Targeted RNA capture sequencing (RNA CaptureSeq) showed in 2012 that even intensively studied human fibroblast transcriptomes contain a deep layer of rare, spliced, mostly noncoding transcription — intergenic lncRNAs present at roughly 0.0006 copies per cell, plus unannotated exons and isoforms in loci as well characterized as p53 and HOX — and that reads previously dismissed as noise correspond to genuine rare transcripts. The authors concluded that the full extent of the dynamically expressed human transcriptome across cells, tissues and developmental stages is far from characterized, that the low copy number of bona fide transcripts implies substantial cell-to-cell differences challenging the notion of a single stable transcriptome per cell type, and that the biological relevance of low-level transcription remained debated. The still-open core is: a complete inventory of rare transcripts and isoforms across the diversity of human cells, a mechanistic/functional account of what the rare low-copy transcript class does — functional regulation, regulated cell-state signatures, or byproducts of bursty transcription — and a quantitative account of the cell-to-cell transcriptomic heterogeneity implied by such low copy numbers, including whether a single stable transcriptome can characterize a cell type.

Background

RNA sequencing (RNA-seq) reads out the set of RNA molecules (the transcriptome) produced by a genome, and transcriptome assembly aims to reconstruct full RNA structures (isoforms) rather than just gene-level abundance. By the late 2000s, tiling-array (microarrays of oligonucleotide probes spaced at close regular intervals along the genome, so transcription is measured at every locus rather than only at annotated genes) and RNA-seq surveys had suggested that far more of the human genome is transcribed than the ~1.5% encoding proteins, including thousands of long noncoding RNAs (lncRNAs — RNA transcripts longer than ~200 nucleotides that do not encode proteins, intergenic, overlapping, or antisense to annotated genes). This ignited the 'pervasive transcription' controversy: van Bakel et al. (2010) argued that most intergenic 'dark matter' reads merely cluster near known genes, while Clark et al. (2011) countered that widespread transcription outside known genes is real; the biological relevance of transcripts expressed at very low levels was explicitly an open debate.

Mercer et al. (2012) attacked the detection problem directly. Conventional RNA-seq of a human fibroblast sample samples each molecule at most once, so transcripts present in only a few cells are seen as sparse 'intermittent' reads that assemblers discard. Their RNA CaptureSeq method hybridized cDNA to tiling arrays covering ~2,265 selected genomic regions (~0.77 Mb), achieving ~380-fold enrichment of alignment coverage across those regions before Illumina sequencing, which brought even the rarest transcripts to saturating coverage (asymptotic discovery rate; the bottom 5th percentile of transcripts still averaged ~161 aligned reads). Applying it to foot and lung fibroblasts, they found 443 spliced intergenic regions expressing complex multi-exon lncRNAs (95.7% of intergenic splice junctions novel), and unannotated exons and isoforms in heavily studied loci including p53 and the HOX clusters, including developmentally regulated isoforms. Requiring spliced structure excluded genomic-DNA contamination and 'spurious' noise, and they showed that a substantial fraction of previously 'unassigned' RNA-seq reads are rescued by these novel transcripts. Crucially, quantitation by RT-PCR implied the intergenic lncRNAs average ~0.0006 transcripts per cell (463-fold below median fibroblast gene expression; HOXA cluster ~0.13 per cell), i.e. they are expressed in only a small subpopulation of cells — some of the rarest transcriptional events then characterized, consistent with rapid-bursting transcriptional dynamics (short episodes of promoter activity producing RNA bursts, as in the transient estrogen-response transcription documented by Hah et al. 2011) and with the first single-cell transcriptome measurements (Islam et al. 2011).

The paper's concluding discussion frames the open problem: (a) the full extent of the human transcriptome dynamically expressed across different cells, tissues and developmental stages 'is still far from being characterized'; (b) the low expression of many bona fide transcripts implies 'substantial transcriptomic differences between cells, even those in clonal cell culture', so each cell may carry 'an individual if not unique transcriptomic signature', which 'challenges the notion that there may be a single, stable transcriptome by which a cell can be characterized'; and (c) the biological function of these rare transcripts — the 'biological relevance of transcripts that are expressed at low levels' — remained contested.

Subsequent large-scale projects transformed the empirical landscape without closing the question. ENCODE (2012) catalogued biochemical activity across the genome; FANTOM5 (2014) built a promoter-level expression atlas (using CAGE, cap-analysis of gene expression, which tags the 5' end of capped RNAs) across hundreds of human primary cell types and tissues; the GTEx Consortium (2020) profiled tissue-level transcriptomes and their genetic regulation; Glinos et al. (2022) applied Oxford Nanopore long-read sequencing (which reads whole RNA molecules end-to-end, resolving full isoforms) to 88 GTEx samples and still discovered over 70,000 novel transcripts of annotated genes; and the Human Cell Atlas program (Regev et al. 2017; Rood et al. 2024) has built single-cell RNA-seq censuses (profiling transcriptomes of individual cells rather than tissue averages) moving toward an integrated draft atlas. On the function side, genome-scale perturbation of noncoding loci became possible (Zhu et al. 2016, paired-guide CRISPR-Cas9 deletion screens), yet functional mechanisms are known for only a small subset of lncRNAs (Statello et al. 2021), and the junk-versus-function debate over low-abundance transcription continues (Palazzo & Lee 2015; Palazzo & Koonin 2020). Cell-to-cell variability is now an accepted, quantifiable feature of tissues, and transcriptional bursting is a well-measured phenomenon (Tunnacliffe & Chubb 2020), but the sense in which a single, stable transcriptome can define a cell type under bursty dynamics remains a live conceptual issue.

Problem Statement

Determine what the deep complexity of the human transcriptome consists of and what it is for, as posed by Mercer et al. (2012) and left unresolved. Three jointly linked questions serve this single objective of converting the expanded transcript catalog into a mechanistic and functional account: (1) Cartography — establish how far the rare-transcript layer discovered by targeted deep capture in fibroblasts (spliced intergenic lncRNAs at ~0.0006 transcripts per cell, unannotated exons and isoforms in loci as characterized as p53 and HOX) generalizes, by mapping the full repertoire of rare and low-abundance transcripts and isoforms across diverse human cell types, tissues, and developmental stages and determining where, if anywhere, discovery saturates — a demonstrated plateau in the rate at which new rare transcripts and isoforms are discovered as additional cell types, tissues and developmental stages are added. (2) Function — determine the biological role of this rare transcript class, distinguishing regulated functional transcripts (specific cellular or regulatory roles, including cell-state signatures) from consequences of stochastic, bursty transcription in subpopulations of cells, recognizing that low abundance alone does not settle this. (3) Heterogeneity — quantify the cell-to-cell transcriptomic variability implied by low per-cell copy numbers and establish whether, and in what sense, a single stable transcriptome can characterize a cell type given rapid, transient transcriptional dynamics. The three parts are jointly posed: (1) an explicit assessment of how far the rare-transcript layer generalizes over a specified sampling design, establishing whether the rate at which new rare transcripts and isoforms are discovered plateaus (no longer accumulating) as the sampled cell types, tissues and developmental stages are added and, where it does not, where discovery ceases to saturate and why; (2) a determination of the biological role that covers the rare-transcript class as a whole rather than individual loci or cell states; and (3) a quantitative account that settles whether a single stable transcriptome can characterize a cell type. A well-evidenced limiting answer to (1) resolves part (1). Throughout, the rare/low-abundance class denotes the low-copy tail of the transcriptome — transcripts whose mean abundance in the sampled cell type is at or below roughly one copy per cell, so that they are absent from most individual cells (the source's intergenic lncRNAs average ~0.0006 transcripts per cell, ~463-fold below the median fibroblast gene; the HOXA cluster ~0.13 per cell) — a class that spans both transcripts detected in only a small subpopulation of cells ('rare' in the per-cell sense) and transcripts of genuinely low mean level, and that is applied consistently to parts (1)-(3).

References

  1. Mercer TR, Gerhardt DJ, Dinger ME, Crawford J, Trapnell C, Jeddeloh JA, Mattick JS, Rinn JL (2012). Targeted RNA sequencing reveals the deep complexity of the human transcriptome. Nature Biotechnology 30(1):99–104. doi:10.1038/nbt.2024. https://doi.org/10.1038/nbt.2024
  2. van Bakel H, Nislow C, Blencowe BJ, Hughes TR (2010). Most 'dark matter' transcripts are associated with known genes. PLoS Biology 8(5):e1000371. doi:10.1371/journal.pbio.1000371. https://doi.org/10.1371/journal.pbio.1000371
  3. Clark MB, Amaral PP, Schlesinger FJ, Dinger ME, Taft RJ, Rinn JL, Ponting CP, Stadler PF, Morris KV, Morillon A, Rozowsky JS, Gerstein MB, Wahlestedt C, Hayashizaki Y, Carninci P, Gingeras TR, Mattick JS (2011). The reality of pervasive transcription. PLoS Biology 9(7):e1000625. doi:10.1371/journal.pbio.1000625. https://doi.org/10.1371/journal.pbio.1000625
  4. ENCODE Project Consortium (2012). An integrated encyclopedia of DNA elements in the human genome. Nature 489(7414):57–74. doi:10.1038/nature11247. https://doi.org/10.1038/nature11247
  5. FANTOM Consortium and the RIKEN PMI and CLST (DGT) (2014). A promoter-level mammalian expression atlas. Nature 507(7493):462–470. doi:10.1038/nature13182. https://doi.org/10.1038/nature13182
  6. Islam S, Kjällquist U, Moliner A, Zajac P, Fan J-B, Lönnerberg P, Linnarsson S (2011). Characterization of the single-cell transcriptional landscape by highly multiplex RNA-seq. Genome Research 21(7):1160–1167. doi:10.1101/gr.110882.110. https://doi.org/10.1101/gr.110882.110
  7. Hah N, Danko CG, Core L, Waterfall JJ, Siepel A, Lis JT, Kraus WL (2011). A rapid, extensive, and transient transcriptional response to estrogen signaling in breast cancer cells. Cell 145(4):622–634. doi:10.1016/j.cell.2011.03.042. https://doi.org/10.1016/j.cell.2011.03.042
  8. The GTEx Consortium (2020). The GTEx Consortium atlas of genetic regulatory effects across human tissues. Science 369(6509):1318–1330. doi:10.1126/science.aaz1776. https://doi.org/10.1126/science.aaz1776
  9. Glinos DA, Garborcauskas G, Hoffman P, Ehsan N, Jiang L, Gokden A, Dai X, Aguet F, Brown KL, Garimella K, Bowers T, Costello M, Ardlie K, Jian R, Tucker NR, Ellinor PT, Harrington ED, Tang H, Snyder M, Juul S, Mohammadi P, MacArthur DG, Lappalainen T, Cummings BB (2022). Transcriptome variation in human tissues revealed by long-read sequencing. Nature 608(7922):353–359. doi:10.1038/s41586-022-05035-y. https://doi.org/10.1038/s41586-022-05035-y
  10. Regev A, Teichmann SA, Lander ES, Amit I, Benoist C, Birney E, Bodenmiller B, Campbell P, Carninci P, Clatworthy M, Clevers H, Deplancke B, Dunham I, Eberwine J, Eils R, Enard W, Farmer A, Fugger L, Göttgens B, Hacohen N, Haniffa M, Hemberg M, Kim S, Klenerman P, Kriegstein A, Lein E, Linnarsson S, Lundberg E, Lundeberg J, Majumder P, Marioni JC, Merad M, Mhlanga M, Nawijn M, Netea M, Nolan G, Pe'er D, Phillipakis A, Ponting CP, Quake S, Reik W, Rozenblatt-Rosen O, Sanes J, Satija R, Schumacher TN, Shalek A, Shapiro E, Sharma P, Shin JW, Stegle O, Stratton M, Stubbington MJT, Theis FJ, Uhlen M, van Oudenaarden A, Wagner A, Watt F, Weissman J, Wold B, Xavier R, Yosef N (2017). The Human Cell Atlas. eLife 6:e27041. doi:10.7554/eLife.27041. https://doi.org/10.7554/eLife.27041
  11. Rood JE, Wynne S, Robson L, Hupalowska A, Randell J, Teichmann SA, Regev A (2024). The Human Cell Atlas from a cell census to a unified foundation model. Nature 637(8048):1065–1071. doi:10.1038/s41586-024-08338-4. https://doi.org/10.1038/s41586-024-08338-4
  12. Zhu S, Li W, Liu J, Chen C-H, Liao Q, Xu P, Xu H, Xiao T, Cao Z, Peng J, Yuan P, Brown M, Liu XS, Wei W (2016). Genome-scale deletion screening of human long non-coding RNAs using a paired-guide RNA CRISPR-Cas9 library. Nature Biotechnology 34(12):1279–1286. doi:10.1038/nbt.3715. https://doi.org/10.1038/nbt.3715
  13. Statello L, Guo C-J, Chen L-L, Huarte M (2021). Gene regulation by long non-coding RNAs and its biological functions. Nature Reviews Molecular Cell Biology 22(2):96–118. doi:10.1038/s41580-020-00315-9. https://doi.org/10.1038/s41580-020-00315-9
  14. Palazzo AF, Lee ES (2015). Non-coding RNA: what is functional and what is junk? Frontiers in Genetics 6:2. doi:10.3389/fgene.2015.00002. https://doi.org/10.3389/fgene.2015.00002
  15. Palazzo AF, Koonin EV (2020). Functional long non-coding RNAs evolve from junk transcripts. Cell 183(5):1151–1161. doi:10.1016/j.cell.2020.09.047. https://doi.org/10.1016/j.cell.2020.09.047
  16. Tunnacliffe E, Chubb JR (2020). What is a transcriptional burst? Trends in Genetics 36(4):288–297. doi:10.1016/j.tig.2020.01.003. https://doi.org/10.1016/j.tig.2020.01.003