Jump to content

Chromosome 11

From Wikipedia, the free encyclopedia
Chromosome 11
Human chromosome 11 pair after G-banding.
One is from mother, one is from father.
Chromosome 11 pair
in human male karyogram.
Features
Length (bp)135,127,769 bp
(CHM13)
No. of genes1,224 (CCDS)
TypeAutosome
Centromere positionSubmetacentric[1]
(53.4 Mbp[2])
Complete gene lists
CCDSGene list
HGNCGene list
UniProtGene list
NCBIGene list
External map viewers
EnsemblChromosome 11
EntrezChromosome 11
NCBIChromosome 11
UCSCChromosome 11
Full DNA sequences
RefSeqNC_000011 (FASTA)
GenBankCM000673 (FASTA)

Chromosome 11 is one of the 23 pairs of chromosomes in humans. Humans normally have two copies of this chromosome. Chromosome 11 spans about 135 million base pairs (the building material of DNA) and represents between 4 and 4.5 percent of the total DNA in cells. The shorter arm (p arm) is termed 11p while the longer arm (q arm) is 11q. At about 21.5 genes per megabase, chromosome 11 is one of the most gene-rich, and disease-rich, chromosomes in the human genome.

More than 40% of the 856 olfactory receptor genes in the human genome are located in 28 single-gene and multi-gene clusters along the chromosome.

Gene

[edit]

Number of genes

[edit]

The following are some of the gene count estimates of human chromosome 11. Because researchers use different approaches to genome annotation their predictions of the number of genes on each chromosome varies (for technical details, see gene prediction). Among various projects, the collaborative consensus coding sequence project (CCDS) takes an extremely conservative strategy. So CCDS's gene number prediction represents a lower bound on the total number of human protein-coding genes.[3]

Estimated by Protein-coding genes Non-coding RNA genes Pseudogenes Source Release date
CCDS 1,224 [4] 2016-09-08
HGNC 1,262 271 666 [5] 2017-05-12
Ensembl 1,301 1,060 811 [6] 2017-03-29
UniProt 1,327 [7] 2018-02-28
NCBI 1,314 860 839 [8][9][10] 2017-05-19

Gene list

[edit]

The following is a partial list of genes on human chromosome 11. For complete list, see the links in the infobox on the right.

  • ACAT1: acetyl-Coenzyme A acetyltransferase 1 (acetoacetyl Coenzyme A thiolase)
  • ACRV1: encoding protein Acrosomal protein SP-10
  • AKIP1: A kinase interacting protein 1
  • ALKBH3 encoding protein AlkB homolog 3, alpha-ketoglutaratedependent dioxygenase
  • AMOTL1: angiomotin-like protein 1
  • AMPD3: encoding enzyme AMP deaminase 3
  • API5: encoding protein Apoptosis inhibitor 5
  • APLNR: Apelin receptor (APJ receptor)
  • APOA4: apolipoprotein A-IV
  • ARCN1 encoding protein Archain 1
  • ART5: encoding protein Adp-ribosyltransferase 5
  • ASRGL1: encoding enzyme L-asparaginase
  • ATM: ataxia telangiectasia mutated (includes complementation groups A, C and D)
  • B3GNT1: encoding enzyme N-acetyllactosaminide beta-1,3-N-acetylglucosaminyltransferase
  • BATF2: encoding protein Basic leucine zipper transcription factor, ATF-like 2
  • BDNF: secretes BDNF, a member of the Neurotrophin family of proteins
  • C11orf1: encoding protein
  • C11orf16: encoding protein Uncharacterized protein C11orf16
  • C11orf49: encoding protein UPF0705 protein C11orf49
  • C11orf52 encoding protein C11orf52
  • C11orf53: encoding protein Chromosome 11 open reading frame 53
  • C11orf54: encoding protein Ester hydrolase C11orf54
  • C11orf80: encoding protein Chromosome 11 open reading frame 80
  • C11orf86: encoding protein Uncharacterized protein C11orf86
  • C1QTNF4 encoding protein C1q and tumor necrosis factor related protein 4
  • C1QTNF5: encoding protein C1q and tumor necrosis factor related protein 5
  • CAPRIN1: encoding protein, cell cycle associated protein 1
  • CCDC88B: encoding protein Coiled-coil domain containing 88b
  • CCDC90B: coiled coil domain containing 90B
  • CCL9: Chemokine (C-C motif) ligand 9
  • CD81: cluster of differentiation 81
  • CDHR5: cadherin related family member 5
  • CLDN25: encoding protein Claudin 25
  • COMMD9: COMM domain-containing protein 9
  • CPSF7: Cleavage and polyadenylation specificity factor subunit 7
  • CPT1A: carnitine palmitoyltransferase 1A (liver)
  • CREBZF encoding protein CREB/ATF bZIP transcription factor
  • DAK: Triokinase/FMN cyclase
  • DDI1: encoding protein DNA-damage inducible 1 homolog 1 (S. cerevisiae)
  • DGAT2 encoding protein Diacylglycerol O-acyltransferase 2
  • DHCR7: 7-dehydrocholesterol reductase
  • DKK3: Dickkopf-related protein 3
  • DPF2: Double PHD fingers 2
  • DRD4: Dopamine receptor D4 at 11p15.5
  • DSCAML1: encoding protein Down syndrome cell adhesion molecule like 1
  • EI24: Etoposide-induced protein 2.4 homolog
  • FAM118B: encoding protein Family with sequence similarity 118, member B
  • FAM76B: Family with sequence similarity 76 member B
  • FAR1: fatty acyl-coA reductase 1
  • FAT3: fat atypical cadherin 3
  • FHIP: FTS and Hook-interacting protein
  • FNBP4: Formin-binding protein 4
  • FOLR3: encoding protein Folate receptor gamma
  • GLB1L3: galactosidase, beta 1-like 3
  • GLYAT: Glycine-N-acyltransferase
  • GLYATL2 encoding protein Glycine-N-acyltransferase like 2
  • GPHA2: Glycoprotein hormone alpha-2
  • GYLTL1B: Glycosyltransferase-like protein LARGE2
  • HBB: hemoglobin, beta
  • HBBP1: encoding protein Hemoglobin, beta pseudogene 1
  • HIKESHI: chromosome 11, open reading frame 73
  • HMBS: hydroxymethylbilane VIIA
  • HRASLS3: adipose phospholipase A2
  • HTATIP2: HIV-1 Tat interactive protein 2
  • HYLS1: Hydroletalus (Finnish heritage disease) related gene
  • HYOU1: hypoxia upregulated protein 1
  • IFITM2 encoding protein Interferon induced transmembrane protein 2
  • IFT46: intraflagellar transport protein 46 homolog
  • IGF2-AS: encoding protein IGF2 antisense RNA
  • INS: insulin gene [11]
  • KDM2A: lysine demethylase 2A
  • KIAA1549L encoding protein KIAA1549-like
  • KRTAP5-6: encoding protein Keratin associated protein 5-6
  • LPXN: leupaxin
  • LRFN4 encoding protein Leucine rich repeat and fibronectin type III domain containing 4
  • MADD: MAP kinase-activating death domain protein
  • MEN1: Multiple endocrine neoplasia type 1
  • MIR210: encoding protein MicroRNA 210
  • MIRLET7A2: microRNA let-7a-2
  • MMP7: Matrix metalloproteinases (MMP family)
  • MOGAT2: monoacylglycerol O-acyltransferase 2
  • MTRNR2L8: encoding protein MT-RNR2-like 8
  • NADSYN1: NAD synthetase 1
  • NAP1L4: nucleosome assembly protein 1-like 4
  • NFRKB: nuclear factor related to kappa-B binding protein
  • NNMT: nicotinamide N-methyltransferase
  • NRGN: neurogranin
  • OR2AG2: encoding protein Olfactory receptor family 2 subfamily ag member 2
  • P53AIP1: p53-regulated apoptosis-inducing protein 1
  • PANO1: encoding protein PANO1
  • PAX6: paired box 6
  • PCNX3 encoding protein Pecanex homolog 3
  • PGA3 encoding protein Pepsinogen 3, group I (pepsinogen A)
  • PIWIL4 encoding protein Piwi like RNA-mediated gene silencing 4
  • PLET1: encoding protein Placenta expressed transcript 1
  • PRR5L: proline rich 5 like
  • PTPRCAP: protein tyrosine phosphatase receptor type C associated protein
  • PTS: 6-pyruvoyltetrahydropterin synthase
  • QSER1: glutamine serine rich protein 1
  • RAG1/RAG2: recombination activating genes
  • RASSF7: encoding protein Ras association domain family member 7
  • RASSF10: encoding protein Ras association domain family member 10
  • RELT: tumor necrosis factor recepteor
  • REXO2: RNA exonuclease 2
  • RNH1: ribonuclease inhibitor 1
  • RNU2-2: encoding protein RNA, U2 small nuclear 2
  • ROM1: retinal outer segment membrane protein 1
  • RPL27A: encoding protein 60S ribosomal protein L27a
  • RPL36A: encoding protein 60S ribosomal protein L36a
  • RSF1: remodeling and spacing factor 1
  • SAA1: serum amyloid A1
  • SAA2: serum amyloid A2
  • SAC3D1: SAC3 domain-containing protein 1
  • SART1: squamous cell carcinoma antigen recognized by T-cells 1
  • SBF2: SET binding factor 2
  • SCGB1D2: secretoglobin family 1D member 2
  • SESN3 encoding protein Sestrin 3
  • SIDT2 encoding protein SID1 transmembrane family member 2
  • SLC17A6: encoding protein Solute carrier family 17 (vesicular glutamate transporter), member 6
  • SLC25A22: encoding protein Solute carrier family 25 member 22
  • SMAP1: small acidic protein
  • SMCO4: encoding protein SMCO4
  • SMPD1: sphingomyelin phosphodiesterase 1, acid lysosomal (acid sphingomyelinase)
  • SNHG1: encoding protein Small nucleolar rna host gene 1
  • SPA17: sperm autoantigenic protein 17
  • SRPRA: Srp receptor alpha subunit
  • Suppression of tumorigenicity 2: encoding protein Suppression of tumorigenicity 2
  • TAF1D: TATA box binding protein associated factor RNA polymerase 1 subunit D
  • TALDO1 encoding protein Transaldolase 1
  • TBRG1: transforming growth factor beta regulator 1
  • TECTA: tectorin alpha (nonsyndromic deafness)
  • TH: tyrosine hydroxylase
  • THRSP: thyroid hormone inducible hepatic protein
  • THYN1: thymocyte nuclear protein 1
  • TIMM10: translocase of inner mitochondrial membrane 10
  • TIMM10B: Mitochondrial import inner membrane translocase subunit Tim9 B
  • TM7SF2: transmembrane 7 superfamily member 2
  • TMEM109: encoding protein Transmembrane protein 109
  • TMEM123: transmembrane protein 123
  • TMEM126B: transmembrane protein 126B
  • TMEM134: transmembrane protein 134
  • TMEM25: transmembrane protein 25
  • TMPRSS4: encoding protein Transmembrane protease, serine 4
  • TP53I11: tumor protein 53 inducible protein 11
  • TRAPPC4: trafficking protein particle complex subunit 4
  • TRIM34: encoding protein Tripartite motif containing 34
  • TRPT1: tRNA 2'-phosphotransferase 1
  • UNC93B1: Unc-93 homolog B1
  • UPK2: uroplakin-2
  • UQCC3: encoding protein Ubiquinol-cytochrome c reductase complex assembly factor 3
  • USH1C: Usher syndrome 1C (autosomal recessive, severe)
  • USP47: ubiquitin specific peptidase 47
  • UVRAG: UV radiation resistance associated
  • VPS26B: vacuolar protein sorting 26 homolog B
  • VSIG2: V-set and immunoglobulin domain containing 2
  • WT1: Wilms tumor protein
  • YIF1A: Yip1 interacting factor homolog A
  • ZFP91-CNTF
  • ZNF408: zinc finger protein 408

Diseases and disorders

[edit]

The following diseases and disorders are some of those related to genes on chromosome 11:

Cytogenetic band

[edit]
G-banding ideograms of human chromosome 11
G-banding ideogram of human chromosome 11 in resolution 850 bphs. Band length in this diagram is proportional to base-pair length. This type of ideogram is generally used in genome browsers (e.g. Ensembl, UCSC Genome Browser).
G-banding patterns of human chromosome 11 in three different resolutions (400,[13] 550[14] and 850[2]). Band length in this diagram is based on the ideograms from ISCN (2013).[15] This type of ideogram represents actual relative band length observed under a microscope at the different moments during the mitotic process.[16]
G-bands of human chromosome 11 in resolution 850 bphs[17]
Chr. Arm[18] Band[19] ISCN
start[20]
ISCN
stop[20]
Basepair
start
Basepair
stop
Stain[21] Density
11 p 15.5 0 230 1 2,800,000 gneg
11 p 15.4 230 461 2,800,001 11,700,000 gpos 50
11 p 15.3 461 745 11,700,001 13,800,000 gneg
11 p 15.2 745 935 13,800,001 16,900,000 gpos 50
11 p 15.1 935 1246 16,900,001 22,000,000 gneg
11 p 14.3 1246 1490 22,000,001 26,200,000 gpos 100
11 p 14.2 1490 1545 26,200,001 27,200,000 gneg
11 p 14.1 1545 1775 27,200,001 31,000,000 gpos 75
11 p 13 1775 2114 31,000,001 36,400,000 gneg
11 p 12 2114 2357 36,400,001 43,400,000 gpos 100
11 p 11.2 2357 2655 43,400,001 48,800,000 gneg
11 p 11.12 2655 2872 48,800,001 51,000,000 gpos 75
11 p 11.11 2872 3035 51,000,001 53,400,000 acen
11 q 11 3035 3197 53,400,001 55,800,000 acen
11 q 12.1 3197 3414 55,800,001 60,100,000 gpos 75
11 q 12.2 3414 3550 60,100,001 61,900,000 gneg
11 q 12.3 3550 3685 61,900,001 63,600,000 gpos 25
11 q 13.1 3685 4037 63,600,001 66,100,000 gneg
11 q 13.2 4037 4186 66,100,001 68,700,000 gpos 25
11 q 13.3 4186 4512 68,700,001 70,500,000 gneg
11 q 13.4 4512 4688 70,500,001 75,500,000 gpos 50
11 q 13.5 4688 4877 75,500,001 77,400,000 gneg
11 q 14.1 4877 5148 77,400,001 85,900,000 gpos 100
11 q 14.2 5148 5257 85,900,001 88,600,000 gneg
11 q 14.3 5257 5474 88,600,001 93,000,000 gpos 100
11 q 21 5474 5690 93,000,001 97,400,000 gneg
11 q 22.1 5690 5934 97,400,001 102,300,000 gpos 100
11 q 22.2 5934 6070 102,300,001 103,000,000 gneg
11 q 22.3 6070 6300 103,000,001 110,600,000 gpos 100
11 q 23.1 6300 6503 110,600,001 112,700,000 gneg
11 q 23.2 6503 6693 112,700,001 114,600,000 gpos 50
11 q 23.3 6693 7167 114,600,001 121,300,000 gneg
11 q 24.1 7167 7316 121,300,001 124,000,000 gpos 50
11 q 24.2 7316 7533 124,000,001 127,900,000 gneg
11 q 24.3 7533 7695 127,900,001 130,900,000 gpos 50
11 q 25 7695 7980 130,900,001 135,086,622 gneg

References

[edit]
  1. ^ Tom Strachan; Andrew Read (2 April 2010). Human Molecular Genetics. Garland Science. p. 45. ISBN 978-1-136-84407-2.
  2. ^ a b Genome Decoration Page, NCBI. Ideogram data for Homo sapience (850 bphs, Assembly GRCh38.p3). Last update 2014-06-03. Retrieved 2017-04-26.
  3. ^ Pertea M, Salzberg SL (2010). "Between a chicken and a grape: estimating the number of human genes". Genome Biol. 11 (5): 206. doi:10.1186/gb-2010-11-5-206. PMC 2898077. PMID 20441615.
  4. ^ "Search results - 11[CHR] AND "Homo sapiens"[Organism] AND ("has ccds"[Properties] AND alive[prop]) - Gene". NCBI. CCDS Release 20 for Homo sapiens. 2016-09-08. Retrieved 2017-05-28.
  5. ^ "Statistics & Downloads for chromosome 11". HUGO Gene Nomenclature Committee. 2017-05-12. Retrieved 2017-05-19.
  6. ^ "Chromosome 11: Chromosome summary - Homo sapiens". Ensembl Release 88. 2017-03-29. Retrieved 2017-05-19.
  7. ^ "Human chromosome 11: entries, gene names and cross-references to MIM". UniProt. 2018-02-28. Retrieved 2018-03-16.
  8. ^ "Search results - 11[CHR] AND "Homo sapiens"[Organism] AND ("genetype protein coding"[Properties] AND alive[prop]) - Gene". NCBI. 2017-05-19. Retrieved 2017-05-20.
  9. ^ "Search results - 11[CHR] AND "Homo sapiens"[Organism] AND ( ("genetype miscrna"[Properties] OR "genetype ncrna"[Properties] OR "genetype rrna"[Properties] OR "genetype trna"[Properties] OR "genetype scrna"[Properties] OR "genetype snrna"[Properties] OR "genetype snorna"[Properties]) NOT "genetype protein coding"[Properties] AND alive[prop]) - Gene". NCBI. 2017-05-19. Retrieved 2017-05-20.
  10. ^ "Search results - 11[CHR] AND "Homo sapiens"[Organism] AND ("genetype pseudo"[Properties] AND alive[prop]) - Gene". NCBI. 2017-05-19. Retrieved 2017-05-20.
  11. ^ INS – insulin – Genetics Home Reference
  12. ^ "Human Genome Project Information Site Has Been Updated". Archived from the original on 2012-07-10. Retrieved 2009-01-04.
  13. ^ Genome Decoration Page, NCBI. Ideogram data for Homo sapience (400 bphs, Assembly GRCh38.p3). Last update 2014-03-04. Retrieved 2017-04-26.
  14. ^ Genome Decoration Page, NCBI. Ideogram data for Homo sapience (550 bphs, Assembly GRCh38.p3). Last update 2015-08-11. Retrieved 2017-04-26.
  15. ^ International Standing Committee on Human Cytogenetic Nomenclature (2013). ISCN 2013: An International System for Human Cytogenetic Nomenclature (2013). Karger Medical and Scientific Publishers. ISBN 978-3-318-02253-7.
  16. ^ Sethakulvichai, W.; Manitpornsut, S.; Wiboonrat, M.; Lilakiatsakun, W.; Assawamakin, A.; Tongsima, S. (2012). "Estimation of band level resolutions of human chromosome images". 2012 Ninth International Conference on Computer Science and Software Engineering (JCSSE). pp. 276–282. doi:10.1109/JCSSE.2012.6261965. ISBN 978-1-4673-1921-8. S2CID 16666470.
  17. ^ Genome Decoration Page, NCBI. Ideogram data for Homo sapience (850 bphs, Assembly GRCh38.p3). Last update 2014-06-03. Retrieved 2017-04-26.
  18. ^ "p": Short arm; "q": Long arm.
  19. ^ For cytogenetic banding nomenclature, see article locus.
  20. ^ a b These values (ISCN start/stop) are based on the length of bands/ideograms from the ISCN book, An International System for Human Cytogenetic Nomenclature (2013). Arbitrary unit.
  21. ^ gpos: Region which is positively stained by G banding, generally AT-rich and gene poor; gneg: Region which is negatively stained by G banding, generally CG-rich and gene rich; acen Centromere. var: Variable region; stalk: Stalk.
[edit]
  • National Institutes of Health. "Chromosome 11". Genetics Home Reference. Archived from the original on August 3, 2004. Retrieved 2017-05-06.
  • "Chromosome 11". Human Genome Project Information Archive 1990–2003. Retrieved 2017-05-06.