• Loading metrics

A Hox-Embedded Long Noncoding RNA: Is It All Hot Air?

  • Licia Selleri , (LSE); (GSB)

    Affiliation Program in Craniofacial Biology; Institute of Human Genetics; Eli and Edythe Broad Center of Regeneration Medicine & Stem Cell Research; Department of Orofacial Sciences & Department of Anatomy; University of California San Francisco, San Francisco, California, United States of America

  • Marisa S. Bartolomei,

    Affiliation Department of Cell and Developmental Biology, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania, United States of America

  • Wendy A. Bickmore,

    Affiliation Medical Research Council Human Genetics Unit, Institute of Genetics and Molecular Medicine, University of Edinburgh, Edinburgh, United Kingdom

  • Lin He,

    Affiliation Division of Cellular and Developmental Biology, University of California Berkeley, Berkeley, California, United States of America

  • Lisa Stubbs,

    Affiliation Institute for Genomic Biology, Department of Cell and Developmental Biology, University of Illinois, Urbana, Illinois, United States of America

  • Wolf Reik,

    Affiliation Department of Epigenetics, The Babraham Institute, Cambridge, United Kingdom

  • Gregory S. Barsh (LSE); (GSB)

    Affiliations HudsonAlpha Institute for Biotechnology, Huntsville, Alabama, United States of America, Department of Genetics, Stanford University School of Medicine, Stanford, California, United States of America

A Hox-Embedded Long Noncoding RNA: Is It All Hot Air?

  • Licia Selleri, 
  • Marisa S. Bartolomei, 
  • Wendy A. Bickmore, 
  • Lin He, 
  • Lisa Stubbs, 
  • Wolf Reik, 
  • Gregory S. Barsh

Over 20 years ago, the discovery of Xist as a critical component of X chromosome inactivation revealed a fundamental role for long noncoding RNAs (lncRNAs) in epigenetic regulation during mammalian development and foreshadowed a fascinating connection between RNA and chromatin modification [13]. In the last decade, the field has exploded, heralded in part by a 2007 landmark paper from the group of Howard Chang [4] describing that knockdown of a lncRNA (Hox Antisense Intergenic RNA [HOTAIR]) was associated with loss of transcriptional repression from a locus on another chromosome in trans. HOTAIR lncRNA—encoded within the HOXC locus, although its expression seemed to be required for normal epigenetic silencing of HOXD genes—became one of the most well-known examples of functional lncRNAs in the field of developmental epigenetics. Interest intensified when a subsequent paper from the Chang lab [5] reported that targeted deletion of the orthologous locus in the mouse (Hotair) caused homeotic transformations underpinned by derepression of HoxD gene transcription in vivo. Discovery of new lncRNAs and exploration of their potential actions and effects during development and disease is a continued source of excitement [6,7]. But questions about the effects and actions of Hotair have been controversial, raised in part by work from the group of Denis Duboule [8]. This debate is addressed directly in the current issue of PLOS Genetics in a manuscript from Duboule’s laboratory [9] that reanalyzes Hotair mutant mice generated by the Chang lab, a formal comment in response to that manuscript from the Chang lab [10], and this perspective.

A precise genomic organization conserved for hundreds of millions of years underlies the complexity of Hox gene regulation. Expression of adjacent Hox genes in specific spatiotemporal domains helps to pattern and maintain the identity of developmental compartments along the body and appendicular axes [8,11]. It is the latter part of that equation—maintenance of developmental identity—for which genomic organization may be so important, and for which epigenetic mechanisms play a critical role, because specific patterns of Hox gene transcription can be faithfully maintained long after development has concluded, sometimes in cultured cells outside the body [12].

These considerations, together with previous studies on the role of noncoding RNA (ncRNA) in the regulation of Hox gene expression in fruitflies [13], provided the backdrop for the work of Rinn et al. [4] to exhaustively profile noncoding (and coding) RNAs transcribed from the four human HOX loci. In flies, ncRNAs were shown to act in cis via transcriptional interference, so a potential trans-acting role for HOTAIR, from the HOXC to the HOXD cluster, came as something of a surprise.

It is in this context—a function for HOTAIR?—that the current work [9,10] pertains. The mouse HoxC cluster is ~140 kb in length; Hotair is a 2-exon transcript that lies in the region between Hoxc11 and Hoxc12. In the 2013 paper from Li et al., mice carrying an ~4 kb deletion that removed both Hotair exons variably exhibited a homeotic posterior transformation with five rather than six lumbar (L) vertebrae (L6 was transformed into the sacral (S)1 vertebra; L6→S1), as well as wrist bone malformations and subtle caudal anterior transformations [5]. The authors also found that Hotair RNA was enriched in RNA immunoprecipitates (RNA-IP) from E11.5 embryos using antibodies against components of the Polycomb repressive complex 2 (PRC2), which methylates histone H3 at lysine 27 (H3K27), or against the Lsd1 complex, which demethylates histone H3 at lysine 4 (H3K4) in vivo [5]. Finally, the authors reported that derepressed genes in Hotair mutant tail tip fibroblasts, including multiple members of the HoxD but not HoxC clusters, exhibited loss of H3K27me3 and gain of H3K4me3 [5]. Taken together, this work has been at the foundation of the viewpoint that Hotair functions as a trans-acting repressor of HoxD gene expression via recruitment of PRC2 [4,5].

But this viewpoint has been controversial. Part of the controversy stems from a 2011 study from the Duboule group [14] in which it was reported that E13.5 embryos or embryo fibroblasts carrying a complete deletion of the mouse HoxC locus exhibited no significant changes in HoxD expression or chromatin marks. At the time, this observation was interpreted as a difference in function between human HOTAIR and mouse Hotair. However, with publication of the 2013 paper from the Chang group [5], claiming a conserved function with regard to trans-acting repression of HoxD genes, there seemed to be a direct conflict between the two sets of observations [5,14].

How could the mice generated and studied by the Chang group, with a 4-kb deletion within the HoxC cluster, exhibit a more severe phenotype with respect to HoxD gene regulation than mice carrying a deletion of the entire HoxC cluster studied by the Duboule group? Li et al. [5] suggested that deletion of the entire cluster might have removed genes with functions that oppose that of Hotair and generously provided their animals (with the 4-kb deletion) to the Duboule group for additional analyses. The current issue of PLOS Genetics features the manuscript describing the results of this further investigation [9]. Although the analyses are extensive, the results are simple to summarize: Amandio et al. [9] report that, in their hands, the Hotair deletion allele generated by the Chang lab does not cause up-regulation of HoxD genes and associated homeotic transformations. Amandio et al. [9] do identify and confirm subtle anterior transformations in caudal vertebrae but suggest that this difference may be caused by changes in HoxC expression due to local effects of the deletion.

How can two groups reach essentially opposite conclusions studying the exact same allele? We consider this apparent paradox with regard to both the whole animal and the transcriptomic phenotypes. First, and as pointed out here in a formal comment from the Chang group [10], their 2013 study [5] was carried out on an inbred background (C57BL/6), whereas the Duboule group deliberately used a mixed background that includes both C57BL/6 and CBA genomes [9,14]. Potential advantages of an inbred background include the ability to compare mutant and nonmutant animals across space and time (although the Chang lab results are based on litters that carry both mutant and nonmutant animals), whereas advantages of a mixed background include the potential for inter-individual stabilization of variably expressed phenotypes due to canalization [15,16]. Indeed, the L6→S1 transformation causing 5 instead of 6 lumbar vertebrae (the most important phenotype from the perspective of evaluating the proposed effect of /Hotair/ on /HoxD/ derepression) is variably expressed. Li et al. reported that 18/31 mutant embryos had 5 lumbar vertebrae compared to 1/11 non-mutants [5]. By contrast, Amandio et al. [9] report that most of the embryos they examined, 8/10 mutant and 8/11 nonmutant, had 5 lumbar vertebrae. In other words, both mutant and nonmutant animals examined by the Duboule lab had the same phenotype described as characteristic for mutants in the Chang lab.

In fact, the boundary between lumbar and sacral vertebrae has been an oft-studied and intensive subject of investigation by mouse geneticists for decades and might best be described as capricious. The total number of lumbar vertebrae (5 or 6) and “sacralization” of L6 vary not only between inbred strains but also within the same strain and can be influenced by sex, maternal diet, age, parity, litter size, and environmental temperature [1719]. Although these considerations do not change the statistical significance (p = 0.0002) of the observation by Li et al. [5], they do call into question its biological significance, as pointed out more than 50 years ago by Earl Green [17], Anne McLaren [20], and Hans Grüneberg [21].

What about the different transcriptomic conclusions? Li et al. [5] generated RNA-Seq data from duplicate cultures of tail tip fibroblasts (mutant, nonmutant, and heterozygous), whereas the RNA-Seq studies of Amandio et al. [9] were obtained from duplicate microdissections of six different E12.5 embryonic regions (forelimb, hindlimb, genital tubercle, and lumbosacral, sacrocaudal, and caudal trunk). Li et al. had made their raw data publicly available, allowing Amandio et al. to present and analyze RNA-Seq reads from both the tail tip fibroblasts and the embryonic fragments with the exact same bioinformatics pipeline, from alignment to annotation to inferences regarding differential expression. An example from the unique read count data is informative: in the study of Li et al., Hoxd10 was the most highly expressed HoxD gene in tail tip fibroblasts, and there were 4,4,7,8,15, and 15 unique reads for duplicates from nonmutant, heterozygous, and mutant cells, respectively. In contrast, in the caudal embryonic tissue (the posterior-most trunk segment where Hotair is highly expressed), there were 1,654, 1,475, 1,289, and 1,315 unique Hoxd10 reads, for duplicates from nonmutant and mutant tissues, respectively. These data represent a general theme of the transcriptome studies from both groups: the tail tip fibroblasts studied by Li et al. [5] express HoxD genes (and, incidentally, Hotair) at very low levels, whereas the embryonic tissue fragments studied by Amandio et al. [9] exhibit expression of HoxD (and Hotair) at much higher levels, allowing for statistically robust comparisons.

Additional studies in both papers [5,9] and additional Hotair alleles in other papers [22] raise additional questions. How could Hotair affect cells or embryonic regions such as the forelimb or lumbar trunk where, according to Amandio et al. (9), its expression was not detectable? What might account for a subtle anterior vertebral transformation (in the caudal region), a phenotype upon which both groups agree? This phenotype is typically caused by loss- rather than gain-of-function alterations in Hox genes. And to what extent are the phenotypes caused by the Hotair deletion due to Hotair itself or cis-regulatory alterations in HoxC gene expression? Regardless, neither these questions nor their potential answers provide much additional help in resolving the somewhat provocative query posed by Amandio et al: Is Hotair homeotic or homeopathic? Readers should reach their own conclusions based on the data [5,9,10], but our collective impression is that Hotair is not quite homeopathic, but it is also not a major determinant of developmental identity.

From the standpoint of mechanism, Hotair RNA was proposed to affect gene expression in trans at least in part by targeting PRC2 to genes. However, questions also remain about the potential biological significance of the Hotair-PRC2 association. This is in part because Polycomb target genes identified by Li et al. [5] in tail tip fibroblasts (by H3K27me ChIP-seq) did not significantly overlap with Hotair target genes identified by Amandio et al. in vivo [9]. Although several labs have observed and/or questioned how PRC2 interacts with Hotair RNA in vitro, an important challenge for the field is to understand what happens in vivo [2325].

One of the reasons research in the lncRNA field has been both fruitful and controversial is the ability of exquisitely sensitive genome-scale technologies to identify transcripts expressed at very low levels, which is part of a more general debate about how to distinguish transcriptional noise from transcription with a biological function [26,27]. Going forward, we believe the developmental genetics community will best be served by studying lncRNAs and epigenetic modulators that yield robust and highly penetrant whole-animal phenotypes, and that may act by a variety of different mechanisms [6,7]. Fortunately, there are many from which to choose [22,28] and evolutionarily informed strategies to guide that choice [29].

We also wish to comment that the specific controversy itself is not as important as the ways in which it has been addressed. The Chang group made all of their data publicly available and provided the mouse they generated to the Duboule group. This should be the norm across the scientific community, but there are frequent exceptions, and the willingness of the Chang group to share their data and resources should be noted and appreciated by all concerned. Furthermore, the authors, reviewers, and editors of the Amandio et al. manuscript recognized the importance of publishing (largely) negative data, which, in this case, is likely to have a substantial impact on the field. Overall, transparency, open access principles, and collegiality are the champions in the debate about Hotair.


  1. 1. Bernstein E, Allis CD (2005) RNA meets chromatin. Genes Dev 19: 1635–1655. pmid:16024654
  2. 2. Lee JT (2012) Epigenetic regulation by long noncoding RNAs. Science 338: 1435–1439. pmid:23239728
  3. 3. Ulitsky I, Bartel DP (2013) lincRNAs: genomics, evolution, and mechanisms. Cell 154: 26–46. pmid:23827673
  4. 4. Rinn JL, Kertesz M, Wang JK, Squazzo SL, Xu X, Brugmann SA et al. (2007) Functional demarcation of active and silent chromatin domains in human HOX loci by noncoding RNAs. Cell 129: 1311–1323. pmid:17604720
  5. 5. Li L, Liu B, Wapinski OL, Tsai MC, Qu K, Zhang J et al. (2013) Targeted disruption of Hotair leads to homeotic transformation and gene derepression. Cell Rep 5: 3–12. pmid:24075995
  6. 6. Arun G, Diermeier S, Akerman M, Chang KC, Wilkinson JE et al. (2016) Differentiation of mammary tumors and reduction in metastasis upon Malat1 lncRNA loss. Genes Dev 30: 34–51. pmid:26701265
  7. 7. Lee S, Kopp F, Chang TC, Sataluri A, Chen B et al. (2016) Noncoding RNA NORAD Regulates Genomic Stability by Sequestering PUMILIO Proteins. Cell 164: 69–80. pmid:26724866
  8. 8. Kmita M, Duboule D (2003) Organizing axes in time and space; 25 years of colinear tinkering. Science 301: 331–333. pmid:12869751
  9. 9. Amandio AR, Necsulea A, Joye E, Mascrez B, Duboule D (2016) Hotair is dispensable for mouse development. PLoS Genet 12(12): e1006232.
  10. 10. Li L, Helms JA, Chang HY (2016) Comment on “Hotair is dispensable for mouse development”. PLoS Genet 12(12): e1006406.
  11. 11. Mallo M, Alonso CR (2013) The regulation of Hox gene expression during animal development. Development 140: 3951–3963. pmid:24046316
  12. 12. Rinn JL, Bondre C, Gladstone HB, Brown PO, Chang HY (2006) Anatomic demarcation by positional variation in fibroblast gene expression programs. PLoS Genet 2: e119. pmid:16895450
  13. 13. Petruk S, Sedkov Y, Riley KM, Hodgson J, Schweisguth F, Hirose S et al. (2006) Transcription of bxd noncoding RNAs promoted by trithorax represses Ubx in cis by transcriptional interference. Cell 127: 1209–1221. pmid:17174895
  14. 14. Schorderet P, Duboule D (2011) Structural and functional differences in the long non-coding RNA hotair in mouse and human. PLoS Genet 7: e1002071. pmid:21637793
  15. 15. Flatt T (2005) The evolutionary genetics of canalization. Q Rev Biol 80: 287–316. pmid:16250465
  16. 16. Waddington CH (1942) Canalization of development and the inheritance of acquired characters. Nature 3811: 563–565.
  17. 17. Green EL (1941) Genetic and Non-Genetic Factors Which Influence the Type of the Skeleton in an Inbred Strain of Mice. Genetics 26: 192–222. pmid:17247002
  18. 18. McLaren A, Michie D (1955) Factors affecting vertebral variation in mice. 2. Further evidence on intra-strain variation. Journal of Embryology and Extracellular Morphology 3: 366–375.
  19. 19. Munro KM, Barnett SA (1969) Variation of the lumbar vertebrae of mice at two environmental temperatures. J Embryol Exp Morphol 21: 97–103. pmid:5765794
  20. 20. McLaren A, Michie D (1954) Are inbred strains suitable for bio-assay. Nature 173: 686–687. pmid:13165622
  21. 21. Gruneberg H (1954) Variation within inbred strains of mice. Nature 173: 674–676. pmid:13165620
  22. 22. Lai KM, Gong G, Atanasio A, Rojas J, Quispe J, Posca J et al. (2015) Diverse Phenotypes and Specific Transcription Patterns in Twenty Mouse Lines with Ablated LincRNAs. PLoS ONE 10: e0125522. pmid:25909911
  23. 23. Cifuentes-Rojas C, Hernandez AJ, Sarma K, Lee JT (2014) Regulatory interactions between RNA and polycomb repressive complex 2. Mol Cell 55: 171–185. pmid:24882207
  24. 24. Davidovich C, Zheng L, Goodrich KJ, Cech TR (2013) Promiscuous RNA binding by Polycomb repressive complex 2. Nat Struct Mol Biol 20: 1250–1257. pmid:24077223
  25. 25. Davidovich C, Wang X, Cifuentes-Rojas C, Goodrich KJ, Gooding AR, Lee JT et al. (2015) Toward a consensus on the binding specificity and promiscuity of PRC2 for RNA. Mol Cell 57: 552–558. pmid:25601759
  26. 26. Ponting CP, Oliver PL, Reik W (2009) Evolution and functions of long noncoding RNAs. Cell 136: 629–641. pmid:19239885
  27. 27. Struhl K (2007) Transcriptional noise and the fidelity of initiation by RNA polymerase II. Nat Struct Mol Biol 14: 103–105. pmid:17277804
  28. 28. Sauvageau M, Goff LA, Lodato S, Bonev B, Groff AF, Gerhardinger C et al. (2013) Multiple knockout mouse models reveal lincRNAs are required for life and brain development. Elife 2: e01749. pmid:24381249
  29. 29. Ulitsky I (2016) Evolution to the rescue: using comparative genomics to understand long non-coding RNAs. Nat Rev Genet 17: 601–614. pmid:27573374