figure 8. Correlation analyses etween Smart-Seq and gene expression levels. Individual human ES cells (ESC, n=8), diluted rain RNA at 1pg and 1pg or more, and individual CTCs (n=6), compared with data from error of the mean is shown for each comparison. Within each Smart-Seq sample (i.e. columns) the matrix cells are colored according to a linear white to red scale, with white at lowest mean correlation red at maximum mean correlation. Note data from ES cells and melanoma respectively. Identical Brain RNA prepared with Smart-Seq and correlate higher with increasing starting amounts in Smart-Seq. ESC RNA-Seq data was downloaded from GEO (GSM672836).

9 Smart-Seq on individual cells Immune SKMEL UACC PM CTC ESC PBL BL BJ W L PMEL MITF TYR MLANA CD53 PTPRC CCL5 CD4 Lineage Markers: melanocytes lymphocytes, monocytes Expression (RPKM, log 2 ) Expression (RPKM, log 1 ) Supplementary Figure 9. Single-cell cdnas from peripheral lood lymphocytes Two individual peripheral lood lymphocytes (PBLs) were prepared using Smart-Seq and shallow sequenced on a MiSeq. These two cells (PBL) were compared with Smart-Seq transcriptomes from melanoma cancer cell lines (SKMEL5, UACC257), primary melanocytes (PM), circulating tumor cells (CTC), human emryonic stem cells (ESC) and transcriptome data from urkitts lymphoma cell lines (BL41, BJAB), white lood cells (W) and lymphnodes (L). Gene expression levels (RPKM, log2) for marker genes of melanocyte and hematopoietic lineages are shown in a heatmap. Despite a shallow sequence depth, we detected markers for hematopoietic lineage in oth PBLs, in contrast to melanocyte lineage markers.

10 SKMEL5 UACC257 PM CTC hesc BL MAGEH1 MAGEA1 MAGEC1 MAGEB1+MA GEB4 MAGEA11 MAGEA8 MAGEF1 MAGED4+MA GED4B MAGED4+MA GED4B MAGEE1 MAGED2 MAGED1 MAGEB6 MAGEL2 MAGEA4 MAGEB1 MAGEB18 MAGEC3 MAGEB16 MAGEB3 MAGEE2 MAGEA9+MA GEA9B MAGEA9+MA GEA9B MAGEB2 MAGEC2 MAGEA1+MA GEA1-MAGEA5+MA GEA5 MAGEA2+MA GEA2B MAGEA2+MA GEA2B MAGEA12 MAGEA6 MAGEA3 Relative Expression ( log 2 ) -5 5 Supplementary Figure 1. Expression of all melanoma-associated antigens (MAGEs) Heatmap showing the expression levels of all MAGE genes in CTC, PM, melanoma cell lines (SKMEL5 and UACC257), and in human ESC and Burkitts lymphoma cell lines BL41 and BJAB (BL).

11 Genomic sites Numer of sites (cumulative) 8, 6, 4, 2, Fraction of sites overlapping dsnp (132) Ti/Tv Minimum numer of samples Transitions, known SNPs Transversions, known SNPs Transitions, novel sites Transversions, novel sites Supplementary Figure 11. Detection of SNPs and mutations in CTCs (a) Genomic sites with support for allelic difference etween single-cell CTCs and reference sequence were sorted according to the QUAL values (Phred scaled proaility of polymorphism). For ins of genomic sites, we analyzed the fraction of sites that were already present in dsnp (uild 132) and the ratio of transitions to transversions (Ti/Tv). In genomic DNA re-sequencing studies, one expects to find ~9% of individual differences to e present in dsnp and a Ti/Tv ratio of ~2.1 (genome-wide) or ~2.8 (exons). We found that sites with QUAL values of 5 or higher (dashed line) had 92% overlap with SNPd and a Ti/Tv ratio of 2.3, suggesting that genotype calls aove that QUAL threshold are reliale. At this threshold we found 4,312 genomic sites using transcriptome data from the six CTCs and requiring that the allele e supported in two or more CTCs. () The numer of sites indentified requiring support in an increasing numer of CTCs. The arplot shows the numer of genomic sites with known or novel transitions and transversions. We found a large difference etween sites supported in one or two cells, likely reflecting a large amount of false positives in calls from a single cell. Requiring support in two or more cells led to more consistent and slowly declining numer of SNP or mutation calls that proaly reflect power. The QUAL threshold used is 5 as indicated with the dashed line in (a).