pmcid stringlengths 6 6 | title stringlengths 9 374 | abstract stringlengths 2 4.62k ⌀ | fulltext stringlengths 167 106k | file_path stringlengths 64 64 |
|---|---|---|---|---|
544934 | WWW design code – a new tool for colour estimation in animal studies | Background The colour of animals' skin, fur, feathers or cuticula has been estimated in a large number of studies. The methods used to do so are diverse, with some being costly and not available to all researchers. In a study to measure plumage colour in a bird species, a new method of creating a colour chart was developed. While colour-charts have their own limitations, these can be minimised when they have the following properties: 1) being readily available to the majority of biologists, 2) containing a large array of colours to allow accurate recording and differentiation of subtle colour differences, 3) low cost, 4) adhering to a world-wide standard, and 5) being available in both hard-copy and digital formats to allow for various analytical methods. The method described below satisfies all of these requirements. Results Colour charts estimated to fit the range of the species' plumage colours were created on the computer screen using web software that allowed for HTML-coding (in this case Dreamweaver™). The charts were adjusted using feathers from dead specimens until a satisfying range of darker and lighter colours were found. The resulting chart was printed out and was successfully used in the field to determine the plumage colour of hand-held birds. Conclusion Access to a computer and printer, and the software to enable the creation of a chart, is within the reach of the vast majority of biologists. The numbers of colours that can be generated should suit most studies, with the advantage of the method being that the chart can be individually tailored to the species under study. HTML colour coding is a worldwide standard, thus the colours used in studies can be described in the methods section of journal articles using the six-digit alphanumeric code. We believe this method is very useful as a low-tech method for future estimation of individual colour. | Background Animal studies from diverse fields – such as morphology, physiology, behaviour, population dynamics, genetics, ageing and sexing – often require the estimation or classification of colour in skin, fur, feathers or cuticula. To make meaningful comparisons within and between studies, and effectively communicate these findings, a standardised method of assigning a colouration is required. Colour can be categorised, scored and ranked without recourse of any aid other than the observer's eye, as has been successfully demonstrated in many studies [ 1 - 4 ]. Aids to indexing and labelling colours have traditionally compared animals to published colour-standard cards, charts or books. These were often generated for other purposes, such as paint making or characterising soil types [ 5 - 8 ]. Such charts have been employed since the early 1800's, with Charles Darwin using Werner's colour charts [ 9 ] on the Beagle expedition [ 10 ]. Newer electronic measuring tools, using a reflectance spectrophotometer to give a high precision estimate of hue, saturation and brightness of the colour [ 11 - 14 ], have been recommended as a reliable and objective way of acquiring detailed data on different aspects of colour [ 15 ]. Lately, photo-processing software (e.g. Adobe Photoshop ® – Adobe Systems Inc., San Jose, CA, USA) has been used in estimating colour brightness, saturation and hue from digitised photos [ 11 , 16 ]. The benefit of using highly technical methods is the ability to gain very detailed and precise information on different aspects of colour that may be impossible to detect using the human eye. However, in many studies such a high level of precision is not required, and the equipment may be impractical in the field or beyond the budget of the research group. Thus despite the option of high-tech measurement methods, visual comparisons to standardised colour charts are still practical and valuable for many field biologists. While colour-charts have their own limitations, these are minimised when they have the following properties: 1) availability to the majority of biologists, 2) large array of colours to allow accurate recording and differentiation of subtle colour differences, 3) low cost, 4) adherence to a world-wide standard, and 5) availability in both hard-copy and digital formats to allow for various analytical methods. Below we describe a method that satisfies all of these requirements. Results and Discussion A starting point for development of this new method was a study on plumage colour in an endemic New Zealand passerine – North Island robin ( Petroica longipes ) – where there was a need for a reliable, species-specific and cost-efficient method to estimate colour (Fig. 1 ; Å Berggren unpublished data). The birds were to be caught and handled in the field, but could not be moved and no samples from the plumage were to be taken. Hence, it was decided to create a specific colour chart, unique for the colours to be compared in this study, and easy to use under fieldwork conditions. It was decided that a computer-generated colour-chart would ideally suit the purposes of the study. Figure 1 The making of a HTML coded colour chart for estimation of animal colour in the field. First the colours relevant for the study are estimated from samples from the species of interest (in this case the New Zealand robin ( Petroica longipes )) (a). From the sample, a set of colours is coded using HTML coding (b), which has a specific code for every colour. In this case the software Dreamweaver™ was used to create nine different shades ranging from light grey (888888) to black (000000) (c). When the chart is made it is printed out and can be enclosed within a plastic covering for protection. This procedure worked well for colour estimation of the species in focus, with individuals' colours being easy to index under field conditions (d). The light emitted from a computer monitor is composed of a particular combination of red, green, and blue light. The proportion of each of these components in the visual colour spectrum can be expressed as a number unique for each specific colour or hue. These colours are coded for using the HTML (Hypertext Markup Language) computer language. This system of colour coding was developed for the Netscape ® web-browser (Netscape Communication Corp.) and has since become the industry standard [ 17 ]. Today, colour coding is one of the most important features in web page design when creating informative and graphically appealing sites [ 18 ]. Each HTML colour-code specifies the composite of a colour with a six-digit alphanumeric code where the first two digits represent the amount of red, the middle two the amount of green, and the last two the amount of blue. Each character may be represented in one of 16 ways (0 – 9 and A – F), creating a vast array of potential colours [ 17 ]. The actual number of colours that can be produced for viewing on the screen is limited by the computer, ranging from 256 colours in older computers to 16.7 million in newer models. These colours can be generated in web browser software such as Netscape ® , by using the composer feature, or through web-design programmes (e.g. Dreamweaver™ – Macromedia Inc. or FrontPage © – Microsoft Corp.). The colours can be displayed on the computer screen and fine-tuned by adjusting the HTML code. For example, between the colours "0000FF" for blue, "00008B" for dark blue and "00C78C" for turquoise blue, a large number of other blues can be created and viewed. When a suitable set of colours have been decided upon, they can be printed out as they present themselves on the screen and saved for future use. The focus species of the plumage study, the North Island robin, has a brown-grey-black plumage (Fig. 1 ) [ 19 ]. Using HTML coding, a colour chart was developed to match the natural variance in the species' brown to black plumage. To ensure the colour range of the feathers was accurately represented in the chart, feathers from a dead specimen were used and compared to the computer generated colours. From this point, an equal number of brighter and darker colour gradations were created, centring on the colour of the feather specimens. The HTML-coding numbering from 0 to 8 (000000 – 888888) resulted in nine shades from black to light grey (Fig. 1 ). This allowed a progression in equal steps from lighter to darker to be displayed sequentially on a printout. This made it possible to hold the bird next to the chart and move it until a colour match was made (Fig. 1 ). The technique worked well and it was possible to get an accurate colour ranking of the darkness of the plumage for the captured 32 birds (Å Berggren unpublished). Conclusions The method of colour-chart creation utilising HTML code satisfies the criteria listed above. Access to a computer and printer, and the software to enable the creation of a chart, is within the reach of the vast majority of biologists. The numbers of colours (and patterns) that can be generated should suit most studies, with the advantage of the method being that the chart can be individually tailored to the species under study. HTML colour coding is a worldwide standard, thus the colours used in studies can be described in the methods section of journal articles using the six-digit hexadecimal code. Comparisons are not limited to a printout of a colour chart, digital images can also be compared and their colours scored using this method. Drawbacks may include identifying the right colours for the chart to accurately match the animals as encountered in the field, a problem equivalent to other printed colour charts. It is also possible that when printing, the printed colours differ from the colour range as displayed on the computer screen. With some people still using older computers, which are not able to display colours coded in newer machines, there is a risk that there is a discrepancy between computers in the colour presented on the screen. This may be a problem when the aim is to compare specific colours between different studies, but not an issue within studies. Fortunately, this problem will decrease with more computer system being able to display the full range of 16.7 million colours. When using the colour-charts in the field, the usual care of indexing individuals under the same lighting conditions should be taken [ 15 ]. As the technique is easy to refine and adjust to the requirements needed for the species, we believe it is very useful as a low-tech method for future estimation of individual colour. We encourage other researchers and field workers to try the method in future colour studies. Authors' contributions JM came up with the initial idea of using web designer tools for creating colour charts. ÅB developed the colour charts using HTML coding and used them in the field as a research tool. ÅB and JM wrote the manuscript, with ÅB doing the major part. All authors read and approved the manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544934.xml |
539049 | Containing the Threat—Don't Forget Ebola | In 2000, Uganda experienced the largest outbreak of Ebola fever ever described. What can we learn from the Ugandan experience to help us prepare for future outbreaks? | On 8 October 2000, the acting district director of health services for the Gulu district in northwestern Uganda received two concurrent reports of an unusual illness with high mortality, occurring in the community and at a local hospital. One report attributed the illness to poisoning at a funeral at a remote village in the far north of the district. At the same time, the medical superintendent of the hospital also reported to the health authorities that he was experiencing a cluster of cases of critically ill patients, and that there had been several deaths, including some nurses. These events heralded what was to become the largest outbreak of Ebola fever so far described, involving 425 cases, of whom 224 died. The development of the epidemic and the measures taken to try and control it have recently been reported by Lamunu and her colleagues [1] . Their report underlines the challenges faced when dealing with such highly contagious and highly virulent infections. (At the request of PLoS Medicine , Lamunu et al. have made a full-text version of their report available on the World Health Organization Web site [2] .) Ebola Virus Ebola virus is a member of the family Filoviridae, which consists of two distinctive species, Marburg and Ebola, both of which cause severe and often fatal haemorrhagic disease in humans and monkeys. The viruses have a distinctive filamentous morphology under the electron microscope and a genome that consists of a nonsegmented, negative-stranded RNA approximately 19 kb in length. Ebola testing (Photo: Public Health Image Library, Centers for Disease Control and Prevention) Three distinct subtypes (genotypes) of Ebola have been described that are pathogenic for humans: Ebola-Zaire, Ebola-Sudan, and Ebola–Côte d'Ivoire. A fourth type, Ebola-Reston, affects only primates but has been identified in animal facilities in the United States, Italy, and the Philippines. The Illness Ebola is transmitted person to person by direct contact with infected body fluids, or by direct inoculation via contaminated instruments such as needles or razors. The incubation period of Ebola haemorrhagic fever is usually between four and 21 days. The illness is characterised by an acute onset of fever, malaise, myalgia, severe frontal headache, and pharyngitis. One of the great difficulties in making the diagnosis is that these symptoms are typical of many acute infective syndromes that occur in Ebola-endemic areas. As the disease progresses patients develop a maculopapular rash, typically at about six days, followed by vomiting and bloody diarrhoea, with uncontrollable haemorrhage from needle sites and body orifices. Death is from shock secondary to blood loss. Treatment is largely supportive, although a recent study has reported promising results with an inhibitor of tissue factor, which may help control the bleeding diathesis [3] . The Ugandan Outbreak Lamunu et al. describe how initial identification of the outbreak was delayed: six weeks elapsed before the Ugandan Ministry of Health was notified. There were several reasons for this delay. In part it could be explained by a weak surveillance system, especially at the local and regional levels. But also, the nonspecific nature of the symptoms meant that the initial, sporadic cases were frequently attributed to malaria or typhoid, and patients turned to local healers for help. Important, too, was the fact that this was the first outbreak of viral haemorrhagic fever in Uganda, and lack of familiarity with the disease caused further delays. It was only when clusters of cases became apparent that wider public health measures were instituted, and the outbreak started to come under control. In this phase, too, there were important lessons to learn. The initial identification of the disease as due to Ebola virus was made in the World Health Organization laboratories in South Africa, but soon thereafter a field laboratory was established, and this proved invaluable in guiding both case management and surveillance activities. Early involvement of specialised agencies, including the Global Outbreak and Response Network of the World Health Organization, was essential. Disseminating up-to-date information through the media, and the local communities, was important in getting the population “on side”. Lessons from the Outbreak The 2000 outbreak in Uganda was the last large outbreak, but other, smaller outbreaks continue to occur. During 2004 alone there have been two further epidemics: in January there were 35 cases in the Congo, with 29 deaths, and in August a smaller outbreak in the Sudan infected 17 patients, of whom seven died [ 4 , 5 ]. In each of these cases the epidemic was brought under control relatively quickly, and the infection was largely localised to the immediately surrounding area. However, the lessons of the Uganda outbreak have obvious resonance with many of the recent concerns that have been raised about the global spread of infectious diseases, be they naturally acquired or related to potential biowarfare. By and large, once an outbreak has been recognised by the public health authorities there are well-tried processes and procedures that come into play that serve to contain further spread of the infection and limit additional cases of the disease. This was shown spectacularly in the case of the SARS outbreak, in which not only was the disease controlled but the novel causative agent was identified, both within a few months. But as Lamunu and colleagues make clear, the most difficult aspect of the outbreak control is the initial recognition of the disease: diagnosis depends on the astute health-care worker who notices an unusual clinical picture, or more usually, an unexpected cluster of cases. Although the viral haemorrhagic fevers have until now been largely confined to their epidemic foci in Africa, cases will continue to occur from time to time in travellers, in whom diagnosis may be delayed. The key lessons from the Gulu outbreak are the extremely high case mortality of Ebola and the importance of instituting rigorous procedures to control cross-infection. These lessons are crucial both for communities in Africa, where public health infrastructures are often suboptimal, and in developed countries, where the infrastructure is sophisticated but can only be deployed once the disease is recognised. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539049.xml |
544883 | The WRKY transcription factor superfamily: its origin in eukaryotes and expansion in plants | Background WRKY proteins are newly identified transcription factors involved in many plant processes including plant responses to biotic and abiotic stresses. To date, genes encoding WRKY proteins have been identified only from plants. Comprehensive search for WRKY genes in non-plant organisms and phylogenetic analysis would provide invaluable information about the origin and expansion of the WRKY family. Results We searched all publicly available sequence data for WRKY genes. A single copy of the WRKY gene encoding two WRKY domains was identified from Giardia lamblia , a primitive eukaryote, Dictyostelium discoideum , a slime mold closely related to the lineage of animals and fungi, and the green alga Chlamydomonas reinhardtii , an early branching of plants. This ancestral WRKY gene seems to have duplicated many times during the evolution of plants, resulting in a large family in evolutionarily advanced flowering plants. In rice, the WRKY gene family consists of over 100 members. Analyses suggest that the C-terminal domain of the two-WRKY-domain encoding gene appears to be the ancestor of the single-WRKY-domain encoding genes, and that the WRKY domains may be phylogenetically classified into five groups. We propose a model to explain the WRKY family's origin in eukaryotes and expansion in plants. Conclusions WRKY genes seem to have originated in early eukaryotes and greatly expanded in plants. The elucidation of the evolution and duplicative expansion of the WRKY genes should provide valuable information on their functions. | Background Transcriptional control is a major mechanism whereby a cell or organism regulates its gene expression. Sequence-specific DNA-binding transcription regulators, one class of transcription factors [ 1 ], play an essential role in modulating the rate of transcription of specific target genes. In this way, they direct the temporal and spatial expressions necessary for normal development and proper response to physiological or environmental stimuli. Comparative genome analysis reveals that genes for transcription regulators are abundantly present in plant and animal genomes, and the evolution and diversity of eukaryotes seem to be related to the expansion of lineage-specific transcription regulator families [ 2 ]. WRKY proteins are recently identified transcriptional regulators comprising a large gene family [ 3 ]. The first cDNA encoding a WRKY protein, SPF1, was cloned from sweet potato ( Ipomoea batatas ) [ 4 ]. Numerous genes for WRKY proteins have since been experimentally identified from more than 10 other plant species, including Arabidopsis thaliana [ 5 , 6 ], wild oats ( Avena fatua ) [ 7 ], orchardgrass ( Dactylis glomerata ) [ 8 ], barley ( Hordeum vulgare ) [ 9 ], tobacco ( Nicotiana tabacum ) [ 10 - 13 ], chamomile( Matricaria chamomilla ) [ 14 ], rice ( Oryza sativa ) [ 9 , 15 ], parsley ( Petroselinum crispum ) [ 16 , 17 ], a desert legume ( Retama raetam ) [ 18 ], sugarcane ( Saccharum hybrid cultivar) [ 19 ], bittersweet nightshade ( Solanum dulcamara ) [ 20 ], potato ( Solanum tuberosum ) [ 21 , 22 ], and wheat ( Triticum aestivum ) [ 9 ]. In addition, over 70 WRKY genes were identified in the Arabidopsis genome by sequence similarity comparisons [ 2 , 23 ]. To date, WRKY genes have not been cloned from species other than plants. The absence of WRKY homologues in the genomes of animals ( Caenorhabditis elegans and Drosophila melanogaster ) and yeast ( Saccharomyces cerevisiae ) [ 2 ] leads to the suggestion that WRKY transcription regulators are restricted to the plant kingdom [ 2 , 3 ]. As genome sequence data for species representing several major eukaryotic lineages are now available, we can re-examine whether WRKY genes are plant-specific or have ancestors predating the appearance of plants. The WRKY family proteins contain one or two highly conserved WRKY domains characterized by the hallmark heptapeptide WRKYGQK and a zinc-finger structure distinct from other known zinc-finger motifs [ 3 ]. To regulate gene expression, the WRKY domain binds to the W box in the promoter of the target gene to modulate transcription [ 5 , 7 , 16 , 24 ]. In addition to the W box, a recent study indicates that the WRKY domain can also bind to SURE, a sugar responsive cis element, as a transcription activator [ 9 ]. In plants, many WRKY proteins are involved in the defense against attack from pathogenic bacteria [ 6 , 22 , 23 , 25 - 27 ], fungi [ 26 ], viruses [ 12 , 26 , 28 ], and oomycetes [ 21 , 26 , 29 ]. Further, WRKY genes are implicated in responses to the abiotic stresses of wounding [ 11 , 30 ], the combination of drought and heat [ 31 ], and cold [ 18 , 20 ]. It is also evident that some members of the family may play important regulatory roles in morphogenesis of trichomes [ 32 ] and embryos [ 8 ], senescence [ 26 , 33 - 35 ], dormancy [ 18 ], plant growth [ 27 ], and metabolic pathways [ 7 , 9 , 32 , 36 ]. Based on the number of WRKY domains and the pattern of the zinc-finger motif, Eulgem et al. [ 3 ] classified members of the WRKY superfamily from the Arabidopsis genome into three groups. Members of Group 1 typically contain two WRKY domains, while most proteins with one WRKY domain belong to Group 2. Group 3 proteins also have a single WRKY domain, but the pattern of the zinc-finger motif is unique. Eulgem et al. [ 3 ] further divided Group 2 into five subgroups, according to the phylogenetic analysis of the WRKY domains. Given the large family of WRKY genes with divergent regulatory functions in important plant processes, it would be desirable to understand the evolutionary origin and gene duplications leading to the multi-member WRKY family. The clarification of the phylogenetic relationships among WRKY genes in model plants will also assist understanding of the functions of these genes in important crops. We have comprehensively searched all currently available sequence data for the existence of WRKY genes outside the plant kingdom. Homologues of WRKY genes are found from two eukaryotic species: Giardia lamblia , a primitive protozoan, and Dictyostelium discoideum , a slime mold. The data indicate an early origin of WRKY genes in eukaryota and tremendous gene amplifications in the plant lineage. We then cataloged the WRKY genes from the rice genome and compared them with Arabidopsis WRKY genes. We also identified WRKY genes from expressed sequence tags (ESTs) and EST-assembled sequence contigs from nineteen plant species. The result suggests that WRKY gene duplication events correlate with the increasing structural and functional complexities in land plants. We propose a model for the evolution of WRKY genes. Results WRKY genes in non-plant eukaryotes We searched for WRKY genes in two comprehensive datasets, GenBank's non-redundant (nr) and dbEST of all species. Together these datasets contain over 13 million sequence records from more than 110,000 organisms [ 37 ]. Homologues of WRKY proteins are not found in the superkingdoms of archaea and eubacteria. In eukaryotes, no WRKY genes are identified from the lineages of fungi and animals. Interestingly, two WRKY homologues were identified from non-plant eukaryotic species, and both have two WRKY domains [see Additional files 1 and 2 ]. The first protein (GenBank accession: EAA40901) is encoded by an intronless gene in the draft genome sequence of Giardia lamblia [ 38 ]. The unicellular protist Giardia is one of the most primitive organisms that represent the earliest branching among extant eukaryotes [ 39 , 40 ]. The second (accession AAO52331) is encoded by the genomic sequence of chromosome 2 of the slime mold Dictyostelium discoideum [ 41 ]. The genomic sequence for the WRKY domains were assembled from sequences generated from three libraries prepared by two groups [ 42 ], indicating that it is not from sequence contamination. The gene contains an intron, which interrupts the coding region between the two WRKY domains. For this species, about 150,000 EST sequences are currently available in GenBank. One EST (accession AU033476) aligns to the WRKY gene, indicating that the gene is expressed. D. discoideum belongs to the Mycetozoa, a lineage more closely related to animals and fungi than to green plants [ 41 , 43 ]. A WRKY gene in a green alga Chlamydomonas reinhardtii is a unicellular green alga with a cell wall. It also has chloroplasts for photosynthesis. The evolutionary position of the species is located before the divergence of land plants [ 44 , 45 ]. The release 1.0 of its genome sequence has approximately 9 × whole genome shotgun coverage [ 46 ]. Since the gene annotation for the release is still at a preliminary stage, we predicted WRKY genes from the genome sequence (see Methods). The sequence similarity search between the genome sequence and Pfam's WRKY domain sequences indicated that the sequence 'Scaffold_1387' may encode WRKY domains. This sequence was then used for further WRKY domain and gene predictions. Despite minor differences in the gene structure prediction, both gene prediction programs FGENESH and GENSCAN agree on the major features of the protein, including the presence of two WRKY domains [see Additional files 1 and 2 ]. Moreover, the predicted peptide sequence of the WRKY domains is identical among all the gene and domain predictions. Sequence alignment by blastn indicates that six ESTs are from the predicted coding regions of the gene; the GenBank accessions for these ESTs are BI727288, AW772895, BM000804, BG846749, BE121978 and BQ821537. A catalog of WRKY genes in rice Rice, one of the most important crops for world agriculture, is recognized as a model monocot for the study of cereal crop genomes. A comprehensive catalog of rice WRKY genes would provide a basis for investigating the evolutionary patterns of the gene family and for transferring knowledge of the functions of these transcription factors from Arabidopsis to rice and from rice to other cereal crops. We identified the members of the WRKY family in rice (Japonica variety) from its published genome sequence [ 47 ]. The WRKY gene identification procedure employed in this study (see Methods) was first tested with the Arabidopsis genome sequence. The procedure successfully identified all reported Arabidopsis WRKY genes [ 3 , 23 ]. The rice genome seems to encode 109 WRKY proteins, four of which have incomplete WRKY domains. The remaining 105 proteins with complete WRKY domains, listed in Additional file 3 , were used for further analysis. The multiple sequence alignment of WRKY domains from rice, Arabidopsis, the green alga, the slime mold and Giardia lamblia , and the conserved WRKY domain patterns can be found in Additional file 2 . Some rice genes encode identical WRKY domains. For example, OsWRKY34 and OsWRKY57 share identical amino acid sequences in the WRKY domains, but the nucleotide sequences for the domains are not identical and they are located in different chromosomes (1 and 4, respectively), indicating that they are distinct genes. Similarly, OsWRKY8 located in Chromosome 6 and OsWRKY76 located in Chromosome 2 also represent two genes. The following genes in parenthesis share the identical WRKY domains and have a high identity of the corresponding coding nucleotide sequences: (OsWRKY9, 101), (12, 98 and 99), (21, 97), (29, 96), (39, 105), (51, 103), (73, 104), (80, 102), and (82, 100). These highly similar genes may represent newly duplicated paralogues. The 105 genes are unevenly distributed in the 12 chromosomes, ranging from 25 genes (the highest number) in Chromosome 1 to two genes (the lowest) in Chromosome 10. Sequence alignment indicates that 60 WRKY genes have one or more matched rice ESTs from the dbEST database (data not shown). Out of the 105 proteins, 13 have two WRKY domains. We assigned the WRKY domains into subfamilies using phylogenetic analysis with already classified AtWRKY genes from A. thaliana [ 3 ] as the reference. Eleven proteins with two WRKY domains are assigned to Group 1 because their C-terminal domains belong to this group. Since the N- and C-terminal domains form distinct clusters, we designated the two domains as 1N and 1C, respectively. Six proteins with a single domain also belonged to Group 1. While OsWRKY15, 16, 73 and 104 have a single domain homologous to Group 1N, OsWRKY13 and 91 contain a single Group 1C domain. Interestingly, both N- and C-domains of the other two double-domain-containing proteins (OsWRKY66 and 67) are always clustered with Group 3 domains. Thirty-five single WRKY domain proteins are also assigned to this group. All together, there are 39 domains or 37 proteins in Group 3. We assigned 49 proteins to three new groups, Group 2_a + 2_b (13), Group 2_c (18), and Group 2_d + 2_e (18). These new groups are reorganized from the five subgroups IIa through IIe in Eulgem et al. [ 3 ] (see details of the classification in Discussion). Domains of OsWRKY 25 and 95 cannot be consistently classified and therefore remain unassigned [see Additional file 3 ]. Interestingly, several variant patterns of the WRKY domains exist in the rice WRKY proteins. Although the WRKYGQK peptide is highly conserved, nine variants with one or two amino acids substituted are observed in 19 domains, most of which belong to Groups 3 and 2_c (Table 1 ). While WRKYGEK and WRKYGKK are two common variants shared by seven (all in Group 3) and five (all in Group 2_c) domains, respectively, each of the other seven different heptapeptides occurs in only one protein. The WRKY domains also contain patterns of zinc-finger motifs that have not been reported in the literature (Table 1 ). No variants are found in domains of Groups 1C and 2_a + 2_b. The WRKY genes encoding the variant domain patterns might be functional, because 10 genes with a total of seven heptapeptide variants and two zinc-finger motif variants have sequenced ESTs, although the DNA binding capacity may be reduced [ 48 ]. Furthermore, ESTs have been sequenced from the gene regions for the variants of WRKYGEK, WRKYGKK, WKKYGQK and C_X6_C_X28_H_X1_C, indicating that these patterns are not artifacts of the gene prediction (Table 1 ). Table 1 Variants of the conserved WRKYGQK peptide and zinc-finger motifs in rice WRKY domains Pattern Domain Group Available ESTs ID a Encoding the domain Variants of WRKYGQK WRKYG E K OsWRKY7 3 OsWRKY8 3 CA755335 Yes OsWRKY65 3 OsWRKY72 3 CF282152, CF330819, CF303772, CF282153, CF330818, CF305084, CF328161 Yes OsWRKY76 3 CA755335 OsWRKY77 3 Yes OsWRKY94 3 WRKYG K K OsWRKY20 2_c OsWRKY27 2_c OsWRKY36 2_c D43156 No OsWRKY46 2_c TC154521, AU093050 No OsWRKY63 2_c TC143003, BE230596, BM419201 Yes WR IC GQK OsWRKY15 1N WR MC GQK OsWRKY16 1N W K KYGQK OsWRKY25 unassigned AU162739 Yes W I KYGQK OsWRKY55 3 W KR YGQK OsWRKY66C 3 AW155482 No W S KY E QK OsWRKY67N 3 CA760141 No WRKY SE K OsWRKY92 3 Variants of zinc-finger motifs C_X5_C_X25_H_ X2 _C OsWRKY6 2_d + 2_e C_ X8 _C_X25_H_X1_C OsWRKY67N 3 CA760141 No C_ X6 _C_X28_H_X1_C OsWRKY68 3 TC103502 Yes a TIGR's TCs or GenBank's accessions. Survey of WRKY genes in land plants Since the genomes of rice and Arabidopsis have numerous WRKY genes whereas the green alga may have only a single copy, it would be interesting to investigate the gene duplication events of WRKY family during the course of evolution from unicellular plant organisms to flowering plants and the relationship between expansion of the WRKY family and the increased structural and functional complexities of the higher plants. Ideally, the complete set of WRKY genes should be identified from species representing different branches on the evolutionary tree of plants for further analysis. Unfortunately, genome sequence is currently not available for most plant species. However, a large number of EST sequences for many plants are publicly available and can be used to roughly estimate the minimum number of WRKY genes in these species. We first surveyed GenBank's dbEST set and found that WRKY genes are widespread in land plants, as over 40 species have expressed WRKY genes (data not shown). We then estimated the number of unique WRKY genes for 17 species using their Gene Indices, which are assembled EST sequence contigs with the minimal redundancy, provided by The Institute for Genomic Research (TIGR) [ 49 ]. The analysis also included ESTs for the moss Physcomitrella patens and the fern Ceratopteris richardii whose Gene Indices are not available [see Additional file 4 ]. For the EST set, redundant ESTs for WRKY proteins were manually removed. Together these 19 species represent different branches on the evolutionary tree of the land plants. While the moss Physcomitrella is an early diverged land plant, the fern is an ancient vascular plant. The conifer Pinus represents the gymnosperm lineage, and the remaining are the evolutionarily more advanced flowering plants [ 50 ]. ESTs encoding WRKY proteins were identified in all the 19 species. Moreover, multiple WRKY genes are represented in the EST or contig sets for most plants including the moss and pine, with the most WRKY genes (109) from soybean [see Additional file 4 ]. Although the actual number of WRKY genes encoded in a plant genome can only be known using the genome sequence, EST datasets are useful to estimate the relative size of WRKY family in plant species whose genome sequences are not available, given sufficient large EST sets sampled from the genomes. If a set of ≥ 50,000 ESTs is considered a large sample, then pine, moss and 12 flowering plants listed in Additional file 4 have enough ESTs for the estimation. The comparison of the number of WRKY genes identified from EST sets with comparable size suggests that the genomes of moss and pine seem to encode much fewer WRKY genes than evolutionarily advanced flower plants. We also compared pine with Arabidopsis in another analysis using ESTs from GenBank's dbEST database (as of 10/28/2002). We identified ESTs for 46 Arabidopsis WRKY genes but only two pine WRKY genes, although Arabidopsis' EST set (176,915) is less than three times bigger than pine's (60,226). The abundance of WRKY ESTs in the total EST set is lower for pine, fern and moss than for flowering plants, as the percentage of WRKY ESTs in the total EST set for the three non-flowering plants is among the lowest [see Additional file 4 ]. The WRKY EST abundance in an EST dataset may be affected by the number of WRKY genes in the species and by the expression levels of WRKY genes in the cells from which ESTs were obtained. For example, WRKY EST abundance for pine is much lower than that for tomato (0.0086% : 0.3546%, or ~ 1 : 40). The low WRKY EST abundance of pine may be partly due to fewer WRKY genes from pine than from tomato (4 : 51, or ~ 1 : 13) [see Additional file 4 ]. It is also possible that pine WRKY genes are lowly expressed. For example, for a tomato WRKY gene the average EST count is > 10, but for pine it is < 2. The identified WRKY genes were phylogenetically classified into five groups [see Additional file 4 ]. In six WRKY genes identified from the moss ESTs, two are homologous to Group 2_c and three belong to Group 2_d + 2_e, indicating an early origin of these groups in land plants. In comparison, genes in Group 3 are only identified in the EST sets of flowering plants but not from EST data of more ancient plants, i.e., moss, fern and pine [see Additional file 4 ]. Phylogeny of the WRKY domains To examine the evolutionary relationships among the WRKY domains, we estimated the phylogeny by using the neighbor-joining program from PHYLIP 3.57 for the amino acid sequences of WRKY domains from G. lamblia , the slime mold, the green alga, Arabidopsis and rice. The phylogenetic relationships were also inferred with the programs of the least squares and parsimony from PAUP 4.0 for the corresponding nucleotide sequences. We also did the same analysis for the rice dataset alone. The topology of trees obtained from these analyses is essentially the same, and the neighbor-joining tree is shown in Figure 1 . Group 2 domains designated by Eulgem et al. [ 3 ] are not monophyletic, but form three distinct clades. These include: 2_a + 2_b, 2_c, and 2_d + 2_e. Moreover, Group 2_a + 2_b and Group 2_c are closely related to Group 1C domains, while Group 3 is clustered with Group 2_d + 2_e. In addition, the rice and Arabidopsis WRKY trees (not shown) consistently clustered WRKY1N domains as a monophyletic subtree and all other domains as a natural clade, supporting the suggestion that Groups 2 and 3 domains are more closely related to the C-terminal domains of Group 1 genes than to the N-terminal domains [ 3 ]. Figure 1 Unrooted phylogenetic tree of the WRKY domains . The tree was reconstructed from the amino acid sequences using the neighbor-joining program from Phylip 3.57. Clades of WRKY domains are labelled according to the classifications of AtWRKY domains by Eulgem et al [3] who proposed three groups and five subgroups in Group 2 (a, b, c, d and e). We suggest classifying WRKY domains into five groups modified from the old system. While Groups 1 and 3 are unchanged, the original subgroup 2_c is promoted to Group 2_c. Subgroups 2_a and 2_b, and subgroups 2_d and 2_e are combined to form two new groups, 2_a + 2_b, and 2_d + 2_e, respectively (see text for details). WRKY domains from G. lamblia are represented by thick and dark-green branches; the slime mold, thick and cyan; the green alga, thick and magenta; Arabidopsis, thin and blue; and rice, thin and red. In flowering plants, genes encoding WRKY domains appear to have been duplicated independently in monocots and dicots. For Group 3 domains, three subsets each of which consists of five or more members only from rice can be distinguished from the phylogram shown in Figure 2 . Similarly, six members of WRKY domains, all from Arabidopsis, are clustered together. Independent domain clusters of either species are also found in other WRKY subfamilies (data not shown). These results suggest that numerous duplications and diversifications for WRKY genes, particularly Group 3 genes, have occurred after the divergence of the monocots and dicots. Indeed, all rice WRKY domains with the sequence WRKYGEK (Table 1 ) are classified as a sub-cluster of the largest rice domain cluster in Group 3 (Figure 2 ), implying that multiple duplication events led to this large cluster in rice. Figure 2 Phylogram of Group 3 WRKY domains from Arabidopsis (AtWRKY) and rice (OsWRKY) . The amino acid sequences were analysed with the neighbor-joining and parsimony algorithms implemented in PHYLIP 3.57. Bootstrap values ≥ 50% are indicated above the nodes for distance analysis. The C-terminal domains, AtWRKY1C, was used as the outgroup. OsWRKY proteins with the variant WRKYGEK are marked by *. Discussion WRKY genes seem to be an innovation in eukaryota after the divergence of eubacteria – archaea – eukaryota. In eukaryotes, the WRKY genes are present in the green plants as well as in the ancient eukaryote G. lamblia and the mycetozoan D. discoideum , but not in fungi and animals. G. lamblia is a primitive unicellular eukaryote diverged ~ 1,500 million years ago (mya) [ 51 ]. Originally thought as plant-specific [ 2 , 3 ], the WRKY transcription factors therefore seem to have an early origin in eukaryotes. As the mycetozoa is closely related to the fungi-animal clade [ 41 , 43 ], the WRKY gene(s) may have been lost prior to the divergence of fungi and animals, but after the split of the slime mold and fungi-animal lineages. Based on the current data, we propose a model for the origin and evolution of the WRKY factor family (Figure 3 ). First, the ancestor of the descendant WRKY genes found in G. lamblia , the slime mold and the green alga seems to be a Group 1 gene encoding two WRKY domains. The conservation of the C- and N-terminal domains suggests that they are derived from a single domain by domain duplication. Therefore we hypothesize that the earliest WRKY factor had one WRKY domain and the gene was innovated post the first appearance of eukaryotes ~ 2,500 mya [ 52 ] but prior to the divergence leading to Giardia protist, ~ 1,500 mya. Second, our data and the previous results by Eulgem et al. [ 3 ] suggest that the WRKY domains of groups 2_a + 2_b, 2_c, 2_d + 2_e and 3 are evolutionarily close to the WRKY1C domain. It seems that Group 1 genes which contain only the C-terminal WRKY domain are ancestors of the descendant WRKY genes in other groups. The N-terminal domain in Group 1 genes may have been lost prior to the gene duplication. As the green alga may have only one WRKY gene which belongs to Group 1, the duplications and diversifications leading to other groups in plants probably occurred some time after the divergence of chlorophytes and streptophytes, ~ 800 mya [ 53 ]. Third, the domain structure conservation [see Additional file 2 ] and the phylogenetic analysis (Figure 1 ) suggest that the three distinct subsets, Groups 2_a + 2_b, 2c, 2_d + 2_e, may be independently evolved from the Group 1 genes which have only the C-terminal domain. In addition, Group 3 genes appear to share a common ancestor with the clade 2_d + 2_e. The identification of 2_c and 2_d + 2_e genes in the moss EST data [see Additional file 4 ] suggests that the duplications of the genes in these groups predate the diversification of bryophytes, ~ 420 mya [ 50 ]. Although the WRKY genes in Group 2_a + 2_b and Group 3 are identified only from flowering plants in the current data, the origin of these genes seems to have occurred prior to the divergence of monocots and dicots, because the characteristic features of the WRKY domains in Group 3 are highly conserved in Arabidopsis and rice. In addition, multiple copies of Group 3 genes may exist in the common ancestor of monocots and dicots, since clusters with nested Arabidopsis and rice sequences are found in the group (Figure 2 ). Figure 3 Model of the origin and duplications of WRKY gene family . The phylogenetic tree of eukaryotes using the archaea as the outgroup is modified from Baldauf and Doolittle [43] and Kenrick and Crane [50]. The solid lines correspond to branches where WRKY homologues are identified, while the thickness of the line represents the relative size of WRKY family for the branch, from the thinnest for one copy in Giardia, the slime mold and the green alga to the thickest for over 100 copies in rice. The broken lines represent branches where WRKY genes are not present or have not been identified. The WRKY gene is symbolized by the box for the WRKY domain and the lines for sequences around the domain. The text in the box indicates the group the WRKY domain belongs to (1, Group 1; 1N and 1C, N- and C-terminal domains of Group 1 proteins; a + b: Group 2_a + 2_b; c: Group 2_c; d + e, Group 2_d + 2_e; 3: Group 3). The major gene duplications and diversifications are shown above the branch. The number shown below the branch is the divergence time (million years ago) of its children branches. The branch length is not scaled to the evolutionary distance. The classification of the WRKY family in Arabidopsis by Eulgem et al. [ 3 ] is not completely based on phylogenetic analysis and therefore does not necessarily reflect the evolutionary relationships among the groups. This is even apparent for the tree of AtWRKY genes built by the authors (see their Figure 3 ). For example, their Group 2 is not monophyletic, but seems to have several ancestors. Obviously it is necessary to implement a new classification scheme for the WRKY family to reflect the evolution of the WRKY domains. Based on phylogenetic analysis (Figure 1 ), conserved domain structures and intron positions of the WRKY domains [see Additional file 2 , B], we suggest a new classification system modified from Eulgem et al. [ 3 ]. Instead of three groups and five subgroups under Group 2 in their classifications, genes are reorganized into five independent groups according to the phylogeny of their WRKY domains, i.e., Group 1, Group 2_a + 2_b, Group 2_c, Group 2_d + 2_e, and Group 3. The relationship between the modified system and the original of Eulgem et al. [ 3 ] is as follows. Groups 1 and 3 are unchanged, while Group 2_c corresponds to the subgroup c of the old Group 2. The original subgroups a and b, and d and e in the old Group 2 are combined to become two new groups, 2_a + 2_b, and 2_d + 2_e, respectively. Our evolutionary analysis of WRKY transcription factors in this study may be important to the understanding of the overall mechanisms of biodiversity in the plant kingdom and the particular functions WRKY genes play in plant regulatory networks. First, the comparative analysis of WRKY factors in lower and higher plants indicates that the WRKY family expands as plants evolve from simpler, unicellular to more complex, multicellular forms. Since WRKY genes seem to play important regulatory roles in plants under abiotic and biotic stresses, and flowering plants which have the largest WRKY family are dominant over non-flowering plants in their distribution on the earth, WRKY genes might be essential for much of the enhanced adaptability of flowering plants to the environment. In comparison with pine, fern and moss, WRKY ESTs of flowering plants seem to be over-represented [see Additional file 4 ], suggesting that the normal functions of flowering plants might depend to a greater extent on the regulatory roles of these transcription factors. It would be interesting to analyze the functions of genes in Group 3, a greatly amplified group in monocots which are most advanced in evolution and most successful in adaptability. Second, the pairs of Arabidopsis WRKY genes, AtWRKY3 and 4, 8 and 28, 11 and 17, 14 and 35, 18 and 60, 24 and 56, and 38 and 62 share similar expression patterns in response to pathogen inoculation and salicylic acid treatment [ 23 ]. Phylogenetic analysis indicates that these pairs of genes are clustered together with high bootstrap value support (data not shown). Thus, the newly duplicated WRKY genes may overlap in functions to better protect the cell or organism from deleterious effects caused by gene mutation or deletion. Moreover, a number of WRKY genes from different phylogenetic groups may be activated by the same physiological or environmental stimulus, such as bacterial pathogen attack [ 6 , 25 , 27 , 54 ], viral pathogen attack [ 23 ], wounding [ 30 ], or senescence [ 33 - 35 ]. The WRKY genes are possibly involved in multiple pathways leading to an array of physiological responses. Nevertheless, the elucidation of the evolution and duplicative expansion of the WRKY genes should provide valuable information on their functions. Conclusions Originally believed to be plant-specific, WRKY transcription factor family has an early origin in eukaryotes and is also present in a slime mold which is more closely related to the lineage of fungi-animals than to plants. WRKY genes have been duplicated many times during evolution in plants, resulting in a large gene family for WRKY proteins in flowering plants. The elucidation of the evolutionary pathway of WRKY family and a new classification system we proposed based on phylogenetic analysis, conserved WRKY domain structures and intron positions should assist the functional characterization of WRKY genes. Methods Datasets The annotated genome sequences of rice ( Oryza sativa spp. japonica ) (OSA1, released on 7/27/2003) and Arabidopsis (ATH1, released on 4/17/2003) were downloaded from TIGR [ 55 ]. OSA1 and ATH1 include nucleotide sequences of genes, mRNA and coding regions, peptide sequences, and the gene structure information such as the start and end of the exons in a gene. For the green alga Chlamydomonas reinhardtii , the genome sequence release 1.0 on 2/4/2003 was used [ 56 ]. We also downloaded Giardia lamblia genome sequence released on 1/1/2003 [ 57 ]. GenBank's Non-Redundant (nr), dbEST and taxonomy datasets were downloaded from National Center for Biotechnology Information (NCBI) [ 58 ]. TIGR's Gene Indices for plant species [see Additional file 4 ] and the slime mold Dictyostelium discoideum were downloaded from TIGR [ 59 ]. These Gene Indices represent non-redundant gene transcripts assembled from publicly available ESTs and annotated sequences [ 49 ]. Pfam's WRKY domain sequences (WRKY-seed) were also downloaded [ 60 ]. WRKY gene identification We searched 'nr' and dbEST datasets for WRKY genes in species outside the plant phyla. The dbEST dataset was also used to survey the expressed WRKY genes in plant species. We aligned the sequences in the datasets with WRKY-seed using BLAST programs [ 61 ]. To determine the taxonomical distribution of WRKY genes from the BLAST output, we constructed a database where the BLAST results, the subject sequences and their associated taxonomy information from NCBI [ 58 ] were stored. The significant hits (E < 10 -4 ) were parsed and manually checked for the presence of the characteristic features of the WRKY domain. To systemically catalog the WRKY genes for rice and G. lamblia , we searched their genome sequences with blastp and PSI-BLAST [ 61 ] using WRKY-seed as the query. For PSI-BLAST, we used the default settings for three iterations. We also searched for WRKY genes with HMMER using the global profile of the WRKY domain [ 60 ]. HMMER, a sequence analysis tool based on profile Hidden Markov models [ 62 ], is available at [ 63 ]. The search results with the threshold of E < 10 -4 for blastp and PSI-BLAST and E < 0.1 for HMMER were manually compared to remove non-WRKY hits. We also used the same strategy to identify the set of WRKY genes from the Arabidopsis genome. To identify WRKY genes from the green alga, we first BLASTed its genome sequence against the WRKY-seed. The significantly aligned sequences (E < 10 -4 ) were then subject to WRKY domain and gene predictions. The WRKY domain was predicted with the Pfam's DNA SEARCH [ 64 ], a web-interface backed by the GeneWise algorithm [ 65 ]. The WRKY gene was predicted by FGENESH using the profile for monocots [ 66 , 67 ] and GENSCAN using the profile for maize [ 68 , 69 ]. We also searched ESTs and EST-assembled contigs for the identified WRKY genes of rice, the green alga, G. lamblia and the slime mold, using blastn. An EST- or contig-hit was accepted if the identity of the alignment was > 96% for > 400 aligned nucleotides (nt), > 97% for 300 ~ 399 nt, > 98% for 200 ~ 299 nt, > 99% for 100 ~ 199 nt, and = 100% for 50 ~ 99 nt. The alignment with < 50 nt was discarded. Analysis of WRKY genes The WRKY domain boundary was defined as by Eulgem et al. [ 3 ]. The peptide sequences of the domains were aligned with ClustalX (v1.81, with default settings) [ 70 ] and the alignment was adjusted based on the conserved features of the WRKY domains. The results were then used to guide the alignment of the corresponding nucleotide sequences. The neighbor-joining algorithm implemented in PHYLIP 3.573c [ 71 ] for amino acid sequences with the pairwise distance computed under the PAM model, and the least square fit and most parsimony algorithms in PAUP* 4.0b10 [ 72 ] for nucleotide sequences were used for phylogenetic tree reconstruction. Authors' contributions LW initiated the study. YZ and LW carried out the analyses, and YZ drafted the manuscript. Supplementary Material Additional File 1 WRKY genes from Giardia lamblia , Dictyostelium discoideum and Chlamydomonas reinhadrtii Click here for file Additional File 2 Multiple alignments, domain classification and sequence conservation patterns of WRKY domains from rice (OsWRKY), Arabidopsis (AtWRKY), the green alga (ChrWRKY), the slime mold (DsWRKY) and Giardia lamblia (GlWRKY) Click here for file Additional File 3 Identified members of the WRKY superfamily in the rice genome Click here for file Additional File 4 Survey of WRKY genes from ESTs or their assembled gene indices for 19 plants and the phylogenetic classification of the genes Click here for file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544883.xml |
549592 | Can Routine Commercial Cord Blood Banking Be Scientifically and Ethically Justified? | Background to the debate: Umbilical cord blood—the blood that remains in the placenta after birth—can be collected and stored frozen for years. A well-accepted use of cord blood is as an alternative to bone marrow as a source of hematopoietic stem cells for allogeneic transplantation to siblings or to unrelated recipients; women can donate cord blood for unrelated recipients to public banks. However, private banks are now open that offer expectant parents the option to pay a fee for the chance to store cord blood for possible future use by that same child (autologous transplantation.) | Nicholas Fisk and Irene Roberts's Viewpoint: There Are Good Reasons to Be Wary of Private Banking No one disputes the merit of public cord blood banking, in which women altruistically donate umbilical cord blood (UCB) for haemopoietic stem cell (HSC) transplantation, in a way similar to bone marrow donation. Unrelated UCB transplants have good outcomes in children and are associated with less graft-versus-host disease than adult marrow or peripheral blood stem cells [ 1 , 2 ]. Public cord blood banks also increase the availability of donor HSCs for ethnic groups underrepresented in bone marrow registries [ 3 ]. Similarly, there is little argument against storing UCB from siblings in families with a known genetic disease amenable to HSC transplantation [ 4 ]. The validity of directed UCB storage in “low risk” families, however, has been widely challenged. After early concerns from the American Academy of Pediatrics [ 5 ] and American College of Obstetricians and Gynecologists [ 6 ], the United Kingdom's Royal College of Obstetricians and Gynaecologists concluded in 2001 that routine, directed commercial UCB storage could not be justified scientifically, was logistically difficult, and therefore could not be recommended [ 7 ]. In 2002, the French National Consultative Ethics Committee for Health and Life Sciences reached similar conclusions [ 8 ]. In Italy the practice has been banned. A recent European Union report highlighted serious ethical concerns about commercial UCB banks and questioned their legitimacy in selling a service of no real use [ 9 ]. So what's wrong with allowing parents who can afford it the biological luxury of storing their child's stem cells? Are commercial UCB banks exploiting the emotional vulnerabilities of parents for financial gain? (Illustration: Giovanni Maki) First, UCB is very unlikely ever to be used. The probability of needing an autologous transplant is less than one in 20,000 [ 9 , 10 ], although commercial providers quote figures at least an order of magnitude higher, often confusing prearranged usage in at risk children with unanticipated use in those at low risk. For acute leukaemia, perhaps the most likely indication for autologous UCB transplantation, improvements in conventional therapy and allogeneic transplantation mean few proceed to autologous transplantation. In any case, there are arguments against the use of autologous UCB, including the presence of pre-leukaemic mutations and the high rate of relapse [ 11 ]. Similar considerations apply to bone marrow failure [ 11 ]. Of current indications for HSC transplantation [ 12 ], only for solid tumours, lymphomas, and auto-immune disorders might autologous UCB find a role, and even here, UCB collections often contain only enough HSCs to reconstitute children (not adults). Other uses for UCB remain speculative since it is unclear whether non-haemopoietic stem cells are present in sufficient numbers for use against degenerative conditions. Even in the uncommon event of a requirement for autologous stem cells, failure to store UCB is unlikely to be disastrous; HSCs could still be harvested from bone marrow or peripheral blood, and multipotent stem cells are increasingly being isolated from other accessible sources (e.g., deciduous teeth). Umbilical cord blood is very unlikely ever to be used. Second, there are important moral issues. The persuasive promotional materials of commercial UCB banks target parents at a vulnerable time, urging them to take this “once in a lifetime opportunity” to “save the key components to future medical treatment” and freeze “a spare immune system” [ 7 ]. Even at a typical cost of several thousand dollars, how could any responsible parent fail to provide for their child's future by preserving “something that may conceivably save his or her life”? As well as enumerating conditions currently treated by HSC transplantation, such literature boasts lengthy lists of diseases potentially amenable to stem cell therapy in the future, including Parkinson disease, diabetes, cancer, and heart disease. Such banks have been said to raise hopes of utopia and to use the promise of “helping children” to disguise a mercantile project CureSource, a commercial UCB bank, believes that banking is a “once in a lifetime opportunity” (Figure: CureSource) Third, collection imposes a considerable logistic burden on the obstetrician or midwife. In addition to consent, parental blood collection, and the associated packaging and paperwork, a large volume of blood has to be collected from the umbilical vessels in utero, requiring multiple syringes under aseptic technique. This may distract professionals from their primary task of caring for the mother and baby at this risky time or, more generally, divert delivery room staff from attending others [ 7 ]. This applies particularly in multiple or operative deliveries, and thus UCB collection is not recommended at complicated births [ 5 ]. These problems do not apply to altruistic donations to public cord blood banks, which can be harvested less intrusively ex utero; inadequate or logistically difficult samples can be discarded or forgone without consequence [ 3 ]. Finally, individual UCB banks need to remain in business long term if cryopreserved stem cells are to be retrieved. The commercial attractiveness of a service paid years in advance is attested to by the burgeoning number of private providers, yet it seems unlikely that all will survive. Indeed, some US providers are already in trouble for infringing collection patents. There remain reservations about whether laboratories will meet national standards and be accredited. There is a further danger that misplaced enthusiasm for commercial auto-collection will undermine the proven utility of altruistic public cord blood banks. Notwithstanding the above, we accept that the utility of UCB storage in low-risk families is very different from the entirely speculative post-mortem cryonics industry. We acknowledge the possibility that autologously stored UCB stem cells may eventually be used. Indeed, recent research documenting the multi-potentiality of UCB mesenchymal lineages [ 13 ] and the in vitro expandability of cord HSC numbers sufficient to transplant an adult [ 14 , 15 ] may even improve such prospects. Private banks, however, must provide clear, honest, and unambiguous information for their customers. EU guidance recommends they be told that the likelihood of stored UCB stem cells being used to treat their child is negligible and that future therapeutic possibilities are very hypothetical [ 9 ]. Roger Markwald and Vladimir Mironov's Viewpoint: No One Has a Second Chance to Collect Their Cord Blood Stem cells may potentially be used in life-saving therapies for degenerative diseases or injuries. Stem cells self-replicate and are multi-potential—they can differentiate into diverse cell types [ 13 ]. While stem cells can come from many sources, our viewpoint is that UCB is an important source of progenitor (stem) cells that can be used as an immediate alternative for bone marrow transplantation and for engineering healthy new cells and tissues. To fully realize this potential will require collection and banking of UCB cells, which are harvested without pain or trauma from placental structures that are normally discarded after birth. We realize that UCB banking (public and private) has sparked controversy. Critics of routine banking question its cost-to-benefit ratio, citing doubts about the clinical relevance of cord stem cells or the likelihood that they will ever be used [ 16 ]. Other critics argue that embryonic stem cells (ESCs) are the better option. The “stemness” of UCB cells is not merely theoretical (as suggested by Steinbrook; [ 16 ]). UCB has two types of multi-potential progenitor cells—HSCs and mesenchymal stem cells. These express different cell surface markers, making it possible to show that HSCs can differentiate into new red and white blood cells and, as with mesenchymal stem cells, can also transdifferentiate in vivo into liver, kidney, brain, bone, skeletal, and cardiac muscle cells [ 13 , 17 , 18 , 19 , 20 ]. While ESCs have the potential to form all types of cells, they are harvested from embryos shortly after fertilisation, raising moral and legal issues not attributed to UCB cells [ 21 ]. The real question is who should pay for umbilical cord blood collection and storage. ESCs also represent an allogeneic source of cells—they are derived from another individual whose tissue type does not match up with the recipient, resulting in immune rejection when transplanted [ 22 ]. We know of no clinical or preclinical animal study that provides hard evidence of functional integration (without immune rejection) of transplanted ESCs. Even with somatic nuclear transfer (cloning), ESCs remain allogeneic, as they still have foreign mitochondrial DNA for which there remains untested potential risk for auto-immune diseases. In contrast, 2,000 allogeneic UCB transplants have been performed, mostly in children, for the treatment of a variety of malignant and nonmalignant conditions [ 22 ]. A London Cord Blood Bank report found that two years after transplantation the survival rate varied between 54% and 69%, depending upon the number of matched units [ 23 ]. For reasons not fully understood, allogeneically transplanted UCB cells have immune tolerance (of HLA mismatch) [ 24 ], and the risk of causing graft-versus-host disease is considered to be acceptable [ 24 , 25 ]. With millions of healthy babies born each year, there is potentially a large UCB supply that can be stored, tissue-typed, and made available at short notice. If saved for potential use by the donor, UCB cells become a source of perfectly matched, autologous stem cells (plus there is a 25% probability of being an exact match for a sibling). Yet the American Academy of Pediatrics came out against UCB banking, saying that the odds (with some exceptions) of a donor ever using a UCB sample were low, between 1/1,000 and 1/200,000. While the chance of a donor benefiting may presently be low, this does not automatically mean that another member of society could not benefit. For people with genetic diseases or cancers, the chances of finding an immune-tolerant donor match would obviously be increased by the expansion of cord blood sampling. Also, at the pace that stem cell research is moving, perhaps there will be new uses for UCB cells in the next decade, especially in the field of tissue engineering [ 26 ]. Importantly, unlike bone marrow, an increase in UCB samples will enhance availability for every ethnic group for tissue matching. What is certain is that no one has a second chance to collect their cord blood. Who should operate cord blood banks—the private or the public sector? There are around 20 private UCB banks in the United States. They charge a collection fee, typically $1,000–$1,500, which includes testing for pathogens and genotyping. Samples are maintained in a frozen state for around $100 a year. An additional $15,000–$25,000 is charged if a sample is used for transplantation (usually covered by health insurance). The cost of UCB cell transplantation is significantly less than bone marrow transplantation, and the risk of graft-versus-host disease is lower [ 24 ]. The private sector, not government, has been the innovator for most new technology related to harvesting, storing, and utilizing cord blood as well as stem cell research. Licensing fees and patent protection are essential to biotechnology companies—they are needed to attract venture capital, build businesses, and develop new technologies. The only alternative in most countries is public cord banks, which suffer from insufficient funding. Any exploitation by companies of the vulnerabilities of expectant parents for financial gain is clearly unacceptable. Federal legislation to establish a national cord blood stem cell bank network—free to all donors—has been introduced in the US Senate and House of Representatives that, if approved, should diminish the risk of exploitation. But unless the network is well-designed from a sociological viewpoint, it could generate a situation where not all cultural and ethnic groups are represented or where benefits are accessible only to families with health insurance or sufficient income to afford transplants. It still remains difficult to find full matches for African, Asian, and Native Americans—mostly because of an insufficient number of UCB donors and the diversity of HLA types in different ethnic groups [ 16 ]. The real question is who should pay for UCB collection and storage—the individual donor, who currently has only a small prospect of using their cord blood, or society as a whole? We believe that it is the job of government to assure that people of all ethnic groups are informed and educated about donating UCB. Then, to facilitate UCB banking and the development of technological innovations for its storage and clinical utility, we recommend a national network that is a mixture of for profit, non-profit, and governmental organizations. Fisk and Roberts's Response to Markwald and Mironov's Viewpoint Markwald and Mironov argue that commercial UCB banking is ethically justified on the grounds that UCB transplantation is effective treatment for many haematological disorders, that autologous UCB is a useful future source of stem cells for the donor, and that there is no second chance to collect these cells. We did not dispute (indeed we acknowledged) the value of UCB HSCs for the treatment of many malignant and non-malignant haematological disorders. However, evidence of their value derives from allogeneic HSCs from public UCB banks [ 27 ]. Like many in the routine UCB collection industry, Markwald and Mironov fail completely to distinguish between public and private banks in their discussion, and further neglect to mention that most transplants have been of allogeneic cells donated altruistically by non-related donor families. Markwald and Mironov state that the real question is who should pay for routine UCB collection and storage. However, they take no account of the considerable logistic burden this imposes, the extremely low chance that autologous cells will ever be used (less than one in 20,000), and the costs of routine UCB collection [ 9 ]. They also fail to mention that autologous UCB HSCs are frequently unsuitable for use for two reasons. First, they cannot cure inherited disorders (e.g., β-thalassaemia major or congenital bone marrow failure syndromes), and second, clinically hidden pre-leukaemic and/or leukaemic cells may be present in UCB at birth in children who years later develop full-blown leukaemia [ 28 ]. In addition, the authors introduce the irrelevant argument of the likely unsuitability of ESC transplants, with which, given the propensity of these transplants to cause teratomas, we agree [ 29 ]. Thus the real questions are, first, why should society in general, or the government as a representative of at least a substantial proportion of society, pay for a service not shown to be of any real use? (After all, as we pointed out, autologous HSCs are rarely required and there is no evidence that UCB can treat degenerative disease in elderly humans.) And second, why should commercial banks be allowed to continue to target vulnerable parents anxious to do the best for their children while making no mention of the low chance of use, of alternative sources of available stem cells (e.g., autologous marrow, a better source of non-haemopoietic stem cells), or of the risks of reducing stocks of allogeneic HSC in public UCB banks? Markwald and Mironov's Response to Fisk and Roberts's Viewpoint We agree with Fisk and Roberts that exploiting the emotional vulnerabilities of expectant parents is unjustifiable—thus we support regulation of UCB banking, monitoring, certification, and informed consent. But we disagree that there is a lack of solid scientific evidence for UCB collection and that “future therapeutic possibilities are very hypothetical.” Research on stem cells is advancing rapidly, and stem cells derived from UCB are emerging as a reasonable first choice for the field of regenerative medicine. Fisk and Roberts are inconsistent in their views. They claim that stem cells collected in UCB units often are not “in sufficient numbers for use against degenerative conditions” in adult life but then acknowledge that “the in vitro expandability of cord HSC numbers is sufficient for transplantation into an adult.” They argue that private and public UCB collections create dramatically different “logistic burdens,” but in our experience, the syringes, paperwork, and level of personal distraction are generally the same for public or private banking. We strongly disagree that private UCB banking has no future. While we anticipate a consolidation phase for this industry, surviving companies should be eager to acquire UCB units collected from competitors. If all stem cell sources were under a state monopoly—without private sector contribution—there would be less incentive or opportunity for fostering innovation in long-term storage, expansion, or phenotype characterization of UCB stem cells. The growth of new biotech companies focused on regenerative medicine would be discouraged, compromised, or undermined by the absence of competition, inadequate access to venture capital, and the typical resistance of state health-care systems and their affiliated medical professionals to innovation. Fisk and Roberts are creating obfuscations by mixing “speculative cryobionic companies” that promise “immortality and eternity” with serious biotech companies and private UCB banks that focus on a realistic commercialisation of UCB stem cells as a platform for promoting new biotech initiatives. The collection and storage of UCB stem cells is an opportunity for society to build a representative collection of UCB units that can improve the chances of identifying suitably matched donors for transplantation. Human ESCs are mired in ethical concerns and concerns about immunological intolerance. Autologous cells from the bone marrow or elsewhere lose their attractiveness if there is a genetic mutation or a progressive loss of “stemness” due to normal aging [ 30 ]. UCB cells offer the best short- and long-term hope for treating sick children with cancers or adults with a variety of diseased organs and tissues. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549592.xml |
526182 | Repetitive Editing and RNA Splicing | null | The so-called “central dogma” of biology—DNA makes RNA makes protein—is a simple statement that subsumes a wealth of complexity. In particular, the past decade has shown that after RNA is made, it is run through a gauntlet of processes that strip it of introns and splice its exons, add a cap and tail, and even chemically modify one or more bases along the way. This last possibility includes deamination—removal of an NH 2 group—from adenosine, converting it to inosine. During translation, the ribosome reads an inosine as a guanosine; thus, an A-to-I edit in RNA can even cause an amino acid change in the resulting protein. A new study in this issue by Stefan Maas and colleagues shows that A-to-I editing is remarkably widespread among human genes, and commonly targets a ubiquitous repetitive sequence, the Alu repeat. A-to-I editing has been recognized for several years, but the known targets have been few, far fewer than the number predicted by measuring the inosine content in messenger RNA. To identify more targets, Alekos Athanasiadis, Alexander Rich, and Maas compared genomic sequences to cDNA sequences. Adenosines are unchanged in genomic DNA, while in complementary DNA (cDNA), which is derived from reverse transcribing mRNA, any adenosines that were converted to inosines during RNA editing show up as guanosines. Thus, A-to-G discrepancies revealed candidate editing sites. To reduce the number of false positives, the researchers confined their search to regions with multiple A-to-G discrepancies. In an initial screen of 3,000 cDNAs, they found 26 A-to-I edited genes. In all but one case, the editing occurred in an Alu sequence. Sequence and structure preferences of editing in Alu repeats There are approximately 1.4 million Alu sequences in the human genome, each about 300 base-pairs in length, which together comprise about 10% of the entire genome. Not all of them occur in genes, but those that do are typically found in transcribed but untranslated regions (introns), either upstream (3′) or downstream (5′) of the translated region. In many genes, they are found in pairs, ordered head to head or tail to tail, separated by a short intervening sequence. Once transcribed, the Alu sequences can pair up, forming a stretch of double-stranded RNA that makes an ideal target for the A-to-I RNA editing machinery, called ADAR (adenosine deaminase acting on RNA). A typical gene contains between one and two dozen Alu sequences. Based on this and the frequency of editing found when analyzing more than 100,000 mRNAs in the human transcriptome, Athanasiadis, Rich, and Maas estimate that the probability that any particular mRNA undergoes A-to-I editing is between 85% and 95%. While the bulk of edited Alu sites are in introns, a small fraction of them are in exons. Here they can lead to alternative forms of the same protein, expressed in different cell types or at different times; this appears to be especially common in the nervous system. Alu editing can also convert introns to exons, and vice versa, through creation or destruction of splice sites. It is possible A-to-I editing may be used to reduce the creation of deleterious new exons, although more work will be needed to explore this possibility, as well as what role, if any, A-to-I editing plays in promoting new exon creation. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526182.xml |
524024 | Optimization of an immunostaining protocol for the rapid intraoperative evaluation of melanoma sentinel lymph node imprint smears with the 'MCW melanoma cocktail' | Background In the management of cutaneous melanoma, it is desirable to complete the regional lymphadenectomy during the initial surgical procedure for wide excision of biopsy site and sentinel lymph node (SLN) biopsy. In this study, we optimized and evaluated a rapid 17 minutes immunostaining protocol. The discriminatory immunostaining pattern associated with the 'MCW Melanoma Cocktail' (mixture of Melan- A, MART- 1, and tyrosinase) facilitated the feasibility of intraoperative evaluation of imprint smears of SLNs for melanoma metastases. Methods Imprint smears of 51 lymph nodes from 25 cases (48 SLNs and 3 non-SLNs, 1 to 4 SLNs/case) of cutaneous melanoma were evaluated. Results Sixteen percent, 8/51 lymph nodes (28%, 7/25 cases) were positive for melanoma metastases in immunostained permanent sections with the 'MCW melanoma cocktail'. All of these melanoma metastases, except 1 SLN from 1 case, were also detected in rapidly immunostained wet-fixed and air-dried smears (rehydrated in saline and postfixed in alcoholic formalin). The cytomorphology was superior in air-dried smears, which were rehydrated in saline and postfixed in alcoholic formalin. Wet-fixed smears frequently showed air-drying artifacts, which lead to the focal loss of immunostaining. None of the 5 SLNs from 5 cases exhibiting capsular nevi showed a false positive result with immunostained imprint smears. Conclusions Melanoma metastases can be detected intraoperatively in both air-dried smears and wet-fixed smears immunostained with the MCW Melanoma cocktail. Air-dried smears rehydrated in saline and postfixed in alcoholic formalin provide superior results and many practical benefits. | Background What is already known on this topic? A rapid intraoperative evaluation of sentinel lymph nodes (SLNs) for melanoma metastases during the interval between the SLN biopsy and the wide excision of the melanoma biopsy site may eliminate the need of an additional surgery for completion of regional lymphadenectomy. What this study adds? Air-dried imprint smears which were postfixed in alcoholic formalin following saline rehydration were optimal for immunocytochemical evaluation with the 'MCW melanoma cocktail'. The rapid evaluation of imprint smears immunostained with the 'MCW melanoma cocktail' is reliable for the intraoperative evaluation of cutaneous melanoma SLNs for melanoma metastases. The prevailing trend in the management of cutaneous melanoma supports the sentinel lymph node (SLN) biopsy as a standard of care [ 1 - 16 ], but a few authors regard it as controversial [ 17 , 18 ]. In a given case where the SLN biopsy is performed and is positive for melanoma metastases, it is usually followed by additional surgery for regional lymphadenectomy. A rapid intraoperative evaluation of SLNs for melanoma metastases during the interval between the SLN biopsy and the wide excision of the melanoma biopsy site may eliminate the need for an additional regional lymphadenectomy surgery. Previously evaluated approaches such as fluorodeoxyglucose-positron emission tomography [ 19 , 20 ], morphological evaluation of frozen sections [ 21 - 24 ], intraoperative morphological evaluation of imprint cytology [ 25 , 26 ], and immunostaining of frozen sections [ 27 ] are not sufficiently sensitive. Imprint smears of lymph nodes can be prepared rapidly. When compared to frozen sectioning, imprint smears are more desirable due to lower cost, quicker process, avoidance of tissue loss in the cryostat, prevention of freezing artifact in the tissue, and the elimination of problems associated with cryo-sectioning of fatty lymph nodes. These advantages have resulted in a preference for imprint smears over frozen sections by many investigators for the evaluation of SLNs in breast carcinoma [ 28 , 29 ]. Although relatively specific, the morphological interpretation of imprint smears alone used for the evaluation of melanoma metastases in SLNs is not very sensitive [ 25 ]. This is predominantly because of the inherent limitations associated with morphological interpretation. Singly scattered cells of melanoma metastases in a sea of numerous other cells are difficult to differentiate from reactive histiocytes, endothelial cells, and other cells with morphology alone. At the current time, frozen-section examination (with or without immunohistochemical evaluation) and the morphological evaluation of imprint cytology smears are the methods available for intraoperative evaluation of SLN in cutaneous melanoma. However, these studies have demonstrated relatively low sensitivity and specificity, discouraging the practical application [ 22 - 27 ]. Conventional immunomarkers such as the S-100 protein and HMB45 suffer a significant drawback because of interference by non-melanoma cells resulting in high signal to noise ratio [ 28 ]. Because of this, rapid and accurate intraoperative evaluation of SLN with immunostained imprint smears was not previously possible. In our previous study, the 'MCW Melanoma Cocktail'- a mixture of monoclonal antibodies- MART-1 {1:500}, Melan- A {1:100}, and Tyrosinase {1:50} (Table 1 ) demonstrated a highly discriminatory immunostaining pattern [ 28 ]. This observation suggested the feasibility of rapid intraoperative evaluation by examining the imprint smears of SLNs immunostained with the cocktail [ 28 - 30 ]. In the current study, we have optimized a protocol for the rapid intraoperative immunostaining of SLN imprint smears from patients with cutaneous melanoma utilizing the 'MCW melanoma cocktail'. Our previous experience suggested that air-dried smears postfixed in alcoholic formalin after saline rehydration demonstrated optimal results for immunostaining [ 31 ]. In this study, in addition to air-dried smears we also evaluated wet-fixed smears for further confirmation. Table 1 The composition of the 'MCW Melanoma Cocktail' ¶ Marker Clone Source *Final Dilution in the cocktail MART-1 M2-7C10 Signet Laboratories, Inc. Dedham, MA 1:500 Melan-A A103 Dako Corporation, Carpinteria, CA 1:100 Tyrosinase T311 Novocastra Laboratories Ltd Newcastle upon Tyne, UK 1:50 * Optimum dilution for each antibody was standardized individually for that batch of antibodies with the sections of known melanoma positive control. The standardized dilution was achieved as final titer in the cocktail by adding 20 μl MART-1, 100 μl Melan-A, and 200 μl Tyrosinase to 9.68 ml of DAKO Antibody diluent (Dako Corporation, Carpinteria, CA). ¶ Adopted from Shidham et al [28]. Material and methods Patients We prospectively studied 51 lymph nodes (48 SLNs and 3 non-SLNs) from 25 patients (range- 1–4 SLNs per patient, mean- 2 per patient) under an IRB approved protocol at Froedtert Memorial Lutheran Hospital / Medical College of Wisconsin, Milwaukee, WI. A standard surgical protocol was used to identify the SLN [ 32 ]. The SLNs were harvested and submitted fresh to pathology for intraoperative and permanent section evaluation. Pathologic Examination (Figure 1 ) Figure 1 Pathological evaluation of sentinel lymph nodes for melanoma metastases. Section number 2, 5, & 8- stained with H & E; 4- immunostained with `MCW melanoma cocktail'; 6- negative control; 1, 3, 7, & 9- unstained. Number of slices of SLN shown (a,b,c) is just for illustration and would vary according to the size of the lymph node. (FPTS, formalin-fixed paraffin-embedded tissue sections; H & E, hematoxylin and eosin stain) For the evaluation of the maximum surface area of the lymph node and most of the capsular area, the lymph nodes were transected perpendicular to the long axis as thin (not thicker than 2 mm) cross sections. Two pairs of imprint smears (one test and one negative control in each pair) were made by gently touching the cut surfaces to glass slides without allowing the imprint to be smeared. One of the pairs (1 test and 1 negative control) was air-dried, rehydrated in saline, and post-fixed in 'alcoholic formalin' [ 31 ] (see video clips as Additional file 1 Higher resolution- (for high speed connection); or Additional file 2 Low resolution- (for low speed connection), the screen shots in PDF file are available as Additional file 3 Screenshots). The other pair was wet-fixed by immersing the imprint smears in 95% ethanol before drying. Fixed smears were rinsed with 95% ethanol and then immunostained with 'MCW melanoma cocktail' (Table 1 ) using a rapid immunostaining protocol (Table 2 ). This rapid protocol required 17 minutes. Additional time required for smear preparation, smear processing, and evaluation of immunostained imprint smears may vary depending on the number of slides controlled by some variables such as the size and the number of SLNs submitted for evaluation. Table 2 Rapid immunostaining protocols. Entirely manual Partially manual and with Autostainer † 1. Re-hydrate air- dried imprint smear with 0.9% saline- 15 seconds (slow~10 dips) 1. Re-hydrate air- dried imprint smear with 0.9% saline- 15 seconds (slow~10 dips) 2. Post-fix the re-hydrated smear in 'alcoholic formalin'*- 5 slow dips and then 1 minute 2. Post-fix the re-hydrated smear in 'alcoholic formalin'*- 5 slow dips and then 1 minute 3. Rinse the post-fixed smear with 95% ethanol: 5 dips 3. Rinse the post-fixed smear with 95% ethanol: 5 dips 4. Hydrate the smear in DW- 30 sec 4. 100% ethanol: 10 dips 5. 3% H 2 O 2 in DW- 1 mt 5. 100% ethanol: 10 dips 6. Protein blocking solution- 1 mt 6. Methanol: 10 dips 7. 'MCW melanoma cocktail'**- 5 mt 7. 50% ¶ H 2 O 2 inMethanol: 1 mt with agitation 8. Rinse in 0.2% Tween 20 in DW 8. Deionized water: 10 dips 9. HRP-linker Antibody***- 5 mt 9. Tris buffer (ph 7.6): 10 dips 10. Rinse in tap water 10. Place smear on Dako Autostainer which automatically applies- 11. Chromogen (DAB)- 3 mt a. Envision blocking Solution †† - 1 mt 12. Rinse in tap water b. 'MCW melanoma cocktail'**- 5 mt 13. Azure B (Blue solution of Diff-Quik ® )- 1 mt c. Envision+ Monoclonal HRP ††† - 5 mt 14. Rinse in tap water d. Chromogen (DAB)- 3 mt 15. Harris Hematoxylin- 30 sec 11. Remove the smear(s) and proceed with the following steps 16. Rinse in tap water 12. Deionized water: 10 dips 17. Dehydrate in ascending concentration of ethanol 13. Azure B- (Blue solution of Diff-Quik ® )- 1 mt 18. Clear in xylene 14. Rinse in tap water 19. Coverslip the smear with the mounting medium 15. Harris Hematoxylin- 30 sec 16. Rinse in tap water 17. Dehydrate in ascending concentration of ethanol 18. Clear in xylene 19. Coverslip the smear with the mounting medium mt, minute; sec, seconds; DW- Distilled water *Alcohol formalin was prepared by adding 50 ml of formalin (38–40% formaldehyde) to 350 ml of 95% ethanol and 100 ml of distilled water [modified and simplified from [31]; **'MCW melanoma cocktail'- Mixture of Melan- A, MART-1, & tyrosinase [28]; *** PowerVision™ Poly-AP anti-Mouse IgG (ImmunoVision Technologies, Co; Daly City, CA); † DakoAutostainer; †† Dako Corporation, Carpinteria, CA; ††† Dako Envision+ (Dako Corporation, Carpinteria, CA); ¶ 3% 10 volume Hydrogen peroxide, USP (Hydrox Laboratories, Elgin, IL, USA). Positive controls were prepared from unfixed fresh melanoma tumor, which was immunoreactive for each of the individual components of the cocktail. The cut surface of the tumor was scraped with one end of glass slide and the scraped material accumulated at the end of the slide was spread between two glass slides similar to the spreading of bone marrow smears [32]. The air-dried smears were processed similar to 'test' slides. They were rehydrated in saline (10 to 20 seconds) and postfixed in alcoholic formalin (1 minute) [30]. The post-fixed smears were rinsed in 95% ethanol and then dehydrated by taking the smears through absolute ethanol, cleared in xylene, and cover-slipped with mounting medium. The cover-slip of positive control was removed by keeping the slide in xylene overnight (for this reason a formal communication to pathology and immunochemistry lab at least 24 hours before the SLN surgery is required). After removing the coverslip the smear was passed through absolute ethanol, 95% ethanol, and then joined with the protocol for immunostaining mentioned above. These coverslipped smears of positive control could be archived at room temperature for long periods of time (personal experience). We have used such smears after removing the coverslip as positive controls up to 1 year later without loosing immunoreactivity for most of the commonly used immunomarkers. The first imprint smears from each pair were used as a 'test' and were immunostained with the 'MCW melanoma cocktail' by rapid protocol. The second imprint smear was used as a 'negative control' and processed in the same manner as the test slide except that Dako diluent ® was used in place of 'the cocktail'. Numerous smears of positive controls (both air-dried and wet-fixed smears) were prepared previously from a melanoma tumor with a known immunoreactivity for each of the three components of 'the cocktail'. These smears were prepared by scraping the cut surface of the fresh melanoma tumor and spreading the scraped material between two slides as described previously [ 33 ]. The air-dried smears were fixed in alcoholic formalin after saline rehydration. Both smears (air-dried, saline rehydrated smears, post-fixed in alcoholic formalin and wet-smears fixed in 95% ethanol) were stored after processing them through ascending grades of alcohol and xylene, followed by mounting with a glass coverslip using mounting medium (Table 2 ). Both air-dried and wet-fixed positive control smears were used during immunostaining for each batch of test smears by removing the coverslip following immersion of the slide in xylene for about 24 hours. This dissolves the mounting medium and separates the coverslip. After removal of the coverslip, the smears were put through absolute ethanol and then descending grades of ethanol to water to be combined with the respective step in the rapid immunostaining protocol (Table 2 ). The immunostained imprint smears were evaluated for melanoma micrometastases. The test smears were compared with corresponding positive controls (air-dried versus wet-fixed). The wet-fixed test smear from a given SLN was compared with a respective air-dried test smear by evaluating the sharpness of immunostaining, morphological details of the immunostained tumor cells, frequency of staining of non-melanoma structures such as mast cells and erythrocytes, air-drying artifact, deterioration in the immunostaining of the cells with unequivocal features of tumor cells, and nonspecific background staining. The results were interpreted by pathologists as positive, indeterminate, or negative for melanoma metastases. For statistical analysis, indeterminate interpretations of immunostained imprint smears were considered negative. This was based on the clinical significance with reference to the intraoperative decision algorithm for the completion of regional lymphadenectomies in SLN positive cases. After the preparation of imprint smears, the slices of SLNs were fixed in 10% formalin and processed for formalin-fixed paraffin-embedded tissue sectioning. These sections were evaluated according to the melanoma protocol (Figure 1 ) and immunostained by the avidin-biotin-peroxidase complex (ABC) method described previously [ 28 ]. Results Wet-fixed smears were difficult to prepare without focal air-drying artifact (Figure 2 , Table 3 ). This difficulty was due to the time required to transfer each of the SLN slices on the glass slide one by one and then immersing the slide (with some of the imprints already dried) in 95% ethanol for wet fixation. This was not a concern while preparing the air-dried smears, as all the imprints were ultimately dried before processing (see Additional files 1 , 2 ,& 3 ). The turnaround time for processing, immunostaining, and evaluating the smears was approximately 28 (range, 24–37) minutes. Figure 2 Comparison of cytomorphological features of immunostained, air-dried smears postfixed in alcoholic formalin after saline rehydration (ADS, 'a') versus wet-fixed smears fixed in 95% ethanol (WFS, 'b' through 'd'). The cytoplasmic immunostaining for the 'MCW melanoma cocktail' does not obscure the nuclei in 'a'. In contrast, immunostaining of shrunken cytoplasm around nuclei in wet-fixed smears obscures the nuclear details (arrows in 'b'). Air-drying artifact is present focally (arrows in 'c') with the presence of non-specific background staining (arrows in 'd'). Table 3 Comparison of air-dried versus wet-fixed imprint smears. S.No. Feature Air-dried imprint smears Wet-fixed imprint smears 1 Ease of preparing imprint smears of SLNs Easy Challenging 2 Air-drying artifact Not applicable Frequent 3 Non-specific background staining Rare Frequent 4 Immunostaining of non-melanoma structures Rare Common 5 Ease of processing, handling, and transporting the smears Easy Difficult 6 Loss of immunoreactivity of melanoma tumor cells due to air-drying artifact Not applicable Possible with potential for false negativity. 7 Sharpness of immunostaining Present Present 8 Shrinkage artifacts Absent Frequent 9 Morphological details of immunostained smears Good Poor 10 Potential loss of sample material on slide during immersion of slide in the fixative Rare Frequent The immunostained tumor cells of melanoma metastases demonstrated a high nuclear/cytoplasmic ratio. The immunostaining was non-granular and cytoplasmic. The cytoplasmic immunostaining pattern facilitated the evaluation of the nuclear details. The cell margins were usually well defined. Unlike the chromatin of mast cells, the nuclear chromatin was not clumped and did not resemble the chromatin of the lymphocytes in the background. Nucleoli were usually prominent (Figure 3 ). Figure 3 Cytomorphological spectrum of tumor cells (arrows) of melanoma metastases from different cases in rapidly immunostained air-dried imprint smears with the 'MCW melanoma cocktail' after saline rehydration and postfixation in alcoholic formalin. The tumor cells are large with well defined borders and show high nuclear to cytoplasmic ratio with non-granular cytoplasmic staining with clear nuclear details. The nuclear chromatin does not resemble the chromatin of adjacent lymphocytes in the background. Because of the brief peroxidase blocking step, the endogenous peroxidase could not be blocked entirely in some cases, leading to the staining of some non-melanoma cells such as mast cells. These were detectable in both test smears and negative control smears. The mast cells showed smaller round nuclei with clumped chromatin. This clumped chromatin was comparable to the nuclear chromatin of adjacent lymphocytes in the background. The staining was coarsely granular. The cell margins of mast cells were usually hazy and ill-defined (Figure 4 ). Figure 4 Morphological spectrum of non-tumor structures in rapidly immunostained air-dried imprint smears with 'MCW melanoma cocktail' after saline rehydration and postfixation in alcoholic formalin. a through e: Mast cells (brown arrows) show low nuclear/cytoplasmic ratio with granular staining of cytoplasm and fuzzy cell borders. The nuclear chromatin is clumped and resembled the chromatin of lymphocytes in the background. f: Non-nucleated ill defined structures (black arrow). g & h: Cells with immunoreactive nucleus (blue arrow). Insets of both g & h- zoomed cells with unequivocally negative cytoplasm but with brown staining of nucleus. Rarely some nuclei demonstrated brown staining (the cocktail immunostaining is cytoplasmic and is not nuclear). The cells with such brown stained nuclei were morphologically consistent with histiocytes (Figure 4f & 4g ). Unequivocal nuclear staining without cytoplasmic immunostaining should be interpreted as negative in immunostained imprint smears. Brown non-nucleated round to irregular material (probably erythrocytes with unblocked endogenous peroxidase) was observed in a few cases (Figure 4h ). The non-specific staining was relatively frequent in wet-fixed smears (versus alcoholic formalin fixed saline rehydrated air-dried smears) and in manually immunostained smears (versus smears immunostained with Autostainer). These structures were usually interpreted as negative with ease. However, this factor could increase the interpretation time for negative cases due to the distraction effect and could prolong the crucial turn around time for intraoperative consultation. Seventeen percent (8 out of 48) lymph nodes (28%, 7/25 cases) were positive for melanoma metastases in immunostained permanent sections. All melanoma metastases, except 1 SLN from 1 case, were demonstrated in both wet-fixed and air-dried imprint smears immunostained with the rapid protocol (sensitivity 89% and specificity 100%). On a case by case basis, 86% (6/7) of positive cases showed metastases in imprint smears immunostained with the 'MCW melanoma cocktail'; and demonstrated a sensitivity of 86%, a specificity of 100%, a negative predictive value of 95%, and a positive predictive value of 100%. Imprint smears, immunostained with the rapid protocol, showed unequivocal melanoma metastases in 1 SLN which was negative by immunohistochemical evaluation of permanent sections. This was one of the SLN from a case with two other unequivocally positive SLNs in permanent sections and rapid immunostained imprint smears. This unequivocal positivity with immunostained imprint smears alone underscored the sampling benefit with imprint cytology. Two SLNs (from 2 patients) interpreted as negative for melanoma metastases by immunohistochemical evaluation of permanent sections, were interpreted as indeterminate with the rapid protocol in both wet-fixed and air-dried smears. After retrospective evaluation, the rare doubtful cells observed in immunostained imprint smears were consistent with mast cells (Figure 4 a through 4e ). In 2 SLNs from 2 other cases, some scattered cells with non-granular immunostaining but with small, inconspicuous nuclei were observed. These cells were not mast cells and were absent in negative controls. They were also present as scattered single cells in permanent sections immunostained with 'the cocktail'. They were interpreted as benign and negative. Discussion Metastases of melanoma tumor cells in SLN could be detected in imprint smears immunostained with the 'MCW melanoma cocktail'. The air-dried imprint smears from different SLNs were easier to prepare than wet-fixed smears. As reported previously, air-dried smears have numerous advantages [ 34 ]. Because of the shrinkage factor associated with wet-fixation, the cellular details are less distinct in wet-fixed smear as compared to air-dried smears. The wet-fixed smears frequently showed non-specific background staining. They also showed air-drying artifact, which frequently compromised the immunoreactivity, resulting in multifocal faint or false negative immunostaining (Table 3 ). Wet-fixed smears with a scant number of tumor cells may be translated into a false negative result because of air-drying artifacts. Imprint smears (both air-dried and wet-fixed) immunostained with the 'MCW melanoma cocktail' showed excellent sensitivity and specificity (the indeterminate interpretations were equivalent to negative results). As compared to this, the alternative intraoperative approaches such as frozen-section alone [ 22 ], immunostaining of frozen sections with a cocktail of Melan- A, HMB-45, & tyrosinase [ 27 ], and the morphological evaluation of imprint smears alone [ 25 , 35 ] demonstrated relatively poor results. This is of practical significance. It facilitates the intraoperative decision to proceed with regional lymphadenectomy during the same anesthetic procedure. Immunostained imprint smears unequivocally showed melanoma metastases in one SLN, but these melanoma metastases were not detected in permanent sections immunostained with 'the cocktail'. Two other SLNs from this case showed melanoma metastases in both immunostained permanent sections and imprint smears. This unequivocally positive result with immunostained imprint smears highlights the benefit of the enhanced sampling with imprint cytology. Imprint smearing facilitates the sampling of two surfaces from each slice, except for the first and the last slice, as compared to only one surface of all slices by any sectioning method (Figure 1 ). In contrast to a sectioning method yielding 3–4 micron sections, which represent a tiny fraction of the lymph node slice, immunostained imprint smears facilitate the evaluation of the entire material sampled as an imprint on the glass slide from the cut surface of the lymph node. The possibility of false positive results due to the cells of capsular nevi was disproved by the negativity of all immunostained imprint smears from 5 SLNs (5 cases) with capsular nevi. The cells in capsular melanocytic nevi located in the capsule and fibrous septa of lymph node did not exfoliate and adhere to slides during preparation of imprint smears. This appears to be due to the greater cohesiveness of the cells in capsular melanocytic nevi than the cells of malignant melanoma. However, scattered cells with non-granular cytoplasmic brown staining which masked the small and inconspicuous nucleus were observed in 2 SLN of 2 cases. Such cells exhibiting benign morphology were also observed as scattered single cells in permanent sections immunostained with 'the cocktail'. These cells were interpreted as negative without significant challenge, but their exact nature could not be established. The possibility of singly scattered nevus cells was considered. Contrary to capsular nevus cells, such cells may be detached easily and picked up by the glass slide during the preparation of imprint smears. In some cases, they may cause an interpretation dilemma leading to indeterminate results even with permanent sections. In 2 patients, 2 SLNs were interpreted as indeterminate with immunostained imprint smears. They were interpreted as negative by the immunohistochemical evaluation of permanent sections. Cytomorphologically, the rare doubtful cells present in immunostained imprint smears were consistent with mast cells. They were also present in the respective negative controls (Figure 4 ). As endogenous peroxidase activity could not be blocked completely during the short endogenous peroxidase blocking step in the rapid protocol, non-melanoma cells such as mast cells may show brown staining in some cases. Familiarity with the morphological spectrum of immunostained tumor cells (Figure 3 ) and other non-specifically stained cells including mast cells (Figure 4 ) in immunostained imprint smears should prevent the indeterminate interpretation of these cells in future. The positive control smears may be prepared from time to time utilizing fresh, unfixed melanoma tumors for long term availability. Alternatively, the smears of melanoma tumor cell lines immunoreactive to individual components of the cocktail may be used after processing and fixing similar to test smears. These smears may be dehydrated and coverslipped (Table 2 ). Coverslipped positive control smears could be archived at room temperature for extended time periods. Coverslips can be removed by immersing the slides in xylene (usually 24 hours) to dissolve the mounting medium and to loosen the glass coverslip from the slide. We have used such smears after removing the coverslip as positive controls up to 1 year after they were originally prepared without affecting immunoreactivity for most of the commonly used immunomarkers including the 'MCW melanoma cocktail' (personal experience). Since a positive control had to be processed in advance by removing the coverslip in xylene, a notice at least one day prior to the intraoperative evaluation was required routinely as a part of the protocol. Imprint smears are easy and quick to make without incurring significant expense. They are faster than frozen sectioning and help prevent the loss of tissue associated with cryosectioning. Frozen-sectioning of lymph nodes is frequently problematic because of fat, either adjacent to or in the lymph node. These problems are circumvented with imprint smears, which would also prevent problems associated with the interpretation of final permanent sections of frozen tissue. For billing and reimbursement purpose, the rapid intraoperative evaluation of immunostained imprint smears may be coded with existing CPT (Current Procedural Terminology) codes- 88329 for the intraoperative consultation, 88161 for the preparation and processing of the imprint smears, and 88342 for the immunostaining of the imprint smears with interpretation [ 36 ]. As a future prospect, a 'cocktail' of directly conjugated individual antibodies (with a peroxidase or similar indicator system) used for a one step immunostaining method resulting in a significant reduction in immunostaining time (up to 6 minutes) with fewer staining steps would simplify the procedure [ 37 ]. As discussed above, the rapid protocol may not block the endogenous peroxidase. Rapid blocking of endogenous peroxidase with specific inhibitors / blocking agents, without affecting the cytomorphology, could prevent the non-specific staining of mast cells. This would improve the interpretation speed and confidence by eliminating the distraction factor of non-specifically stained cells thus reducing the chances of indeterminate interpretations and simplify the learning curve. In summary, air-dried imprint smears which were postfixed in alcoholic formalin following saline rehydration were optimal for immunocytochemical evaluation with the 'MCW melanoma cocktail'. Wet-fixed smears did not compromise the immunoreactivity of 'the cocktail', but they were difficult to prepare without air drying artifact and non-specific background staining. Capsular melanocytic nevi did not show false positive results. The rapid evaluation of imprint smears immunostained with the 'MCW melanoma cocktail' is reliable for the intraoperative evaluation of cutaneous melanoma SLNs for melanoma metastases. List of abbreviations ABC, avidin-biotin-peroxidase complex; DAB, Diaminobenzidine Hydrochloride; H&E, hematoxylin and eosin; MCW, Medical College of Wisconsin; SLNs, Sentinel Lymph Nodes. Competing interests None. Authors' contributions VS conceived, designed, carried out the entire study in addition to the standardization of MCW melanoma cocktail protocol and preparation of manuscript. RK participated in its design and coordination. GD performed the immunohistochemical staining including standardization of the MCW melanoma cocktail with VS and SK. WD organized the clinical participation including recruiting of cases and arranging patients consent. All authors read and approved the final manuscript. Supplementary Material Additional file 1 Higher resolution - (for high speed connection) Click here for file Additional file 2 Lower resolution - (for low speed connection) Click here for file Additional file 3 Screen shots Click here for file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524024.xml |
535897 | Prevention of genital herpes in a guinea pig model using a glycoprotein D-specific single chain antibody as a microbicide | Background Genital herpes (GH) is a recurrent sexually transmitted infection (STI) that causes significant morbidity and is also the major source of herpes simplex virus (HSV) in cases of neonatal herpes. Vaccination is a current goal which has had limited success so far in preventing GH and microbicides offer an attractive alternative. Treatment of primary disease cannot prevent establishment of latent infections and thus, cannot prevent subsequent recurrent disease. Recently, many of the molecular events leading to entry of HSV into cells have been elucidated, resulting in the description of a number of herpesvirus entry mediators (HVEMs) that interact with HSV glycoprotein D (gD) on the surface of virions. Described here is a strategy for interrupting the spread of HSV based on interfering with these interactions. The hypothesis addressed in the current report was that single chain antibody variable fragments (scFv) that interrupt associations between gD and HVEMs would not only prevent infection in vitro but could also be used as microbicides to interfere with acquisition GH. Results and Conclusions Here we show that a scFv derived from a particular hybridoma, DL11, not only inhibits infection in vitro but also prevents development of GH in a guinea pig model when applied intravaginally in an inert vehicle. Comparison of different anti-gD single chain antibodies supported the hypothesis that the activity of DL11-scFv is based on its ability to disrupt the associations between gD and the two major receptors for HSV, nectin-1 and HveA. Further, the results predict that bacterial expression of active single chain antibodies can be optimized to manufacture inexpensively a useful microbicidal product active against HSV. | Background GH is generally caused by HSV type 2 (HSV-2), though HSV type 1 (HSV-1) is increasingly recognized as a significant cause of primary infections [ 1 ]. Throughout the last few decades there were substantial advances in understanding the epidemiology of genital HSV infections. Primary infection is almost always self-limited but on healing virus is not eliminated from the host but rather, viral genomes remain in a latent (dormant) state in sensory neurons innervating initially infected skin and mucous membranes [ 2 , 3 ]. The significance of latency is that it is a reservoir of infection that can periodically reactivate, causing virus to travel down nerve fibers to skin or mucous membranes in the dermatome of primary infection. This may be manifest clinically as recurrent GH or more frequently, causes unrecognized shedding of infectious HSV [ 4 - 7 ] which despite being unrecognized is responsible for the majority of new HSV-2 infections [ 8 ]. The epidemiology is further complicated by the fact that many primary infections are asymptomatic or unrecognized, which has the important implication that the first clinical presentation of GH, often referred to as the initial episode, may be caused by a recurrence of a prior asymptomatic primary infection [ 9 ]. In the latter half of the 20 th century, there were great strides in antiviral therapy for GH but unfortunately, treating primary disease does not prevent establishment of infection [ 10 ] and thus, cannot prevent subsequent recurrent disease. Barrier contraception provides some protection but its efficacy remains unclear [ 11 ] owing to the complex nature of HSV pathogenesis, in which virus is shed frequently and asymptomatically from multiple sites below the waist [ 5 ]. Hence condoms are not as effective at preventing transmission of GH as they are for other sexually transmitted infections. Vaccination is a current goal which has had limited success to date. A recent trial of a glycoprotein D-based sub-unit vaccine protected only double (HSV-1 and 2) seronegative women but not men [ 12 ]. Further, protection was mainly measured by prevention of primary disease rather than infection. It is known that treating primary disease does not prevent establishment of latency and consequently, the long term efficacy of this vaccine against subsequent recurrences remains unknown. Thus, the number of strategies for preventing sexual transmission of GH is limited. Recently, there has been considerable interest in topical microbicides as a potentially attractive alternative to vaccination for prevention of sexually transmitted infections, including GH [ 13 ]. Women are able to control the use of vaginal microbicides and several products are currently being used or tested, including acid buffers and sulphated polymer-based inhibitors or surfactants [ 14 ] like nonoxynol-9 (N-9) [ 13 ]. N-9 has been used as a spermicide for 30 years and was thought to provide some protection against gonorrhea and chlamydia, a view was recently proven to be erroneous [ 14 ]. A major factor limiting the efficacy and long-term viability of N-9 and similar chemical compounds as topical agents is their irritant effects on the vaginal epithelium [ 15 ]. Further, recent data suggest that N-9, contrary to prior belief, is not effective at protecting against HIV but rather it was shown to increase rather than decrease the risk of acquiring HIV in some populations studied, particularly women at high risk of infection [ 14 ]. An evolving strategy that may be useful for preventing specific sexually transmitted viral infections is blocking of virus entry into cells or subsequent inhibition cell-to-cell spread. Many of the molecular events leading to entry of HSV into cells have now been unraveled, resulting in the description of two prominent cell-surface herpesvirus entry mediators (Hve-A and nectin-1, also known as Hve-C) that interact with HSV glycoprotein D (gD) on the surface of virions [ 16 - 20 ]. In a recent study [ 21 ], nectin-1 was shown to be expressed in the vaginal epithelium of humans throughout the estrous cycle. In contrast, in mice nectin-1 was expressed in vaginal epithelium only during the stage of estrous at which they are susceptible to HSV. Using a mouse model of GH, pre-incubation of HSV-2 with soluble recombinant nectin-1 was shown to block entry of virus through vaginal mucosa [ 21 ], suggesting the importance of nectin-1 in mediating entry of HSV into the female genital tract. Hve-A and nectin bind to conformationally overlapping regions of gD and we were able exploit this information together with the results of prior studies that had mapped the sites on gD recognized by a panel of monoclonal antibodies [ 22 - 26 ]. Antibody DL11 was of particular interest because it binds to an epitope on gD that blocks the interactions between gD and both Hve-A and nectin-1 [ 19 ] (figure 1 ). We show here that a single chain antibody variable fragment (scFv) constructed from DL11 neutralizes HSV infection in vitro, inhibits cell-to-cell spread of virus and can be used to prevent infection in a guinea pig model of GH. Figure 1 Panel A: Hypothetical model illustrating the antigenic structure of gD and demonstrating the clustering of antigenic sites into seven groups, as defined by locations of amino acids bound by various monoclonal antibodies. Disulphide bonds location defined by braces. Diagram adapted with permission from Nicola et al, 1998 [22]. Of particular relevance to this study are the locations of sites VII (amino acid residues 11–19), which is bound by antibody 1D3, and site Ib, a discontinuous epitope that includes residues 222 to 252 that is bound by antibody DL11. Panel B: Diagram showing interface between N-terminal amino acids of gD and HveA and the approximate residues (blue) bound by monoclonal antibody 1D3 and, by inference, 1D3 scFv (adapted with permission from Connolly et al, 2003 [19]. Results Construction and expression of single chain antibodies against gD Four from the panel of anti-HSV gD hybridomas available were selected for scFv construction based on the known locations of their epitopes [ 22 ] (summarized in figure 1 ) and knowledge about the neutralization properties of the antibodies produced by them. Of particular note are the properties of DL11, which neutralizes both HSV-1 and HSV-2 in the absence of complement and antibody binding to its conformational epitope is known to disrupt the interactions of gD both with Hve-A and nectin-1. Also 1D3 binds to a group VII neutralizing epitope that directly interferes with the interface between gD and HveA (figure 1B ). A fifth scFv cassette, against carcinoembryonic antigen (CEA) was excised from a plasmid encoding an anti-tumor chimeric T-cell receptor, kindly provided by Hinrich Abken (Cologne University, Germany). For production of cDNAs, individual V L and V H regions from each hybridoma were reverse transcribed using primers near the V H -C H and V L -C L junctions. For PCR cloning these primers were paired with a panel of degenerate primers derived from V H or V L signal sequences (Table 1 ) that were able to amplify all hybridoma heavy and light chains tested so far (14/14) irrespective of antibody class or subclass (data not shown). PCR products were sequenced directly to facilitate design of new primer sets allowing, on re-amplification of hybridoma cDNAs, elimination of degenerate primer sequences introduced in the first reaction and introduction of 2/3 of a 15 amino acid hinge region comprising three repeats of four glycine and one serine residues (Figure 2 ). V L and V H are not covalently linked in nature but flexible hinges of this design and length were shown previously [ 27 ] to allow reconstruction of antibody binding sites when V L and V H are linked end-to-end (figures 3 , 4 ). Finally, the PCR products containing the overlapping hinge regions were ligated, PCR amplified and the resultant scFv cassette was TA cloned into pCR2.1TOPO. To generate the desired single chain antibodies, the cassettes were subcloned into the bacterial expression vector pET101-D. An antibody modeling algorithm, verified by the locations of the complementary determining regions, was used to predict the 3-D structures of all four of the anti-gD single chain antibodies. The results were consistent with reconstitution of the original antigen binding sites (e.g. figure 3 , DL11; others not shown). Table 1 Degenerate PCR primers used for amplification of V L (kappa) and V H (gamma). Nomenclature Primer sequences used for PCR reactions Signal sequence/framework primers Kappa 1 GGTGATATCGTGATRACMCARGATGAACTCTC Kappa 2 GGTGATATCWTGMTGACCCAAWCTCCACTCTC Kappa 3 GGTGATATCGTKCTCACYCARTCTCCAGCAAT Kappa 4 CTGWTGTTCTGGATTCCTG Kappa 5 GTGCTCTGGATTCGGGAA Kappa 6 TCAGCTTCYTGCTAATCAGTG Kappa 7 TGGGTATCTGGTRCSTGTG Kappa 8 GTTTCMAGGTRCCAGATGT Kappa 9 TGTTTTCAAGGTRCCAGATGT Kappa 10 CTSTGGTTGTCTGGTGTTGA Kappa 11 TGCTKCKCTGGGTTCCAG C region kappa primer TGGTGGGAAGATGGA Signal sequence/framework primers Gamma 1 GAGGTGAAGCTGCAGGAGTCAGGACCTAGCCTGGTG Gamma 2 AGGTVMAACTGCAGVAGTCWGG Gamma 3 AGGTVVAGCTGCAGVAGTCWGG Gamma 4 ACTGCAGGTRTCCACTCC Gamma 5 RCTACAGGTGTCCACTCC Gamma 6 GCYACAGMTGTCCACTCC Gamma 7 ACTGCAGGTGTCCTCTCT Gamma 8 RCTRCAGGYGTCCACTCT Gamma 9 CCAAGCTGTGTCCTRTCC Gamma 10 TGTTGACAGYCVTT CCKGGT Gamma 11 TAYTTTAAAARGTGTCMAGTGT Gamma 12 CTYTTAAAAGGKGTCCAGWG Gamma 13 CYTTTAMATGGTATCCAGTGT Gamma 14 ATGGCAGCWGCYCAAAG Gamma 15 CTTTTAAAAGWTGTCCAGKGT Gamma 16 CTTCCTGATGGCAGTGGTT C region gamma primer TAACCCTTGACCAGGCATCC Key to degenerate nucleotides: R = A+G; M = A+C; W = A+T; K = G+T; S = G+C; Y = C+T; H = A+T+C; B = G+T+C; D = G+A+T; N = A+C+G+T; V = G+A+C Figure 2 Panel A: Structure of an scFv cassette spliced using a (Gly 4 Ser) 3 hinge. Panel B. Alternative glycine codons were used in the overlapping region of the hinge to avoid production of completely overlapping regions, thereby generating a sub-optimal (Gly 4 Ser) 2 hinge. Figure 3 Single plain view of a 3-D model of DL11 scFv, showing its predicted structure. Panel A: Strand view, colored by group, demonstrates relative orientation of the kappa (top) and gamma (bottom) chains, which shows the positions of residues to which the (Gly 4 Ser) 3 hinge is attached. Panel B: Wireframe image illustrating hinge attachment sites on one side of the molecule (linked by dashed line) and clustering on the opposite side (inside the circle) of the complementary determining regions (CDRs) predicted by the Kabat antibody database. The clustering of CDRs suggests correct conformation of the molecule with formation of an antigen binding site. Figure 4 Western blot demonstrating expression of DL11 scFv by E, Coli. BL21 cells were transfected with p-TOPO10 containing the scFv cassette. Bacterial lysates were purified using a nickel chelation column and the reaction with anti-V5 of total lysates and various fractions from the column are shown. Lane 1, unpurified total bacterial lysate; lane 2, nickel column flow through; lanes 3 and 4, saline washes; lane 6, eluate from Ni beads; lane 7, bacterial supernatant; lane 8, scFv remaining on nickel column after elution; lane 9: supernatant from un-induced bacteria. Bacterial expression and extraction of anti-gD single chain antibodies The single chain antibodies were expressed in E. Coli strain BL21 using pET101-D (Invitrogen), which attaches hexa-His and V5 tags to expressed proteins for their isolation and identification. Bacteria were induced with IPTG, centrifuged and the supernatants tested for the presence of scFvs by western blotting using anti-His antibody (figure 4 ). Bacterial pellets were sonicated in phosphate buffered saline to release inclusion bodies and proteins were solubulized by addition of 6 M guanidine (BL21). Nickel bead chelation was used to extract the His-tagged protein. Western blots of eluates from nickel beads (e.g. DL11 scFv from DL21; Fig. 4 , lanes 6 and 7) identified scFvs that were released by this procedure. They were generally isolated at concentrations of 500–1000 µg/ml from BL21. Re-folding and intra-chain disulphide bond formation were maximized by gradually reducing guanidine concentration by step-wise dialysis from 6 M initially to 3 M, then 2 M, 1 M, 0.5 M and finally 0 M, with addition of L-arginine and oxidized glutathione (GSSG) in final two steps [ 28 ]. The ability of the single chain antibodies produced in this way to bind their target antigen was tested by determining their reaction with plastic bound gD by ELISA. Binding ratios were calculated in relation to the background binding of CEA scFv (e.g. DL11-based scFv; figure 5 ) Figure 5 Binding of scFv to plastic bound gD. Binding ratios of DL11 scFv to gD compared with an irrelevant (CEA) scFv at the same protein concentrations. Selected anti-gD single chain antibodies neutralize HSV in vitro The hypothesis that selected single chain antibodies can block infection of cells in vitro by reacting with an epitope that disrupts the interface between gD and HVEMs was tested by comparing the activities of the various bacterially expressed anti-gD scFv in a Vero cell-based HSV-1 plaque reduction assay. A scFv directed against an epitope on carcinoembryonic antigen was included as an unrelated control. The results showed that pre-incubation of virus with DL11 and 1D3 scFvs inhibited plaque formation with decreasing efficiency. DL6 scFv showed minimal but reproducible activity (data not shown), whereas the other scFvs tested (DL2 and CEA) had no plaque reducing capability at all (figure 6 ). Against HSV-2, only DL11 showed neutralizing activity in a similar plaque reduction assay (data not shown), confirming the type common nature of its epitope. In addition to inhibition of plaque formation, pre-incubating HSV-2 with 100 µg/ml DL11 caused an 80% reduction in plaque numbers and a ~50% reduction (figure 7 ) in the size of plaques (0.95 ± 0.3 mm with DL11scFv vs. 1.9 ± 0.4 mm without, respectively). The same was true for HSV-1 and DL11 (not shown). It was concluded that DL11scFv could not only block infection of cells with HSV but also was able to inhibit cell-to-cell spread of virus. Figure 6 Specific reduction of HSV-2 plaque numbers by incubation of virus with anti-gD scFv. Vero cells were pre-incubated with approximately 120 PFU HSV-2, strain G with single chain antibodies generated from hybridomas D11 (¦), 1D3 (?), DL2(?) and an irrelevant CEA-specific construct (?). Figure 7 Reduction in plaque size in the presence of DL11 scFv. Mean plaque size in absence of scFv (Panel A) was 1.9 ± 0.4 mm compared with 0.95 ± 0.3 mm in presence of 100 mg/ml DL11 scFv (Panel B). Figures represent mean of 100 plaques ± standard deviation. Bar = 1 mm. Protection against HSV type 1 and type 2 GH by administration of a DL11-based single chain antibody before infection with virus The HSV type-common and startling in vitro activities of single chain antibodies derived from hybridoma DL11 prompted us to examine the ability of DL11scFv to protect against vaginal HSV disease, using a well established guinea pig model of GH [ 29 , 30 ]. The vehicle selected for these preliminary studies was 1% carboxymethylcellulose because this is an inert compound that is used for its viscosity in our routine plaque assays. A pilot experiment was done with HSV-1, in which BL21 produced DL11 and DL2 single chain antibodies (0.5 mg/ml) were each instilled into the vaginas of guinea pigs (1 ml/animal). Approximately 20 seconds later the guinea pigs were challenged with 5 × 10 6 PFU HSV-1, strain SC16 and monitored for development and severity of primary disease. The result (figure 8 ) showed that DL11-based scFv completely protected the animal from lesion development whereas DL2-based scFv appeared to have, as expected, no effect. Figure 8 Effect of DL11 scFv on HSV-1 genital disease in guinea pigs. Panel A: Blisters of GH 5 days after instillation of HSV-1 into vaginal vault. Several areas of ulceration with surrounding erythema are visible bilaterally (e.g. arrows); Panel B: Complete protection against HSV-1 by prior instillation, immediately before HSV challenge, of 1 ml CMC containing DL11 scFv (500 µg/ml). Next a more ambitious test of microbicidal activity was attempted, using HSV-2 rather than HSV-1 and a longer interval (20 minutes) between scFv instillation and challenge (Table 2 ). Two groups of 20 guinea pigs were each administered either DL11 or DL2 (control) scFv (1 ml/guinea pig). All animals were challenged with 10 6 PFU of HSV-2, strain G and monitored daily as before. All except one animal were completely protected by DL11 scFv compared with DL2 scFv, all of which developed moderate to severe disease, scored as described in methods (p = <0.0001; Mann Whitney test). Table 2 Prevention of GH in guinea pigs by DL11 scFv. Pre-treatment Severity of lesions Mean lesion score (n = 20) 500 µg/ml DL2 scFv 20 minutes prior to infection 2, 2, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4. 3.55 ± 0.153 * 500 µg/ml DL11 scFv 20 minutes prior to infection 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0. 0, 0, 0, 3. 0.15 ± 0.15 * * p < 0.0001 (Mann-Whitney test) Discussion In practice, the rapidity of isolation and cloning of scFv into a bacterial expression vector (approximately one week) by the procedure described here allowed expeditious activity assessments to be made for the different constructs. Bacterial protein expression systems are widely used for the production of recombinant proteins and problems are often encountered with disulphide bond formation. Whilst there are no covalent bonds between heavy and light chain sequences in immunoglobulin hypervariable domains, intra-chain disulphide bonds can, to varying degrees among different antibodies, influence conformation of the antigen binding site [ 31 ]. Thus, failure to form intra-chain disulphide bonds has the potential to disrupt antigen binding and could also detrimentally affect stability of the molecule. For this reason, a previously reported method [ 28 ] was adapted to promote formation of intra-chain disulphide bonds in vitro, using protein extracted from bacterial inclusion bodies. After isolation of inclusion bodies by sonication of bacteria, proteins were solubulized with 6 M guanidine hydrochloride, the concentration of which was gradually reduced to zero by a stepwise daily dialysis routine. In all cases we tried, this procedure generated soluble hexa-His tagged single chain antibody fragments, which have an approximate molecular weight of 34 kD, at concentrations of approximately 750 – 1000 µg/ml. Precipitation of proteins in the final dialysis step tended to occur above concentrations of 1000 µg/ml and was prevented by careful monitoring of the sample volume to ensure concentrations stayed below this critical threshold. Entry of HSV into cells is known to be mediated through interactions between gD and 3-O-sulfated heparan and one or more specific entry mediators, HveA, nectin-1 and nectin-2 [ 32 ]. Overall, the results of plaque reduction assays in vitro were compatible with the hypothesis that significant interference with the binding of gD with HVEMs can be achieved with a single chain variable fragment selected according to the known properties of their parent antibodies. This is the first direct evidence that neutralization of HSV can be a property of certain antigen binding domains alone, a corollary of which is that no other regions of the antibody need be required to neutralize the virus. Of the five scFvs tested in this study, the most effective was produced from antibody DL11, which is known to interfere with binding both to HveA and nectin-1. Neutralization in vitro by a scFv constructed from antibody 1D3, which binds to a site on gD that overlaps the binding site for HveA, was significantly less efficient than that seen with DL11 scFv, presumably because, in the presence of 1D3 scFv, HSV was still able to utilize nectin-1 as a receptor. As expected, scFvs derived from DL2, a non-neutralizing monoclonal antibody, and a scFv reactive against CEA showed no activity in the plaque reduction assay. The weak activity of the construct made from DL6 correlates with the known weak neutralizing property of the native antibody, which is presumed to be the result of conformational changes induced it's by its binding to gD. The data presented here correlate with the prior finding that mice can be protected against HSV-2 by topical administration of antibody [ 33 , 34 ] and a subsequent report from Zeitlin et al [ 35 ] that mice were protected against HSV-2 transmission by intravaginal administration of an IgG 2a monoclonal anti-gD antibody and its IgA switch variant. Here, these observations are extended in several respects. The epitope on gD recognized by the most effective scFv, that constructed from antibody DL11, was defined as one that interferes with binding of gD with two major mediators of herpesvirus entry into cells, namely HveA and nectin-1. In this respect, attention is drawn to the recent report of Linehan et al [ 21 ] that nectin-1 is expressed in the genital tracts of mice and humans and soluble nectin-1 can block entry of HSV into vaginal epithelium. These data [ 21 ] together with the unprecedented protection of guinea pigs by DL11scFV shown here strongly implicate nectin-1 as a critical mediator of HSV the entry genital epithelium and in fact it is suggested here that nectin-1, rather than other herpesvirus entry mediators, likely play a dominant role in genital tract infection. The protective activity of a scFv established with certainty that the constant regions of anti-gD antibody molecules are not required for protection against HSV. This finding has the important consequence of eliminating the complement binding activity of IgG, which will greatly limit the potential for unwanted inflammatory side effects of topically administered anti-gD preparations, an important advantage if they are to be used clinically. The specific nature of anti-herpes scFv and the ability to choose an inert formulation has two potential advantages over other microbicides. First is selected high specific activity against HSV and second is that they are not irritating to the genital tract. Their murine derivation is not anticipated to be a problem with topical use, but humanization of the hypervariable region is possible by grafting the complementary determining regions onto a human framework, This is an option should their systemic use ever be considered. Of particular interest may be the use of microbicidal gels prior to delivery for the prevention of neonatal herpes. The inert nature of single chain antibodies, combined with a suitable vector, should enable their widespread use in this context among HSV-2 seropositive mothers. These are important considerations given the high prevalence of GH and its frequent asymptomatic nature. In summary, we believe that single chain antibodies against HSV merit further study and development as topical microbicides. The production of active molecules in bacteria makes their use a feasible and relatively inexpensive prospect. Conclusions Single chain antibodies against HSV gD could be synthesized readily from several IgG secreting hybridomas using degenerate immunoglobulin heavy and light chain immunoglobulin primers that hybridized to regions flanking the complementary determining regions, which determine antigen specificity. Two mechanisms of interference with infection were evident when DL11 scFv was examined in detail. First, the number of plaques produced by virus could be inhibited by up to 90% when reacted with HSV prior to infection of Vero cells, indicating that scFv neutralized virus prior to establishment of productive infection. This result also suggested that nectin-1 and HveA, the binding of which are both blocked by DL11, are the main mediators of virus entry into Vero cells. 1D3, which interferes specifically with the interface of gD with HveA, was effective to a lesser extent. Second, in addition the striking ability of DL11 scFv to neutralize virus inoculums, this particular construct reduced plaque size significantly, from which it was concluded that cell-cell spread of HSV was also inhibited. This observation could have implications for therapeutic use of single chain antibodies in the future and may have enhanced the performance of DL11 scFv as a microbicide in the guinea pig model. This result was mediated by suboptimal scFv concentrations for virus neutralization, implying that lower concentrations of DL11 scFv may be required to interfere with intercellular spread of virus than to block entry. The finding that DL11 scFv was active for 20 minutes, the maximum time tested, when instilled into the vaginal vault was considered encouraging for future development of scFv as microbicides and the observation merits further consideration of the vehicle used. Slow release formulations may be appropriate depending on their cost. Overall, it appears that selected single chain antibodies are promising candidates for interfering with binding of gD to HVEMs and studies in a guinea pig model of GH suggest that they may comprise a plausible strategy for preventing transmission of GH. Methods Generation of scFvs Single chain antibodies were constructed from four anti-gD secreting hybridomas, DL11, DL6, DL2 and 1D3. An additional scFv, directed against carcinoembryonic antigen (CEA) served as an independent control. Messenger RNAs from 5 × 10 5 - 10 6 hybridoma cells were isolated using Trizol (Invitrogen, CA) and cDNAs were generated by reverse transcription with Taq polymerase ('Expand High Fidelity Taq polymerase' ; Roche, IN). RT was primed with anti-sense oligonucleotides designed to anneal either to mouse kappa light chain or heavy chain constant region sequences, just downstream of the J-C junction (table 1 ). Light and heavy chain hypervariable regions (V L and V H ) were amplified by priming 'sense' PCR reaction products with panels of oligonucleotides (OGNs) designed from Kabat database sequences to be complementary to kappa (light chain) and gamma (heavy chain) signal or framework sequences (table 1 ). In practice, pools of 11 degenerate OGN sequences were found to be sufficient to prime 100% of kappa chain reactions (14/14 hybridomas regardless of subclass). Similarly, a pool of 14 degenerate OGNs successfully amplified the gamma chains from these hybridomas. From each hybridoma, the resulting V L and V H cDNAs were sequenced and new specific primers were designed each of which included 2/3 of the fifteen amino acid (Gly 4 Ser) 3 flexible hinge region, allowing the variable regions to be amplified and spliced together reconstituting the antigen binding site on reconformation (figures 2 , 3 ). To prevent complete overlap of the complementary hinge sequences, which would result in the introduction of a sub-optimal 10 amino acid (Gly 4 Ser) 2 intervening segment, alternative glycine codons were used in each component of the hinge. Four of the scFvs were TA cloned into the bacterial expression vector pET101/D-TOPO (Invitrogen, Carlsbad, CA) which generates hexa-His tagged proteins after expression in vitro. Expression of single chain antibodies in bacteria Proteins were expressed in IPTG-induced E. Coli BL21 [DE3] (Invitrogen), released by sonication in PBS and inclusion bodies were separated by centrifugation. Proteins in inclusion bodies were solubulized with 6 M guanidine HCl and purified by metal chelation. A stepwise dialysis procedure with addition of GSSG (oxidized glutathione; Sigma) and L-arginine in the final two steps was used to assist in the formation of intra-chain disulphide bonds in order to optimize re-conformation and stability of the scFvs [ 28 ]. Protein concentrations were measured using the BCA method (Pierce). ELISA to quantify binding of scFv to gD Microtiter plate wells were coated with soluble gD (6 µg/ml) and then blocked with 1% skimmed milk. After incubation with serial two-fold dilutions of scFv, binding was detected with anti-V5, the alternative tag on the scFv, because the recombinant gD used in the assay was, like the single chain antibodies, tagged with hexa-His. Binding ratios were calculated in relation to an irrelevant (CEA-specific) scFv. Virus growth, titration and plaque neutralization assays HSV-1 (strain SC16) and HSV-2 (strain G) were grown and titrated in Vero cells as described [ 36 , 37 ]. Titers were determined using a standard plaque assay [ 38 ]. Cells were grown and maintained in Dulbecco modified Eagle medium supplemented with 10% (growth medium; GM) or 1% (maintenance medium; MM) fetal bovine serum. A plaque reduction assay was done in Vero cells to assess the neutralizing capabilities of each scFv. Briefly, 100–200 plaque forming units (PFU), diluted in MM, of either HSV-1 (strain SC16) or HSV-2 (strain G) were incubated at room temperature for 1 hour with serial ten-fold dilutions of each scFv in a total volume of 1 ml. After gentle shaking with 3 × 10 6 Vero cells for a further 1 hour the samples were plated in 6 cm dishes (Nunc) in a total volume of 5 mls of GM containing 2% carboxymethylcellulose (CMC). Plaques were enumerated after 3 days incubation at 37°C in a 5% CO 2 atmosphere. Guinea pig model of GH The microbicidal properties of scFv were tested using a guinea pig model of GH. Female outbred Hartley guinea pigs weighing 350–400 grams were obtained from Charles River laboratories (Wilmington, MA). Prior to inoculation of each guinea pig with virus, the introitus was opened with a calcium alginate swab moistened in physiological saline and 1 ml of 1% CMC containing either DL2 scFv or DL11 scFv at a final concentration of 500 µg/ml, was instilled using a pipette with a plastic tip. CMC was used as a vehicle to facilitate retention of the scFv in the vaginal vault. At various times thereafter, animals were challenged with 10 6 PFU HSV-1 (strain SC16) or HSV-2 (strain G). Over the ensuing two weeks lesions were scored on a scale of 0–4 (0 = no lesion; 1 = erythema and swelling only; 2 = small vesicles <2 mm; 3 = coalescent or large vesicles >2 mm; 4 = ulceration and maceration). All experiments were done according to the guidelines laid down in The NIH Guide for Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee. Competing interests The author(s) declare that they have no competing interests. Authors' contributions AS conceived and coordinated the work described and wrote the manuscript. JC was responsible for the experiments described and SKD provided technical support. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535897.xml |
529266 | Length of sick leave – Why not ask the sick-listed? Sick-listed individuals predict their length of sick leave more accurately than professionals | Background The knowledge of factors accurately predicting the long lasting sick leaves is sparse, but information on medical condition is believed to be necessary to identify persons at risk. Based on the current practice, with identifying sick-listed individuals at risk of long-lasting sick leaves, the objectives of this study were to inquire the diagnostic accuracy of length of sick leaves predicted in the Norwegian National Insurance Offices, and to compare their predictions with the self-predictions of the sick-listed. Methods Based on medical certificates, two National Insurance medical consultants and two National Insurance officers predicted, at day 14, the length of sick leave in 993 consecutive cases of sick leave, resulting from musculoskeletal or mental disorders, in this 1-year follow-up study. Two months later they reassessed 322 cases based on extended medical certificates. Self-predictions were obtained in 152 sick-listed subjects when their sick leave passed 14 days. Diagnostic accuracy of the predictions was analysed by ROC area, sensitivity, specificity, likelihood ratio, and positive predictive value was included in the analyses of predictive validity. Results The sick-listed identified sick leave lasting 12 weeks or longer with an ROC area of 80.9% (95% CI 73.7–86.8), while the corresponding estimates for medical consultants and officers had ROC areas of 55.6% (95% CI 45.6–65.6%) and 56.0% (95% CI 46.6–65.4%), respectively. The predictions of sick-listed males were significantly better than those of female subjects, and older subjects predicted somewhat better than younger subjects. Neither formal medical competence, nor additional medical information, noticeably improved the diagnostic accuracy based on medical certificates. Conclusion This study demonstrates that the accuracy of a prognosis based on medical documentation in sickness absence forms, is lower than that of one based on direct communication with the sick-listed themselves. | Background The increasing rate of sick leave experienced in most Western countries challenges insurance companies, employers, and public authorities to identify measures to reduce burdens at the individual, workplace and societal levels. To reduce the expenses of sick leave and the risk of expulsion from work, the Norwegian government introduced legislation in 1993 that anticipated early and more vigorous interventions of the Norwegian National Insurance Scheme [ 1 ]. The Norwegian Public Report no. 27 [ 2 ], 2000, underscored the importance of early intervention by the National Insurance Offices (NIOs). A major challenge for the NIOs is to identify newly sick-listed individuals at risk of prolonged sick leave, and who are therefore potential candidates for rehabilitating interventions. The selection process is currently based on information in medical sickness certificates supplied by access to the register of previous sickness benefits. A medical sickness certificate (Sickness Certificate 1; SC1) is required if sick leave exceeds 3 days, and after 8 weeks an extended medical certificate is mandatory (Sickness Certificate 2; SC2) [ 3 ]. In addition to diagnosis and certified period, the majority of SC1s contain information on the occupation and employee, whereas information on chronic disease, previous sick leave episodes, prognosis and comments are more scattered. SC2s include updated medical information on work ability, planned diagnostics and treatments, and on the prognosis. The value of this information as a guideline for selective intervention has, however, never been established, either as an indicator of potential prolonged absence, or as an indicator of the need for occupational or vocational rehabilitation [ 4 ]. Based on the current practice with identifying sick-listed individuals at risk of long-lasting sick leaves, the objectives of this study were to inquire diagnostic accuracy of predictions within the NIOs, and to compare their predictions with the self-predictions of the sick-listed. Methods In October and November 1997 and March and April 1998, newly sick-listed persons with musculoskeletal or mental disorders (ICPC, L- and P- diagnoses) [ 5 ] were included consecutively if they were certified sick for longer than 2 weeks (Figure 1 ). Five hundred persons were included in each period. The study took place in the cities of Tromsø and Harstad in Northern Norway. The total length of sickness benefits was registered during the following year in the National Sickness Benefit Register. Missing data on the length of sick leave reduced the number of included subjects to 993. The mean ages of these 391 men and 602 women were 41.4 and 39.7 years, respectively. Musculoskeletal disorders were the main reason for sick leaves (83% of the cases). Figure 1 Flow-chart. Flow-chart of inclusion, and the different assessments of expected length, of the included sick leaves after 2 and 8 weeks of sick leave. A total of 495 randomly selected persons received a questionnaire on the expected length of their ongoing sick leave period. The answer categories were: less than 4 weeks, 4 to 7 weeks, 8 to 11 weeks, 12 to 15 weeks, 16 to 25 weeks, 26 to 51 weeks, and at least 1 year. Some 152 persons (30.7%), called the responder group, returned the questionnaire with this question filled in. Based on SC1s available after 14 days of sick leave, two NIO officers without formal medical competence, but experienced in working with sick-listed persons, and two experienced physicians working part time as insurance medical officers (NIO medical consultants), assessed the expected length in each of the 993 ongoing sick leave cases. In 496 randomly chosen cases, the NIO assessors had additional access to information on sick leave periods during the previous 3 years. Of potentially 1986 assessments in each profession, the officers and medical consultants had 18 and 25 missing assessments, respectively. SC2s became available in 322 of the 459 cases where sick leave exceeded 8 weeks, and the NIO assessors reassessed these cases. Reproducibility of assessments by medical consultants were analysed in 20 cases reassessed by the two NIO medical consultants, and assessed by another eight of their colleagues. Observed length of sick leaves The reference standard lengths of individual sick leaves within 1 year were collected from the National Sickness Benefit Register. Sick leaves interrupted by only 1–2 days without sickness benefits, typically on weekends, were registered as a single period. The observed length of sick leave thus comprised the total period of continuous full-time or part-time absence due to sickness within 1 year. Statistics The diagnostic accuracy of predicted lengths was compared on the basis of sensitivity, specificity, likelihood ratio and the area under the receiver operating characteristics curves (ROC area) [ 6 , 7 ]. The non-parametric standard error and 95% CI for the ROC area were calculated in SPSS-11. The ROC curve represents plots of the true-positive rate (sensitivity) and the false positive rate (1 – specificity) at the average of two consecutive categories of the assessments (>= 0 weeks, >= 4 weeks, >= 8 weeks etc). The ROC curves of the mean assessment by NIO officers and medical consultants include even intermediate points representing half categories. The predictive validity is presented as sensitivity, specificity, positive predictive value (PPV) and likelihood ratio at different thresholds, cut-offs, in predicted length [ 8 ]. Reliability of predicted length was analysed with agreement between assessors, the kappa value [ 9 , 10 ]. Approval The Regional Ethical Committee approved the protocol, and the Norwegian Data Inspectorate licensed the necessary register of sick-listed subjects. Results The mean observed continuous sickness absence was 100.8 days (median 48 days). Sick leaves in females lasted a mean of 105.1 days, compared to 94.6 days in men (medians 55 and 43 days, respectively). The mean length among persons with musculoskeletal disorders was 90.2 days in 335 males and 108.6 days in 489 females. The mean length among persons with mental disorders was 120.6 days in 56 males and 90.0 days in 113 females. The mean length of the sick leave in the responder group was 107.4 days (95% confidence interval, CI, 88.7–126.1 days), compared to 92.4 days in the 343 non-responders. Stratified analysis revealed longer mean sick leaves among responders 40 years and younger, of 109.3 days (95% CI 81.4–134.5 days), compared to the 79.3 days (95% CI 65.6–93.1 days) in non-responders. Stratification on gender or musculoskeletal or mental disorders did not reveal any significant differences in the length of sick leave between responders and non-responders. All assessors, including the sick-listed themselves, systematically overestimated the length of short sick leaves (lasting 4–11 weeks) and underestimated the length of long sick leaves (exceeding 16 weeks; Table 1 ). The proportions of sick leaves lasting longer than 8, 12 or 26 weeks did not differ significantly between the responder group and the rest. Table 1 Categorical distribution of observed and predicted length of sick leave. Observed and predicted length of sick leaves in seven categories for all participants (n = 993) compared to the responder group (n= 152). The assessments of National Insurance medical consultants and officers are grouped according to proportions of persons predicted in each category. All participants Proportion according to Responder group Proportion according to Length of sick leave categories Observed length % Assessed by medical consultants % 95% CI Assessed by officers % 95% CI Observed length % 95% CI Assessed by medical consultants % 95% CI Assessed by officers % 95% CI Assessed by sick-listed % 95% CI < 4 weeks 31.7 27.6 25.7–29.7 18.9 17.2–20.7 29.6 22.5–37.5 32.2 27.0–37.8 20.9 16.4–25.9 25.0 18.3–32.7 4–7 weeks 22.0 41.8 39.6–44.0 36.8 34.7–39.0 25.0 18.3–32.7 40.9 35.3–46.7 33.4 28.1–39.1 36.2 28.6–44.4 8–11 weeks 12.9 20.3 18.6–22.2 25.4 23.5–27.4 7.2 3.7–12.6 18.3 14.1–23.1 26.8 21.9–32.2 15.1 9.8–21.8 12–15 weeks 6.2 7.0 5.9–8.3 13.7 12.2–15.3 3.9 1.5–8.4 6.3 3.8–9.7 13.6 9.9–18.0 10.5 6.1–16.5 16–25 weeks 9.3 1.0 0.6–1.6 1.9 1.3–2.6 13.2 8.2–19.6 0.7 0.1–2.4 2.6 1.2–5.2 5.9 2.7–10.9 26–51 weeks 6.8 0.7 0.4–1.2 0.7 0.4–1.2 9.9 5.6–15.8 0.3 0.0–1.8 0.0 0.0–1.2 1.3 0.2–4.7 >= 52 weeks 11.1 1.5 1.0–2.1 2.5 1.9–3.3 11.2 6.7–17.3 1.3 0.4–3.2 2.6 1.2–5.2 5.9 2.7–10.9 Receiver operating characteristics of prediction The sick-listed subjects predicted sick leaves equal to or longer than 12 weeks more accurately than the NIO medical consultants and officers, as shown by the ROC curve in Figure 2 . The differences in ROC area between responders and non-responders were most marked among younger subjects and in females (Table 2 ). Generally, the length of sick leave was predicted more accurately in older subjects than in younger subjects, and better in males than in females. Access to past history of sick leaves improved the ROC area of NIO consultants from 60.6% (95% CI 51.3–69.9%) to 75.4% (95% CI 68.2–82.6%) in male sick-listed, but did not improve the ROC area in assessments of female sick-listed. Figure 2 ROC curves of identifying sick leaves lasting at least 12 weeks. The ROC curve of ability to identify sick leaves lasting at least 12 weeks, plotted at the average of two consecutive categories, in length predicted by sick-listed (n = 152), and mean length predicted by National Insurance officers and medical consultants in the responder group (n = 149, 150) and for all the data (n= 972, 975). The points representing cut-offs in predicted length >= 4 weeks (red), >= 8 weeks (pink) and >= 12 weeks (blue) are identified. Table 2 ROC area of identifying sick leaves lasting at least 12 weeks. The ability to identify sick leaves lasting at least 12 weeks in the responder group (n = 152) and in all participants (N = 993), presented as ROC area, calculated from length of sick leave predicted by sick-listed, and mean length predicted by National Insurance medical consultants and officers. The range of the individual National Insurance ROC areas is presented for all participants. Medical consultants Officers Self-assessed Responders n = 152 Responders n = 149 All participants n = 972 Responders n = 150 All participants n = 975 Sick-listed n ROC area ROC area ROC area Range individual ROC area ROC area Range individual N 95% CI 95% CI 95% CI ROC area 95% CI 95% CI ROC area All 152 80.9 55.6 64.6 59.6–64.2 56.0 61.4 55.6–65.6 993 73.7–86.8 45.6–65.6 60.8–68.3 46.6–65.4 57.7–65.1 17–40 years of age 78 508 76.4 65.6–87.2 43.0 28.8–57.2 57.2 51.7–62.8 54.5–57.8 48.9 35.4–62.5 57.4 52.0–62.9 51.9–57.8 41–67 years of age 74 485 85.7 77.2–94.2 68.3 54.8–81.8 70.7 65.8–75.6 63.4–70.1 62.5 49.4–75.6 65.1 60.1–70.1 56.1–73.4 Males 56 90.9 63.0 68.7 62.8–68.3 59.6 63.6 56.5–71.8 391 83.4–98.4 47.3–78.7 62.8–74.6 44.3–74.9 57.5–69.8 Females 96 74.7 50.9 62.0 57.7–61.5 54.0 60.3 52.9–61.8 602 64.8–84.7 38.1–63.8 57.2–66.8 42.0–65.9 55.6–64.9 Changing the observed length to be identified from 12 weeks to 8 or 26 weeks did not significantly change the diagnostic accuracy as assessed by the ROC area. The sick-listed identified sick leaves lasting 8 weeks or longer with a ROC area of 79.5% (95% CI 72.2–85.6%), and sick leaves lasting 26 weeks or longer with a ROC area of 75.5% (95% CI 67.9–82.1%). Sick-listed persons with mental disorders or with neck, or shoulder and arm disorders, were most accurate in their assessment (Figure 3 ). This was in contrast to NIO assessors, who demonstrated the lowest predictive ability in these diagnostic groups, particularly in responders. The impact on diagnostic accuracy of knowing the occupation was small. Figure 3 ROC area in different diagnostic groups. ROC area representing ability to identify sick leaves 12 weeks or longer in different diagnostic groups, calculated on length predicted by sick-listed, and mean of lengths predicted by NIO assessors. The ROC area are presented with blue bars of 95% CI in the responder group (n = 152/), and red bars without horizontal lines between upper and lower individual ROC area of the NIO assessors for all sick leaves (n = /958). Sensitivity, specificity, predictive value and likelihood ratio The sick-listed subjects predicted their sick leaves with higher sensitivity and PPV than the NIO assessors (Tables 3 , 4 ). Male sick-listed predicted sick leaves lasting at least 12 weeks with a sensitivity of 0.82% (95% CI 0.60–0.95) and a PPV of 0.78 (95% CI 0.56–0.93) using predicted length of at least 8 weeks. The corresponding sensitivity and PPV of female sick-listed were both 0.61 (95% CI 0.44–0.77). Table 3 Predictive validity – identifying sick leaves lasting at least 12 weeks. Predictive validity of identifying sick leaves that lasted at least 12 weeks, using 8 weeks as the cut-off in length as predicted by the sick-listed, medical consultants and officers. The prediction based on the Sickness Certificate 2 (SC2) used a cut-off in predicted length of at least 12 weeks. Sensitivity, specificity, PPV, and likelihood ratio data for NIO assessors are presented as means with 95% CI. Predicted length n assessments Sensitivity (95% CI) Specificity (95% CI) Likelihood ratio (95% CI) PPV 1 (95% CI) PPV adjusted to prevalence 33.4% (95% CI) Sick-listed 152 0.69 (0.56–0.84) 0.80 (0.70–0.87) 3.4 (1.9–6.2) 0.68 (0.54–0.79) 0.63 (0.49–0.76) Medical consultants Responder group 301 0.35 (0.26–0.44) 0.78 (0.71–0.84) 1.6 (1.0–2.5) 0.49 (0.38–0.61) 0.44 (0.33–0.56) Medical consultants All participants 1961 0.42 (0.38–0.45) 0.75 (0.73–0.77) 1.7 (1.4–1.9) 0.45 (0.41–0.49) Officers Responder group 302 0.53 (0.44–0.62) 0.59 (0.51–0.66) 1.3 (0.9–1.8) 0.44 (0.36–0.53) 0.39 (0.31–0.48) Officers All participants 1968 0.53 (0.49–0.57) 0.60 (0.58–0.63) 1.3 (1.2–1.5) 0.40 (0.37–0.43) Medical consultants SC2 637 0.85 (0.82–0.88) 0.44 (0.36–0.52) 1.5 (1.2–1.9) 0.82 (0.79–0.86) 0.43 (0.39–0.48) Officers SC2 636 0.88 (0.86–0.91) 0.33 (0.26–0.41) 1.3 (1.1–1.7) 0.80 (0.77–0.84) 0.40 (0.35–0.44) 1 The prevalence of sick leaves lasting at least 12 weeks was 38.2% in the responder group ( n = 152), 33.4% for all participants ( n = 993), and 72.1% in the SC2 group ( n = 322). Table 4 Predictive validity – identifying sick leaves lasting at least 26 weeks. Predictive validity of the ability to identify sick leaves lasting at least 26 weeks, using 8, 12 or 26 weeks, as cut-offs in length as predicted by the sick-listed, medical consultants or officers. Sensitivity, specificity, PPV and likelihood ratio data for NIO assessors are presented as means for length predicted on Sickness Certificates 1 and Sickness Certificates 2 (SC2). Predicted length n assess-ments Sensitivity (95% CI) Specificity (95% CI) Likelihood ratio (95% CI) PPV 1 (95% CI) PPV adjusted to prevalence 17.9%(95% CI) Sick-listed >= 8 weeks 152 0.69 (0.50–0.84) 0.69 (0.60–0.77) 2.2 (1.3–3.9) 0.37 (0.25–0.51) 0.33 (0.21–0.47) Sick-listed >= 12 weeks 152 0.50 (0.32–0.68) 0.83 (0.75–0.90) 3.0 (1.5–6.1) 0.44 (0.28–0.62) 0.40 (0.24–0.58) Sick-listed >= 26 weeks 152 0.28 (0.14–0.47) 0.98 (0.94–1.00) 16.9 (3.5–160) 0.82 (0. 48–0.98) 0.78 (0.44–0.95) Consultants >= 8 weeks 1961 0.44 (0.39–0.49) 0.72 (0.70–0.74) 1.6 (1.3–1.9) 0.25 (0.22–0.29) Consultants >= 12 weeks 1961 0.20 (0.16–0.24) 0.92 (0.90–0.93) 2.4 (1.8–3.2) 0.35 (0.28–0.41) Consultants >= 26 weeks 1961 0.07 (0.04–0.10) 0.99 (0.98–0.99) 2.8 (1.5–5.4) 0.54 (0.38–0.69) Officers >= 8 weeks 1968 0.55 (0.50–0.60) 0.58 (0.56–0.61) 1.3 (1.1–1.5) 0.22 19.3–24.9 Officers >= 12 weeks 1968 0.26 (0.21–0.31) 0.83 (0.81–0.85) 1.6 (1.3–2.0) 0.25 (0.20–0.29) Officers >= 26 weeks 1968 0.06 (0.04–0.09) 0.98 (0.97–0.98) 1.5 (0.9–2.6) 0.34 (0.23–0.48) SC2 Consultants >= 12 weeks 637 0.89 (0.86–0.93) 0.29 (0.25–0.34) 1.3 (1.1–1.5) 0.44 (0.40–0.48) 0.22 (0.18–0.25) Consultants >= 26 weeks 637 0.24 (0.19–0.29) 0.96 (0.93–0.98) 5.9 (3.4–11.0) 0.79 (0.68–0.87) 0.56 (0.41–0.70) Officers >= 12 weeks 636 0.89 (0.85–0.93) 0.21 (0.17–0.25) 1.1 (0.9–1.3) 0.41 (0.37–0.45) 0.20 (0.16–0.23) Officers >= 26 weeks 636 0.28 (0.22–0.34) 0.90 (0.87–0.93) 2.8 (1.9–4.3) 0.64 (0.54–0.73) 0.38 (0.28–0.49) 1 The prevalence of sick leaves lasting at least 26 weeks was 21.1% in the responder group ( n = 152), 17.9% for all the data ( n = 993), and 38.5% in the SC2 group ( n = 322). Duration of at least 8 weeks was the preferable cut-off in predicted length, to identify sick leaves lasting at least 12 weeks (Table 3 ). A predicted length of at least 12 weeks reduced the sensitivity in all the data to 0.17 in medical consultants and 0.25 in officers. The corresponding improvement in PPV was modest, reaching 0.54 in medical consultants and 0.45 in officers. Using a predicted length of at least 4 weeks would have markedly reduced the specificity (Figure 2 ). The sensitivity of identifying sick leaves lasting at least 26 weeks was generally low when medical consultants and officers predicted on the basis of SC1s. (Table 4 ). The sensitivity was improved somewhat by introducing SC2 information, but the effects on likelihood ratio and PPV if prevalence corrected, were minor. According to the results, the effects of the different predictive strategies can be illustrated by considering a program designed to intervene in all cases where the subject is expected to be sick-listed for more than 12 weeks at 14 days of sick leave. Out of every 1000 sick-listed persons, 333 will be sick-listed for more than 12 weeks according to the prevalence in this study. The random selection of 333 persons will include 111 true positives, while 333 persons selected by officers will include 133 of the 333 persons that will be sick-listed at least 12 weeks. The evaluation of 1000 sick-listed individuals thus increases the number of true positives by 22 in a selection of 333 sick-listed persons. The alternative strategy of asking the sick-listed themselves will include 210 true positives in a selection of 333 persons. Reliability and reproducibility of the predicted length Agreement between medical consultants in their initial prediction of sick leaves lasting at least 12 weeks, was fair, with a kappa of 0.31 (95% CI 0.20–0.43). The corresponding kappa value between officers was 0.05 (95% CI -0.05–0.14). In the prediction of sick leaves lasting at least 12 weeks based on the SC2, agreement was moderate between medical consultants (kappa = 0.42, 95% CI 0.29–0.54) and fair between officers (kappa = 0.26, 95% CI 0.10–0.42). The corresponding agreements in the prediction of sick leaves lasting at least 26 weeks were moderate between medical consultants (kappa = 0.55, 95% CI 0.40–0.70) and fair between insurance officers (kappa = 0.31, 95% CI 0.17–0.47). The differences in diagnostic accuracy, between the two participating medical consultants and their eight colleagues in the reproducibility group, were not significant. Discussion The results of the present study question any practical value of using information in medical sickness certificates in predicting the length of sick leave, as is the current practice in Norwegian NIOs. Instead, the sick-listed themselves predicted their length of sick leaves far more accurately, but this information is not routinely sought. Representativeness The officers in the present study were selected from experienced officers who had shown an interest in the field of sick leave. This might introduce a bias of overestimating the officers' general ability to predict the length of sick leaves. The performances of the two medical consultants were representative of eight of their colleagues who participated in the reproducibility part of the study. We therefore consider the diagnostic accuracy of the assessors to be representative of their professional groups, or at least not underestimated due to bias. Although the diagnostic accuracy varied within each group, the main conclusion of better predictive ability among the sick-listed, was challenged neither by comparing with the mean length predicted by assessors, nor by comparing with the best-performing NIO assessor. The distributions of gender and diagnosis among the 993 persons included in the study were comparable with those in the National Sickness Benefits Register. The findings of longer sick leaves in women with musculoskeletal disorders, and longer sick leaves in men with mental disorders, are consistent with the Register and other studies [ 11 - 13 ]. The low responder rate among the sick-listed introduced a possible selection bias, although we could not identify any selection bias in gender, age, diagnosis or occupation [ 14 ]. If there was a selection towards more predictable sick leaves, this should have been reflected in the assessments of officers and medical consultants. The general trend of lower diagnostic accuracy of NIO assessors in the responder group indicates that if any selection bias contributes to the results, it is an underestimate of the self-predictive ability. Why did the sick-listed make better predictions? If the lengths of sick leaves were predominantly related to loss of function caused by sickness, in line with the legislation, we would expect that the medical consultants' professional competence would favour them in predictions of the lengths of sick leaves. The differences we observed between medical consultants and officers in mean ROC area, were however minor. Furthermore, we could not demonstrate any significant differences in diagnostic accuracy between medical consultants and officers when aggregate information on disease, treatment, function related to work, and prognoses were available in the SC2. The improvement in ROC area with this aggregated information was minor, with the area just reaching 70%, which is considered borderline useful for some purposes [ 7 ]. The result is in line with Bjørndal's findings of low prognostic impact of the SC2 [ 15 ], and is supported by findings of a low predictive power of symptoms and signs in neck and shoulder disorders [ 16 ]. The better prediction of the length of sick leave by the sick-listed themselves, is supported by studies that have identified different non-disease determinants of sick leave, such as job satisfaction [ 17 ], attitudes towards pain [ 18 ], irreplaceability [ 19 ] and psychosocial work environment [ 20 - 22 ]. Studies identifying that at least the initial sickness certification is predominantly patient controlled [ 23 , 24 ] indicate the competence of the sick-listed. Self-rated health seems to be an independent predictor of return to work [ 17 ], disability pension [ 25 ] and early retirement [ 26 ]. Our findings can be interpreted as indicating that the subjective perception of sickness and work ability is more predictive of the length of sick leave, than the apparently more objective description in medical terms. The differences in predictive ability were especially significant in persons with mental and neck disorders, while the NIO assessors performed equal to the sick-listed in the more clear-cut injuries with more standardised treatment and prognosis. Mental disorders, with high prevalence in the population, and an increasing cause of absence [ 27 ], are of special interest [ 13 ]. This increasing prevalence of sick leaves indicates the presence of factors separate from the diagnosis criteria. It seems that the more clear-cut the disease and the recommended treatment, the lesser the gain in predictive ability achieved by asking the sick-listed, and vice versa. The modest gain in predictive ability caused by introducing more medical information by the inclusion of the SC2 supports this interpretation. A more complete description of symptoms and treatment does not necessarily give better prognostic information when this includes little knowledge of the consequences related to occupation, and the effects of treatment are undocumented or, at best, marginal. Diagnostic accuracy – practical implication The Norwegian NIO is obliged by legislation to perform early intervention on the sick-listed in an effort to reduce the length of sick leave and the risk of expulsions from work. Limited resources and the large number of sick-listed individuals make selection desirable before any intervention is initiated. An alternative to selection on the basis of medical certificates is to communicate directly with the sick-listed themselves. This selection for intervention by NIOs might be seen as screening. The aim is to reach – at an acceptable cost – as many as possible of those that might profit from intervention. The potential individual gain by intervention will be greater when longer lasting sick leaves can be anticipated, and greater the sooner individual intervention programs are established. The marginal predictive ability and modest agreement between NIO assessors questions the use of resources in selection based on information from medical certificates. The predictions of medical consultants tend to be better than those of officers, but not to an extent that makes it more meaningful to use medical consultants in the selection process, rather than officers. With limited resources for intervention, it might be more cost effective to identify those whose sick listing will last longer than 26 weeks instead of 12 weeks. Based on self-reporting, eight out of ten would be true positives, and one fourth of the individuals would be reached. To reach the same number of true positives at 14 days of sick leave, the ratio of true positives would be reversed from eight out of ten, to two or three out of ten, if the selection were based on medical certificates. In the search for tests predicting long-lasting sick leaves, such as The Örebro Musculoskeletal Pain Questionnaire [ 28 ], the present study indicates that the results of any such tests should be compared with the results of crude self-estimated length. Conclusions Sick-listed individuals predicted their length of sick leave far more accurately than did NIO medical consultants and officers based on information from sickness certificates and the history of past sick leaves. The predictions of sick-listed males were better than those of females, and older persons predicted better than younger persons. The availability of more information, as through the SC2, had only a minor effect on the predictive ability of the medical consultants and officers. Neither reliability nor validity of their predictions was satisfactory. This study demonstrates the need to re-consider the diagnostic usefulness of documentation on sickness absences, and supports a change in strategy from collecting more medical information to more direct communication with the sick-listed themselves, for effective and early interventions to prevent long sick leaves and expulsions from work. Competing interests The author N.F. is part-time employed as National Insurance medical consultant. Authors' contributions NF was in charge of designing and running the study, and performed most of the analyses and the writing of this manuscript. RJ actively supervised all parts of the study, and OHF contributed to planning and writing. All authors read and approved the final version of the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529266.xml |
539061 | A Signaling Pathway Involving TGF-β2 and Snail in Hair Follicle Morphogenesis | In a common theme of organogenesis, certain cells within a multipotent epithelial sheet exchange signals with their neighbors and develop into a bud structure. Using hair bud morphogenesis as a paradigm, we employed mutant mouse models and cultured keratinocytes to dissect the contributions of multiple extracellular cues in orchestrating adhesion dynamics and proliferation to shape the cluster of cells involved. We found that transforming growth factor β2 signaling is necessary to transiently induce the transcription factor Snail and activate the Ras-mitogen-activated protein kinase (MAPK) pathway in the bud. In the epidermis, Snail misexpression leads to hyperproliferation and a reduction in intercellular adhesion. When E-cadherin is transcriptionally down-regulated, associated adhesion proteins with dual functions in signaling are released from cell-cell contacts, a process which we demonstrate leads to Ras-MAPK activation. These studies provide insights into how multipotent cells within a sheet are stimulated to undergo transcriptional changes that result in proliferation, junctional remodeling, and bud formation. This novel signaling pathway further weaves together the web of different morphogens and downstream transcriptional events that guide hair bud formation within the developing skin. | Introduction Mammalian development involves the morphogenesis of complex three-dimensional structures from seemingly uniform sheets or masses of cells. A simple bud-like structure initiates the formation of many organs, including lungs, spinal cord, mammary glands, and hair follicles [ 1 ]. The multipotent, adhering epithelial cells are typically attached to an underlying basal lamina that polarizes the epithelial sheet and separates it from surrounding mesenchyme. Budding morphogenesis is guided by a reciprocal exchange of signals between epithelium and mesenchyme to specify the identity of the organ that will form and to govern its growth. At the helm of these molecular communication pathways are Wnts, bone morphogenic proteins (BMPs), transforming growth factor βs (TGF-βs), and fibroblast growth factors (FGFs). Through activation of cell surface transmembrane receptors, these external signaling molecules trigger distinct cascades of intracellular events that culminate in changes in gene expression, growth, and differentiation [ 2 ]. How this constellation of signals collaborates in tailoring each budding process so that it executes a distinct morphogenetic program has yet to be comprehensively defined. However, the process appears to be patterned at the initial stages of bud formation, since the relative importance of these pathways and their downstream effectors differ as buds begin to develop and cell fates are specified. The development of a bud requires a number of coordinated changes in the behavior of the targeted cells within an epithelial sheet. The process must be accompanied by alterations in the proliferation, polarity, shape, and adhesiveness of selected cells, as well as by modifications in their underlying basal lamina. Thus, extracellular epithelial-mesenchymal crosstalk must be intricately orchestrated to couple the determination of distinct cell fates with the contemporaneous remodeling of the physical and structural properties of the cell. Among the few dispensable organs, hair follicles offer an excellent model system to study epithelial bud formation. Mammalian skin epithelium begins as a single sheet of multipotent ectodermal cells. During development, specialized mesenchymal cells populate the skin in a spatially defined pattern to initiate the complex epithelial-mesenchymal crosstalk that will specify the bud [ 3 ]. Once committed, a small cluster of epithelial cells, the placode, instructs a group of underlying mesenchymal cells to condense and form the nascent dermal papilla, which will be a permanent fixture of the hair follicle. Subsequent exchanges between the placode and nascent dermal papilla result in further growth of the follicle into the underlying dermis, or down-growth, and eventual differentiation into the six concentric layers of the mature follicle. Previously, we delineated how two respective epithelial and mesenchymal signals, Wnts and the BMP-inhibitory factor noggin, function in concert to induce lymphoid enhancer factor-1/β-catenin (LEF-1/β-catenin)-mediated gene transcription within the follicle placode [ 4 ]. The downstream changes elicited through convergence of these two early signaling pathways include down-regulation of the gene encoding E-cadherin, the prototypical epithelial cadherin that forms the transmembrane core of intercellular adherens junctions (AJs) [ 5 ]. We subsequently showed that when E-cadherin is transgenically elevated in mouse skin, hair follicle morphogenesis is blocked, suggesting that E-cadherin down-regulation is a critical event in governing the adhesion dynamics necessary for budding morphogenesis [ 4 ]. Like LEF-1, E-cadherin also binds to β-catenin. At sites of cell-cell contact, however, E-cadherin-β-catenin complexes recruit α-catenin, which in turn coordinates the associated actin polymerization dynamics necessary to stabilize nascent AJs and integrate the cytoskeleton across an epithelial sheet [ 6 , 7 , 8 ]. α-Catenin also binds to the class III Lin-1, Isl-1, Mec-3 (LIM) protein Ajuba (a member of the zyxin family of proteins), which appears to function dually in both adhesion and in activation of the Ras-mitogen-activated protein kinase (MAPK) pathway [ 9 , 10 ]. Through these links, AJs appear able to couple adhesion with cytoskeletal dynamics as well as with nuclear and cytoplasmic signaling. This provides a framework for conceptualizing why E-cadherin levels appear to impact upon a plethora of developmental processes (reviewed in [ 11 ]). As we probed more deeply into the underlying mechanisms governing E-cadherin promoter activity, we were intrigued by the close proximity of the LEF-1/β-catenin binding site to a site known to bind the Snail/Slug family of zinc finger transcriptional repressor proteins [ 12 , 13 , 14 , 15 ]. Both activity of Snail and down-regulation of E-cadherin play pivotal roles in epithelial to mesenchymal transitions (EMTs), typified by the transformation of polarized, adhering epithelial cells into motile mesenchymal cells [ 16 , 17 ]. Bud formation differs from an EMT in that E-cadherin activity needs to be down-regulated but not prevented, so that adhesive junctions are remodeled rather than quantitatively impaired. Supportive of an underlying ability to fine-tune cadherin expression at the transcriptional level, Snail seems to have an additive effect with LEF-1/β-catenin in negatively modulating E-cadherin promoter activity [ 4 ]. In the present study, we discovered that Snail is expressed briefly at an early stage of hair bud formation, when E-cadherin down-regulation and activation of proliferation take place. Thereafter, Snail disappears and remains absent during subsequent follicle down-growth and maturation. This exquisite pattern appears to be functionally relevant since altering it in vivo correspondingly affects features associated with hair bud formation, including down-regulation of E-cadherin, increased proliferation, and repressed terminal differentiation. Although the temporal spike of Snail in the hair bud is reflected at the mRNA level and seems to follow Wnt signaling and BMP inhibition, LEF-1/β-catenin activation does not appear to induce Snail gene expression in embryonic skin keratinocytes. In contrast, we provide in vitro, transgenic (Tg), and gene targeting evidence to show that TGF-β2 and small phenotype– and mothers against decapentaplegic–related protein 2 (SMAD2) signaling are upstream inducers of Snail gene expression in skin epithelium. In the absence of TGF-β2 signaling and Snail gene expression, hair placodes can form, but further follicle down-growth is blocked. Our studies point to the view that Snail likely functions downstream of cell fate specification, at a stage where the bud begins to exhibit enhanced proliferation and migration. Results Snail mRNA and Protein Are Expressed Transiently at the Hair Bud Stage of Follicle Morphogenesis Although Snail family members are most frequently associated with EMTs, they also participate in many malignant processes involving a down-regulation but not a quantitative abrogation of intercellular junctions [ 18 ]. The range of developmental processes in which Snail family members have been implicated thus includes the type of epithelial remodeling that is observed in hair follicle bud formation. Given our prior observation that exogenously added Snail can participate with LEF-1/β-catenin in down-regulating E-cadherin expression in keratinocytes [ 4 ], coupled with the established requirement for LEF-1/β-catenin in hair follicle morphogenesis [ 4 , 19 ], we turned to addressing whether Snail/Slug family members might also participate in the process. PCR analyses identified transient Snail mRNA expression during a period of skin embryogenesis when waves of hair follicles are forming (unpublished data).To pinpoint specifically where Snail mRNA is expressed in the developing skin, we conducted in situ hybridization using a cRNA probe unique to the Snail 3′ untranslated region (UTR). Embryonic day 17.5 (E17.5) was chosen, since the multiple waves of follicle morphogenesis occurring at this time enabled us to evaluate Snail expression at different stages of the process. As shown in Figure 1 A, specific hybridization was detected within the epithelium of nascent hair buds. By contrast, as follicles progressed further through their development (e.g., germ and peg stages), they exhibited no signs of hybridization ( Figure 1 A). The transient nature of Snail mRNA expression during follicle development was most apparent in hybridized skin sections containing follicles from two different waves of morphogenesis (as shown in Figure 1 ). Hybridizing hair buds from a later wave appeared juxtaposed with nonhybridizing follicles from an earlier wave. Figure 1 Snail Is Expressed Exclusively in the Hair Bud during Morphogenesis Embryos were either frozen in OCT embedding compound (A, F, and H) or embedded in paraffin (C, D, E, and G), and then sectioned (8 μm). (A) In situ hybridizations with Snail sense or antisense cRNA probes. Black dotted lines demarcate the basement membrane that separates the epidermis (epi) from dermis (der). Arrows point to Snail RNA expression, restricted to the hair bud stage of follicle morphogenesis. It was not seen in later hair germ or peg stages. (B) Expression of Snail protein coincides with hair development. Protein extracts were prepared from keratinocytes transfected with empty expression vector (K14), containing the K14 promoter or with the vector driving HA-tagged Snail (K14- Snail ); or from whole skin from E13.5 to P5 animals, including newborn (nb). Equal amounts of proteins were then resolved by SDS-PAGE through 12% gels and subjected to Western blotting using either an affinity-purified Snail polyclonal antiserum, which we generated, or anti-tubulin (loading control). (C–E) Immunohistochemistry shows expression of Snail protein in the nuclei of cells within the hair and skin. (C) E13.5 skin with a single layered epidermis (epi) shows no Snail expression. (D) The first morphological sign that cells have adopted a hair follicle fate is a cluster of cells called a placode in E16.5 skin. Snail is not expressed at this stage of development. (E) Snail is expressed in the hair bud of E17.5 skin but not in later stages of development such as the germ or peg. (F) Immunofluorescence with anti-Ki67 (green) identifies the proliferating cells of the skin, restricted to the basal layer of the epidermis and developing hair follicles. Anti-β4 int labeling reveals the presence of the hemidesmosomal integrin β4, restricted to the base of cells adhering to the underlying basement membrane. The white dotted line marks the outermost surface of the skin. (G) Immunohistochemistry with pMAPK marks a subset of proliferating cells within the epidermis and hair bud. Anti-pMAPK labeling was consistently robust within the hair bud. (H) Immunofluorescence with anti-laminin 5 (lam5), which demarcates the basement membrane, and anti-E-cadherin (E-cad), a component of AJs. At the leading edge of the growing bud, cell-cell borders show markedly diminished anti-E-cadherin labeling (arrowheads). To determine whether this transient nature of Snail mRNA expression is reflected at the protein level, we generated an antibody against the N-terminal sequence that resides upstream of the more conserved zinc finger domains. As judged by Western blot analysis, the antibody did not detect endogenous proteins from cultured keratinocytes, but it did yield a band of the expected size from keratinocytes transiently expressing a hemagglutinin (HA)-tagged Snail protein ( Figure 1 B). The antibody also recognized a band corresponding to the size of endogenous Snail (approximately 28 kDa) in lysates from embryonic mouse skin, the temporal appearance of which corresponded to the waves of hair follicle morphogenesis from E15.5 to newborn when over 90% of the hair on the mouse is formed ( Figure 1 B). Consistent with the Western blot data, immunohistochemical analysis did not detect Snail in single-layered E13.5 epidermis ( Figure 1 C) nor in the placode, which is the earliest morphological sign of the commitment of multipotent cells of the embryonic ectoderm to a hair cell fate ( Figure 1 D). Consistent with the in situ hybridization results, anti-Snail antibody labeled only hair buds and not follicles at more mature stages of development ( Figure 1 E). Taken together, the anti-Snail antibody appeared to be specific for its target protein. It did not detect other Snail family members known to be expressed in keratinocytes and/or skin (unpublished data). Furthermore, the immunohistochemical data paralleled our Snail in situ hybridization data revealing transient Snail expression at the hair bud stage ( Figure 1 A). As judged by immunohistochemistry, Snail protein was localized to the nuclei of the hair bud cells ( Figure 1 E). This feature was consistent with Snail's known function as a transcriptional repressor [ 12 , 13 ]. Additionally, anti-Snail labeling was detected in only three of the four major waves of follicle morphogenesis. Snail was not found in the buds of guard hairs that are the earliest of all hairs to form (at E13.5), and which constitute less than 5% of the mouse coat (unpublished data). As judged by immunofluorescence with antibodies against the proliferating nuclear antigen Ki67, the timing of Snail expression coincided with the stage at which the developing follicle enhanced its proliferation and down-growth ( Figure 1 F). Immunohistochemistry with antibodies against the active (phosphorylated) form of MAPK (pMAPK) marked a subset of the proliferating (Ki67-positive) cells, and pMAPK-positive cells were enriched in the hair bud ( Figure 1 G). The timing of Snail induction and Ki67 and pMAPK enrichment in the hair bud appeared to follow closely the induction of LEF-1/β-catenin activity, known to initiate in the hair placode stage [ 20 ]. However, like placodes, hair buds exhibited down-regulation in E-cadherin expression ( Figure 1 H; see also [ 4 ]). Sustained Expression of Snail Results in Epidermal Hyperproliferation and Differentiation Defects in Tg Mouse Skin The striking spike of Snail expression coincident with hair bud formation and enhanced proliferation prompted us to examine the consequences of ectopically expressing Snail elsewhere in mouse skin epidermis. To distinguish Tg from endogenous Snail, we used the HA-epitope, shown previously not to alter Snail's transcriptional activity [ 12 ]. Of 20 K14-Snail[HA] Tg animals generated, three expressed the transgene and all exhibited analogous phenotypes. Mice that integrated the transgene at the single-cell stage died at or shortly after birth. The three surviving full-Tg founder mice harbored transgene integrations that gave stable transmission of mosaic Snail gene expression through the germline. Progressively poor health necessitated our sacrificing most offspring from these lines within a year of birth. As Snail Tg animals grew, they became distinguished by their small size, short tails, and flaky skin ( Figure 2 A). Histological analyses of 3-d old (P3) mice revealed mosaic patches marked by epidermal thickening ( Figure 2 B). The mosaic morphology was reflected at the level of Tg Snail protein, with only the hyperthickened regions expressing nuclear HA-tagged Snail ( Figure 2 C). These hyperthickened areas were marked by excessive proliferation, as revealed by antibodies against the proliferating nuclear antigen Ki67 ( Figure 2 D and 2 E). Activated, pMAPK-positive cells were also prevalent in these areas ( Figure 2 F and 2 G), as were cells expressing keratin 6, a keratin induced in the suprabasal layers of hyperproliferative skin ( Figure 2 H and 2 I). Figure 2 Misexpression of Snail in Mouse Skin Epidermis Results in Hyperproliferation Three different surviving Tg founder mice harbored a K14-Snail transgene that was integrated into a locus that resulted in inheritable, mosaic expression of the transgene in skin epidermis. All displayed similar abnormalities, as did their offspring. (A) P16 WT and K14-Snail Tg mice. Insets denote magnified tail segments, which displayed a mosaic, flaky appearance in Tg mice. Size differences appeared with age, and are likely due to K14-promoter activity in the tongue and oral epithelium, resulting in progressive defects and reduced food intake. Hence, skin sections from young (P3) mice were analyzed (B–I). (B) Hematoxylin- and eosin-stained Tg skin section. Double arrows demarcate the border of mosaic histology, with seemingly normal epidermis (epi) and a mature hair follicle (hf) at left and hyperthickened epidermis at right. (C) Immunofluorescence of Tg skin section labeled with antibodies as color-coded on frame. Double arrows demarcate the border of mosaic anti-Snail (green), revealing Snail expression coincident with regions of hyperthickened epidermis (at left) and absent in regions of normal epidermis (at right). (D–I) Sections of P3 WT or Tg skin (affected region) subjected to either immunofluorescence (D, E, H, and I) or immunohistochemistry (F and G) with antibodies as indicated on the panel. Anti-keratin 5 indicates K5, normally restricted to the basal layer of the epidermis; anti-keratin 6 detects keratin 6, expressed in postnatal epidermis under conditions such as wounding, in which hyperproliferation occurs. All other antibodies are as in the legend to Figure 2 . Comparison of D and E provide representative examples that illustrate that pMAPK is found in only a subset of all proliferating (Ki67-positive) cells. Note also the presence of Ki67- (E) and pMAPK-positive (G) cells in some suprabasal areas; Ki67-positive cells colabeled with anti-Snail (E). Expression of the Snail transgene did not block terminal differentiation in the hyperproliferative epidermis, but it distorted it markedly ( Figure 3 A– 3 H). Typical of most hyperproliferating conditions, Snail expression led to a large expansion in layers with spinous and granular morphology. Additionally, however, was a marked and variable expansion of keratin 5 (K5), normally restricted to the innermost basal layer (see Figure 3 ). Although the failure of Snail -null mice to develop past gastrulation [ 21 ] precluded our ability to study the loss of Snail function in skin development, a good correlation emerged between the expression of Snail protein and the extension of K5, Ki67, and pMAPK suprabasally (compare data in Figures 2 and 3 ). Figure 3 Alterations in the Differentiation Program and Basement Membrane Organization in Snail-Expressing Tg Epidermis (A–H) Immunofluorescence of skin sections from P3 WT and Tg mice. Shown are affected areas of Tg skin; in areas where Snail protein was not expressed, stainings were normal. Sections were labeled with antibodies as indicated and color-coded on each frame. Antibodies are against markers of normal epidermal differentiation, and include K5 (a basally expressed keratin), K1 (a suprabasal keratin, expressed in spinous layer cells), involucrin (Inv; a suprabasally expressed cornified envelope protein found in upper spinous and granular layer cells), loricrin (Lor; a cornified envelope protein expressed in the granular layer), and filaggrin (Fil; a protein that bundles keratin filaments in the granular layer and stratum corneum). Note abnormal extension of anti-K5 suprabasally, often present in anti-K1 positive suprabasal Tg cells. (I–N) Immunohistochemistry (I and J) or immunofluorescence (K–N) of sections of P30 Wt (I, K, and M) and Tg (J, L, and N) (affected areas) skins using the antibodies indicated. Note that with age, affected areas of the Tg epidermis became increasingly undulating, often exhibiting papilloma-like invaginations (J). Insets in I and J are magnified views of the boxed areas, illustrating the absence (Wt) or presence (Tg) of nuclear anti-cyclin D staining. With age, affected areas of the Tg epidermis also displayed perturbations within the basement membrane, as judged by antibody labeling against either basement membrane (K and L) or hemidesmosomal (M and N) components. Double arrows in L demarcate mosaic zones, revealing that perturbations were restricted to hyperthickened, i.e., Snail-positive zones (to left of double arrows). Other abbreviations are as noted in the legend to Figure 2 . The changes in hyperproliferation and differentiation were not initially accompanied by gross signs of epithelial invaginations. With age, however, epidermal folds and undulations developed in areas where Snail was expressed, and proliferative markers persisted in these regions ( Figure 3 I and 3 J; anti-cyclin D staining). The undulations were accompanied by partial dissolution of the underlying basement membrane ( Figure 3 K and 3 L). Aberrant staining was also observed with antibodies against components of the hemidesmosomes, which provide strong adhesion of basal epidermal cells to the underlying basal lamina ( Figure 3 M and 3 N). Interestingly, similar types of alterations occur in the basement membrane in the hair bud of embryonic and newborn mice when Snail is normally expressed. The fact that the basement membrane separating the epidermis from the dermis is altered only in the adult Tg animals suggests the involvement of intermediary factors not as readily available in the epidermis as they are in the follicle. Possible Links between Epidermal Hyperproliferation and Down-regulation of AJ Proteins in Snail Tg Mice Given that the E-cadherin promoter is a direct target for Snail-mediated repression in vitro [ 4 , 12 , 13 ], and that E-cadherin was down-regulated in Snail-expressing hair buds, we examined the status of E-cadherin and other AJ proteins within regions of hyperproliferative epidermis where Tg Snail was present ( Figure 4 A). In these regions, immunofluorescence staining of E-cadherin and α-catenin were markedly diminished. In contrast, the intensity of antibody staining for two other AJ proteins, β-catenin and Ajuba, was still strong. Interestingly, however, despite appreciable immunofluorescence, localization of β-catenin and Ajuba appeared to be largely cytoplasmic rather than at cell-cell borders ( Figure 4 A insets). Figure 4 Snail-Mediated Remodeling of AJs Contributes to Hyperproliferation (A) Immunofluorescence of skin sections from P30 Wt and Tg mice. Shown are affected areas of Tg skin; in areas where Snail protein was not expressed, stainings were normal. Antibodies used are against AJ proteins and include E-cadherin (E-cad), the transmembrane core protein; β-catenin (β-cat), which binds E-cadherin at AJs and which can also participate as a transcription cofactor when associated with LEF-1/TCF proteins in the nucleus; α-catenin (α-cat) which binds to both β-catenin and Ajuba, a close relative of zyxin; and Ajuba, which can associate with proteins that bind to the actin cytoskeleton, as well as with Grb-2, a mediator of the GTP nucleotide-exchange protein Sos, involved in activation of the Ras-MAPK signaling cascade. In Snail-expressing Tg regions, there was a reduced staining with anti-E-cad and anti-α-cat and a more diffuse staining with anti-Ajuba. Insets in the panels for β-catenin and Ajuba staining are magnified views of the boxed areas. Arrows mark membrane localization of the protein and asterisks mark cells with elevated levels of cytoplasmic β-catenin or Ajuba. (B) Western blot analyses of protein extracts from P30 Wt and Tg back and ear skins. Antibodies are as in (A) except anti-P-cad, which detects P-cadherin, whose expression in the hair follicle was not affected, and anti-tubulin, which detects tubulin, a control for equal protein loadings. Note that the reductions seen in E-cadherin and α-catenin are likely to be underestimates of the actual differences in affected regions, since the Tg skin expressed Snail mosaically. (C) In the presence of elevated Snail, α-catenin levels can be restored by overexpression of E-cadherin. Keratinocytes were transfected with either HA-tagged Snail (Snail[HA]; images on the left) or Snail(HA) and Ecad(HA) (images on the right). 2 d after transfection, cells were switched from low-calcium growth medium to high-calcium medium for 6 h to induce AJ formation. Cells were stained with antibodies as indicated on the panels. Arrowheads point to sites of intercellular contact between a Snail-transfected keratinocyte and its neighboring untransfected cell. (D) Reintroduction of E-cadherin in keratinocytes expressing Snail returns pMAPK to basal levels. Keratinocytes were transfected with control vector (K14), or Snail(HA), or Snail(HA) + E-cad(HA) . After 2 d, cells were serum starved for 4 h and whole cell lysates were made and Western blotted with antibodies to pMAPK, HA to recognize the HA-tagged Snail and E-cadherin protein, 20or tubulin as a loading control. (E) Ajuba interacts with Grb-2 under conditions where α-catenin levels are reduced. Protein extracts were made from skins of P30 Wt and K14-Snail Tg P30 mice (blots on the left) and of newborn Wt and K14-Cre/α-catenin (fl/fl) conditionally null animals (blots on the right) [ 7 ]. Equal amounts of protein extracts were treated with anti-Grb-2 antibody (+) or control isotype antibody (–), and following centrifugation, immunoprecipitates were subjected to SDS-PAGE and Western blot analysis with anti-Ajuba and anti-Grb-2 antibodies. Note the presence of Ajuba only under conditions where levels of α-catenin and other AJ proteins were aberrantly low or absent. (F) Transgene expression of excess Ajuba or the Grb-2-interacting domain (pre-LIM) of Ajuba in keratinocytes results in the activation of the Ras-MAPK pathway. Primary newborn mouse keratinocytes were transfected with either the empty K14 expression vector (K14), or the expression vector driving Snail, full length Ajuba, or the pre-LIM domain of Ajuba in the absence or presence of a peptide inhibitor (inh) that disrupts the interaction between Grb-2 and Sos. 48 h posttransfection, protein extracts were prepared and subjected to SDS-PAGE and Western blot analyses with antibodies against pMAPK, total MAPK, Ajuba (also recognizing the smaller, pre-LIM domain), and Snail. Architectural differences in the epidermis made Western blot analyses somewhat difficult to gauge. However, in regions such as ear skin, where the highest levels of Snail protein were expressed, the effects were accentuated. In both back skin and ear skin, overall levels of E-cadherin and α-catenin were reduced, under conditions where β-catenin and Ajuba levels remained unchanged relative to controls ( Figure 4 B). Taken together, these data were consistent with our results obtained from immunofluorescence microscopy. A priori, the decrease in α-catenin levels could be due to either direct transcriptional repression by Snail or perturbations in AJ formation caused by the decrease in E-cadherin gene expression. To distinguish between these possibilities, we tested whether α-catenin levels could be restored by exogenous expression of E-cadherin in Snail-expressing keratinocytes. As shown in Figure 4 C, transiently transfected keratinocytes expressing HA-tagged Snail displayed a loss of E-cadherin and α-catenin at cell-cell borders. Coexpression of exogenous HA-tagged E-cadherin not only enabled cell-cell border localization of E-cadherin protein, but also rescued the cell-cell border staining of α-catenin ( Figure 4 C). The ability to restore α-catenin expression and localization under these conditions argues against the notion that Snail transcriptionally represses α-catenin. Rather, the findings are consistent with a previous report that E-cadherin is required for the translation of α-catenin mRNA [ 22 ]. Despite the reductions in AJ markers, Tg skin still displayed sealed membranes and intercellular junctions that were largely intact, as judged by ultrastructural analyses (unpublished data). In this respect, the skin epithelium resembled that of the hair bud, where the down-regulation in junction proteins is permissive for cell-cell remodeling without abrogating intercellular adhesion. The similarities between Snail Tg epidermis and hair buds extended to the hyperproliferative state, leading us to wonder whether the down-regulation of AJ proteins might contribute to this condition. Given the increase in pMAPK staining in Snail Tg epidermis (see Figure 2 G), we used pMAPK levels as our assay to test whether the loss of E-cadherin contributed to the Snail-mediated increase in proliferation. Consistent with our in vivo observations, transfected keratinocytes expressing Snail exhibited a substantial increase in pMAPK levels relative to control cells ( Figure 4 D). Coexpression of E-cadherin with Snail appeared to abrogate this effect. Together, these findings raised the possibility that an AJ-associated protein that is normally sequestered at the plasma membrane may participate in a proliferation signaling pathway when AJs are deconstructed. Numerous studies have correlated a down-regulation of E-cadherin with a translocation of β-catenin to the nucleus and a transactivation of genes that are regulated by the LEF-1/T cell factor (TCF) family of DNA binding proteins [ 23 , 24 , 25 ]. The presence of nuclear cyclin D in hyperproliferative Snail Tg epidermis was particularly intriguing since prior studies have reported cyclin D gene as a direct target of TCF/β-catenin transcription [ 26 ]. This said, we did not detect nuclear β-catenin in our Tg epidermis, and mating the Snail Tg mice against the TOPGal reporter mouse [ 20 ] gave no signs of ectopic LEF-1/Tcf/β-catenin activity (unpublished data). We next turned to the presence of cytoplasmic Ajuba for a possible mechanistic link to the proliferative increase in our Snail Tg epidermis. In addition to its documented ability to bind α-catenin [ 10 ], Ajuba can also associate with growth factor receptor-bound protein-2 (Grb-2)/son of sevenless (Sos), the nucleotide exchange factor for Ras, which is upstream from activation of MAPK [ 9 ]. Given the increase in pMAPK staining in Tg skin, we examined the possibility that Ajuba might have changed its binding partner in Snail-expressing epidermis. Interestingly, Ajuba was readily detected in anti-Grb-2 immunoprecipitates of protein lysates from skins of Snail Tg mice but not from the corresponding wild-type (WT) animals ( Figure 4 E). When these experiments were repeated with α-catenin -null epidermis, a similar Grb-2-Ajuba association was detected, and again, this interaction was not detected in the protein extracts from control littermate skin ( Figure 4 E). Together, these data demonstrate that the reduction in α-catenin levels, either by Snail-mediated down-regulation of E-cadherin or by α-catenin conditional targeting, allows Ajuba to interact with Grb-2/Sos. If the competition between Grb-2/Sos and α-catenin for Ajuba is functionally relevant to the hyperproliferative state of a keratinocyte, then overexpression of Ajuba would be expected to bypass the competition and promote activation of the Ras-MAPK pathway in WT keratinocytes. Indeed, when serum-starved keratinocytes were transiently transfected with an Ajuba expression vector, the levels of pMAPK were not only elevated but also comparable to those transfected with the K14-HASnail transgene ( Figure 4 F). This activation was abolished when cells were treated with a small peptide inhibitor that specifically interrupts the Grb-2/Sos interaction ( Figure 4 F; see lanes marked “inh”) [ 27 ]. Ajuba's pre-LIM domain is the segment that associates with Grb-2's Src-homology 3 domain [ 9 ]. When this domain was overexpressed in serum-starved keratinocytes, a comparable elevation in pMAPK was observed ( Figure 4 F). As expected, the small peptide inhibitor that interrupts the Grb-2/Sos association blocked the effects. These data suggested that by elevating cytosolic Ajuba levels, Ajuba's pre-LIM domain may associate with Grb-2/Sos in a manner that stimulates its nucleotide exchange activity and leads to activation of the Ras-MAPK pathway. Although this pathway provides one mechanism by which Snail expression and proliferation may be coupled in skin epithelium, proliferative circuitries involving AJs are known to be complex and often interwoven. Future studies will be needed to systematically dissect these putative intricacies at a molecular level. Probing the Regulation of Snail Gene Expression Reveals an Essential Link to TGF-β2 Signaling in the Developing Hair Bud The temporal spike of Snail mRNA expression in the hair bud prompted us to consider what factor(s) may be regulating the Snail gene. A variety of extracellular signals have an impact on the cell type-specific expression of different Snail family members, and many of them, including Wnts, BMPs, FGFs, and TGF-βs, also affect hair bud development [ 2 , 16 , 28 ]. Since Snail is not expressed in cultured skin keratinocytes that secrete active BMPs and FGFs (see Figure 1 B), we focused our attention on Wnt and TGF-β signaling as more likely candidates for Snail induction in this cell type. Previously, we showed that effective transmission of a Wnt-3a signal in cultured keratinocytes can be achieved through their exposure to the BMP inhibitor noggin, which induces LEF-1 expression [ 4 ]. In vitro, these conditions down-regulated the E-cadherin promoter and induced a LEF-1/β-catenin-sensitive reporter gene, TOPFLASH [ 4 ]. In contrast, Snail expression was not induced by these conditions ( Figure 5 A). Thus, despite essential roles for Wnts and noggin in hair follicle specification [ 4 , 29 , 30 ], our studies did not support an essential role for these signals in governing Snail expression in keratinocytes. Figure 5 TGF-β2, but Not Wnt/noggin, Transiently Induces Snail Expression In Vitro (A) Failure of Wnt and noggin signaling to induce Snail in cultured keratinocytes. Primary mouse keratinocytes were treated with Wnt- and/or noggin-conditioned medium (+) or the corresponding control medium (–). These conditions are known to activate the LEF-1/β-catenin reporter TOPGal and down-regulate the E-cadherin promoter (see [ 4 ] for details). Using Western blot analyses, cellular proteins were then analyzed for Snail, LEF-1, β-catenin, and tubulin. Proteins from keratinocytes transfected with K14- Snail were used as a positive control for Snail expression. (B) TGF-β2 can induce Snail protein. Primary keratinocytes were treated for the indicated times with recombinant TGF-β2 (+) or heat inactivated TGF-β2 (–).Total cellular proteins were then isolated and analyzed by Western blot for Snail, pSMAD2 (reflective of activated TGF- signaling), and tubulin. Note the activation of Snail expression, peaking at 2 h post-TGF-β2 treatment and then disappearing thereafter. (C) Snail mRNA expression is transiently induced by TGF-β2. The experiment in (B) was repeated, and this time, total RNAs were isolated from keratinocytes treated with TGF-β2 for the indicated times. RT-PCR was then used with (+) or without (–) reverse transcriptase (RT) and with primer sets specific for Snail and GAPDH mRNAs. Note that Snail mRNA expression also peaked at 2 h, paralleling Snail protein. (D) TGF-β2 treatment results in enhanced activity of a Snail promoter-β-galactosidase reporter. Keratinocytes were transfected with a β-galactosidase reporter driven by a Snail promoter that is either WT (wt prom) or harbors a mutation in a putative pSMAD2/pSMAD4 binding site (mt prom). At 2 d posttransfection, cells were treated with either TGF-β or heat-inactivated TGF-β2 (inact) for the times indicated. β-galactosidase assays were then conducted, and results are reported as fold increase over a basal level of activity of 1. The experiment was repeated three times in triplicate, and error bars reflect variations in the results. TGF-β1 has been shown to induce Snail family members in hepatocytes and heart [ 15 , 31 ]. In keratinocytes, however, TGF-β1 inhibits keratinocyte growth and seems to be involved in triggering the destructive phase of the cycling hair follicle [ 32 ]. Of the loss-of-function mutations generated in each of the TGF-β genes, only the TGF-β2 null state blocked follicle development at the hair bud stage [ 32 ]. Thus, we turned towards addressing whether TGF-β2 might be involved in regulating Snail expression in keratinocytes isolated from the basal layer of the epidermis. Though there is no cell culture system available to specifically study placodal cells, these keratinocytes are their progenitors and are the closest approximation available to study the behavior of epithelial cells of the placode. Interestingly, treatment of cultured keratinocytes with as little as 5 ng/ml of TGF-β2 caused a rapid and transient induction of Snail ( Figure 5 B). Following this treatment, Snail protein was detected within 30 min, peaked at 2 h, and then declined thereafter. The induction of Snail appeared to be specific for the active form of the growth factor, as pretreatment of TGF-β2 for 10 min at 100 °C obliterated the response [ Figure 5 B, lanes marked (–)]. By contrast, although TGF-β receptor activation remained elevated during the duration of the experiment (as measured by the sustained phosphorylation of the downstream effector SMAD2) Snail expression could not be maintained ( Figure 5 B). Thus, although Snail expression correlated with phosphorylated SMAD2 (pSMAD2) induction, its decline seemed to rely on secondary downstream events. The rapid kinetics of Snail expression were reflected at the mRNA level, suggesting that Snail promoter activity in keratinocytes might be sensitive to TGF-β2 signaling ( Figure 5 C). To test this possibility, we engineered a transgene driving the β-galactosidase reporter under the control of approximately 2.2 kb of promoter sequence located 5′ from the transcription initiation site of the mouse Snail gene. At 2 d after transient transfection, keratinocytes were treated with TGF-β2 (t = 0) and then assayed for transgene activity over the same time course in which we had observed Snail protein induction. The results of this experiment are presented in Figure 5 D. Within 0.5 h of TGF-β2 treatment, Snail promoter activity had increased 3-fold, and by 2 h, it peaked to approximately 10-fold over control levels ( Figure 5 D). Thereafter, Snail promoter activity rapidly returned to the basal levels seen in unstimulated keratinocytes. The kinetics of Snail promoter activity closely paralleled those observed for Snail protein induction. Moreover, the stimulatory effects appeared to be specific to TGF-β2, since they were abrogated either by heat inactivation of the TGF-β2 protein or by mutation of a putative SMAD binding element located about 1.8 kb 5′ from the Snail transcription start site ( Figure 5 D). Taken together, these results suggested that in keratinocytes, TGF-β2 signaling results in a pSMAD2-dependent transient activation of the Snail gene, and that maintenance of Snail protein relies, in part, upon sustained promoter activity. The brevity of Snail gene and protein induction in TGF-β2 treated cultured keratinocytes resembled the temporal appearance of Snail mRNA and protein at the initiation of hair follicle morphogenesis in embryonic mouse skin. To test whether TGF-β2 might be required for Snail induction in hair bud formation in vivo, we first analyzed whether TGF-β2 was expressed in or around the hair bud. Consistent with previous observations [ 33 ], an anti-TGF-β2 antibody labeled developing hair buds ( Figure 6 A). This labeling appeared to be specific as judged by the lack of staining in follicle buds from mice homozygous for a TGF-β2 null mutation ( Figure 6 A; [ 34 ]). Moreover, the downstream effector of TGF-β2 signaling, pSMAD2, was also expressed in WT, but not TGF-β2 -null, hair buds ( Figure 6 B). Together, these data underscore the importance of the TGF-β2 isoform despite expression of both TGF-β1 and TGF-β2 in developing hair buds at this stage. Figure 6 TGF-β2 Is Necessary to Induce Snail Expression and Regulate Proliferation and E-Cadherin in the Hair Bud (A–D) Skins from TGF-β2 WT or KO E17.5 embryos were analyzed for expression of TGF-β2 protein (A), which is present in the epidermis and dermis as previously described [ 33 ] and in the hair bud, pSMAD2 (B), Snail (C), and Snail mRNA (D). Arrows point to the hair buds. (E) Western blot analyses of Snail expression in the skins of 2-wk-old K14-Smad2 transgenic (SMAD2 TG) and WT littermate (WT) mice. Antibody to tubulin was used as a control for equal protein loadings. The K14-Smad2 Tg mouse was previously shown to possess activated TGF-β signaling [ 35 ]. (F–G) Proliferation markers Ki67 (F) and pMAPK (G) are diminished in TGF-β2 -null hair relative to its WT counterpart. (H–J) TGF-β2 -null hair fails to down-regulate E-cadherin (H). Wnt and noggin signaling pathways are still intact in the TGF-β2 null hair as nuclear LEF-1 (I) and nuclear β-catenin (J) are still expressed. To further explore the possible relation between Snail and TGF-β2, we examined the status of Snail expression in TGF-β2 -null hair buds. As judged by immunohistochemistry, Snail protein was absent from E17.5 skin of TGF-β2- null embryos but not from that of control littermates ( Figure 6 C). This effect appeared to be exerted at the transcriptional level, since Snail mRNAs were also not found in TGF-β2 null hair buds under conditions in which the signal was readily detected in the hair buds of littermate skin ( Figure 6 D). Conversely, in 2-wk-old K14-Smad2 Tg mice, which display elevated TGF-β signaling in skin [ 35 ], Snail protein was readily detected by Western blot analyses, where it was not found in postnatal skin ( Figure 6 E). Taken together, these results provide compelling evidence that TGF-β2 is functionally important for inducing Snail gene expression in a pSMAD-dependent manner in developing hair buds. Whether pMARK activity also contributes to Snail induction was not addressed in the present study [ 15 ]. Although some hair buds still formed in TGF-β2 null skin, their number was reduced by approximately 50% [ 32 ]. Thus, although the pathway mediated by TGF-β2 signaling impacts the earliest step of epithelial invagination, it does not appear to be essential for bud morphogenesis. Consistent with this notion, basement membrane remodeling still took place in the TGF-β2 -null buds, as judged by immunofluorescence with antibodies against β4 integrin, an integral component of keratinocyte-mediated adhesion to its underlying basement membrane ( Figure 6 F). In contrast, TGF-β2 signaling appeared to be an important factor for the early proliferation that occurs in the developing hair buds, as judged by anti-Ki67 and anti-pMAPK immunofluorescence ( Figure 6 F and 6 G). If TGF-β2 stimulates Snail expression in developing buds, loss of this morphogen would be expected to affect the expression of genes that are typically repressed by Snail. Since a major target for Snail-mediated repression is the E-cadherin gene [ 12 , 13 ], we investigated the status of E-cadherin in TGF-β2 -null buds. As shown in Figure 6 H, hair buds in TGF-β2 null skin displayed elevated immunofluorescence staining relative to their WT counterparts. Previously we demonstrated that the concerted action of the extracellular signals Wnt and noggin are required for the generation of a LEF-1/β-catenin transcription complex to repress E-cadherin transcription at the onset of hair fate specification. As shown in Figure 6 I and 6 J, both WT and TGF-β2 null buds exhibited nuclear LEF-1 and β-catenin localization, signs that the Wnt-noggin signaling pathway was intact. These data suggest that during hair follicle morphogenesis, TGF-β2 functions subsequently to Wnt/noggin-mediated determination of hair fate. Moreover, through activation of Snail gene expression, TGF-β2 appears to work in tandem with these other morphogens to down-regulate E-cadherin levels, which contributes to the activation of proliferative circuitries. Discussion During budding morphogenesis, intersecting signaling networks from the epithelium and mesenchyme govern transcriptional, adhesive, polarity, and motility programs in these select groups of cells. The dynamic nuclear and cytosolic changes that take place during this time form the cornerstone for organ morphogenesis. Two major challenges in understanding the mechanisms underlying a particular budding process are to order the temporal sequence of external cues involved and then to dissect how the cells of the developing bud translate these signals into the downstream events of cellular remodeling, proliferation, and differentiation. Our studies here provide some insights into how these events are orchestrated during hair bud formation in developing skin. Signaling during Early Hair Follicle Morphogenesis Recent studies on hair bud morphogenesis suggest that Wnt signals likely from the epithelium and BMP inhibitory signals from the underlying mesenchyme converge to produce an active transcription factor complex involving β-catenin and LEF-1, which in turn plays a key role in specifying the hair follicle fate [ 4 , 29 , 30 , 36 , 37 ]. Sonic hedgehog (Shh) and TGF-β2 signaling also play essential roles in follicle morphogenesis, but in contrast to β-catenin null skin, in which follicle invaginations are absent [ 30 ], some hair buds still form in the absence of LEF-1, Shh, or TGF-β2 [ 32 , 38 ]. These likely reflect the first wave of follicle (i.e., guard hair) morphogenesis, which accounts for a small number (fewer than 5%) of hairs and is under distinct regulatory control. Guard hairs form in the absence of LEF-1 and TGF-β2, and we have found that they also fail to express Snail at the budding stage of development (unpublished data). How E-cadherin is regulated in guard hairs remains to be determined. Several candidates include other Snail family members such as Slug or twist, or alternatively, transcription factors involving β-catenin and a different member of the LEF-1/TCF/ Sry -type HMG box (commonly known as SOX) family [ 39 , 40 ]. Further investigation will be required to determine whether the signaling pathway we have elucidated here is a theme with multiple variations. TGF-βs are known to promote withdrawal of keratinocytes from the cell cycle [ 41 ]. Hence, when TGF-β2 protein was detected at the transition between the growing and destructive phases of the adult hair cycle, research initially and naturally focused on a role for this family member in cessation of growth and/or triggering apoptosis ([ 42 ] and references therein). However, in contrast to TGF-β1 -null skin, which exhibits an extended growing phase of postnatal hair follicles, TGF-β2 -null skin displays an embryonic block in follicle bud progression [ 32 ]. Although this phenotype is consistent with TGF-β2's embryonic expression patterns [ 33 ], about 50% of TGF-β2 null buds appear unable to progress to the down-growth phase, a feature that cannot be explained readily on the basis of previously established effects of TGF-βs. Our finding that TGF-β2 is upstream from Ki67 expression and MAPK activation lends further support to the notion that hair follicle keratinocytes at this early stage of development react to TGF-β2 signaling in a fashion opposite to that typically expected for TGF-β factors. This said, based upon pSMAD2 immunohistochemistry, the immediate steps of downstream signaling appeared to be intact. Thus, we surmise that the proliferative outcome is likely to be rooted in differences in the repertoire of activated SMAD target genes. In this regard, the positive effects of TGF-β2 on proliferation within the hair bud may be more analogous to what has been seen in progression of squamous cell carcinoma to metastatic carcinoma [ 43 ] rather than that typically observed for keratinocytes [ 44 , 45 , 46 ]. The prior identification of the Snail gene as a potential target of TGF-β signaling [ 15 ] was intriguing, given the temporal wave of Snail gene expression that occurs in the developing hair bud. The additional correlation between epithelial hyperproliferation and Snail transgene expression further fostered our interest in a possible link between TGF-β2 and Snail. Our functional studies demonstrate that without TGF-β2, Snail expression is abolished in the mutant hair buds, and conversely, in K14-Smad2 skin, Snail is ectopically activated. Moreover, our in vitro studies indicate that even sustained TGF-β2 exposure may cause only a transient induction of Snail, offering a possible explanation as to why Snail is so briefly expressed during hair follicle morphogenesis. An additional point worth mentioning is that prolonged expression of Tg Snail in postnatal skin resulted in morphological and biochemical signs of epithelial to mesenchymal transitions (unpublished data), underscoring why transient Snail expression may be so important during normal hair follicle morphogenesis [ 18 ]. At first glance, the sparsity in hair coat of K14- Snail Tg mice seemed indicative of a defect in follicle formation (see Figure 2 A). Closer inspection, however, revealed that not all hairs penetrated the hyperthickened Tg epidermis. Several factors may contribute to the seemingly normal follicle development in these mice. One obvious factor is the K14 promoter, which is elevated in the basal layer of the epidermis and the outer root sheath (ORS) of the hair follicle, but is markedly down-regulated in developing embryonic hair buds as well as in the postnatal hair progenitor cells. The K14 promoter is also less active in the ORS than epidermis and hence this might also account for the lack of apparent response of the ORS to ectopic Snail. Additional contributing factors could be the multiplicity of Snail family members and their differential expression, the saturation, and/or diversity of regulatory mechanisms that govern AJ formation, migration, and proliferation in the follicle ORS. Distinguishing between these possibilities must await the generation of mice harboring skin-specific loss-of-function Snail mutations. Links between Signaling, Transcriptional Cascades, Epithelial Remodeling, and Proliferation in the Hair Bud Previously, we discovered that early during hair follicle morphogenesis, E-cadherin gene expression is down-regulated concomitantly with the invagination of developing bud cells into the skin [ 4 ]. Because the timing of this event correlated with the activation of a LEF-1/β-catenin transcription factor complex [ 20 ], we were intrigued by the presence of a putative LEF-1/TCF binding site in the E-cadherin promoter. This prompted an investigation that subsequently led to our discovery that LEF-1/β-catenin can contribute to repression of E-cadherin gene expression in skin keratinocytes [ 4 ]. In the course of these studies, we also noted that Snail can also contribute to this process in keratinocytes in vitro, and our present studies revealed that Snail is expressed at the right place and time to be physiologically relevant in the process. In noggin -null embryonic skin, LEF-1 expression and subsequent activation of the LEF-1/β-catenin reporter gene is abrogated in the developing placodes. The corresponding failure of E-cadherin down-regulation underscores the importance of Wnt/noggin signaling in regulating this event in follicle morphogenesis [ 4 ]. Conditional gene targeting studies will be necessary to establish whether Snail family members also contribute to the down-regulation in E-cadherin gene expression that occurs during follicle formation. However, it is intriguing that K14-Snail Tg epidermis displayed a marked down-regulation in E-cadherin expression, thereby demonstrating its potential to do so in skin. Our prior findings showed that by elevating E-cadherin levels or by conditionally ablating α-catenin, hair follicle morphogenesis can be impaired [ 4 , 7 ]. The marked epidermal hyperproliferation seen in the K14-Snail Tg skin, coupled with the converse suppression of proliferation and Snail in TGF-β2 -null hair buds led us to wonder whether the down-regulation of E-cadherin during follicle morphogenesis might have a direct impact on elevating the proliferative state of these cells. Our Tg studies suggested that, at least in part through its regulation of E-cadherin , Snail is able to influence the subcellular localization of a variety of AJ-associated proteins. One of these appears to be Ajuba, which was previously shown to have the dual capacity to bind Grb-2 as well as α-catenin [ 9 , 10 ]. Our studies revealed that in skin keratinocytes that either harbor a conditional null mutation in α-catenin or that overexpress Snail, Ajuba develops an interaction with Grb-2 that is otherwise not observed in WT keratinocytes. The corresponding abilities of either Snail -transfected or Ajuba -transfected keratinocytes to exhibit elevated activation of the Ras-MAPK pathway suggest that the Grb-2 association of Ajuba under conditions of reduced levels of AJ proteins may be directly relevant to the parallel in hyperproliferation. In stable epithelial (i.e., Madin-Darby canine kidney, or MDCK) cell lines, Snail has been shown to block cell cycle progression and promote motility and shape changes for invasion [ 47 ]. While our in vivo studies are consistent with a role for Snail in motility and epithelial remodeling, they differ with respect to Snail's apparent proliferative effects. A priori, this could be simply due to variations in the response of different cell types to Snail expression. Alternatively, these differences may be relevant to the benefit of using mouse models to reveal functions not always recapitulated in stable cell line models. Future studies should highlight the underlying reasons for these opposing results. Irrespective of these differences, our in vivo studies do not stand alone, as there are many situations in which a down-regulation in AJ proteins correlate with enhanced proliferation. In fact, a myriad of diverse mechanisms have been implicated in activating epithelial proliferation upon down-regulation of AJ proteins [ 7 , 23 , 24 , 48 ]. Sifting through these converging pathways is likely to be a difficult and painstaking process. This said, by identifying the status of different players involved in specific cell types and at specific stages in development, our mechanistic understanding of how intercellular remodeling is linked to proliferation in epithelial morphogenesis should begin to surface in the future. Elucidating the molecular mechanisms through which these networks converge is also a prerequisite for understanding how these processes go awry during tumorigenesis. Materials and Methods Reagents Primary antibodies used were against: E-cadherin (M. Takeichi, Kyoto University, Japan); α-catenin, β-catenin, pMAPK, tubulin (Sigma, St. Louis, Missouri, United States), Ajuba (G. Longmore, Washington University, St. Louis, Missouri, United States); β4 integrin/CD104 (BD Pharmingen, San Diego, California, United States), laminin 5 (R. Burgeson, Harvard University, Cambridge, Massachusetts, United States), K5, K1, loricrin (Fuchs Lab), involucrin, fillagrin (Covance, Berkeley, California, United States), MAPK, pSMAD2 (Cell Signaling, Beverly, Massachusetts, United States); Grb-2 (Santa Cruz Biotech, Santa Cruz, California, United States); P-cadherin (Zymed Laboratories, South San Francisco, California, United States); HA (Roche Biochemicals), vimentin (Chemicon, Temecula, California, United States), Ki67 (Novo Castra, Newcastle Upon Tyne, United Kingdom), keratin 6 (P. Coulombe, John Hopkins University, Baltimore, Maryland, United States), cyclin D (Oncogene, San Diego, California, United States), and TGF-β2 (L. Gold, New York University, New York, New York, United States). FITC-, Texas Red-, or HRP-conjugated secondary antibodies were from Jackson ImmunoResearch (West Grove, Pennsylvania, United States). Biotinylated secondary antibodies were from Vector Labs (Burlingame, California, United States). Dilutions were according to the manufacturer's recommendation. The Snail antibody was generated in Guinea pigs by inoculating them with the N-terminal sequence of murine Snail fused to GST (Covance, Princeton, New Jersey, United States). Recombinant human TGF-β2 was purchased from R&D (Minneapolis, Minnesota, United States). Heat inactivated TGF-β2 was generated by heating the recombinant protein at 100 °C for 10 min. Mice The K14-Snail Tg mouse was generated by digesting the pcDNA3-mm Snail -HA plasmid (G. de Herreros, Universitat Pompeu, Fabra, Barcelona, Spain) with BamHI and NotI and subcloned into the K14 vector [ 49 ]. The linearized construct was injected into the nucleus of embryos from CD1 mice. The K14-Smad 2 Tg mouse was reported in Ito et al., 2001. The TGF-β2 knockout (KO) mouse was described in [ 34 ]. The shh KO mouse [ 38 ] and TOPGal mouse [ 20 ] have previously been reported. Western blot and immunoprecipitation Protein extracts from primary keratinocytes were generated either by lysing cells in lysis buffer (1% NP-40, 1% sodium deoxycholate, 20 mM Tris-Cl [pH 7.4], 140 mM NaCl containing 1 mM sodium vanadate, 2 mM phenylmethylsulfonyl fluoride, and protease inhibitors) or directly in Laemmli bβuffer and boiled. For skin tissue: Frozen tissue was pulverized in a liquid nitrogen-cooled Gevebesmascher and the powder scraped into a chilled microcentrifuge tube. RIPA buffer (1% Triton X-100 in PBS with 10 mM EDTA, 150 mN NaCl, 1% sodium deoxycholate, and 0.1% SDS) and protease inhibitors or Laemmli buffer was added. The cell suspension was sonicated three times for 15 s and centrifuged at 14,000 rpm at 4 °C. The supernatant was separated from the pellet and used in the experiments. Extracts subjected to immunoprecipitation were precleared with Protein G Sepharose (Amersham, Piscataway, New York, United States) and incubated with antibody with rocking overnight at 4 °C. Protein G Sepharose was added and samples were incubated for 1 h at 4 °C with rocking. Samples were washed three times for 5 min each in lysis buffer, and the Protein G Sepharose-antibody-antigen pellet was resuspended in Laemmli buffer and boiled for 10 min. Samples were run on SDS-PAGE and transferred to nitrocellulose membrane (Schleicher and Schuell Bioscience, Keene, New Hampshire, United States). Western blot signals were developed using the enhanced chemiluminescence kit from Amersham Cell culture Primary keratinocytes were culture in low-calcium medium as previously described [ 4 ]. Transient transfections were carried out with FuGENE6 reagent (Roche, Indianapolis, Indiana, United States) according to the manufacturer's protocol. Measurement of β-galactosidase or luciferase levels in promoter activity studies were carried out with the Galacto-Lite assay kit (TROPIX, Bedford, Massachusetts, United States) and the Dual luciferase (Promega, Madison, Wisconsin, United States), respectively. Runella luciferase was cotransfected into cells to correct for transfection efficiency. Experiments were done in triplicate and repeated at least three times. Measurements were done on a luminometer (MGM Instruments, Hamden, Connecticut, United States). For experiments measuring phosphorylation of MAPK, keratinocytes were serum starved for 3 h prior to harvesting of cells by incubation in medium lacking serum. Treatment of cells with Wnt- and noggin-conditioned medium was previously described [ 4 ]. Constructs The 2.2-kb murine Snail promoter was generated by PCR using a forward primer with an XbaI linker sequence, 5′-
TCTAGAATTGTTTGCTGCTGTATGGTCTTC-3′, along with a reverse primer with a BglII linker sequence, 5′-
AGATCTGTTGGCCAGAGCGACCTAG-
GTAG-3′, and mouse genomic DNA as a template. The PCR product was purified with the Gel Extraction Kit (Qiagen, Valencia, California, United States) and ligated into pCRII-TOPO TA vector (Invitrogen, Carlsbad, California, United States). The promoter was verified by sequencing and digested with XbaI and BglII and subcloned into the pβ-gal BASIC vector (BD Biosciences Clontech, Palo Alto, California, United States). The point mutations in the SMAD binding element was generated with the Quik-Change Kit (Stratagene, La Jolla, California, United States) using the forward primer 5′-
GGGCGGGCTTAGGTGTTTTCATTTACTCTTGAGGAAAAGCTTGGC-3′ and the reverse primer 5′-
GCTTTT-
CCTCAAGAGTAAATGAAAACACCTAAGCCCGCCCTGCCC-3′. The probes for the Snail in situ hybridization were generated against the 3′ UTR by PCR using the forward primer 5′-
ACCTTCTCCCGCATGTCCTTGCTCC-3′ and the reverse primer 5′-
CTGCTGAGGCATGGTTACAGCTGG-3′, and genomic DNA as a template. The PCR product was gel purified and ligated into pCRII-TOPO TA vector. The pre-LIM domain of Ajuba was generated essentially as described [ 9 ], but was fused to GFP by subcloning from the pEGFP-N1 20 vector (BD Biosciences Clontech) In situ hybridization The pCRII-TOPO TA vector containing a region of the 3′ UTR of Snail was used as a template to generate digoxigenin-labeled sense and antisense riboprobes (Roche). The respective probes were obtained by XhoI and BamH1 digestions. In situ hybridizations were performed on 10-μm thick sections of E17.5 mouse embryos. The sections were fixed with 4% PFA for 10 min at room temperature, prehybridized at room temperature for 4.5 h, hybridized with the probe (2 μg/ml) at 55 °C for 12–14 h, blocked with 10% NGS, and treated with anti-dig Fab-AP antibody (Roche #1093274) at a 1:2,500 dilution for 3 h. The sections were incubated with NBT and BCIP until adequate signal was detected. Immunofluorescence and immunohistochemistry Tissue samples for immunofluorescence were frozen in OCT and sectioned 10 μm thick on a cryostat. Sections were fixed in 4% paraformaldehyde for 10 min at room temperature, blocked, and stained with antibodies. Tissue samples for immunohistochemistry were fixed in 4% paraformaldehyde, dehydrated, and embedded in paraffin. Samples were sectioned on a microtome (10 μm thick) and rehydrated prior to staining with antibody. Samples stained with Snail, pMAPK, pSmad2, and cyclin D were antigen unmasked with 10 mM sodium citrate (pH 6) in an Antigen Retriever 2100 (Pickcell Laboratories, Leiden, Netherlands). The DAB substrate kit (Vector Labs) was used according to manufacturer's instructions to develop the signal. RT-PCR RNA was extracted from keratinocytes or skin tissue with Trizol (Invitrogen) according to the manufacturer's protocol. cDNA was generated using oligo-dT primers and the Superscript II kit (Invitrogen). The primers used for PCR were Snail forward: 5′-
CAGCTGGCCAGGCTCTCGGT-3′; Snail reverse: 5′-
GCGAGGGCCTCCGGAGCA-3′; GAPDH forward 5′-
CGTAGACAAAATGGTGAAGGTCGG-3′; and GAPDH reverse: 5′-
AAGCAGTTGGTGGTGCAGGATG-3′. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539061.xml |
521174 | Genetic Analysis of Lice Supports Direct Contact between Modern and Archaic Humans | Parasites can be used as unique markers to investigate host evolutionary history, independent of host data. Here we show that modern human head lice, Pediculus humanus, are composed of two ancient lineages, whose origin predates modern Homo sapiens by an order of magnitude (ca. 1.18 million years). One of the two louse lineages has a worldwide distribution and appears to have undergone a population bottleneck ca. 100,000 years ago along with its modern H. sapiens host. Phylogenetic and population genetic data suggest that the other lineage, found only in the New World, has remained isolated from the worldwide lineage for the last 1.18 million years. The ancient divergence between these two lice is contemporaneous with splits among early species of Homo, and cospeciation analyses suggest that the two louse lineages codiverged with a now extinct species of Homo and the lineage leading to modern H. sapiens. If these lice indeed codiverged with their hosts ca. 1.18 million years ago, then a recent host switch from an archaic species of Homo to modern H. sapiens is required to explain the occurrence of both lineages on modern H. sapiens. Such a host switch would require direct physical contact between modern and archaic forms of Homo. | Introduction One of the most intensely debated topics in evolutionary biology pertains to the origin of modern Homo sapiens. The debate concerns the precise manner in which anatomically modern humans arose from archaic ancestors. Empirical studies tend to support one of two prominent models of human origins, the Recent African Replacement model ( Stringer and Andrews 1988 ) or the Multiregional Evolution model ( Wolpoff et al. 1994 ). The Recent African Replacement model, as originally proposed, suggests that modern humans arose from an archaic ancestor in Africa ca. 130,000 years ago, and then replaced archaic humans in Asia, Africa, and Europe without introgression between archaic and modern humans. The Multiregional Evolution model (as proposed by Wolpoff et al. [1994] and revisited by Wolpoff et al. [2000] ) suggests that gene flow existed not only among populations of modern Homo sapiens, but also between modern H. sapiens and archaic forms of Homo (e.g., Homo neanderthalensis and Homo erectus ), which led to some degree of regional continuity. Both models can be subdivided into many variants. There are two common variants of the Multiregional Evolution model. In one variant, the transition from archaic to modern humans occurs incrementally across a large geographic region (i.e., both within and outside Africa); in the other variant, the transition from archaic to modern humans arises first in Africa then spreads through gene flow outside of Africa. This latter variant is very similar to a Diffusion Wave model recently put forth by Eswaran (2002) . Both types of models of human origins (the Recent African Replacement and Multiregional Evolution models) have been examined with both human fossil and genetic data, but no single model or variant has been supported by all the data. Fossils provide the only source of data available for most species of archaic humans and are therefore crucial to understanding the origin of modern humans. Unfortunately, missing taxa and fragmentary fossils limit our ability to reconstruct human evolutionary history based solely on fossil data. Molecular (DNA sequence) data have provided additional insight into the recent evolutionary history of humans, but these data are limited mainly to extant human populations. Ancient DNA was recently sequenced from H. neanderthalensis ( Krings et al. 1997 , 1999 , 2000 ) and a 24,000-year-old specimen of modern H. sapiens ( Caramelli et al. 2003 ), but even these ancient DNA studies do not agree on hypotheses of modern human origins ( Templeton 2002 ; Serre et al. 2004 ). Only a few ancient specimens have been examined molecularly, and additional sequences are slow to emerge. Furthermore, DNA may never be retrieved from some specimens because it is difficult, if not impossible, to liberate sequenceable DNA from poorly preserved ( Krings et al. 1997 ) or very old ( Paabo and Wilson 1991 ) fossil material. Therefore, the degree to which we can reconstruct human evolutionary history depends, in part, upon additional types of data. Several recent studies have inferred portions of human evolutionary history from the evolutionary history of their parasites ( Chan et al. 1992 ; Ho et al. 1993 ; Ong et al. 1993 ; Escalante et al. 1998 ; Ashford 2000 ; Leal and Zanotto 2000 ; Hoberg et al. 2001 ). Parasites can be a powerful tool for reconstructing host evolutionary history because they provide data that are independent of host data. For example, human papillomaviruses ( Chan et al. 1992 ; Ho et al. 1993 ; Ong et al. 1993 ), tapeworms ( Hoberg et al. 2001 ), and malarial parasites ( Escalante et al. 1998 ) each have evolutionary origins in Africa, consistent with most human fossil and molecular data. Human T-cell leukaemia/lymphoma virus (HTLV) sequences show that most human viral strains are closely related to those of Old World apes and monkeys ( Leal and Zanotto 2000 ). In contrast, some Native American strains of HTLV have closer affinities to viral strains from Asian primates, suggesting a dual origin for this virus in humans ( Leal and Zanotto 2000 and references therein). Ashford (2000) recently reviewed the use of parasites as markers of human evolutionary history, pointing out that five parasites of humans (lice, tapeworms, follicle mites, a protozoan, and bedbugs) have closely related taxonomic pairs that suggest periods of host geographic isolation. Unfortunately, none of these five pairs has been studied rigorously with the primary goal of inferring host evolutionary history. Of these parasites, the ones most likely to provide the greatest insight into human evolutionary history are those that are known to have had a long-term coevolutionary association with their hosts, such as lice (Insecta: Phthiraptera) ( Page 2003 ). Lice are obligate parasites of mammals or birds that complete their entire life cycle on the body of the host; they cannot survive more than a few hours or days off the host ( Buxton 1946 ). Mammal lice are closely tied to their hosts in both ecological ( Reed and Hafner 1997 ) and evolutionary ( Hafner et al. 1994 ) time. The lice found on primates are quite host specific, with most species occurring only on a single species of host ( Durden and Musser 1994 ). Host specificity is reinforced by the fact that primate lice require direct physical contact between hosts for transmission ( Buxton 1946 ; Durden 2001 ; Canyon et al. 2002 ; Burgess 2004 ). Host specificity often goes hand in hand with long-term coevolutionary patterns between hosts and parasites ( Page 2003 ), making primate lice excellent candidates for inferring host evolutionary history. Humans are parasitized by two species of lice: head/body lice (Pediculus humanus), the focus of this paper, and pubic lice (Pthirus pubis), which serve as a phylogenetic outgroup in this study. P. humanus is found in two forms (head and body lice) that are morphologically similar, but ecologically distinct. Body lice live primarily in clothing and move onto the skin to feed once or twice a day. Head lice are confined to the scalp and feed more frequently. Body lice vector the bacteria responsible for epidemic typhus, trench fever, and relapsing fever; head lice are not known to vector any agent of human disease under natural conditions ( Buxton 1946 ). Recent molecular work by Leo et al. (2002) showed that, despite the ecological differences between head and body lice, the two forms are not genetically distinct. Kittler et al. (2003) confirmed this finding but also discovered two deeply divergent clades within P. humanus that are uncorrelated with the head and body louse forms. The divergent clades of lice stand in contrast to mitochondrial sequence data from extant human populations, which coalesce to a single lineage very rapidly. The shallow coalescence in human mitochondrial sequence data is likely the result of a recent population bottleneck and subsequent population expansion ( Rogers and Harpending 1992 ), which obscures much of the evolutionary history of humans prior to the bottleneck. The deep divergences within P. humanus have the potential to reveal aspects of human evolutionary history that cannot be recovered from human DNA markers. We reconstructed the evolutionary history of P. humanus and several outgroup taxa using both morphological and molecular data. First, we used louse morphological data to test for patterns of cospeciation between primate lice and their hosts. Then we collected molecular data from a subset of the same taxa and calculated divergence dates for nodes in the louse phylogeny. This broad phylogenetic approach allowed us to date the origin of the human louse, P. humanus, and to date the two divergent lineages within the species. Finally, we collected population genetic data for P. humanus to compare with population-level characteristics of extant humans. Taken together, our phylogenetic and population-level data provide a well-resolved picture of the evolutionary history of P. humanus, which can be used to indirectly infer human evolutionary history. Specifically, we compared three distinct models of modern human origins (Recent African Replacement without Introgression, Multiregional Evolution, and Diffusion Wave) to see which model best fits the data from human lice. Results Phylogenetic Analyses and Divergence Estimates Both the morphological and molecular data sets produced a single phylogenetic relationship for the louse species in Figure 1 . The phylogeny shows that Pediculus species on chimpanzees and humans are sister taxa, which together with Pthirus form a clade that is sister to Pedicinus, the most basal member of the ingroup ( Figure 1 ). Bootstrap support for these relationships is high. Reconciliation analysis using Treemap v. 2.0 (M. A. Charleston and R. D. M. Page, software distributed by authors) revealed significant congruence ( p < 0.01) between the louse and primate phylogenies, thus validating the assumption of cospeciation ( Kittler et al. 2003 ). Reconciliation analysis using Treemap showed four cospeciation events and one host switch. One particular node of cospeciation determined that as cercopithecoid and hominoid primates diverged 20–25 million years ago (MYA) ( Benefit 1993 ; Leakey et al. 1995 ), Pedicinus diverged from the lineage leading to Pediculus and Pthirus. Since the nodes of cospeciation in congruent host and parasite trees are contemporaneous, the louse tree can be calibrated using the host tree. Figure 1 Phylogeny of Primate Lice from Morphological and Molecular Data The phylogeny is a strict consensus of morphology and a 1,525-bp fragment of COI and Cytb. Branch lengths were determined from the molecular data. Numbers in parentheses are bootstrap values from molecular and morphological data, respectively. Divergence dates are direct estimates from mtDNA data (see text). Louse images from light microscopy were taken by VSS. We used the date of 22.5 ± 2.5 MYA to calibrate the split between Pedicinus and Pthirus + Pediculus in the louse tree. This, in turn, yielded a divergence time of 11.5 MYA for the Pthirus / Pediculus split and 5.6 MYA for the split between Pediculus schaeffi and P. humanus ( Table 1 ). Our estimated divergence between chimp and human lice (5.6 MYA) is strikingly similar to the 5.5 MYA estimates for the chimp/human divergence based on both mitochondrial and nuclear sequence data ( Stauffer et al. 2001 ). To test the original calibration date of 22.5 MYA, we used the molecular estimate of the chimp/human split (5.5 MYA; Stauffer et al. 2001 ) to reverse calibrate the louse tree. This younger calibration point resulted in divergence estimates that were nearly identical to those from the previous calibration. For example, the 5.5 MYA calibration resulted in an estimated divergence of 22.65 MYA for the split between Pedicinus and Pthirus + Pediculus. Estimates of divergence time error were calculated from bootstrapped data sets ( Table 1 ). Other studies have shown that louse mitochondrial DNA (mtDNA) sequences evolve at a rate two to three times faster than that of host sequence rates ( Page 1996 ; Page et al. 1998 ). The lice in this study are evolving at ca. 2.3 times the rate of their primate hosts, when nucleotide substitutions are estimated under a best-fit model of sequence evolution. Table 1 Mean (± Standard Deviation), Minimum, and Maximum Estimates of Divergence Times (in Millions of Years) of Louse Lineages from 100 Bootstrapped Data Matrices Divergence dates were calculated without rate-smoothing methods because our data do not depart from the assumptions of a molecular clock (see text). Direct estimates of divergence times were calculated using the original data set (1,525-bp mtDNA) rather than the bootstrapped matrices. Trees were calibrated with a cercopithecoid/hominoid primate divergence of 22.5 MYA (from fossil evidence) except for the Pedicinus dates (asterisks), which were calibrated from the 5.5 MYA estimate of the human (P. humanus) -chimp (P. schaeffi) divergence date from molecular data (see text) Phylogenetic analysis revealed two divergent clades within P. humanus (6% uncorrected sequence divergence for cytochrome oxidase subunit I [COI] and cytochrome b [Cytb]). One of the two lineages in our data set is worldwide (WW) in distribution ( Figure 2 , Worldwide clade), contains both head and body louse forms, as determined by discriminant function analysis ( Figure 3 ), and has a most recent common ancestor (MRCA) 0.54 MYA ( Table 1 ). Even within this WW clade head and body lice are not reciprocally monophyletic, and a constraint to enforce such monophyly can be rejected using a Shimodaira-Hasegawa test ( p < 0.01) ( Shimodaira and Hasegawa 1999 ). The other lineage ( Figure 2 , New World clade) is restricted to the New World (NW), contains only the head louse form, and has a MRCA only 0.15 MYA. The MRCA of all P. humanus was 1.18 MYA, which predates by a considerable margin the origin of modern H. sapiens based on mtDNA (≤0.20 MYA; Cann et al. 1987 ; Vigilant et al. 1991 ; Ingman et al. 2000 ) as well as fossil evidence (0.15–0.16 MYA; White et al. 2003 ). Figure 2 Molecular Phylogeny of P. humanus from Geographically Diverse Human Populations This species exhibits distinct “head” and “body” forms, which differ in ecology, and slightly in size. Head lice (black lettering) are smaller than body lice (red lettering) and are confined to the scalp, whereas body lice live primarily in clothing. Haplotypes shown in green were found in both head and body lice. There are no fixed genetic differences between the head and body forms, suggesting a lack of reproductive isolation, despite the fact that the two forms can be distinguished using discriminant function analysis of morphological data. These results are consistent with experimental data showing that head lice can transform morphologically into body lice within a few generations ( Levene and Dobzhansky 1959 ). The Worldwide clade (red branches) shares a MRCA ca. 0.54 MYA and the geographically restricted New World clade (blue branches) has a much younger MRCA, ca. 0.15 MYA. Asterisks denote samples from Leo et al. (2002) Figure 3 Plot of the First and Second Canonical Discriminant Functions for Specimens of Adult Head/Body Lice (P. humanus) and Pubic Lice (Pthirus pubis) Solid points denote reference specimens of known identity acquired from museum collections. Unfilled points denote newly collected specimens used in the molecular analyses. In all cases the discriminant function analysis successfully classified each unknown case with a probability of >0.95. Our estimate of the age of the MRCA for P. humanus (1.18 MYA) is much older than that reported by Kittler et al. (2003) , which was only 0.53 MYA based on mtDNA. Their estimate of 0.53 MYA was determined using a mtDNA sequence from a specimen of the chimp louse, P. schaeffi, that is quite aberrant when compared to other primate lice. Phylogenetic analysis of the Kittler et al. Cytb data (downloaded from GenBank), combined with our own data, shows that the Kittler et al. sequence for P. schaeffi is 40% divergent from P. humanus and 40% divergent from our own sequence of P. schaeffi. Phylogenetic analysis places their specimen of P. schaeffi outside all other primate lice and even outside the rodent louse ( Figure 4 ), whereas our specimen of P. schaeffi is sister to P. humanus, based on both morphology and molecular data. We think that the Kittler et al. specimen has been attributed to the species P. schaeffi in error. In contrast to the mitochondrial data reported by Kittler et al. (2003) , our analysis of their nuclear elongation factor (EF1-alpha) sequences produces a MRCA for P. humanus that is ca. 2 MYA. Similarly, 18S rRNA sequences for P. humanus from Yong et al. (2003) , combined with an 18S rRNA sequence from P. schaeffi, provide a MRCA for P. humanus that is ca. 2 MYA (for GenBank accession numbers, see Supporting Information ). Together, these mitochondrial and nuclear markers support a MRCA for P. humanus greater than 1.18 MYA, which is an order of magnitude older than the MRCA for its human host. Figure 4 Neighbor-Joining Tree Using a Best-Fit Model of Nucleotide Substitution (Tamura-Nei + Γ) for a Combined Data Set of Cytb Sequences from Our Study and from Kittler et al. (2003) The clades of P. humanus identified by Kittler et al. (2003) are nearly identical to those from our data, with the exception of their basal African clade, which was not represented in our data set. One clade contains both head lice and body lice and is WW in distribution. Another clade is comprised solely of head lice from the NW (our data) and Europe (samples from Kittler et al. 2003 ), and the most basal clade contains isolates 4, 18, and 33 from Kittler et al. (2003) , which are head lice from Africa. The size of the triangles representing the three clades are proportional in size to the number of taxa within the clade. This phylogeny is rooted with a divergent louse, Dennyus hirundinus, which is a bird louse in the suborder Amblycera. Note the placement of the Kittler et al. (2003) specimen of P. schaeffi, which falls outside all other primate lice and the rodent louse Fahrenholzia. Population Genetic Analysis of P. humanus We can calculate an expected date of mitochondrial coalescence for P. humanus if we assume for the moment that the entire population of lice mated at random (i.e., panmixia). The estimate of expected coalescence is based on the effective female population size (N ef ), which was estimated from the sample of all P. humanus specimens to be 1.1 million female lice from the equation Θ = 2N ef μ. The estimate of N ef provides an expected coalescence time for the two divergent mitochondrial lineages of P. humanus of 1.10 million generations or ca. 0.11 MYA, which is an order of magnitude younger than the observed divergence time of 1.18 MYA. In a large randomly mating population consisting of 1.1 million female lice, one would expect to maintain two distinct haplotypes for only ca. 0.11 million years (MY). This suggests that we can reject panmixia if we assume that N ef prior to the bottleneck was roughly similar to what we see today. If estimates of N ef were drastically higher (ca. 60 million female lice) prior to the bottleneck, then expected time to coalescence could be much longer. The F st value, a measure of genetic population differentiation, calculated for the WW and NW clades was 0.96, indicating substantial population structure, which also supports the rejection of panmixia. The WW clade of P. humanus shows evidence of a recent population expansion (Fu and Li's D* = −2.80 [ Fu and Li 1993 ]; p < 0.02). We estimated the date of this population expansion from the mismatch distribution. The estimate was calculated by comparing the average pairwise difference within the WW clade of P. humanus (4.21 mutations) to the pairwise difference between P. humanus and P. schaeffi (220 mutations), which diverged 5.6 MYA. The population expansion of the WW clade is estimated to be 0.11 MYA, similar to the estimated date of population expansion of modern humans out of Africa ca. 0.10 MYA ( di Rienzo and Wilson 1991 ; Rogers and Harpending 1992 ; Harpending et al. 1993 ). In contrast, the NW clade of P. humanus does not exhibit the signature of a recent population expansion (Fu and Li's D* = 0.17), but instead shows a more stable population size. Contemporaneous Divergences in Pediculus and Archaic Homo spp. The age of the MRCA of P. humanus dates to 1.18 MYA (for mtDNA), which is roughly midway between the estimated ages of H. neanderthalensis (0.60 MY) and H. erectus (1.8 MY). We used a maximum likelihood (ML) analysis to test whether our two divergent lineages of lice could have diverged in tandem with H. sapiens and H. neanderthalensis (Neandertals). H. neanderthalensis is the only other species of Homo for which DNA sequence data are available ( Krings et al. 1999 ). The test evaluated whether relative branch lengths (scaled according to mutation rate) in the host tree, specifically for the branch between H. sapiens and H. neanderthalensis, are consistent with the parasite DNA sequence data ( Huelsenbeck et al. 1997 ). In cospeciating assemblages, host and parasite branch lengths are highly correlated due to a shared evolutionary history ( Page 1996 ). A likelihood ratio test (LRT) rejected ( p < 0.0001) the H. sapiens / H. neanderthalensis split as a node of cospeciation with the two clades of P. humanus because the branch length between H. sapiens and H. neanderthalensis is far too short to explain the louse DNA sequence data. In other words, the split between H. sapiens and H. neanderthalensis is too recent to have been contemporaneous with the divergence of the two lineages of lice. If one artificially lengthens the branch between H. sapiens and H. neanderthalensis to approximate the split between H. sapiens and H. erectus (anywhere from 1.2 to 1.8 MYA), the LRT fails to reject this hypothesis of cospeciation. Discussion Morphological and molecular data agree that primates and their lice have been cospeciating for over 20 MY. Indeed, it is this cospeciation that permits us to use host fossil evidence to calibrate portions of the louse phylogenetic tree. This has resulted in the discovery of two extant lineages of human lice that diverged 1.18 MYA. This ancient divergence is surprising because humans, and presumably their lice, are thought to have passed through a population bottleneck ca. 0.05–0.10 MYA ( Rogers and Harpending 1992 ). Such bottlenecks reduce genetic diversity by eliminating uncommon haplotypes, thereby making it less likely that multiple haplotypes survive bottleneck events. For example, mtDNA sequences from human populations coalesce to a single lineage very quickly (≤0.20 MYA), presumably the result of the population bottleneck. The deep divergences found in P. humanus could conceivably be the result of sequencing a nuclear copy of a mitochondrial gene. However, several lines of evidence strongly suggest otherwise. Because we amplified two different mitochondrial genes (COI and Cytb) that show the same divergent lineages and similar percent sequence divergences, copies of both mitochondrial genes would have had to enter the nucleus simultaneously, which is unlikely. In addition, we amplified each gene with a nested set of overlapping primers, and we never amplified more than one gene copy, even during bouts of cloning. Nucleotide base composition for our COI and Cytb data do not deviate from the mean values for all louse COI and Cytb sequences in GenBank (unpublished data), which would not be the case for a nuclear copy of a mitochondrial gene. Finally, the deep divergences seen in our mitochondrial genes are confirmed by preliminary analyses of nuclear data (EF1-alpha and 18S rRNA, unpublished data). Therefore, we are confident that the DNA sequences used in this study are mitochondrial in origin, and we must attempt to explain the occurrence of such ancient mitochondrial haplotypes in human lice. Gene Trees and Ancient Polymorphisms Gene trees (e.g., mitochondrial lineages) can be considerably older than species trees, and therefore our louse mitochondrial lineages could predate the actual origin of the species P. humanus (i.e., its speciation time). It is useful to determine an expected time to coalescence from the estimated N ef of 1.1 million female lice, even though this estimate seems high for a parasite of humans, who themselves have had very small effective population sizes (as few as 10,000 individuals) and recently went through a population bottleneck ( Rogers and Harpending 1992 ). Although we do not necessarily expect human and louse effective population sizes to be directly correlated, it is difficult to imagine that humans could have maintained such a large effective population of lice during a bottleneck event. Regardless, the expected time to coalescence was estimated to be 0.10 MYA, an order of magnitude younger than the observed divergence time of 1.18 MYA. The deeper gene tree that our data provide also could have been produced either by balancing selection or by subdivision of the louse population into several distinct groups with very limited gene flow. A Fu and Li test does not detect balancing selection when both lineages of P. humanus are evaluated together ( p = 0.11); therefore, we must consider the alternative explanation of extensive population subdivision. Population Substructure and Host Geographic Isolation Substantial isolation among populations of lice on modern H. sapiens could disrupt gene flow and allow the retention of very old lineages, making the age of P. humanus seem much older than it actually is. However, there is no evidence of such pervasive geographic isolation in the modern human hosts of these lice. Other species of lice have been shown to have substantial geographic substructure (i.e., isolation) even when hosts show no geographic isolation ( Johnson et al. 2002 ). If populations of P. humanus are more highly subdivided than those of their hosts, then we might expect P. humanus to have retained ancient mitochondrial polymorphisms, even through host bottleneck events. One prediction of this hypothesis would be that both clades of P. humanus (the WW and NW clades) would show signs of the recent population expansion of humans during the last ca. 0.10 MY. However, only the WW clade shows evidence of this event, which very closely matches the timing of human population expansion. Because the WW clade is commonly found worldwide, and shows a population expansion concurrent with that of modern H. sapiens, we conclude that this lineage has a common evolutionary history with modern H. sapiens. In contrast, the NW clade appears to have diverged from the WW lineage 1.18 MYA, and has had a distinctly different evolutionary history. We are left unable to explain the retention of two ancient louse lineages, each with a different evolutionary history, within the confines of a single host, modern H. sapiens. Given the history of cospeciation between primate lice and their hosts, it is necessary to look beyond modern H. sapiens to determine whether the two divergent lineages of P. humanus are legacies of a more ancient divergence. Contemporaneous Divergences in Pediculus and Archaic Homo spp. ML analyses rejected H. neanderthalensis as having diverged from H. sapiens contemporaneously with the two divergent lineages of lice. The mitochondrial MRCA of Neandertals and humans is 0.60 MYA ( Krings et al. 1997 ), which is only about half as old as the MRCA of the two ancient lineages of P. humanus, 1.18 MYA. The same ML test failed to reject the codivergence of these lice with H. erectus and H. sapiens when their divergence was set anywhere between 1.2 and 1.8 MYA. Therefore, the deep divergence within P. humanus is entirely consistent with a cospeciation event within the genus Homo ca. 1.2–1.8 MYA, but not 0.60 MYA. Unfortunately, no DNA sequence data exist for H. erectus or any other archaic species of Homo to enable a more direct test of cospeciation. There is much debate regarding the past 2 MY of hominid evolution. However, one area of broad agreement is that, prior to 2 MYA, our ancestors were confined to Africa, then left the continent ca. 1.8 MYA. This first migration out of Africa resulted in archaic species of Homo that were widespread in distribution, and at times both contemporaneous with, and geographically isolated from, the lineage leading to modern H. sapiens ( Figure 5 ). The 1.18 MY of isolation required to preserve the two ancient louse lineages must have occurred, in part, among these archaic species of Homo. It should be noted here that some interpretations of the Multiregional Evolution model do not necessarily consider modern H. sapiens to be a distinctly different species from archaic humans (e.g., H. erectus and H. neanderthalensis ). We refer to them as “species” mostly for convenience of writing. Whereas the WW lineage has population genetic characteristics that are similar to those of modern H. sapiens, the geographically restricted NW lineage does not. It likely evolved on a now extinct species of Homo only to switch to modern H. sapiens very recently. For example, Figure 5 depicts one possible scenario where the NW lineage evolved on H. erectus and switched to modern H. sapiens. Interestingly, Hoberg et al. (2001) reported that two species of tapeworms of humans diverged ca. 0.78–1.71 MYA, and one of the two species, Taenia asiatica, is entirely restricted to Asia. This is consistent with the depiction in Figure 5 , if one assumes that T. asiatica evolved on H. erectus. Although divergence dates are not available, it is intriguing that some Native American strains of HTLV have closer affinities to Asian primate strains than to other human strains of HTLV, suggesting an independent Asian origin of this virus in humans. One must still explain how these parasites came to be on modern H. sapiens, but taken together, the parasitological evidence (especially the deep divergences in tapeworms and lice) suggests that they might have evolved on H. erectus and switched recently to H. sapiens. If true, this implies that H. erectus was contemporaneous with modern H. sapiens in eastern Asia, as suggested by Swisher et al. (1996) , and it begs a discussion of recent human origins. Figure 5 Temporal and Geographical Distribution of Hominid Populations Redrawn from Stringer (2003) This figure depicts one view of human evolutionary history based on fossil data. Other interpretations differ primarily in the taxonomy and geographical distribution of hominid species. The temporal distribution of the two divergent lineages of P. humanus is superimposed on the hominid tree to show host evolutionary events that were contemporaneous with the origin of P. humanus. Whereas the NW lineage is depicted on H. erectus in this figure, several alternative hypotheses are consistent with our data when other evolutionary histories of hominids are considered (unpublished data). The WW clade is shown in red and the NW clade in blue (see text for descriptions of clades). Recent Human Origins Explanations of modern human origins are dominated by two competing models, Recent African Replacement and Multiregional Evolution. The Recent African Replacement model assumes that anatomically modern H. sapiens arose as the result of a speciation event ca. 0.13 MYA, and then replaced without introgression (i.e., admixture) non-African archaic humans (e.g., H. neanderthalensis and H. erectus ). In contrast, the sensu stricto Multiregional Evolution model assumes that modern H. sapiens evolved from early African descendants (up to 2 MYA). Characteristics of modernity were spread geographically through intercontinental gene flow, but local regional characteristics were maintained through admixture between modern and archaic forms ( Wolpoff et al. 1994 ). Most debates on modern human origins in the recent literature focus on one central question: “Was there admixture (i.e., introgression) between modern and archaic humans?” Both the Recent African Replacement and Multiregional Evolution models have been proposed with numerous variations that include introgression, one of which was recently put forth by Eswaran (2002) . Eswaran's Diffusion Wave model proposes that a diffusion wave of modern H. sapiens left Africa (ca. 0.13 MYA) and replaced archaic humans through a process of introgression, natural selection, and gradual demic expansion. The evolutionary history of P. humanus is somewhat consistent with all three models of modern human origins mentioned above; however, the number of ad hoc assumptions required to reconcile host and parasite evolutionary histories varies among the three views of human origins. Each model can account for the deep divergence between the two clades of P. humanus because each recognizes divergences between archaic species of Homo ca. 2 MYA. However, the model that best fits the louse data must account not only for the 1 MY of isolation between archaic and modern forms of lice, but also for a recent population expansion in just one louse lineage, the WW clade. The model must also explain how archaic louse DNA might have been incorporated into the lice of modern H. sapiens. The sensu stricto model of Multiregional Evolution ( Wolpoff et al. 1994 ) predicts continual gene flow between the geographically separated populations of humans following their early migration out of Africa (ca. 2 MYA), which is inconsistent with the louse data. This intercontinental gene flow among humans is required in the Multiregional Evolution model to maintain the continuity of the species H. sapiens. This scenario does not provide the louse populations with the degree of isolation necessary (ca. 1.18 MY) to maintain the two divergent louse lineages, unless we assume that gene flow between human populations was considerably greater than gene flow among their populations of lice. There is no reason to assume such a disparity in gene flow between hosts and parasites. The Multiregional Evolution model also predicts that we should detect the same genetic fingerprint of recent population expansion in both clades of P. humanus, which we do not. The Recent African Replacement model provides the isolation necessary between archaic and modern forms, because it assumes that modern H. sapiens left Africa ca. 0.10 MYA, more than a million years after archaic species of Homo left Africa, and that the modern and archaic humans remained distinct (i.e., no introgression). Furthermore, it explicitly assumes a recent population expansion in modern H. sapiens, which would account for the population expansion seen in our WW clade. However, in the strict sense, this model also predicts that modern H. sapiens replaced archaic forms of humans without introgression (i.e., hybridization), which leaves no obvious mechanism for archaic louse DNA to reach the lice of modern H. sapiens. This lack of host introgression implies, but does not require, a lack of direct physical contact between modern and archaic humans. It is conceivable that direct contact between modern and archaic humans was sufficient to allow the lice to switch hosts without making the assertion that the hosts were interbreeding. Therefore, the Recent African Replacement model is fairly consistent with the louse data, so long as one assumes some level of direct contact (e.g., fighting, sharing/stealing of clothing, etc.) between modern and archaic humans. Eswaran's Diffusion Wave model (2002) is similar to the Multiregional Evolution model in that it permits some level of introgression between modern and archaic humans. However, it is also similar to the Recent African Replacement model in that it assumes the same recent population expansion of modern humans out of Africa (ca. 0.10 MYA), thus providing both the isolation and population expansion necessary to accommodate our louse data. The additional assumption of introgression between modern and archaic forms of humans, which is proposed to have occurred only in the last 0.10 MY, provides a ready vehicle that would have transported archaic louse DNA into the modern louse population. Eswaran's model applied to lice suggests that at the beginning of the diffusion wave of modern H. sapiens leaving Africa (ca. 0.13 MYA), modern and archaic humans had distinct types of lice owing to ≥1 MY of isolation. As modern humans began to replace archaic forms, direct contact between hosts during introgression allowed archaic lice to switch to modern H. sapiens hosts. As previously stated, the recent literature addressing human origins boils down to models that do not permit introgression (strict-sense replacement models) and those of many types that do (admixture models, including variants allied with both the Multiregional Evolution and Recent African Replacement models). All things being equal, our parasite data are most consistent with a limited amount of admixture between modern and archaic humans, because this process presents the opportunity for host switching. However, introgression between modern and archaic humans over a protracted period of time would erode the isolation required to maintain the two louse lineages that we have observed. For example, some variants of the Multiregional Evolution model reject a single origin of modernity in Africa ca. 0.13 MYA in favor of a piecemeal acquisition of modern traits over a long period of time. This long-term admixture is precisely what would disrupt the isolation required to maintain the two louse lineages. Eswaran's Diffusion Wave model, on the other hand, confines admixture to the last ca. 0.10 MY. Our data cannot directly address whether host introgression occurred, because nonsexual, direct contact between hosts is sufficient for parasite transmission. We are confident that “direct contact” would be required for a host switch because these obligate parasitic lice cannot move between individuals without direct physical contact ( Buxton 1946 ; Durden 2001 ; Canyon et al. 2002 ; Burgess 2004 ) and furthermore, they die within 24 h of being removed from their host. However, an examination of Pthirus pubis, the human pubic louse, might shed light on the subject of human admixture because unlike head and body lice, pubic lice are primarily transmitted during intercourse. If our scenario involving lice switching from H. erectus to H. sapiens were true, then the host switch would have brought together two long-separated taxa of lice. It is impossible to know whether this long separation affected the reproductive compatibility of the two louse taxa once reunited. Discriminant function analysis shows no morphological differences between members of the two divergent molecular haplotypes of head lice (see Figure 3 ). There are other well-defined species of lice (e.g., see Johnson et al. 2002 ) whose populations show even greater sequence divergence (19% uncorrected sequence divergence) and yet have no discernible morphological differences between populations. It is likely that the two long-separated types of lice have experienced some level of introgression since their secondary contact on modern H. sapiens. The recency of this introgression of archaic louse DNA into modern lice also accounts for the younger coalescence time for the NW clade (0.15 MYA) compared to the WW clade (0.54 MYA). Presumably, the archaic form (i.e., morphotype) of louse either was extirpated along with its host or was assimilated into modern P. humanus. Regardless of the mechanism, ancient louse lineages can be found among the lice of modern H. sapiens. A recent review by Ashford (2000) reported five parasites that occur on humans as closely related pairs of taxa (lice, tapeworms, follicle mites, a protozoan, and bedbugs). The fact that there are five such pairs caused Ashford to ask, were humans once two distinct populations that rejoined after a long separation? The ancient divergences seen in mitochondrial data from P. humanus are clearly consistent with some level of long-term host isolation, and preliminary evidence from nuclear markers (EF1-alpha and 18S rRNA) reveals similarly ancient divergences (unpublished data). Furthermore, the two tapeworm species from humans showed amazingly concordant divergences ( Hoberg et al. 2001 ) and distributional patterns. Our data suggest that the isolation Ashford refers to may be between species of Homo rather than within modern H. sapiens itself. We conclude that the parasites may be very useful in the study of human evolutionary history, because they represent an independent marker of human evolution that has yet to be studied in detail. Materials and Methods Specimen collection and preparation. We collected human head and body lice (P. humanus) from many localities, ranging from remote areas such as the Papua New Guinea highlands to metropolitan areas like Boston ( Table 2 ). We also obtained P. schaeffi (from chimpanzees), Pthirus pubis (from humans), Pedicinus hamadryas (from baboons), and Fahrenholzia pinnata (from a rodent) to use as outgroup taxa in phylogenetic analyses. All lice were preserved in 95% EtOH and stored at −80 °C. DNA was extracted from lice by separating the thorax and abdomen and placing both in digestion buffer (Qiagen DNeasy tissue kit; Qiagen, Valencia, California, United States). Digestion proceeded for 48 h at 55 °C, then followed the manufacturer's protocol. After digestion, each louse was reassembled on a microscope slide as a voucher specimen corresponding to each DNA sequence. Voucher specimens were deposited in the Price Institute of Phthirapteran Research (PIPeR) collection at the University of Utah. Table 2 Specimens of P. humanus and Outgroup Taxa Examined in This Study, Their Collection Locality, and Number of Specimens Examined a From this study b From Leo et al. (2002) c Downloaded from GenBank (see Supporting Information for accession numbers) Phylogenetic analyses: morphological data. Considerable morphological variation exists among different species of primate lice. We examined 155 unordered morphological characters for 113 specimens of P. humanus (from humans), P. schaeffi (from chimpanzees), Pthirus pubis (from humans), Pthirus gorillae (from gorillas), Pedicinus hamadryas (from baboons), and F. pinnata (from a rodent). Morphological data were scored in the software package MacClade v. 4.05 (W. P. Maddison and D. R. Maddison; Sinauer, Sunderland, Massachusetts, United States), and heuristic searches consisting of random stepwise addition (1,000 replicates) and tree bisection/reconnection branch swapping were performed in PAUP* v. 4.0b10 (D. L. Swofford; Sinauer). Branch support was estimated with bootstrapping (tree bisection/reconnection swapping, 1,000 replicates). The complete data matrix is available from TreeBASE ( http://www.treebase.org/ ) as study accession number SN1969. Primate-louse cospeciation. The morphological data set (six species, 155 characters) was compared to the host phylogeny ((((human, chimp), gorilla), baboon), rodent) using reconciliation analysis in Treemap v. 2.0 with default parameters. Treemap determines whether the two phylogenies are more congruent than expected by chance based on randomizations of both the host and parasite phylogeny. Significant congruence between host and parasite phylogenies is interpreted as being the result of a shared evolutionary history (i.e., repeated bouts of cospeciation). Discriminant function analysis. We examined the morphology of P. humanus lice in detail to test for morphological correlates of the differences detected at the molecular level. Busvine (1978) examined a large series of head and body louse specimens and found no discrete morphological differences between the two forms. However, he noted that several morphological characters related to size and shape might be useful in this regard. To test this hypothesis we measured head width, thoracic width, total body length, and second-leg tibia length from a series of 50 slide-mounted adult museum specimens collected by earlier workers prior to our study. The microhabitat (head or body, i.e., clothing) from which these museum specimens were collected was well documented. Canonical discriminant analysis was used to build a predictive model to attempt to distinguish between the head and body forms of P. humanus. The predictor variables were used to build a set of discriminant functions that maximized variation among groups while minimizing within-group variation. The first two canonical discriminant functions explained 100% of the variation within the data. These discriminant functions, which were built using existing museum specimens, were then applied to our newly collected specimens in a blind test to determine whether the specimens could be identified as head or body lice from morphology alone. We were able to classify our samples as head or body lice with a probability of ≥0.95. Indeed, the assignment of adult specimens proved to be 100% accurate when checked against microhabitat data for the new specimens. Phylogenetic analyses: molecular data. Fresh specimens suitable for the collection of molecular data were obtained for five of the six species of lice. We sequenced 1,525 combined base pairs (bp) of the mitochondrial (mtDNA) genes COI (854 bp) and Cytb (671 bp) from 69 individuals of P. humanus, P. schaeffi, Pthirus pubis, Pedicinus hamadryas, and F. pinnata. PCR primers were as follows: (5′−3′) COI, C1-J-1718 GGAGGTTTTGCTAATTGATTAG and H7005 CCGGATCCACNACRTARTANGTRTCRTG; Cytb, L11122 GAAATTTTGGGTCWTTRCTNGG and H11823 GGCATATGCGAATARGAARTATCA. PCR parameters included 94 °C for 30 s, 48 °C for 30 s, and 72 °C for 1.5 min (five cycles), then 30 cycles of 94 °C for 30 s, 52 °C for 30 s, and 72 °C for 1.5 min. Amplified fragments were sequenced in both directions, assembled using Sequencher v. 4.1 (GeneCodes, Ann Arbor, Michigan, United States), and deposited in the NCBI database (see Supporting Information ). To ensure that we were not amplifying nuclear copies of mitochondrial genes, we performed additional PCR amplifications using nested sets of overlapping primers. The computer program modelTest ( Posada and Crandall 1998 ) was used as a guide to determine a best-fit ( Cunningham et al. 1998 ) ML model for the molecular data. This model (GTR+I+G) was incorporated into ML branch and bound and heuristic searches in PAUP* with 100 bootstrap replicates. An LRT was used to compare ML estimates from a clock-enforced and an unconstrained analysis. Our data did not depart significantly from the assumption of a molecular clock. Dating nodes in the louse phylogeny. We used the significant cospeciation shown between the primate and louse phylogenies (see Results ) as a basis for dating nodes in the louse phylogeny. We used a calibration point of 22.5 ± 2.5 MYA for the split between Pedicinus and Pediculus + Pthirus. The date is based on fossil evidence ( Benefit 1993 ; Leakey et al. 1995 ) of the split between cercopithecoid primates that host only lice in the genus Pedicinus and hominoid primates that host only lice in the genera Pthirus and Pediculus ( Durden and Musser 1994 ). All reconciliations of the host and parasite trees in our cospeciation analysis determined that this particular host/parasite node represents a cospeciation event (see Results ). Using the computer software r8s (M. J. Sanderson, software distributed by the author), we constrained the divergence of Pedicinus + Pthirus/Pediculus to 22.5 ± 2.5 MYA and allowed all other nodes in the louse tree to be determined from our DNA sequence data. Error estimates on divergence dates were calculated by generating 100 bootstrapped data matrices in Phylip (J. Felsenstein, software distributed by the author). Each of these bootstrapped datasets was calibrated with the same 22.5 ± 2.5 MYA divergence. Cospeciation within Homo An ML-based analysis was used to determine whether deep divergences in the louse tree were contemporaneous with divergences of now extinct species of Homo. We calculated the branch length between H. sapiens and H. neanderthalensis (the only species of Homo for which DNA sequence data are available) based on the human mitochondrial hypervariable region II of the D loop (see Supporting Information ). This value was scaled according to the average distance between these two taxa and their sister taxon (chimpanzee), providing a relative branch length within the primate tree (e.g., the branch between human and Neandertal is one-fifth the length of the branch that unites them with the chimpanzee). This relative branch length was incorporated into a louse constraint tree, in effect forcing the two clades of P. humanus to conform to a prescribed relative branch length. The resulting likelihood score was compared with the unconstrained tree score using an LRT (d.f. = taxa − 2). If the constrained and unconstrained tree scores are not significantly different, then the host tree topology and branch lengths describe the parasite DNA sequence data as well as the parasite tree itself ( Huelsenbeck and Crandall 1997 ). However, if a significant difference is detected, then the host tree does not fit the parasite data well enough to be explained by cospeciation, and the hypothesis of cospeciation is rejected. Population genetic analyses. Population genetic analyses were performed on a pruned dataset, which contained only specimens of P. humanus (i.e., no outgroup taxa). The computer software package DnaSP ( Rozas and Rozas 1999 ) was used to generate mismatch distributions, to calculate Fu and Li's D* statistic for the P. humanus clades, and to calculate additional population parameters (e.g., F st and Θ). Ten additional haplotypes from Leo et al. (2002) were used in these population-level analyses (see Supporting Information ). To estimate an expected time to coalescence for P. humanus, we used estimates of theta (Θ, from the software package DnaSP [ Rozas and Rozas 1999 ]) and mutation rate (μ) for P. humanus (see below) to calculate louse N ef from the equation Θ = 2N ef μ. The mutation rate (μ = 9.0 × 10 −9 substitutions per site per generation) was calculated by determining the expected number of substitutions per site between P. humanus and P. schaeffi under the Tamura-Nei + Γ model of nucleotide substitution. This mutation rate for P. humanus is roughly five to six times faster than that of human mtDNA, excluding the D-loop ( Ingman et al. 2000 ), when both mutation rates are scaled to absolute time (i.e., number of substitutions per site per year). N ef was then used to determine the expected time to coalescence (in generations) given the formula 2N ef (1 − 1/ n ), where n is the number of haplotypes detected in the population. One can also ask the similar question, what is the probability that two lineages, which are expected to differ by k e substitutions, actually differ by k * or more substitutions? The expression used, (k e /(1+ k e ) ^ k * ), is derived from the geometric distribution (see, for example, Golding and Strobeck 1982 ) and for our data suggests that the large number of substitutions found between the WW and NW clades is far greater than that which is expected ( p < 0.0006). Supporting Information Accession Numbers GenBank ( http://www.ncbi.nlm.nih.gov/Genbank/ ) accession numbers for items discussed in the text are as follows: the EF1-alpha sequences from Kittler et al. (2003) , AY316794–AY316834; the 18S rRNA sequences for P. humanus from Yong et al. (2003) , AY236410–AY236418, AF139478–AF139482, AF139484, AF139486, and AF139488; the 18S rRNA sequence from P. schaeffi, AY695939; the Cytb sequence from P. schaeffi, AY316793; the human mitochondrial hypervariable region II of the D loop, M76311, AY195756, AY217615, AF282972, AF142095, X97709, X98472, X93336, X93337, X93347, and X93348; the ten haplotypes from Leo et al. (2002) used in population-level analyses, AF320286; the divergent louse Dennyus hirundinus shown in Figure 4 , AF545694 and U96434. The amplified PCR fragments discussed in Materials and Methods have been deposited in the NCBI database under accession numbers AY695939–AY696069. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC521174.xml |
516041 | Expression and importance of matrix metalloproteinase 2 and 9 (MMP-2 and -9) in human trophoblast invasion | Background The aim of this study was to examine the invasiveness of first trimester trophoblasts according to the secretion profile of MMP-2 and -9 at different gestational stages, and to test the similarity between primary trophoblast cell-culture and the JAR choriocarcinoma cell-line. Methods First trimester trophoblasts were divided into two groups: 6–8 weeks (early) and 9–12 w (late) of gestation. The two trophoblast groups and JAR cells were cultured in medium, with various concentrations of forskolin and Epidermal Growth Factor (EGF). Proteolytic activity was detected by zymography and invasiveness was assessed by Matrigel invasion assay. Student's T-test was used for statistical analysis. Results In 6–8 w trophoblast, proMMP-2 was only slightly dominant over proMMP-9 (53.2% vs. 46.8% respectively), whereas in 9–12 w, proMMP-9 was clearly dominant over proMMP-2 (61.7% vs.38.3% respectively). In JAR cells proMMP-2 was strongly dominant (90.2% vs.9.8% respectively). In JAR cells forskolin significantly increased proMMP-2 and -9 secretion (128.5% +/- 12 and 183.2% +/- 27.9 of control, respectively). EGF had a dual effect on JAR cells: at 8 ng/ml both proMMP-2 and -9 were increased (133.5% +/-15 and 223.9% +/- 32.4 of control, respectively) while at 80 ng/ml both proMMP-2 and -9 were decreased (65.1% +/- 18.3 and 66.6% +/- 37 of control, respectively). Forskolin significantly increased both proMMP-2 and -9 secretion in 6–8 w and 9–12 w trophoblasts (125.9% +/- 6.3,128.4% +/- 6.4; 169.7% +/- 20.3, 120.3% +/- 4.5 of control, respectively). EGF also significantly increased both proMMP-2 and -9 secretion in 6–8 w and 9–12 w trophoblasts (141.22% +/- 14.8, 138.8% +/- 10.3; 168.3% +/- 18.2, 117.3 +/- 3.8 of control, respectively). Both forskolin and EGF increased trophoblast cells invasiveness in all groups. The invasive ability of trophoblast cells, induced by forskolin, was reduced by MMP-2 antibody in: JAR cells, 6–8 w and 9–12 w trophoblasts. Likewise trophoblast invasion induced by EGF was reduced by MMP-2 antibody in all groups. However the invasive ability induced by forskolin or EGF was inhibited by MMP-9 antibody only in trophoblasts from 9–12 w. Conclusions First trimester trophoblasts express differential gelatinase secretion profile according to the gestational week. In JAR and early trophoblasts (6–8 w) MMP-2 is the main gelatinase and the key enzyme in trophoblast invasion. Thereafter in late first trimester trophoblasts (9–12 w), both MMP-2 and -9 participate in trophoblast invasion. | Background Successful implantation depends on the ability of the embryo to degrade the basement membrane of the uterine epithelium and to invade the uterine stroma. Cytotrophoblastic cells (CTB) are derived from trophoectodermal cells of the blastocyst. CTB ensue to become either the villous cytotrophoblastic cells which will proliferate and differentiate by fusion to form the syncytiotrophoblast, or they will stream out of the syncytiotrophoblast to form mononuclear multilayered invasive extravillous cytotrophoblastic cells. The temporal and spatial regulation of trophoblast invasion is mediated in an auto-and paracrinic way by trophoblastic and uterine factors [ 1 ]. Several factors have been studied, including hormones, cytokines and growth factors [ 2 , 3 ]. Trophoblast invasion is facilitated by degradation of the extracellular matrix of the endometrium/decidua by various proteinases, among them, the matrix metalloproteinases (MMPs). The tissue inhibitors of matrix metalloproteinases (TIMPs) inhibit the activity of the MMPs by binding to the highly conserved zinc-binding site of active MMP [ 4 ]. Successful implantation and trophoblast invasion are closely linked to the expression of MMPs, which are able to degrade basement membranes. The gelatinases (gelatinase A: MMP-2: 72-kDa and gelatinase B: MMP-9: 92-kDa) which degrade collagen IV, the main component of the basement membrane, are expressed by trophoblast cells and are therefore regarded as key enzymes in the invasion process [ 5 ]. Several studies have shown that MMP-2 and MMP-9 synthesis and activation are required for trophoblast invasion [ 1 , 5 - 9 ]. Some studies have found either MMP-9 [ 5 , 7 , 8 ], or MMP-2 [ 9 - 11 ] to be more pronounced during the first stage of trophoblast invasion. However, the exact changes in protease expression during the first trimester are still not clear. Xu et al [ 11 ] found differential expression of MMP-2 and -9 in first trimester trophoblast cells, with MMP-2 being the main gelatinase secreted until 9 w and hereafter MMP-9. In this study, trophoblast cells from first trimester were therefore divided into two groups according to their MMP secretion profile. The aim of this study was to examine the expression and importance of MMP-2 and -9 in human trophoblast invasion, and to test the similarity between primary trophoblast cell culture and the JAR choriocarcinoma cell-line. The JAR cell-line serves as a widely used model for 1 st trimester trophoblasts [ 12 - 14 ]. The limited availability of 1 st trimester trophoblast tissue often requires the use of such a model, and therefore a comparison study between JAR cell-line and 1 st trimester trophoblasts is of significant importance to ensure similarity. This study shows a differential, dynamic importance of each gelatinase in trophoblast invasion during the 1 st trimester. Methods Cell culture The JAR (Jar, HTB 144) human choriocarcinoma line was established from a trophoblast tumor of the placenta (1988 American Type Culture Collection Catalogue). The JAR cells were a generous gift from Prof. Hochberg A Department of Biological Chemistry, Hebrew University, Jerusalem, Israel. JAR cells (1 × 10 4 cells/well) were cultured in M-199 medium (Beit-Ha'Emek, Israel) containing 10% Fetal Calf Serum (FCS, Beit-Ha'Emek, Israel) and penicillin/streptomycin (Beit-Ha'Emek, Israel). Cell culture was maintained in a humidified atmosphere containing 5% CO 2 at 37°C. After 24 hours of culture to facilitate cell attachment, the medium was removed, and M-199 medium with 1.5% serum supplemented with antibiotics was added. The cells were cultured with various concentrations of: a) Forskolin 1–100 μM, b) Epidermal Growth Factor (EGF), 0.8–80 ng/ml (Sigma). Control consisted of M-199 with 1.5% serum alone. Cells were cultured for an additional 48 hours, and media were removed for analysis of MMP secretion and stored at -20°C until use. Cell count was performed with XTT in order to normalize MMP secretion to cell number. Isolation and cultivation of human cytotrophoblast cells Human trophoblast cells were obtained from legal abortions (6 to 12 weeks of gestational age), with the approval of the local ethical committee (in compliance with the Helsinki Declaration) and the consent of the participating patients. Trophoblast cells were isolated as described previously in detail elsewhere [ 8 , 9 , 15 , 16 ] with modifications. Briefly, tissues were digested by 0.25% trypsin (Sigma) and DNase I (Sigma) at 37°C, then trophoblast cells were separated from blood cells and decidua on a discontinuous Percoll gradient (Sigma) and immunopurified with magnetic antibody CD45RB (DAKO, Denmark). The cells were plated at 1–2 × 10 5 cells/well in 96-well plates or in Transwell plates (Corning) with M-199 medium supplemented with 1.5% FCS and 1% penicillin/streptomycin and kept in 5% CO2 at 37°C. Inducers (10 μM forskolin or 8 ng/ml EGF (chosen as working concentrations after a dose-response study in JAR cell-line) were added to medium, and after 48–72 h media were collected for analysis of MMP secretion and cell count was performed. This method supplies a 95–98% purity of trophoblastic cells, including all trophoblastic sub-groups. We verified the purity of trophoblast cells by using immunohistochemistry with specific antibodies to cytokeratin 7 (positive) and vimentin (negative), commonly used for indication of trophoblast purity [ 7 , 8 , 15 ]. Figure 1 shows representative pictures of this analysis. Figure 1 Representative immunohistochemical analysis of isolated placental cells (6–8 w) after 24 h in culture. (A) Cells stain positive for anti-human cytokeratin 7 (1:100, Biogenics). (B) Cells stain negative for anti-human vimentin (1:200, Zymed). Magnification: ×100. Cell count assay Evaluation of cell proliferation was performed with XTT Reagent kit (XTT, cell proliferation kit, Beit-Ha'Emek, Israel) according to manufactures protocol. This is based on the activity of mitochondria enzymes in live cells, reducing tetrazolium salts, XTT, into colored formazan compounds, which can be detected colorimetric with a spectrophotometer at 450 nm (ELISA reader). Dye absorbance is proportional to the number of cells in each well. Substrate-gel-electrophoresis (zymography) In order to detect proteolytic activity in conditioned media (CM) collected after 48–72 h culture, substrate-gel-electrophoresis (Zymography) on gels containing gelatin as the substrate were used as was described by our previous manuscript [ 4 ]. Briefly, CM, was diluted in sample buffer (5% sodium dodecyl sulphate (SDS), 20% glycerol in 0.4 mol/l Tris, pH 6.8 containing 0.02% Bromophenol Blue without 2-mercaptoethanol) and electrophoresed, through a 10% polyacrylamid gel containing 0.5% gelatin (50 mg/ml). Afterwards gels were washed twice in 2.5% Triton X-100 for 15 min. and incubated for 24 h at 37°C in 0.2 mol/l NaCl, 5 mmol/l CaCl 2 , 0.2% Brij 35 and 50 mmol/l Tris, pH 7.5. The buffer was decanted and the gels stained with Coomassie Blue G in 30% methanol and 10% acetic acid for 10 min at room temperature on a rotary shaker. Stain was washed out with water until clear bands were seen. Areas where proteolytic activity degraded the gelatin were seen as absence of staining. Identification of each gelatinase band was done in accordance to their molecular weight and commercial standards (gelatinize A and B, 7 μl; Oncogene Science, Cambridge, MA, data not shown). These bands (proMMP) were quantified using the BioImaging gel documentation system (Dinco & Renum, Jerusalem, Israel) endowed with TINA software (Raytest, Staubenhardt, Germany). MMP secretion was expressed as percent of control. Matrigel invasion assay Matrigel invasion assay was prepared in our laboratory with modifications as described in detail elsewhere [ 17 , 18 ]. Briefly, diluted 1:10 Matrigel (1 mg/ml) (BD Biosciences, Beit-Ha'Emek, Israel) in serum free cell culture media was added to upper chamber of 24-well transwell plate, and incubated at 37°C 3–4 h for gelling. JAR Cells were harvested from tissue culture flasks by Trypsin/EDTA, washed and resuspended in 1.5% FCS in M-199 medium and added to upper wells at a density of 10 5 cells/well in 200 μl medium, while 500 μl medium was added to lower well. 1 st trimester trophoblast were cultured in upper wells at a density of 2 × 10 5 cells/well in 100 μl medium. The same density of cells, in the absence or presence of activators, was seeded in a well without transwell and counted at time of the invasion assay, as reference of total cells. Preliminary studies found no significant matrigel-mediated changes in multiplication rates between 6–8 w and 9–12 w trophoblasts, whether seeded on matrigel or at plastic bottom of well (data not shown). Activators (10 μM Forskolin or 8 ng/ml EGF) and inhibiting MMP-2 or MMP-9 antibodies (Oncogene Cat. IM33L, Cat. IM09L; concentration as recommended by manufacture) were added to medium in upper and lower wells. Plates were incubated at 37°C for 36–48 hours, and then non-invaded cells on top of the transwell were scraped off with a cotton swab. The amount of invaded cells in the lower well as a percent of total seeded cells was evaluated with XTT Reagent kit. The percent of invasion was calculated as: Invasion was expressed as Invasion Index (Percent of control). Statistical methods Results are expressed as mean ± SEM of 5–10 independent experiments, each treatment performed in duplicates. Statistical analysis was performed using the SPSS statistical software. Student's t-test and "one way analysis of variance" (ANOVA) were used when appropriate. P < 0.05 was considered significant. Immunohistochemistry Immunohistochemistry was performed as previously described [ 19 ] using the Histostain-Plus kit (Zymed laboratories Inc., USA). Briefly, cultured cells were fixed with cytospray for 20 min and quenched with 3% hydrogen peroxidase in methanol to eliminate endogenous peroxidase activity. The slides were washed, blocked and incubated at room temperature with primary antibodies (mouse anti-human cytokeratin-7 (1:100, clone OVTL12/30, Biogenics) and mouse anti-human vimentin (1:200, clone V9, Zymed laboratories Inc., USA). Secondary antibody used: Histostain-Plus broad-spectrum biotinolated second antibody (Zymed laboratories Inc., USA). Slides were then developed with a substrate-chromagen solution of aminoethyl carbazole (Zymed laboratories Inc., USA). Results Relative secretion profile of proMMP-2 and proMMP-9 in 6–8 W, 9–12 W trophoblasts and in JAR cells without treatment JAR cells (1 × 10 4 cells/well), 1 st trimester trophoblast cells 6–8 w and 9–12 w (1–2 × 10 5 cells/well) were incubated for 48 hours, then media collected and gelatinase secretion analyzed by zymography. Figure 2 summarizes the results. In 6–8 w trophoblast, proMMP-2 secretion was only slightly dominant (statistically not significant) compared to proMMP-9, 53.2% vs. 46.8% respectively (SEM ± 4.3). In 9–12 w trophoblasts the picture was different, with proMMP-9 being dominant (P < 0.05) over proMMP-2, 61.7% vs.38.3% respectively (SEM ± 4.6). In JAR cells proMMP-2 was dominant (P < 0.05), whereas proMMP-9 only had a small contribution to the gelatinase secretion, 90.2% vs.9.8% respectively (SEM ± 1.4). Figure 2 (A) Representative secretion pattern of proMMP-2 (72 kD) and proMMP-9 (92 kD) in 6–8 w, 9–12 w trophoblasts and in JAR cells without treatment as examined with zymography. (B) Bar graph describing the relative percentage of gelatinases secretion, representing mean ± SEM from 5 independent experiments. Black bars represents proMMP-2 and white bars represent proMMP-9, *P < 0.05. Dose-dependent effect of forskolin and EGF on the proMMP-2 and -9 secretion by JAR cells JAR cells (1 × 10 4 cells/well) were incubated 48 hours in the absence or presence of forskolin (1 μM, 10 μM or 100 μM) or EGF (0.8 ng/ml, 8 ng/ml or 80 ng/ml) and media was analyzed by zymography for gelatinase secretion. Fig 3 summarizes the results. 10 μM forskolin significantly enhanced secretion of proMMP-2 compared to control (128.5% ± 12.0, P < 0.05). Forskolin (1 μM and 10 μM) significantly enhanced proMMP-9 secretion compared to control (131.3% ± 11.1, and 183.2% ± 27.9, P < 0.05, respectively) (Fig. 3A ). 8 ng/ml EGF significantly enhanced secretion of proMMP-2 (133.5% ± 15.0, P < 0.05) and of proMMP-9 (223.9% ± 30.4, P < 0.05) compared to control. 80 ng/ml EGF, on the contrary, decreased proMMP-2 and proMMP-9 secretion compared to control (65.1% ± 18.3, and 66.6% ± 3.7, P < 0.05) (Fig. 3B ). Figure 3 Dose-dependent effect of Forskolin and EGF on gelatinase secretion in JAR cells. JAR cell (1 × 10 4 /well) were incubated 48 hours with or without Forskolin (1 μM, 10 μM or 100 μM) or EGF (0.8 ng/ml, 8 ng/ml or 80 ng/ml) and media collected for measurement of Gelatinase secretion. ( A) Bar graph, representing mean ± SEM of 10 independent experiments, of cells treated with forskolin. ( B) Bar graph, representing mean ± SEM of 10 independent experiments, of cells treated with EGF. Black bars represent proMMP-2 and gray bars represent proMMP-9. *P < 0.05 vs. control. Effect of forskolin on proMMP-2 and proMMP-9 secretion by JAR cells, 1 st trimester trophoblast cells 6–8 w and 9–12 w of gestation JAR cells (1 × 10 4 cells/well), 1 st trimester trophoblast cells 6–8 w and 9–12 w (1–2 × 10 5 cells/well) were incubated for 48 hours in the absence or presence of forskolin 10 μM. Figure 4A shows representative zymography gels. Figure 4B and 4C summarizes the results. Gelatinase secretion was enhanced by forskolin in all cell groups: Forskolin significantly increased proMMP-2 secretion in JAR cells (144.3% ± 8.8, P < 0.05), in 6–8 w trophoblast (125.9% ± 6.3, P < 0.05) and in 9–12 w trophoblast (169.7% ± 20.3, P < 0.05) as compared to control (Fig. 4B ). Forskolin also significantly increased proMMP-9 secretion in JAR cells (226.6% ± 50.7, P < 0.05), in 6–8 w trophoblast (128.4% ± 6.4, P < 0.05), and in 9–12 w trophoblast (120.3% ± 4.5, P < 0.05) as compared to control (Fig. 4C ). Figure 4 Secretion of proMMP-2 and proMMP-9 after 48 hours incubation of JAR cells, 1 st trimester trophoblast cells 6–8 week or 1 st trimester trophoblast cells 9–12 week in medium in absence or presence of 10 μM Forskolin. ( A) Representative zymography gels. ( B) Bar graph, representing mean ± SEM from 10 independent experiments detecting proMMP-2 (72 kD). ( C ) Bar graph, representing mean ± SEM from 10 independent experiments detecting proMMP-9 (92 kD). White bars represent control medium of cells without treatment. Black bars represent medium from cells with forskolin treatment. *P < 0.05 vs. control. Effect of EGF on proMMP-2 and proMMP-9 secretion by JAR cells, 1 st trimester trophoblast cells 6–8 w and 9–12 w of gestation JAR cells (1 × 10 4 cells/well), 1 st trimester trophoblast cells 6–8 w and 9–12 w (1–2 × 10 5 cells/well) were incubated for 48 hours in the absence or presence of EGF 8 ng/ml. Figure 5A shows representative zymography gels. Figure 5B and 5C summarizes the results. EGF significantly increased proMMP-2 secretion in JAR cells (130.4% ± 13.1, p < 0.05), in 6–8 w trophoblast (141.22% ± 14.8, P < 0.05) and also in 9–12 w trophoblast (168.3% ± 18.2, P < 0.05) as compared to control (Fig. 5B ). EGF significantly increased proMMP-9 secretion in JAR cells (187.8% ± 27.3, P < 0.05), in 6–8 w trophoblast (138.8% ± 10.3, P < 0.005), and in 9–12 w trophoblast (117.3% ± 3.8, P < 0.05) as compared to control (Fig. 5C ). Figure 5 Secretion of proMMP-2 and proMMP-9 after 48 hours incubation of JAR cells, 1 st trimester trophoblast cells 6–8 week or 1 st trimester trophoblast cells 9–12 week in medium in absence or presence of 8 ng/ml EGF. ( A ) Representative zymography gels. ( B) Bar graph, representing mean ± SEM from 10 independent experiments detecting proMMP-2 (72 kD). ( C ) Bar graph, representing mean ± SEM from 10 independent experiments detecting proMMP-9 (92 kD). White bars represent control medium of cells without treatment, black bars represent medium from cells with EGF treatment. *P < 0.05 vs. control. Effect of forskolin on cell invasion properties in JAR cells, 1 st trimester trophoblast cells 6–8 w and 9–12 w of gestation JAR cells (10 5 cells/well), 1 st trimester trophoblast cells 6–8 w and 9–12 w (2 × 10 5 cells/well) were incubated for 36–48 hours in the absence or presence of forskolin 10 μM on top of Transwell wells containing a transwell membrane coated with matrigel. Forskolin (10 μM) significantly enhanced trophoblast invasion in all cell groups. Forskolin increased cell invasion in JAR cells (110.6% ± 3.4, P < 0.05) (Fig. 6A ), in 6–8 w 1 st trimester trophoblast (189.7% ± 14.2, P < 0.05) (Fig. 6B ) and in 9–12 w (302.4% ± 56.0, P < 0.05) as compared to control (Fig. 6C ). The addition of inhibitory MMP-2 antibody significantly decreased invasion of control cells of JAR cells (86.5% ± 3.6, P < 0.05) and of 6–8 w (73.8 ± 10.6, P < 0.05) but did not affect 9–12 w trophoblasts. In forskolin-induced cells the presence of inhibitory MMP-2 antibody caused a significant decrease in invasion of JAR cells compared with induced cells alone (96.6% ± 1.5 versus 110.6% ± 3.4, respectively P < 0.05), of 6–8 w trophoblasts compared with induced cells alone (114.1% ± 24.6 versus 189.7% ± 14.2, respectively, P < 0.05) and of 9–12 w trophoblasts compared with induced cells alone (188.8% ± 18.4 versus 302.4% ± 56.0, respectively, P < 0.05) (Figure. 6A,6B and 6C ) Figure 6 Cell invasion ability of JAR cells, 1 st trimester trophoblast cells 6–8 week or 1 st trimester trophoblast cells 9–12 week tested with Transwell Invasion Assay. ( A) represents JAR cells, ( B ) 1 st trimester trophoblast cells 6–8 week and ( C ) 1 st trimester trophoblast cells 9–12 week. Cells were treated with 10 μM Forskolin and incubated 36 hours on Matrigel coated membrane, with or without MMP-2 or MMP-9 inhibitory antibodies. Cells that invaded the membrane to lower well were counted with XTT. Results represent mean ± SEM from 10 independent experiments. Black bars represent control (cells without treatment or cells treated with Forskolin) with no antibodies. Gray bars represent cells (without treatment or treated with Forskolin) with addition of MMP-2 inhibitory antibody. White bars represent cells (without treatment or treated with Forskolin) with addition of MMP-9 inhibitory antibody. ANOVA for all groups results in p < 0.05, post test confirmed the t test results. *P < 0.05 The addition of inhibitory MMP-9 antibody surprisingly increased invasion of control cells of 6–8 w and 9–12 w trophoblast (132.0 ± 8.3 and 134.9 ± 15.2, respectively, P < 0.05) as compared to control, and only in 9–12 w forskolin-induced cells caused a significant decreased invasion compared with induced cells alone (201.5% ± 6.3 versus 302.4% ± 56.0, respectively, P < 0.05). ANOVA post test confirmed the t test results (Fig. 6A,6B and 6C ). Effect of EGF on cell invasion properties in JAR cells, 1 st trimester trophoblast cells 6–8 w and 9–12 w of gestation JAR cells (10 5 cells/well), 1 st trimester trophoblast cells 6–8 w and 9–12 w (2 × 10 5 cells/well) were incubated for 36–48 hours in the absence or presence of EGF 8 ng/ml on top of Transwell wells containing a transwell membrane coated with matrigel. The results of EGF resembled those of forskolin. EGF enhanced trophoblast invasion in all cell groups. EGF increased cell invasion in JAR cells (112.6% ± 2.9, P < 0.05) (Fig. 7A ), in 6–8 w 1 st trimester trophoblast (157.9% ± 10.4, P < 0.05) (Fig. 7B ) and in 9–12 w (192.4% ± 10.5, P < 0.05) compared to control (Fig. 7C ). In EGF-induced cells the presence of inhibitory MMP-2 antibody caused a significant decrease in invasion of JAR cells (100.2% ± 0.8 versus 112.6 ± 2.9, P < 0.05), of 6–8 w trophoblasts (129.7% ± 8.0 versus 157.9 ± 10.4, P < 0.05) and of 9–12 w trophoblasts (161.1% ± 22.0 versus 192.4 ± 10.5, P < 0.05) compared with induced cells alone (Figure. 7A,7B and 7C ). Figure 7 Cell invasion ability of JAR cells, 1 st trimester trophoblast cells 6–8 week or 1 st trimester trophoblast cells 9–12 week tested with Transwell Invasion Assay. (A) represents JAR cells, (B) 1 st trimester trophoblast cells 6–8 week and (C) 1 st trimester trophoblast cells 9–12 week. Cells were treated with 8 ng/ml EGF and incubated 36 hours on Matrigel coated membrane, with or without MMP-2 or MMP-9 inhibitory antibodies. Cells that invaded the membrane to lower well were counted with XTT. Results represent mean +SEM from 10 independent experiments. Black bars represent control (cells without treatment or cells treated with EGF) with no antibodies. Gray bars represent cells (without treatment or EGF treated) with addition of MMP-2 inhibitory antibody. White bars represent cells (without treatment or EGF treated) with addition of MMP-9 inhibitory antibody. ANOVA for all groups results in p < 0.05, post test confirmed the t test results. * P < 0.05. The addition of inhibitory MMP-9 antibody affected 9–12 w EGF–induced cells and caused a significant decreased invasion compared with induced cells alone (169.7% ± 18.1 versus 192.4 ± 10.5, respectively, P < 0.05), ANOVA post test confirmed the t test results (Figure. 7A,7B and 7C ). The inhibitory affect of MMP-2/-9 antibody on control cells was described in the previous section. Discussion Trophoblastic invasion of the endometrium is highly regulated by interrelated reactions between invasion-promoting factors, such as cytokines, growth factors, MMP-2 and MMP-9, and invasion-inhibiting factors such as TIMPs. In the current study, the secretion and activity of MMP-2 and MMP-9 in human cytotrophoblastic cells from different weeks of gestation was measured and compared with a choriocarcinoma cell-line. A differential secretion profile of proMMP-2 and -9 was found between 6–8 w and 9–12 w as expressed by a shift in the relative proportion of each gelatinase. Elevated proMMP-9 secretion in 9–12 w was observed, compared with both 6–8 w and JAR cells. These results are consistent with those of Xu et al [ 11 ], who found MMP-2 production to be dominant between 6–8 weeks of gestation and then declining, whereas MMP-9 production significantly increased from 8 to 11 weeks, with a shift of dominant gelatinase from MMP-2 to MMP-9 from 9 week of gestation. MMP-2 is predominant in human preimplantation embryos [ 20 , 21 ], whereas MMP-9 is dominant in the third trimester [ 6 ]. Niu et al [ 22 ] reported a dominance of MMP-2 secretion over MMP-9 from 1 st trimester villous tissue, and a dramatically decrease in MMP-2 levels in the second trimester. These results support our findings of a dynamic gelatinase secretion profile during the 1 st trimester. In our study, MMP secretion was induced by two separate signal pathways: PKA and PTK. Several factors with importance in embryo implantation act via the cAMP-protein kinase A (PKA) signal transduction pathway, including human chorion gonadotropin (hCG), the primary signal of an implanting pregnancy [ 23 ]. Forskolin is a prototypical stimulator of the cAMP pathway by direct activation of adenylate cyclase [ 24 ]. In this study, forskolin was found to significantly enhance proMMP-2 secretion in JAR choriocarcinoma cell-line, 6–8 w and 9–12 w trophoblasts. Forskolin also enhanced proMMP-9 secretion, in choriocarcinoma cells and in 1 st trimester trophoblast (from both groups). This indicates that forskolin might influence trophoblast cells invasiveness by enhancing the secretion of gelatinases. Zymography measures all forms of MMP (active and inactive) and therefore does not reliably represent the true physiological activity, which is influenced by the presence of MMP inhibitors and activators. In order to examine the role of the gelatinases in the implantation process, we examined cell invasion after PKA stimulation in JAR cells, 6–8 and 9–12 w trophoblasts. Our results showed a significant increase in invasive ability in JAR cells and in early and late 1 st trimester trophoblasts, after stimulation with forskolin. In order to detect the contribution of each gelatinase to this invasive process, inhibitory antibodies to MMP-2 and MMP-9 were added to cell culture and cell invasive ability examined. MMP-2 inhibitory antibody caused a significant decrease in cell invasion in JAR cells, 6–8 w and 9–12 w trophoblasts treated with forskolin, whereas MMP-9 inhibitory antibody only caused a decrease in 9–12 w trophoblasts. This indicates, that most probably MMP-2 and not MMP-9 is the key-enzyme in the invasion process of JAR cells and early 1 st trimester trophoblasts stimulated by forskolin, whereas MMP-9 together with MMP-2 plays a role in late 1 st trimester trophoblasts. Only a few published data describe the relationship between forskolin and trophoblast invasion. Human chorionic gonadotropin acts via cAMP, and is considered a sign of differentiation of trophoblasts to syncytiotrophoblast. The hCG receptor was shown to be expressed on invasive trophoblast and in choriocarcinoma cells, and hCG was found to increase in vitro invasion and migration of a trophoblastic cell line, an effect that was also mimicked by forskolin [ 23 ]. Several studies demonstrated a positive correlation between hCG level and successful implantation [ 25 ] or inappropriate implantation/ invasion associated with the development of preeclampsia [ 24 ] or gestational trophoblastic tumors [ 27 ]. To the best of our knowledge, our study is the first to report the effect of forskolin on MMPs in first trimester trophoblastic cells. In addition, this study also distinguishes between the contributions of each of the gelatinases to the invasive capacity enhanced by forskolin in trophoblastic cells. Epidermal Growth factor (EGF) plays a major role in placental implantation, growth and differentiation and is regarded a paracrinic factor modifying the implantation process. EGF acts on trophoblasts via a specific receptor (EGFR) from the tyrosine kinase receptor family [ 28 ]. EGF is secreted from the endometrium during the implantation window, in which the embryo also expresses EGFR [ 29 ], and expressed in placenta from 1 st throughout third trimester [ 30 ] In this study EGF stimulated secretion of proMMP-2 in JAR cells, 6–8 w and 9–12 w trophoblasts and also enhanced the secretion of proMMP-9 in JAR, in 6–8 w trophoblasts and in 9–12 w trophoblasts. Our results therefore indicate, that MMP-2 and MMP-9 secretion by trophoblastic cells may be stimulated through the PTK pathways during the first trimester. We noted that EGF at a high concentration (80 ng/ml), in contrast, decreased proMMP-2 and -9 secretion in JAR cells. This result corresponds with previous published data regarding a dual, concentration dependent effect of EGF on cell functions [ 31 , 32 ]. In order to examine the role of the gelatinases in the implantation process, we examined cell invasion after PTK stimulation in JAR cells, 6–8 and 9–12 w trophoblasts. Our results showed a significant increase in invasion ability in JAR cells and in early and late 1 st trimester trophoblasts, after stimulation with EGF. In EGF-stimulated JAR cells and in 6–8 w and 9–12 w trophoblasts inhibitory MMP-2 antibody decreased cell invasion, whereas inhibitory MMP-9 antibody caused a significant decrease in invasion only in 9–12 w trophoblasts. We thereby showed, that in EGF stimulated cells as well, MMP-2 is the key-enzyme in the invasion process in vitro in JAR cell and in early 1 st trimester trophoblasts, whereas in late 1 st trimester trophoblasts both MMP-2 and MMP-9 have a role. EGF was found to induce changes in morphology and to increase invasive capacity of first trimester trophoblasts, whereas later gestational cytotrophoblasts (2 nd trimester), whose invasive capacity is diminished, are much less affected [ 29 ]. EGF was also found to increase MMP-9 secretion by cytotrophoblasts [ 33 ]. In JAR control cells inhibitory, MMP-2 antibody also decreased invasion, whereas MMP-9 antibody had no affect, indicating that the basic invasive ability of these choriocarcinoma cells is mainly due to MMP-2 and not to MMP-9. Inhibition of MMP-2 in control cells also decreased invasion in 6–8 w trophoblast, but not in 9–12 w, indicating again the importance of MMP-2 in early (6–8 w) trophoblast invasiveness. Surprisingly inhibition of MMP-9 in 6–8 w and 9–12 w trophoblasts without treatment caused an increase in invasion. We speculate that this may be due to a release of other proteinases, including other MMPs, since MMP-9 is known to dimerize [ 34 , 35 ] All in all we found, that in early 1 st trimester trophoblasts (6–8 w), MMP-2 is the major gelatinase participant in cell invasion, whereas in later 1 st trimester trophoblasts (>9 w) both MMP-9 and MMP-2 most probably participate in cell invasion, however we cannot exclude the possibility of other MMP family members participating in this process. Isaka et al [ 9 ] have shown that invasive ability of early first trimester trophoblast was inhibited by MMP-2 antibody in a dose dependent manner, thereby suggesting that the invasive ability of trophoblasts may be regulated by the enzyme activity of gelatinases, especially MMP-2. This study supports ours in the involvement of MMP-2 in trophoblast cell invasion. MMP-2 was found to be located in invasive evCTB in 1 st trimester placenta [ 9 , 10 , 36 ], whereas MMP-9 was located in the non-invasive vCTB [ 9 ]. In contrast, several studies have found MMP-9 to be the key-enzyme in trophoblast invasion in vitro [ 1 , 5 , 7 ]. We speculate, that the main reason for this controversy of results comes from the dynamic gelatinase expression during the 1 st trimester, as earlier discussed. The choice of pathway for stimulation of cell invasion may also contribute to a difference in results. We used stimulation through the PKA and PTK pathways; whereas other groups used the PKC pathway [ 7 , 8 ]. Various stimulators, inducing different signal pathways, are likely not to affect the same enzymes in an identical manner, and thereby can result in varying dominant enzymes. We found, that MMP-2 is also the key-enzyme in JAR cell invasion; therefore JAR cells resemble early 1 st trimester trophoblasts in cell invasive ability and in MMP secretion profile and differ from late 1 st trimester trophoblast in these parameters. It has been documented, that JAR-trophoblast cells have the ability to invade in vivo [ 37 , 38 ]. We chose this cell-line because it provides a large number of uniform cells and preserve the ability to differentiate into syncytiotrophoblast-like cell in vitro [ 39 , 40 ]. Other studies showed different compartment in vitro between choriocarcinoma cell-lines and human first trimester trophoblast in the regulation of invasion [ 36 , 37 ]. The study of JAR cell invasion may therefore represent only partly the aspects and mechanisms of the in vivo situation of invasion, where many cell types are involved. Conclusions We showed that forskolin and EGF stimulate proMMP-2 and -9 secretion from trophoblasts, and that there is a differential, dynamic importance of each gelatinase in trophoblast invasion during 1 st trimester. We suggest that MMP-2 is the key-enzyme in JAR and early 1 st trimester (6–8 w) trophoblast invasion, whereas both MMP-2 and -9 are important for late (9–12 w) trophoblast invasion. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC516041.xml |
526196 | Appraising and applying evidence about a diagnostic test during a performance-based assessment | Background The practice of Evidence-based Medicine requires that clinicians assess the validity of published research and then apply the results to patient care. We wanted to assess whether our soon-to-graduate medical students could appraise and apply research about a diagnostic test within a clinical context and to compare our students with peers trained at other institutions. Methods 4 th year medical students who previously had demonstrated competency at probability revision and just starting first-year Internal Medicine residents were used for this research. Following an encounter with a simulated patient, subjects critically appraised a paper about an applicable diagnostic test and revised the patient's pretest probability given the test result. Results The medical students and residents demonstrated similar skills at critical appraisal, correctly answering 4.7 and 4.9, respectively, of 6 questions (p = 0.67). Only one out of 28 (3%) medical students and none of the 15 residents were able to correctly complete the probability revision task (p = 1.00). Conclusions This study found that most students completing medical school are able to appraise an article about a diagnostic test but few are able to apply the information from the article to a patient. These findings raise questions about the clinical usefulness of the EBM skills possessed by graduating medical students within the area of diagnostic testing. | Background Evidence-based medicine (EBM) has been described as the "conscientious, explicit, and judicious use of current best evidence in making decisions about the care of individual patients" [ 1 ]. Thus, to practice EBM clinicians need to critically appraise articles in the medical literature and then apply the evidence to specific patients. In the area of diagnostic testing, EBM requires the use of Bayesian inference so that appraised evidence can be used in the evaluation of a specific patient [ 2 ]. Nearly all medical schools now provide their students with instruction on EBM [ 3 ]. Students at the University of Iowa Carver College of Medicine receive instruction on EBM during required course work during their two preclinical years. Medical students are introduced to EBM in a series of lectures in the first year, including one on the evaluation of diagnostic tests that introduces students to the concept of test characteristics and probability revision. Other lectures focus on critiquing the medical literature. In later semesters, students are asked to complete evidence-based projects and are given further training on the use of Bayes' Theorem. In addition, our students have a required two-week Laboratory Medicine clerkship taken during the third or fourth year. With the aid of a clinically oriented textbook [ 4 ], students on this rotation are expected to master the concepts underlying diagnostic testing. To complete this clerkship, students need to demonstrate comprehension of test performance characteristics and probability revision by passing an exam in which they are asked to calculate the sensitivity, specificity, positive predicative value, and negative predictive value for two different testing procedures. In undertaking this research, we wanted to assess the critical appraisal and probability revision skills of our medical students shortly before graduation within the domain of diagnostic testing. We specifically wanted to evaluate medical students who had previously demonstrated competency at probability revision in a classroom setting. Because EBM is designed to support the delivery of clinical care, transfer of these skills from the classroom to the exam room seemed an appropriate measure of instructional success. We were also interested in comparing the skills of our students nearing graduation to physicians just entering the Internal Medicine residency at our medical center with the intention of comparing the skills of our students to those of students who had trained at other medical schools. Methods The setting The data were collected during a performance-based assessment utilizing standardized patients (SPs). During this assessment, subjects had a series of 15 minute encounters with SPs followed by a 10 minute post-encounter activity, thus making each station 25 minutes in length. A 25 minute non-SP based station was integrated into this assessment during which subjects were asked to read and appraise an article about a diagnostic test and apply the information to a preceding SP encounter. The subjects Two different groups of subjects participated in this research. The first group was composed of medical students who were assigned to the Psychiatry clerkship in the late winter and spring of 2003. The second group was composed of all incoming first-year Internal Medicine residents who were in the process of orienting to the residency in June 2003 in preparation for their clinical duties. The task Subjects were asked to critically appraise a research study about a diagnostic test using a worksheet derived from the article about diagnostic tests published in Users' Guides to EBM series [ 5 ]. They were asked to assess the validity of the study by identifying the reference standard, whether there was independent and blinded comparison with the reference standard, whether the results of the test being evaluated influenced the decision to perform the reference standard, and whether all subjects underwent the reference standard as well as the test being evaluated. Subjects were also asked to evaluate whether the setting of the study was similar to a community setting in which they would anticipate using the new diagnostic test. Lastly, the subjects were asked to identify the results of the study in terms of the sensitivity and specificity of the test. Thus, in critically appraising the study, subjects were asked to answer a series of 6 questions derived from the published diagnostic test EBM user's guide. After they had assessed the study, the subjects were asked to apply the results of the study by revising the probability of disease given a specified pre-test probability and a test result. To assist with this calculation subjects were provided with calculators. The articles The fourth year medical students, who were participating in their Psychiatry clerkship, were asked to assess an article about a questionnaire to aid in the diagnosis of Major Depression and Panic Disorder [ 6 ]. The Internal Medicine residents read an article about a blood test to aid in the diagnosis of congestive heart failure [ 7 ]. While the articles focused on different diseases and diagnostic tests, the studies were similar from the critical appraisal perspective. Both articles described research on a new diagnostic test that had been undertaken in carefully selected clinical settings to avoid spectrum bias. Clinical experts who were blinded to the result of the test being evaluated were used as gold standards by both studies. The protocols used by both projects avoided referral bias. Lastly, both articles had been recently published by major medical journals. Analysis For this study, the calculated posttest probability was considered correct if it was ±5% of the correct answer derived from Bayes' Theorem. The performances of the medical students and the Internal Medicine residents were compared using Fisher's Exact Test for dichotomous outcomes and t-test for continuous outcomes. Cronbach's alpha was used to calculate the internal reliability of the 6-item critical appraisal worksheet for each group of subjects. The analyses were undertaken using NCSS 2004 (Kayesville, UT). An alpha of 0.05 was used and all tests were 2-tailed. Approval from the institutional review board was obtained for this project. Results Thirty-eight medical students on the Psychiatry clerkship completed the critical appraisal exercise. Twenty-eight of these students were 4 th year medical students who had demonstrated mastery of Bayesian probability revision during a preceding Laboratory Medicine clerkship. The data from these students were analyzed for this report. The other 10 students were dropped from the analysis because they were either M3s or M4s who had not yet completed their Laboratory Medicine clerkship. Twenty-two first year Internal Medicine residents participated in this performance-based assessment and completed the EBM appraisal task and Bayesian inference exercises during their clinical skills assessment. For this report we used the data from the 15 residents who were recent graduates of US medical schools. The data from the 7 non-US graduates were dropped from the analysis. Twenty-six of the 28 medical students (93%) correctly identified whether there was an independent, blind comparison of the test with a reference standard (Table 1 ). Seventeen (61%) correctly identified the reference standard. Eighteen (65%) correctly assessed whether referral bias was present, and twenty-eight (90%) were able to comment on the generalizability of the study. Twenty-five (89%) of the students were able to identify the sensitivity and specificity of the test. The internal reliability of the 6-item critical appraisal questionnaire was 0.59. On average, a medical student correctly answered 78% of the questions related to the appraisal of the article. However, only one of the 28 students (4%) was able to correctly revise the pretest probability. Table 1 Percentage of learners successfully completing tasks relevant to critical appraisal of a diagnostic test journal article % Answering Correctly Critical Appraisal Guides Medical Students (n = 28) IM Residents (n = 15) P value Independent, Blinded Comparison With Reference Standard 93% 100% 0.53 Identification of the Reference Standard 61% 53% 0.75 Assessment of Referral Bias 64% 73% 0.74 Completeness of Testing of Subjects 74% 80% 0.72 Generalizability of Results to Typical Practice Settings 89% 93% 1.0 Identification of Sensitivity/Specificity 89% 87% 1.0 All 15 residents (100%) correctly identified whether there was an independent, blind comparison of the test with a reference standard. Eight (53%) correctly identified the reference standard. Eleven (73%) correctly assessed whether referral bias was present, and fourteen (93%) were able to comment on the generalizability of the study. Thirteen (87%) of the Internal Medicine residents were able to identify the sensitivity and specificity of the test for congestive heart failure. The internal reliability of the 6-item appraisal work sheet was 0.66 for this group of subjects. On average, a resident, who had just recently graduated from a US medical school, correctly answered 81% of the questions related to critical appraisal but none of the 15 residents (0%) were able to revise the pretest probability given a test result. Comparison of medical students and residents Overall, the performances of the two groups of subjects were very similar. We did not find any significant differences in their ability to critically appraise an article about a diagnostic test. Medical students, on average, correctly answered 4.7 of the 6 questions related to appraisal while residents correctly answered 4.9 of these questions (p = 0.67). The two groups also showed similar performance on each of the 6 items on the worksheet as shown in Table 1 (p < 0.05 for each) Both groups also showed similar poor performance when asked to revise a pretest probability of disease given the result of a diagnostic test (p = 1.0) using data provided as probabilities. Discussion Over the past decade, EBM has become a major driving force world wide, impacting medical education, policymaking, and research. The teaching of evidence-based medicine has been increasingly integrated into curricula at all levels of medical education as advocated by the Medical School Objectives Program developed by the Association of American Medical Colleges (AAMC) [ 8 ]. Like most medical schools, the Carver College of Medicine at the University of Iowa has integrated EBM into its curriculum and our data indicates that students who are soon to graduate can demonstrate proficiency at critically appraising an article about diagnostic testing. But in a simulated clinical encounter, few students are able take the last step of using EBM in diagnosing medical illness. We found almost uniform failure of our students to correctly revise a pretest probability of disease given a test result despite their earlier demonstration of competency with Bayes' Theorem on their Laboratory Medicine clerkship. This finding raises questions about whether our students can fully utilize their EBM training in the clinical setting. Our finding that incoming residents demonstrate similar levels of skill at critically appraisal and also are unable to revise a pretest probability implies that our medical students' skill deficit is not solely due to a local curricular problem. Because the residents had only just graduated from 12 different US medical schools, this finding suggests that many medical school graduates are able to critically appraise articles on diagnostic testing but few are able to take the next step- that of revising the probability of disease given a test result. There are few other assessments of the EBM skills of graduating medical students using simulated clinical encounters. In an earlier study, 3 rd year medical students demonstrated good performance at critically appraisal [ 9 ], a finding which is similar to ours. These results generally replicated classroom-based studies on the success of critical appraisal instruction [ 10 ]. To our knowledge, only one other study has investigated whether students are able to integrate critically appraised information about a diagnostic test into clinical decision making. The results of this earlier study also raised concerns of students' abilities to transfer their EBM skills to simulated clinical encounters [ 11 ]. Our study has a number of limitations. The first is the very small sample size. However, it is unlikely that a larger sample size would change our conclusion that by the end of medical school students have largely mastered critical appraisal of an article on diagnostic testing but are unable to use this information to revise a patient's probability of disease. A second limitation is that these data were collected at only one medical school. However, as we find the same pattern of competencies in the recently graduated students who are entering our Internal Medicine program it is likely that our findings apply to many other medical schools. A third limitation is that we had our two groups of subjects critically appraise two different articles although they were very similar from this perspective. A final limitation is that it is possible that our subjects would have demonstrated competency in probability revision if we had provided them with a Bayes nomogram or computer spreadsheet. But we wanted to assess whether our students could apply EBM skills to a clinical encounter without any other external supports except for a simple calculator. In the same way, we do not allow our students to take handbooks or other work aids into their 15 minute OSCE encounters with standardize patients. The poor performance of our students and residents at probability revision is worrisome although previous studies have shown that many clinicians do not master Bayesian inference. Nearly 25 years ago, Casscells documented that few students or faculty at Harvard Medical School were able to correctly complete a probability revision problem [ 12 ]. Eddy duplicated this finding in a second group of physicians [ 13 ]. However, some cognitive psychologists suggest that humans have most likely always used Bayesian inference in order to survive in our uncertain world. They argue that it is the probability format of the numbers and not the inference task that makes most people fail at the task [ 14 ]. A promising line of research suggests that learners show sustained mastery of Bayesian inference using probabilities if they are taught how to first translate probabilities into natural frequencies [ 15 ]. Whether this will prove to be the solution deserves study. Conclusions Currently, most of our medical students are able to critically appraise research articles about diagnostic testing but few are able to apply this information at the patient level using Bayesian inference. Because we are able to document the same pattern of skills in entering Internal Medicine residents, we expect this lack of competence with probability revision to be wide spread amongst medical learners. This raises concerns about the clinical utility of the EBM training many students are currently receiving within the domain of diagnostic testing. Competing interests The authors declare that they have no competing interests. Authors' contributions GB conceived of the study. GB, SV, JT, EF participated in study design. GB, EF, RF participated in data collection. GB performed the data analysis. GB, SV, ST, EF, RF participated in drafting the manuscript. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526196.xml |
387271 | Peace Lessons from an Unlikely Source | How much is the aggression we observe in nonhuman primates the result of culture, and will the answer provide insights into our own violent behaviour? | Upon arrival from Europe, now more than two decades ago, I was taken aback by the level of violence in the American media. I do not just mean the daily news, even though it is hard getting used to multiple murders per day in any large city. No, I mean sitcoms, comedies, drama series, and movies. Staying away from Schwarzenegger and Stallone does not do it; almost any American movie features violence. Inevitably, desensitization sets in. If you say, for example, that Dances with Wolves (the 1990 movie with Kevin Costner) is violent, people look at you as if you are crazy. They see an idyllic, sentimental movie, with beautiful landscapes, showing a rare white man who respects American Indians. The bloody scenes barely register. Comedy is no different. I love, for example, Saturday Night Live for its inside commentary on peculiarly American phenomena, such as cheerleaders, televangelists, and celebrity lawyers. But SNL is incomplete without at least one sketch in which someone's car explodes or head gets blown off. Characters such as Hans and Franz (“We're going to pump you up!”) appeal to me for their names alone (and yes, I do have a brother named Hans), but when their free weights are so heavy that their arms get torn off, I am baffled. The spouting blood gets a big laugh from the audience, but I fail to see the humor. Did I grow up in a land of sissies? Perhaps, but I am not mentioning this to decide whether violence in the media and our ability to grow immune to it—as I also have over the years—is desirable, or not. I simply wish to draw attention to the cultural fissures in how violence is portrayed, how we teach conflict resolution, and whether harmony is valued over competitiveness. This is the problem with the human species. Somewhere in all of this resides a human nature, but it is molded and stretched into so many different directions that it is hard to say if we are naturally competitive or naturally community-builders. In fact, we are both, but each society reaches its own balance between the two. In America, the squeaky wheel gets the grease. In Japan, the nail that stands out gets pounded into the ground. Does this variability mean, as some have argued, that animal studies cannot possibly shed light on human aggression? “Nature, red in tooth and claw” remains the dominant image of the animal world. Animals just fight, and that is it? It is not that simple. First, each species has its own way of handling conflict, with for example the chimpanzee ( Pan troglodytes ) being far more violent than that equally close relative of ours, the bonobo ( P. paniscus ) ( de Waal 1997 ). But also within each species we find, just as in humans, variation from group to group. There are “cultures” of violence and “cultures” of peace. The latter are made possible by the universal primate ability to settle disputes and iron out differences. There was a time when no review of human nature would be complete without assertions about our inborn aggressiveness. The first scientist to bring up this issue, not coincidentally after World War II, was Konrad Lorenz (1966) . Lorenz's thesis was greeted with accusations about attempts to whitewash human atrocities, all the more so given the Nobel Prize winner's native tongue, which was German. But Lorenz was hardly alone. In the USA, science journalist Robert Ardrey (1961) presented us as “killer apes” unlikely to ever get our nasty side under control. Recent world events have done little to counter this pessimistic outlook. The opposition argued, of course, that aggression, like all human behavior, is subject to powerful cultural influences. They even signed petitions to this effect, such as the controversial Seville Statement on Violence ( Adams et al. 1990 ). In the polarized mind-set of the time, the issue was presented in either-or fashion, as if behavior cannot be both learned and built upon a biological foundation. This rather fruitless nature/nurture debate becomes considerably more complex if we include what is usually left out, which is the ability to keep aggression under control and foster peace. For this ability, too, there exist animal parallels, such as the habit of chimpanzees to reconcile after fights by means of a kiss and embrace. Such reunions are well-documented in a multitude of animals, including nonprimates, such as hyenas and dolphins. They serve to restore social relationships disturbed by aggression, and any animal that depends on cooperation needs such mechanisms of social repair ( Aureli and de Waal 2000 ; de Waal 2000 ). There are even indications that in animals, too, cultural influences matter in this regard. This may disturb those who write culture with a capital C , and hence view it as uniquely human, but it is a serious possibility nonetheless. Nonhuman culture is currently one of the hottest areas in the study of animal behavior. The idea goes back to the pioneering work of Kinji Imanishi, who in 1952 proposed that if individuals learn from one another, their behavior may over time grow different from that of individuals in other groups of the same species, thus creating a characteristic culture (reviewed by de Waal 2001 ). Imanishi thus brought the culture concept down to its most basic feature, that is, the social rather than genetic transmission of behavior. Since then, many examples have been documented, mostly concerning subsistence techniques, such as the sweet potato washing of Japanese macaques ( Macaca fuscata ) and the rich array of tool use by wild chimpanzees, orangutans ( Pongo pymaeus ), and capuchin monkeys ( Cebus spp.) ( Whiten et al. 1999 ; de Waal 2001 ; Hirata et al. 2001 ; Perry et al. 2003 ; van Schaik et al. 2003 ). However, much less attention has been paid to social culture , which we might define as the transmission of social positions, preferences, habits, and attitudes. Social culture is obviously harder to document than tool use. In human culture, for instance, it is easy to tell if people eat with knife and fork or with chopsticks, but to notice if a culture is egalitarian or hierarchical, warm or distant, collectivistic or individualistic takes time and is difficult to capture in behavioral measures. A well-documented monkey example of social culture is the inheritance of rank positions in macaque and baboon societies. The future position in the hierarchy of a newborn female can be predicted with almost one hundred percent certainty on the basis of her mother's rank. Females with relatives in high places are born with a silver spoon in their mouth, so to speak, whereas those of lowly origin will spend their life at the bottom. Despite its stability, the system depends on learning. Early in life, the young monkey finds out against which opponents it can expect help from her mother and sisters. When sparring with peer A she may utter screams that recruit massive support to defeat A. But against peer B she can scream her lungs out and nothing happens. Consequently, she will come to dominate A but not B. Experiments manipulating the presence of family members have found that when support dwindles dominant females are unable to maintain their positions ( Chapais 1988 ). In other words, the kin-based hierarchy is maintained for generation after generation through social rather than genetic transmission. Returning to the issue of aggressive behavior, here the effects of social culture can be felt as well. Without any drugs or brain lesions, one experiment managed to turn monkeys into pacifists. Juveniles of two different macaque species were placed together, day and night, for five months. Rhesus monkeys ( Macaca mulatta ), known as quarrelsome and violent, were housed with the more tolerant and easy-going stumptail monkeys ( M. arctoides ) ( Figure 1 ). Stumptail monkeys easily reconcile with their opponents after fights by holding each others' hips (the so-called “hold-bottom” ritual), whereas reconciliations are rare in rhesus monkeys. Because the mixed-species groups were dominated by the stumptails, physical aggression was rare. The atmosphere was relaxed, and after a while all of the monkeys became friends. Juveniles of the two species played together, groomed together, and slept in large, mixed huddles. Most importantly, the rhesus monkeys developed peacemaking skills on a par with those of their more tolerant group mates. Even when, at the end of the experiment, both species were separated, the rhesus monkeys still showed three times more reconciliation and grooming behaviors after fights than typical of their kind ( de Waal and Johanowicz 1993 ). Primates thus can adopt social behavior under the influence of others, which opens the door to social culture. Figure 1 Stumptail Monkeys Stumptail monkeys ( Macaca arctoides ) are among the most conciliatory members of the genus Macaca. They are heavily built, yet remarkably friendly and tolerant, such as here: the alpha male is eating attractive food unperturbed by an entire audience around him. When stumptail monkeys were housed with a less tolerant macaque, they modified the latter species' behavior into a more pacific direction. (Photograph by Frans de Waal, used with permission.) Not unlike rhesus monkeys, baboons have a reputation for fierce competition and nasty fights. With the study by Robert Sapolsky and Lisa Share published in this issue of PLoS Biology , we now have the first field evidence that primates can go the flower power route ( Sapolsky and Share 2004 ). Wild baboons developed an exceptionally pacific social tradition that outlasted the individuals who established it. For years, Sapolsky has documented how olive baboons ( Papio anubis ) on the plains of the Masai Mara, in Kenya, wage wars of nerves, compromising their rivals' immune systems and pushing up the level of their blood cortisol ( Sapolsky 1994 ). An accident of history, however, selectively wiped out all the male bullies of his main study troop. As a result, the number of aggressive incidents dropped dramatically. This by itself was not so surprising. It became more interesting when it was discovered that the behavioral change was maintained for a decade. Baboon males migrate after puberty, hence fresh young males enter troops all the time, resulting in a complete turn-over of males during the intervening decade. Nevertheless, compared with troops around it, the affected troop upheld its reduced aggression, increased friendly behavior, and exceptionally low stress levels. The conclusion from this natural experiment is that, like human societies, each animal society has its own ecological and behavioral history, which determines its prevalent social style. It is somewhat ironic that at a time when researchers on human aggression are increasingly attracted, albeit with a far more sophisticated approach, to the Lorenzian idea of a biological basis of aggression ( Enserink 2000 ), students of animal behavior are beginning to look at its possible cultural basis. There is no reason for animals with a development as slow as a baboon (with adulthood achieved in five or six years) not to be influenced in every way by the environment in which they grow up, including the social environment. How this influence takes place is a point of much debate, and remains unclear in the case of the peaceful male baboons in the Masai Mara. Given their mobility, the males themselves are unlikely transmitters of social traditions within their natal troop. Therefore, Sapolsky and Share look at the females for an answer—female baboons stay all their lives in the same troop. By reacting positively to certain kinds of behavior, for example, females may be able to steer male attitudes in a new direction. This complex problem is hard to unravel with a single study, especially in the absence of experimentation. Yet, the main two points of this discovery are loud and clear: social behavior observed in nature may be a product of culture, and even the fiercest primates do not forever need to stay this way. Let us hope this applies to humanity as well. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC387271.xml |
528839 | A two-layered mechanical model of the rat esophagus. Experiment and theory | Background The function of esophagus is to move food by peristaltic motion which is the result of the interaction of the tissue forces in the esophageal wall and the hydrodynamic forces in the food bolus. The structure of the esophagus is layered. In this paper, the esophagus is treated as a two-layered structure consisting of an inner collagen-rich submucosa layer and an outer muscle layer. We developed a model and experimental setup for determination of elastic moduli in the two layers in circumferential direction and related the measured elastic modulus of the intact esophagus to the elastic modulus computed from the elastic moduli of the two layers. Methods Inflation experiments were done at in vivo length and pressure-diameters relations were recorded for the rat esophagus. Furthermore, the zero-stress state was taken into consideration. Results The radius and the strain increased as function of pressure in the intact as well as in the individual layers of the esophagus. At pressures higher than 1.5 cmH 2 O the muscle layer had a larger radius and strain than the mucosa-submucosa layer. The strain for the intact esophagus and for the muscle layer was negative at low pressures indicating the presence of residual strains in the tissue. The stress-strain curve for the submucosa-mucosa layer was shifted to the left of the curves for the muscle layer and for the intact esophagus at strains higher than 0.3. The tangent modulus was highest in the submucosa-mucosa layer, indicating that the submucosa-mucosa has the highest stiffness. A good agreement was found between the measured elastic modulus of the intact esophagus and the elastic modulus computed from the elastic moduli of the two separated layers. | Introduction The majority of previous mechanical studies on visceral organs, including the blood vessels, have considered them as homogenous tubes; i.e., a single layer structure. Most visceral organs are, however, multilayered, e.g. the arteries consist of intima, media and adventitia and the gastrointestinal tract has circumferential and longitudinal muscle layers, submucosa and mucosa layers. The esophagus represents a very interesting biomechanical model since it is the only organ that can be separated into two layers without damage to either layer. Hence, the muscle layers can be separated from the mucosa-submucosa layer by dissection, leaving two intact tubes. Separation experiments of the esophagus in guinea pigs and rabbits showed that the submucosa-mucosa layer had larger residual strains and opening angles than the muscle layer [ 1 - 3 ]. Considering the multi-layered composite structure and the difference in zero-stress state between the layers, the stress distribution in the wall is expected to be non-homogeneous. Hence, the material constants likely differ between the layers. Such a finding impacts our understanding of biological tissue remodelling and the function of mechanosensitive receptors located in various layers of the wall [ 4 - 6 ]. Therefore, data on the strain and stress distribution in the layers will facilitate the understanding of the relationship between the stress, remodelling of the tissue and sensory responses. To pursue this line of study, however, it is necessary to know how the stress and strain in the esophagus can be computed for each layer, and how the composite can be put together to give the overall observed mechanical properties. In this study we recognize that the esophagus consists of mucosa-submucosa and muscle layers. We analyze these layers as elastic shells. Each layer has its own zero-stress state, and its own elastic constants. We will determine the material properties of each layer separately. Specifically, the material properties in the individual layers will be computed from the pressure-diameter relation and zero-stress state with the method of analysis presented below. We will then propose a simple model to combine the two layers to predict the overall behavior of the esophagus under certain hypotheses. The limitations and implications of the model will be discussed. Materials and methods Eight male Wistar rats, weighing 380–420 grams, were used in the study. Approval of the protocol was obtained from the Danish Animal Experiment Committee. The animals were anesthetized with sodium pentobarbital (50 mg kg -1 ip). Papaverine (15 mg) was injected into the tail vein to relax the visceral muscles and to euthanize the rat. The cervical segment of the esophagus was dissected free from its adjacent tissue. Next, the thoracic and abdominal cavities were opened. After pouring cold Krebs solution into the thoracic cavity, the esophagus was quickly dissected free from adjacent tissues and its in situ length was measured. A 2-cm-long segment from the middle part, intended for the distension test, was marked. The length of this segment and that of the entire esophagus was measured. The entire esophagus was then cut at the proximal and distal ends including the very first part of the stomach, and immediately placed in calcium-free Krebs solution containing 6% dextran and 0.25% ethylene glycol-bis (β-aminoethyl ether)-N,N,N,N-tetraacetic acid (EGTA). The solution was aerated with a gas mixture of 95% O 2 -5% CO 2 at pH of 7.4. After careful removal of all extra-esophageal tissue, the length was measured in vitro . Pressure-diameter experiments The middle part of the intact esophagus was mounted in an organ bath containing the Ca +2 -free Krebs solution. The segment was stretched to the in vivo length and fixed. The distal end was closed whereas the proximal end was cannulated and connected to a fluid container. After preconditioning the tissue with pressures up to 8 cmH 2 O, a ramp test was performed where the pressure was changed continuously at a rate of 2 cmH 2 O per minute up to a pressure of 8 cmH 2 O. A video camera (Sony CCD camera) monitored the changes in diameter and length during the distension and images were grabbed by a PC. After the test of the intact esophagus, the muscle and submucosa-mucosa layers were gently separated into two tubes. The tubes were studied separately using the same procedure outlined above. The only difference was that the maximum pressure was set to 6 cmH 2 O. The zero-stress state of the esophagus The zero-stress state of the esophagus was obtained in accordance with the method used for blood vessels [ 7 ]. Briefly, six rings of 1 mm length were cut from the intact esophagus and from the separated layers and were then cut in the radial direction to obtain the zero-stress state. The choice of the ring length was based on pilot studies. The radial cut caused the rings to open up into sectors. The shape of each ring segment at the zero-stress state was photographed 60 minutes after the radial cut to allow the creep to subside. Data analysis The morphometric measurements were made using SigmaScan Pro image analysis software (Jandel Scientific, Germany). The data were obtained from the images of the tubes in the distended state, rings in the no-load state and sectors in the zero-stress state. In the distended state for both the intact esophagus and the separated tubes, images were analyzed for each 0.5 cmH 2 O increment. The outer diameter was measured at each pressure level and averaged over three locations. At the no-load and zero-stress states, the inner and outer circumferential lengths were measured along with the thickness and area of the wall and layers (for calculation of inner, outer or mid-layer circumference). The opening angle was defined as the angle subtended between two radii drawn from the midpoint of the inner wall to the tips of the inner wall of the open sector. The stresses and strains of the esophagus and its sublayers in the pressurized state were determined under the assumption that the geometric configuration of the lumen is cylindrical, the wall of the esophagus is incompressible, and the material in each layer is homogenous. Based on the above measurements and assumptions, parameters of the esophagus such as the luminal radius (r i-p ), the wall thickness (H p ), the mid-wall circumference (C m-p ) at a given pressure were computed as r i-p = [(r 2 o-p - A n / πλ 1 )] 1/2 , H p = r o-p - r i-p , and C m-p = 2 π (r i-p + H p /2). The outer radius (r o-p ) of the intestine was computed according to the outer diameter (D o ). The circumferential Green's strains and Kirchhoff's stress were computed according to the equations: where Cm-z is the mid-wall circumference at the zero-stress state. where The tangent modulus can be estimated from the slope of the stress-strain relation as The tangent modulus given by Eq. (3) corresponds to Young's modulus in the linear stress-strain regime. Integration of Two-Layers: An Analytical Model We assume that the circumferential stress-strain relationships for the inflation experiment obey Hooke's law for each layer σ θθ ( sm ) = E θθ ( sm ) e θθ ( sm ) σ θθ ( m ) = E θθ ( m ) e θθ ( m ) (4) where σ , E and e indicate the Cauchy stress, Tangent modulus and Green strain, respectively. θθ , sm and m indicate the circumferential direction and the submucosal and muscle layers, respectively. In general, the circumferential stress is a function of circumferential and longitudinal strain. In the present analysis, we assume that the cross-modulus is small such that the longitudinal term is negligible in comparison with the circumferential term and hence eq. (4). We further assume that the esophagus is a circular cylinder. The basic equations of equilibrium and deformation are given in Flugge [ 8 ]. Let x denote the longitudinal axis, θ the circumferential axis and z the radial coordinate. N θ denotes the tensile membrane stress resultant in θ direction. The displacement in the x, θ , and z directions of a point on the neutral axis surface are denoted by u, v, and w, respectively. The displacements of any point, A, is denoted by u A , v A , w A as follows: 1) u A = displacement along the generator, positive in the direction of increasing x; 2) v A = displacement along a circle of radius a + z, positive in the direction of increasing θ and 3) w A = radial displacement, positive outward. According to the Bernoulli-Kirchhoff hypothesis (all points lying on one normal to the neutral surface before deformation remain on the normal after deformation), we have where a indicate the neutral axis. The circumferential strain which is assumed to be small is given by The circumferential membrane stress resultant N θ is given by By substitution of Eqs. (4), (5), and (6) into Eq. (7) and noting that in the inflation experiments u, v, and w do not change with θ and w does not change with x, we obtain Eq. (8) can be integrated to yield where I stands for intact esophagus; a I , a sm and a m are the neutral axes for the submucosa (1.35) and muscle (1.15). Equation (9) can be solved in terms of E I as Hence, we can compare measured E I from Eq. (3) with E I computed from E sm and E m as given by Eq. (9b). Statistical Analysis The data were assumed to be representative of a normal distribution. The results are expressed as means ± SE. Student's t test and analysis of variance were used to detect possible differences between curves obtained from the intact esophagus and the two sublayers. The results were regarded as significant if P < 0.05. Results The esophagi shortened by approximately 30% after excision. However, the segments were stretched to the in vivo length before the mechanical tests. The length was fixed during the distension protocol. The outer radius, wall thickness, and circumferential Green's strain as function of pressure is shown in figure 1 . The radius and the Green's strain increased as function of pressure in the intact as well as the separated esophagus. At pressures higher than 1.5 cmH 2 O the muscle layer had a higher radius and strain than the mucosa-submucosa layer. The strain for the intact esophagus and for the muscle layer was negative at low pressures indicating the presence of residual strains in the tissue. The thickness of the intact wall and the separated layers decreased as function of pressure. The mucosa-submucosa was the thinnest layer. Figure 1 Outer radius (top graph), wall thickness (middle graph) and circumferential Green's strain in the intact esophagus and the muscle and mucosa-submucosa sublayers as function of pressure. Values are means ± SE. Cross-sectional views of the esophagus and its two layers were obtained at the no-load state and zero-stress state. Upon reducing the no-load state to the zero-stress state by cutting the ring radially, the opened ring expanded itself into a sector with an opening angle of about 140° for the intact esophagus (figure 2 , top). Separation of the mucosa-submucosa layer from the muscle layer resulted in the release of compressive forces in the mucosa-submucosal layer and tensile forces in the muscle layer. After separation the opening angle of the muscle and the mucosa-submucosa approached 45 and 90°, respectively. Statistically significant differences in opening angles were found between the intact segment and the mucosa-submucosa (p < 0.05), between the intact segment and the muscle layer (p < 0.05), and between the two separated layers (p < 0.05). In comparison to the specimens in the state of zero bending moment, the buckling of the mucosa became less apparent but still present after separation. Figure 2 Opening angle after anterior cut in the intact esophagus and after separation into the muscle and mucosa-submucosa layers (top) and the stress-strain relationship in sense of Kirchhoff and Green (middle). Values are means ± SE from 8 rats. The bottom graph shows the tangent modulus as function of stress for the intact esophagus and after separation into the muscle and mucosa-submucosa layers The stress-strain data are depicted in figure 2 (middle) in the sense of Kirchhoff stress and Green strain. A non-linear (exponential) curve was used to fit the data. The stress-strain curve for the submucosa-mucosa layer was located to the left of the curves for the muscle layer and for the intact esophagus at strains higher than 0.3, indicating that the submucosa-mucosa has the highest stiffness. Figure 2c shows the relationship between the tangent modulus and stress for the various layers. It can be seen that there is a linear relationship between the tangent modulus and the stress as a result of the exponential nature of the stress-strain relationship. Furthermore, the modulus is significantly higher in the submucosa layer than in the muscle layer and the intact esophagus as predicted from figure 2 (middle). Figure 3 shows a composition of the predicted (Eq. 11) and measured Young's modulus for the intact esophagus. It can be seen that the agreement is good in the low stress-strain regime (pressure < 4 cmH 2 O) where the assumption of linear stress-strain relationship is most justified. Figure 3 The elastic modulus as function of pressure for the intact esophagus. The curve with the error bar is the experimental date whereas the other curve is the theoretical curve. The theoretical fit is within one SD for the low pressure regime. Discussion Scope of Study and Major Findings The peristaltic transport of swallowed material by the esophagus to the stomach is a neuromuscular function affected by a number of neuromuscular factors [ 9 - 14 ]. The nervous system and the contractile muscle behavior of the esophagus have been studied extensively [ 10 , 15 ] but the mechanics of the esophagus tissues is lagging far behind. The stress-strain-velocity history of the tissues of the esophagus is unknown. Since the esophagus is a tube, traditionally in mechanical analysis the wall material is treated as homogeneous without further analysis into layers. The effect of the two layers on the overall mechanics of the esophagus is examined in this article. The unique structure of the esophagus as a composite of submucosa and muscle layers allows the separation of the two layers and an experimental determination of the constitutive properties of each layer. This is difficult to carry out in the blood vessels or in other organs. The coronary arteries can be separated at the external elastic laminae but only by tearing the adventitial layer [ 16 ]. The results show that the material properties differ between the esophageal intact wall, the muscle layer, and the submucosa layer. The submucosa layer is the stiffest. A Two-Layer Model A major difference between the structures of the esophagus and blood vessels is the ease with which the wall can be separated into layers. This fact allowed us to obtain the stress-strain relation and the zero-stress state of the esophageal tissue layers, as reported above. In contrast, in spite of the extensive effort on theoretical and experimental bilayer models of arteries by Berry et al [ 17 ], Demiray and Vito [ 18 ], Maltzahn et al [ 19 , 20 ], and Rachev [ 21 ], the mechanical properties and the zero-stress states of the bi-layers of arteries are still largely unknown. For the esophagus, a practical question is: Can we regard the wall of the esophagus as a homogeneous tissue. Or must we treat it as composed of a mucosa-submucosa layer and a muscle layer? Or must we model it with even more detailed structures? Or can we model it as simply two concentric, non-interacting layers. The answer depends on the purpose of our investigation: What features of the organ does the investigator wish to know. In this article, we examined both the bi-layer model and the monolayer model, present a comparison and propose a simple model to explain the interaction of the two layers. The opening angle of the inner submucosa layer is larger than that of the outer muscle layer which agrees with our previous report [ 1 , 3 ] and layered artery [ 22 , 23 ]. However the opening angles were largest in the intact layer and smallest in the muscle layer which differs somewhat from those obtained in guinea-pig [ 1 ] and in rabbit [ 3 ]. The difference may be species related but also due to experimental technique. In the previous studies esophageal rings were first cut radially and then separated circumferentially [ 1 , 3 ]. In the present study we first separated the inner submucosal tube from the outer muscle tube and then cut the ring radially in each layer. We believe that this procedure minimizes any damage inferred by the cutting. We plan to investigate the differences in experimental protocol to rule what causes the differences in opening angle between the previous studies and this study. The stress-strain curve for the submucosa-mucosa layer was shifted to the left of the curves for the muscle layer and for the intact esophagus at strains higher than 0.3, indicating that the submucosa-mucosa has the highest stiffness (figure 2 ). This corresponds to the finding of a lower Green strain at pressure above 1.5 cmH2O (figure 1 ). The difference found below strain 0.3 and pressure 1.5 is due to that the submucosa-mucosa is compressed in the intact esophagus. Hence, when we study this layer after separation, it has a higher strain at low loads when compared to the other specimens. Furthermore, the submucosa-mucosa layer is rich in collagen. Thus, a contributing factor to the observed difference between layers may be that collagen during stretch first uncrimps with little resistance, then at higher loads has a high stiffness [ 7 ]. We observed that the stress-strain relationship of the intact esophagus and its two layers is exponential. The tangent modulus, which is the slope of the stress-strain relationship, varies exponentially with the strain (according to the stress-strain figure) and linearly with the stress. Hence, it is simpler to examine the modulus as function of stress. For a nonlinear stress-strain relationship it is meaningless to specify the modulus unless a stress or strain level is prescribed. Fung proposed that the slope of the tangent modulus-stress relationship, α , can be used as a measure of stiffness [ 24 ]. The data in this study clearly shows that the mucosa-submucosa layer is the stiffest which is in accordance with previous experience and the fact that submucosa contains large amounts of collagen. Therefore, the esophageal wall should be modelled as at least a two-layered composite system, as has also been proposed for arteries [ 18 - 20 , 22 , 23 , 25 , 26 ]. Limitations of Study A limitation of the study is that only uni-axial data were obtained in this study. Intuitively the circumferential direction seems to be the most important for cylindrical organs. Since longitudinal changes may also be important for esophageal function, future studies should implement bi- or tri-axial data. Another limitation is that the analytical model is restricted to the linear stress-strain regime. Furthermore, the esophagus and its layers are assumed to be cylindrical tubes, and that the esophageal tissue is incompressible. The last assumption is possibly true in the pressure range studied and it is also known from yet unpublished studies that the esophagus attains circular geometry both at the inner and outer surfaces even at low pressures. The linearity assumption needs to be generalized. We also assumed that each layer was homogeneous, though it is well known that the muscle layer is composed of longitudinal and circumferential muscle bundles. Hence, the muscle layer can be modeled into further sublayers. Conclusions and significance of research We have developed an analytical tool that can be used to analyze the mechanics of bilayered organs. The model was used to study the esophagus. The model may be useful for studying the mechanical properties of other organs that can be separated into layers. There are two immediate implications of the results in this study for the understanding of esophageal function and for clinical practice. It is well known that pain may arise from the esophagus and that the receptors involved in the mechanotransduction are located at different positions in the wall. Detailed information about the stress and strain distributions in the layers is therefore important for the interpretation of receptor-mediated responses. Furthermore, the stress reduction during loading (caused by the residual stresses in the layers) may serve as a mechanism to reduce damage to the esophagus during excessive loadings caused by swallowing of large objects or by acute esophageal obstruction. For the esophagus the model may be applied to the study of remodeling of the individual layers in response to disease. For example, in systemic sclerosis the muscle layers in esophagus are slowly replaced by fibrotic tissue, creating a passive conduit (fall pipe). It is already known that the esophageal stiffness increase in patients with systemic sclerosis [ 27 , 28 ] but we know very little about the mechanical remodeling in each layer. Authors contributions Yanhua Fan carried out the experimental work, the measurements, and read and approved the final manuscript. Hans Gregersen designed the study, partly analyzed the data, and drafted the manuscript. Ghassan S. Kassab suggested the analysis, analyzed the data, drafted and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC528839.xml |
548673 | Examination of Cholesterol oxidase attachment to magnetic nanoparticles | Magnetic nanoparticles (Fe 3 O 4 ) were synthesized by thermal co-precipitation of ferric and ferrous chlorides. The sizes and structure of the particles were characterized using transmission electron microscopy (TEM). The size of the particles was in the range between 9.7 and 56.4 nm. Cholesterol oxidase (CHO) was successfully bound to the particles via carbodiimide activation. FTIR spectroscopy was used to confirm the binding of CHO to the particles. The binding efficiency was between 98 and 100% irrespective of the amount of particles used. Kinetic studies of the free and bound CHO revealed that the stability and activity of the enzyme were significantly improved upon binding to the nanoparticles. Furthermore, the bound enzyme exhibited a better tolerance to pH, temperature and substrate concentration. The activation energy for free and bound CHO was 13.6 and 9.3 kJ/mol, respectively. This indicated that the energy barrier of CHO activity was reduced upon binding onto Fe 3 O 4 nanoparticles. The improvements observed in activity, stability, and functionality of CHO resulted from structural and conformational changes of the bound enzyme. The study indicates that the stability and activity of CHO could be enhanced via attachment to magnetic nanoparticles and subsequently will contribute to better uses of this enzyme in various biological and clinical applications. | Background Magnetic materials have been used with grain sizes down to the nanoscale for longer than any other type of material [ 1 ]. This is attributable to a number of factors including a large surface area to volume ratio and the possibility of immobilizing a biological entity of interest [ 2 ]. In the last decade increased investigations and development were observed in the field of nanosized magnetic particles [ 2 ]. Here the term nanoparticles is used to designate particulate systems that are less than 1μm, and effectively below 500 nm [ 2 ]. Due to their magnetic character, magnetite (Fe 3 O 4 ) nanoparticles can be attracted by a magnetic field and are easily separable in solution. Similarly, substances to which they have been attached can be separated from a reaction medium, or directed by an external magnetic field to site specific drug delivery targets [ 2 ]. Magnetic nanoparticles have been widely used in the immobilization of many bioactive substances such as proteins, peptides, enzymes [ 3 - 6 ], and antibodies [ 7 ]. Magnetite is one of the most commonly used magnetic materials because it has a strong magnetic property and low toxicity [ 4 ]. The binding of magnetic particles to bioactive substances involves a number of interactions including the interactions between organic ligand, and the interactions between the amino acid side chains of proteins and the metals centers. Such bindings pave the way for the coupling of biomolecular entities of enhanced stability. Recently reported work in the area of enzyme immobilization has described the catalytic activity of yeast alcohol dehydrogenase [ 3 ] and lipase [ 4 ] directly bound to magnetite nanoparticles, via carbodimiide activation without the use of a ligand. This binding method offers tremendous scope because of its simplicity and high efficiency. Cholesterol oxidase is a flavin-enzyme (with a FAD prosphetic group) that produces hydrogen peroxide according to the reaction 1. Cholesterol + O 2 → 4 - Cholesten - 3 - one + H 2 O 2 (1) The structure of cholesterol oxidase reveals deeply buried active sites occupied by water molecules in the absence of its substrate steroids [ 8 ]. Cholesterol oxidase is industrially and commercially important for application in bioconversions for clinical determination of total or free serum cholesterol [ 9 - 12 ] and in agriculture [ 13 ]. Its activity can be determined by following the appearance of the conjugated ketones, the formation of hydrogen peroxide in a coupled test with peroxidase, or by measuring the oxygen consumption polarographically [ 13 ]. Several studies on its kinetic properties have appeared [ 13 - 15 ]. More recently, Cholesterol biosensor based on entrapment of cholesterol oxidase in a silicic sol-gel matrix at a Prussian Blue modified electrode has been developed [ 15 ]. However, this method of enzyme immobilization raises concerns on reduced surface area for enzyme binding and pore-diffusion resistance [ 2 ]. Immobilization of enzymes onto inorganic material surfaces is of vital importance in enzymatic reactions, especially in biosensor applications. Information on the activity and availability of cholesterol oxidase bound to Fe 3 O 4 magnetic nanoparticles will contribute to the basic understanding of its activity and function. The present study proposes to investigate the direct binding of cholesterol oxidase to Fe 3 O 4 magnetic nanoparticles. The sizes and structure of the nanoparticles were characterized using TEM and FTIR spectroscopy. The stability, activity, and kinetic behavior of bound cholesterol were also examined. Results and discussions Particle size and structure TEM micrographs of "bare" magnetic nanoparticles and CHO-functionalized magnetic nanoparticles are shown in Figure 1a and 1b . The "bare" particles were very fine with a diameter ranging from 9.7 to 56.4 nm. The size of the particles after binding to CHO was globally the same as the "bare" particles. Figure 2 shows the size distribution of the particles. However, some spots of agglomerated particles were visible as seen in figure 1b . These agglomerates cause an increase in maximum particle size. The overall sizes of the particles after binding to magnetic nanoparticles were between 9.7 and 166 nm suggesting a perceptible agglomeration in association with the binding process. A possible explanation is that the binding of magnetic nanoparticles was not only a monomolecular process but may involve the binding of several CHO molecules on a single Fe 3 O 4 particle. It could also be envisaged that CHO molecules formed aggregates to bind several magnetic nanoparticles. Another possible factor in the agglomeration process is the centrifugation process involved in the separation of the supernatant from the Fe 3 O 4 -CHO. It is obvious that the centrifugation tend to bring particles together as a compact material. The effect of agglomeration at this stage can be reduced by separating the Fe 3 O 4 -CHO by an external magnetic field. Since the particles are released after removal of the magnetic field, they may fall separately apart from each other, and are less likely to agglomerate. Figure 1 Transmssion Electron micrographs of Fe 3 O 4 magnetic nanoparticles (a) and Fe 3 O 4 -CHO (b). Figure 2 Distribution of the particle sizes on the electron micrographs. The values denote the averages of duplicate measurements. Binding efficiency The unbound enzyme was determined by assaying the protein content in the supernatant. It was found that the percentage of cholesterol oxidase bound was between 98 and 100%, irrespective of the amount of particles. The amounts of Fe 3 O 4 nanoparticles used were 14.4, 17.2 and 20 mg/mL, corresponding to CHO/Fe 3 O 4 weight ratios of 0.01, 0.08 and 0.007, respectively. These results show that in all the binding operations, there were sufficiently available amount of particles to bind the enzymes till complete saturation. In a previous study [ 4 ], it was found that increasing the amount of Fe 3 O 4 nanoparticles, that is reducing the weight ratio of CHO to Fe 3 O 4 below 0.033 caused an increase in lipase binding up to 100%. This was not observed in this study, possibly because of the difference in the binding mechanism, due to differences in the structure of the enzyme. However, the percentage of bound CHO (98–100%) shows that the binding process was successful. Binding confirmation The binding of CHO to magnetic nanoparticles was confirmed by FTIR analysis. Figure 3 (a, b, and c) shows the FTIR spectra for "bare" Fe 3 O 4 , Fe 3 O 4 -CHO, and CHO in water, respectively. A characteristic band of NH 2 was observed at 1618 cm -1 in the "bare" Fe 3 O 4 nanoparticles. The NH 2 group can be associated with NH stretch at 3400 cm -1 which is not visible here, because of a possible hindrance by OH stretch from water. However, this band was not apparent in the spectra of Fe 3 O 4 -CHO suggesting that the binding of CHO to the nanoparticles involved this amino group and the carboxylic groups of CHO after being activated by Carbodiimide, as suggested by [ 4 ]. Peaks at 3032 cm -1 and 1445 cm -1 are more visible in Figure 3a (bare particles) and perceptible in Figure 3b (Fe 3 O 4 -CHO) and could be assigned to traces of residual ammonium hydroxide. The characteristic bands of proteins at 1647 and 1541 cm -1, and 1645 and 1541 cm -1 , in the spectra of Fe 3 O 4 -CHO, and CHO, respectively shows that cholesterol oxidase was effectively present in the samples, confirming the binding of cholesterol oxidase to Fe 3 O 4 nanoparticles. The negative peak at 3400-2799 cm -1 is possibly due to a reduced amount of water in the sample compared to the water used for background subtraction. The characteristic bands of proteins in the Fe 3 O 4 -CHO spectra were very weak compared to those in the spectra of cholesterol oxidase in water. The weakness of the peaks is due to the limited amount of CHO bound to the nanoparticles, in comparison to the amount dispersed in water. Figure 3 FTIR spectra of Fe 3 O 4 magnetic nanoparticles (a) in nanopure water and Fe 3 O 4 -CHO (b), and pure CHO (c) prepared in phosphate buffer and then dissolved in nanopure water for FTIR analysis. Cholesterol oxidase activity and binding kinetics The kinetic parameters of the enzymatic reactions estimated by the Lineweaver-Burk plots of the initial rates of cholesterol oxidase from experimental data are presented in Figure 4 . The Michaelis-Menten constants V max and K m for CHO were determined to be 0.67 μmol/min mg and 2.08 mM for the free enzyme and 1.64 μmol/min mg and 0.45 mM for the immobilized enzyme, respectively. The V max value of the bound CHO was 2.4 fold higher than that of the free, and the K m value of the bound CHO was 4.6 fold lower than that of the free CHO. The low K m reflects the high affinity to substrate [ 4 ]. The high affinity of the enzyme to the substrate may be explained by the fact that when binding onto the surface of the nanoparticles, the enzyme rearranged itself to present a better conformation. Since the secondary and tertiary structure of cholesterol oxidase play important roles in its activity [ 9 ], the rearrangement in structure and conformation may result in better availability of its active sites. The increase in affinity of the enzyme to the substrate upon binding to Fe 3 O 4 nanoparticles contributed to an enhancement of the activity of the enzyme. Figure 4 Lineweaver Burk plots of the initial rates of CHO (■) and (◆) Fe 3 O 4 -CHO at pH 7.4, from experimental data. Effect of pH The effect of pH on the activities of the free and bound CHO was investigated in the pH range of 6–8.5 at 25°C and presented in Figure 5 . In the pH range between 6 and 7.4 the activities of the free and bound CHO were quite similar and reached a maximum at pH 7.4. The activity then decreased from pH 8 to 8.5. In this range, the activity of the bound CHO was much higher than its free counterpart. This shows that the bound enzyme showed better tolerance to the variation of solution pH. The similarities in these activities in the pH range of 6 to 7.4 indicate that in these conditions, CHO did not suffer from any major activity constraint. Rather, this pH range appears to be suitable for CHO activity. It is well known that the ability of the amino acids at the active sites of the enzyme to interact with the substrate depends on their electrostatic state [ 16 ]. The decrease in activity observed at pH 8 and 8.5 shows that CHO faces some limitations as the pH increased toward more alkaline conditions. If the pH is not appropriate, the charge on one or all of the required amino acids is such that cholesterol can neither bind nor react properly to produce 4-cholesten-3-one. Figure 5 Effect of pH on the activities of free (■) and bound CHO (◆). Thermal stability The thermal stability of free and bound CHO was investigated after 40 min of storage in the temperature range of 25–70°C (Figure 6 ). There was no apparent change in activity in the free CHO as well as in the bound CHO, in the temperature range of 25–37°C. Above this temperature range, the residual activity decreased in both systems. However, the bound CHO showed higher retained activity than the free CHO. The remaining activity at 60°C was about 2 fold that of the free CHO. This proved that the thermal stability was significantly improved upon binding of CHO to magnetic nanoparticles. Table 1 shows the inactivation rates constants ( k ) at temperatures where the inactivation experiments were observed. The rate constants increased with increasing temperature and were higher for the free CHO than for bound CHO. As stated above, the binding to nanoparticles suggests a better resistance of the enzyme to temperature. We hypothesize that the bound enzyme could possibly undergo a conformational change and a spatial rearrangement that could slow down the folding process and denaturation of the enzyme. Figure 6 Thermal stability of free CHO (■) and Fe 3 O 4 -CHO (◆) at pH 7.4. The samples were stored at 50, 60, or 70°C for 40 min and the activities were then measured at 25°C. Table 1 Inactivation rate constants ( k ) of the "bare" and bound CHO at various temperatures Temperature (°C) Free CHO Fe 3 O 4 -CHO k (min -1 ) k (min -1 ) 50 3.4 × 10 -2 4.6 × 10 -3 60 9.3 × 10 -2 5.6 × 10 -2 70 2.8 × 10 -1 1.9 × 10 -2 Effect of temperature on enzyme activity and stability The effect of temperature on the activity of the free CHO was examined by measuring its relative activity when stored at various temperatures (Figure 7 ). It can be observed that at 37°C, the enzyme retained its activity for about 80 minutes before showing a slight decrease. At 50°C the activity decreased continuously to 35% after 110 min. A more severe decrease in activity occurred at 60 and 70°C, resulting in a complete loss of activity after 60 and 70 min, respectively. The decrease in activity may be attributed to a dramatic change in the structure of the enzyme that hindered the availability of the active sites, with a possible denaturation of the enzyme itself. The effect of temperature on the activities of free and bound CHO at pH 7.4 are displayed in the Arrhennius plots (Figure 8 ). Only temperatures (50, 60 and 70°C) at which perceptible changes in activity were observed were studied. The activation energies were calculated to be 13.6 and 9.3 KJ/mol for free and bound CHO, respectively. The low activation energy related to the bound CHO suggests that when bound to the magnetic nanoparticles, CHO seems to acquire a better orientation that reduces the energy barrier for activity. Figure 7 Effect of various temperatures on the activity Fe 3 O 4 -CHO at pH 7.4. Figure 8 Arrhennius plots of the initial plots of the oxidation rates of cholesterol by free CHO (■) and Fe 3 O 4 -CHO (◆) for samples at 50, 60, or 70°C. Storage stabilities The stability and activity of the enzyme are naturally reduced during storage. Figure 9 shows the storage stabilities of free and bound CHO at 25°C at pH 7.4. After 15 days, no residual activity was observed in free CHO. However, the residual activity of bound CHO was 59% during the same time period, and 27% after 30 days indicating a considerable enhancement on its stability. It has been argued that this higher stability of the bound enzyme was due to its fixation on the surface of magnetic nanoparticles, preventing the auto-digestion and thermal inactivity [ 3 ]. Another plausible explanation is that the binding of CHO on Fe 3 O 4 nanoparticles might allow a better spatial orientation of the FAD prosphetic groups and the side chains of CHO providing a better stability to the enzyme. Figure 9 Storage stability of free CHO (■) and Fe 3 O 4 -CHO (◆). The activities measurements were performed at pH 7.4, at 25°C Materials and methods Materials Cholesterol oxidase (EC 1.1.3.6), Nocardia sp. was purchased from VWR international (Pittsburgh, USA). Carbodiimide-HCl (1-ethyl-3-(3-dimethyl-aminopropyl), ammonium hydroxide reagent, Triton X-100, TRIS (Hydroxymethyl) aminomethane HCL, 4-cholesten-3-one, bovine serum albumin (BSA), iron (II) chloride tetrahydrate 97 %, and iron (III) chloride hexahydrate 99% were obtained from Sigma-Aldrich, St Louis (USA). The Biorad Protein Assay Dye Reagent Concentrate was purchased from Biorad Laboratories (Hercules, CA). Acetonitrile was obtained from EMD Chemicals, (New Jersey, USA). Preparation of magnetic nanoparticles Magnetic nanoparticles (Fe 3 O 4 ) were prepared by chemical co-precipitation of Fe 2+ and Fe 3+ ions in a solution of ammonium hydroxide followed by a treatment under hydrothermal conditions [ 4 , 5 ]. Iron (II) chloride and iron (III) chloride (1:2) were dissolved in nanopure water at the concentration of 0.25 M iron ions and chemically precipitated at room temperature (25°C) by adding NH 4 OH solution (30%), at a control pH (10–10.4). The suspensions were heated at 80°C for 35 min under continuous mixing and separated by centrifuging several times in water and then in ethanol at 2800 rpm. The purification step was used to remove impurities from Fe 3 O 4 nanoparticles. The particles were finally dried in a vacuum oven at 70°C. The dried particles exhibited a strong magnetic attraction to a magnetic rod. Attachment of cholesterol oxidase onto magnetic nanoparticles 50–70 mg of magnetic nanoparticles was added to 1 mL of phosphate buffer (0.05 M. pH 7.4). The mixture was sonicated for 15 min after adding 0.5 mL of carbodiimide solution (0.02 g/mL in phosphate buffer (0.05 M. pH 7.4). Following the carbodiimide activation, 2 mL of cholesterol oxidase (0.25 mg/mL) was added and the reaction mixture was sonicated for 30 min at 4°C in a sonication bath and the mixture was centrifuged at 3000 rpm [ 17 ]. The precipitates containing Fe 3 O 4 nanoparticles and Fe 3 O 4 bound cholesterol oxidase (Fe 3 O 4 -CHO) were washed with phosphate buffer pH 7.4 and 0.1 M Tris, pH 8.0, 0.1 M NaCl and then used for activity and stability measurements. NaCl was added to enhance the separation of the magnetic nanoparticles [ 3 ]. Determination of immobilization efficiency The amount of protein in the supernatant was determined by a colorimetric method at 595 nm using the Biorad Protein Assay Reagent Concentrate with bovine serum albumin (BSA) as the protein standard. The amount of bound enzyme was calculated from: A = ( C i - C s )* V (2) Where A is the amount of bound enzyme, Ci and Cs is the concentration of the enzyme initially added for attachment, and in the supernatant, respectively (mg-mL -1 ), V is the volume of the reaction medium (mL). Characterization The size of Fe 3 O 4 nanoparticles and Fe 3 O 4 -CHO was characterized by transmission electron microscopy (TEM, JEM 1200 EXII, JEOL USA) and structure by Fourier Transform Infrared (FTIR) spectroscopy (Biorad FTS 6000, Cambridge, MA). The samples for TEM analysis were prepared by placing a drop of the magnetic nanoparticles dispersed in nanopure water onto a copper grid and evaporated in air at room temperature. Before preparing a sample onto the copper grid, the dispersed solution was sonicated for 4 min to obtain better particle dispersion. The binding of CHO onto the magnetic nanoparticles was investigated using FTIR. CHO and Fe 3 O 4 -CHO samples in phosphate buffer and Fe 3 O 4 particles were dissolved in nanopure water for FTIR analysis. Activity measurement The activity of bound CHO was determined by measuring the initial oxidation rates of cholesterol by cholesterol oxidase at given temperature following the increase of 4-cholesten-3-one concentration at 240 nm, using a Beckman Du Spectrometer. A solution of cholesterol was prepared by dissolving 4.8 g of cholesterol in 10 mL of 2-propanol. A phosphate buffer solution (0.05 M. pH 7.4) containing 4% of Triton-100 was added to the mixture to result in a 0.26 M cholesterol solution. The mixture was gently heated until the solution was clear. To start the enzymatic reaction, 5 ml of cholesterol solution was added to 15 mL centrifuge test tubes containing Fe 3 O 4 -CHO, and mixed by vortex. A solution of free CHO of the same concentration was used to evaluate the activity of the free enzyme. The solution was incubated at various temperatures (25–70°C) at specific intervals of time (1 h) and centrifuged at 3000 rpm for 5 min to separate the supernatant from Fe 3 O 4 -CHO. 10 μL aliquots of the supernatant were then taken and the concentration of 4-cholesten-3-one was assessed. Before measuring the amount of 4-cholesten-3-one in a sample, the activity of the free enzyme was stopped by adding an equal volume of acetonitrile to the reacting solution [ 18 ]. Each kinetic measurement was the average of duplicate replications. Thermal stability of free and immobilized enzyme The thermal stability of free and Fe 3 O 4 -CHO were determined by measuring the residual activity of the enzyme at 25°C, after being exposed to different temperatures (25–70°C) in phosphate buffer (0.05 M, pH 7.4) for 40 min. Aliquots of the reacting solution were taken at time intervals (every 30 min for 7 hours) and assayed for enzymatic activity as described above. The first order inactivation rate constant, k was calculated from the equation: In A = In A 0 - kt (3) where A 0 is the initial activity, A is the activity after a time t (min), k is the reaction constant. Effect of temperature on enzyme activity The effect of temperature on the free CHO and Fe 3 O 4 -CHO was estimated by determining the concentration of 4-cholesten-3-one in samples at various temperatures. A solution of cholesterol was added to the various centrifuge test tubes containing bound or free enzymes. The test tubes were stored in a water bath at specific temperatures (25, 37, 50, 60, and 70°C). At time intervals, the concentration of 4-cholesten-3-one was determined by spectrophotometric analysis. Storage activity The storage stability was evaluated by determining the concentration of 4-cholest-en-3-one at room temperature at time intervals (5 days). Test tubes containing Fe 3 O 4 -CHO or free enzyme solution were stored at 25°C in phosphate buffer (0.05 M. pH 7.4) for 30 days. Thereafter, 5 mL of cholesterol was added. The storage stability of the free and bound cholesterol oxidase was determined by assaying for their residual activity. Determination of kinetics parameters The kinetic parameters of free CHO and Fe 3 O 4 -CHO, K m and V max were determined by measuring initial rates of oxidation of cholesterol (1.3–5.2 mM) by CHO (0.25 mg/mL) in phosphate buffer pH 7.4 at 25°C. Conclusions Magnetic nanoparticles were synthesized by thermal co-precipitation of ferric and ferrous chlorides. The binding of CHO to the particles was confirmed by FTIR spectroscopy and the size characterized by TEM. The binding efficiency was between 98 and 100% irrespective of the amount of particles used. Kinetic studies of the free and bound CHO revealed that the stability and activity of CHO were significantly improved upon binding to nanoparticles. Furthermore, the bound enzyme exhibited a better tolerance to pH, temperature and substrate concentration. The activation energy indicated that the binding of CHO onto Fe 3 O 4 magnetic nanoparticles reduced the energy barrier for CHO activity. As a result of the binding to the magnetic nanoparticles, the storage stability of CHO was considerably enhanced. This higher stability of the Fe 3 O 4 -CHO is attributable to its possible fixation on the surface of the particles preventing auto-digestion and thermal inactivity. In addition, the binding on Fe 3 O 4 nanoparticles might allow a better spatial orientation of the FAD prosphetic groups and the side chains of CHO to provide better stability to the enzyme. The overall improvements observed in activity, stability, and functionality of CHO resulted from structural and conformational changes of the bound cholesterol oxidase. The study may be useful in improving the stability and activity of cholesterol oxidase, and will contribute to more efficient use of this enzyme. List of Abbreviations used CHO: Cholesterol oxidase TEM: Transmission electron microscopy FTIR: Fourier Transform Infrared BSA: Bovine serum albumin Authors' contributions Drs Gilles K Kouassi and Joseph Irudayaraj were the primary authors. They were responsible for the concept, experimental plan, and analysis. Dr Gregory McCarty was the secondary author and contributed to the overall effort. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548673.xml |
529299 | Musculo-skeletal pain among 40- and 45-year olds in Oslo: differences between two socioeconomically contrasting areas, and their possible explanations | Background The objective of the study was to compare the prevalence and severity of musculo-skeletal pain between two socioeconomically contrasting areas in Oslo, Norway, and to explore possible explanatory factors. Methods Questionnaire survey, carried out as part of The Oslo Health Study in 2000–2001. Data from 821 persons (40 and 45 year old) living in a less affluent inner city area (called east) were compared with 854 persons living in an affluent area of the city (called west). Bivariate comparisons (chi square test) and multiple regression analyses were performed to investigate differences between the samples. Results 61 % in east and 56 % in west (p < 0.05) reported pain/stiffness in muscles/joints during the last four weeks. 30 % in east versus 19 % in west (p < 0.001) reported extensive pain. The between area difference in extensive pain was partially explained by physical inactivity, mental health problems and being of non-Western origin. Conclusion Musculo-skeletal pain is reported by 55–60 % of middle aged persons in Oslo during a four week period, and must be considered a normal phenomenon. Poor social conditions, inactivity, mental health problems and being an immigrant imply increased risk of more severe symptoms with a concomitant demand of health care. | Background In affluent societies like Norway, living conditions as well as general health status have improved during the last decades. In spite of this, social health inequities still exist, and recent analyses from Oslo even indicate an increase during the last 30 years [ 1 ]. Life expectancy for men living in the least affluent city area is 69 years, compared to 76 for men in the most affluent area [ 2 ]. Well-known risk factors like smoking, physical inactivity and overweight, as well as the incidence of atherosclerotic disease and several forms of cancer show similar correlation [ 3 ]. Some claim that future research on this topic should concentrate exclusively on interventions [ 4 ]. In Great Britain as well as Holland this has been highlighted for some time [ 5 ], and in Holland a national strategy for tackling health inequities has been developed [ 6 ]. Nevertheless, we need to continuously keep an eye on trends, as well as on causal and maintaining factors. And even if we have ample data on socioeconomic inequity regarding mortality and morbidity, we know far less about the dimensions of disease severity and patients' coping ability, in Oslo as elsewhere [ 7 ]. The aim of the present study was to investigate differences in prevalence and severity of musculo-skeletal pain between middle aged inhabitants of two socioeconomically contrasting areas in Oslo, based on a recent and comprehensive data collection, and to examine some possible explanatory factors. We chose to study musculoskeletal pain, as this is a major cause of disability in the industrialised world [ 8 ]. In Norway, musculoskeletal pain generates 15–20 % of consultations in primary care, and is one of the main reasons for sick leave and social security [ 9 ]. Methods The data collection was part of the Oslo Health Study, a joint collaboration between the Oslo City Council, the University of Oslo and the Norwegian Institute of Public Health, which was conducted from May 2000 to September 2001. All residents born in 1924/25, 1940/41, 1955, 1960 and 1970 (n = 41353) received the three-page main questionnaire by mail, as an invitation to participate in a health screening. At the screening station a simple clinical examination and a blood test were performed, and the questionnaire was handed in. Two supplementary questionnaires were given out: one identical for all age groups, and one in four different versions. Participants were asked to fill in the supplementary questionnaires at home and return them by mail. Two reminders were sent to non-respondents. An overview of all topics covered in the questionnaires (in English) can be obtained from . In the present study we analysed data from persons born in 1955 and 1960, who lived either in the inner eastern part of Oslo or in the outer western part (see below). We used data from the main questionnaire as well as from the age specific supplementary questionnaire. The variables included from the main questionnaire were: marital status, educational level, employment status, disability pension, social assistance, country of origin, physical exercise, alcohol intake, smoking habits, general health status, mental health problems, and musculo-skeletal disorders. Country of origin was recoded as Western (Western Europe, North America, Australia) or non-Western (Eastern Europe, North Africa, Sub-Saharan Africa, Middle East, Indian subcontinent, Eastern Asia, Pacific, Middle America, South America) [ 10 ]. Mental health problems were assessed by the following question: Below is a list of various problems: Have you suffered from any of the following during the last week, including today? Put a cross for every problem . Choices: Not troubled, slightly troubled, quite a lot troubled, much troubled (values 1–4). The values were summarised and divided by the number of answers, and a mean value of 1,85 or more was used as a marker of mental health problems [ 10 ]. Musculo-skeletal pain was explored by the following question: Have you suffered from pain and/or stiffness in muscles and joints in the course of the last four weeks? Choices: Not troubled, somewhat troubled, very troubled (values 1–3) for the alternatives neck/shoulders, arms/hands, upper back, lower back, hips/legs/feet and elsewhere. The values were summarised and divided by the number of answers. A mean value of 2 or more was used as an indicator of extensive pain/stiffness in muscles or joints [ 10 ]. The variables from the age specific supplementary questionnaire included were: own income, household income, muscular pain/stiffness last 4 weeks, duration of muscular pain/stiffness, satisfaction with health care, and belief in own coping ability. The east and west areas Oslo's local authority districts can be ranked according to: level of income, education, employment, disability pension, housing standard, number of non-western immigrants, and mortality [ 11 , 12 ]. According to this ranking, three districts in the western part of the city are on top, indicating the best socioeconomic conditions. These are the districts Vindern, Røa and Ullern, here called west. Three districts in the inner eastern part take on the least favourable positions: Sagene-Torshov, Grünerløkka-Sofienberg and Gamle Oslo, here called east. Per January 1 st 2000, west had 67296 inhabitants and east 80 668. (Since the study was done, the city of Oslo has reorganized the local authority districts. Vindern and Røa are joined under the name Vestre Aker, and the names of two others are changed to Sagene and Grünerløkka). We have chosen to compare these two areas, because they are strongly contrasted regarding living conditions. Statistical analyses Statistical analyses were performed using SPSS version 11.0. Bivariate comparisons of categorical variables were examined by the chi square test. Multiple regression analyses (stepwise) were performed to estimate the explanatory power of independent variables. A 5 % level of significance was chosen. Results The main questionnaire was completed by 821 forty- and 45 year olds living in east (50.7 % women) and 854 living in west (62.9% women), corresponding to a response rate of 39.0 % in east and 43.9 % in west. Some returned the questionnaire without attending the health screening, meaning that 1348 persons completed the supplementary questionnaires. There was no significant difference regarding full time employment, and frequent use of alcohol was more common in west. For all other socioeconomic and lifestyle variables, as well as general and mental health, east came out poorer (Table 1 ). Table 1 Oslo Health Study 2000–2001, 40- and 45- year olds Demographic variables, lifestyle, and self-reported health in east, west, and the whole city (percent). East West City Education =< 9 years 14.0 2.0 8.3 Education > 12 years 57.3 87.8 65.8 Single status 15.4 8.0 9.5 Employment full time 67.4 71.2 1 72.8 Disability pension 10.5 2.7 5.6 Social benefit 6.9 0.5 2.6 Own income < 200 000 30.2 19.6 24.6 Own income > 400 000 8.5 30.0 20.0 Household income < 200000 21.0 4.0 12.0 Household income > 500000 22.7 66.9 46.2 Non-western country of origin 24.5 5.4 16.8 No hard exercise 35.1 19.5 28.7 Alcohol at least once a week 49.4 66.9 52.9 Daily smoking 41.4 22.4 32.5 Less than good health 29.3 11.7 21.2 Mental health problems 23.1 9.4 12 1 Not significant difference east versus west All other variables: significant difference, p < 0.001 (chi square test) The proportion having experienced muscular pain/stiffness during the last four weeks, being very troubled by muscular pain/stiffness in various body parts, or reporting extensive pain/stiffness was higher in east. No difference was found regarding pain duration. Participants in west were more satisfied with health care and more confident in own coping ability (Table 2 ). Table 2 Oslo Health Study 2000–2001, 40- and 45- year olds Musculo-skeletal disorders in east, west, and the whole city (percent). East West City Pain/stiffness in muscles/joints last four weeks 61.4 55.9* 58.4 Very troubled by pain/stiffness in arms/hands 6.7 4.0* 5.3 Very troubled by pain/stiffness in neck/shoulders 13.6 7.5 ++ 10.5 Very troubled by pain/stiffness in upper back 7.6 3.1 ++ 5.3 Very troubled by pain/stiffness in lower back 11.4 4.8 ++ 8.0 Very troubled by pain/stiffness in hips/legs/feet 10.6 4.8 ++ 7.6 Very troubled by pain/stiffness elsewhere 3.8 1.3 + 2.4 Extensive pain/stiffness in muscles and/or joints 30.0 18.9 ++ - Duration > 3 years 44.2 42.5 43.4 Satisfied with health care (quite satisfied/very satisfied) 28.5 38.9 + 32.2 Confident in coping ability (quite sure/very sure) 74.9 86.2 ++ 76.7 * significant difference east versus west (chi square test) p < 0.05 + p < 0.01 ++ p < 0.001 Female gender, living in east, low education, low own income, non-Western country of origin, no hard exercise and mental health problems were all correlated to extensive muscular pain/stiffness (Table 3 , left column). Female gender, no exercise, non-Western origin and mental health problems still implied increased risk of extensive muscular pain/stiffness when the other variables were adjusted for. Low education and living in east no longer showed an independent correlation with extensive pain/stiffness after adjustment (Table 3 , right column). Table 3 Oslo Health Study 2000–2001, 40- and 45- year olds living in areas east and west Odds ratio for much pain/stiffness in muscles and/or joints, related to demographic variables, lifestyle and mental distress. Logistic regression analyses. Odds ratio (95 % CI), unadjusted Odds ratio (95 % CI), adjusted 1 Female sex 1.45 (1.15 – 1.83) 1.85 (1.40 – 2.45) Single status 1.02 (0.82 – 1.29) Area east 1.88 (1.49 – 2.38) 1.22 (0.92 – 1.62) Low education 2.11 (1.43 – 3.16) 1.09 (0.67 – 1.78) Low income 1.34 (1.01 – 1.79) Low household income 1.1 (0.72 – 1.68) Non-western country of origin 4.41 (3.25 – 5.99) 3.17 (2.17 – 4.63) Daily smoking 0.81 (0.64 – 1.04) No hard exercise 2.00 (1.55 – 2.59) 1.37 (1.01 – 1.85) Mental health problems 3.87 (2.92 – 5.14) 2.70 (1.94 – 3.76) 1 Variables included into analyses: sex, area, education, country of origin, exercise, mental health problems. We performed the logistic regression analyses for men and women separately (data not shown). Non-western origin was the most important predictor of extensive pain/stiffness in men (OR 3.36, 1.93 – 5.83) and mental health problems in women (OR 3.04, 95 % CI 1.99 – 4.66). When we performed the analyses for respondents of Western and non-Western origin separately (data not shown), mental health problems were the most important independent predictor for extensive pain/stiffness for both groups (OR 2.87, 95 % CI 1.97 – 4.19 for Western, OR 2.2, 1.1 – 4.5 for non-Western). Discussion In both areas around 60 % reported pain/stiffness in muscles/joints during the previous four weeks: 61.4 % in east and 55.9 % in west, a statistically significant difference of little clinical relevance. We do not know the prevalence among non-respondents, but as The Oslo Health Study implied a comprehensive data collection on many topics, it is unlikely that muscular problems in particular should influence response rate extensively. In a questionnaire survey we carried out in the same areas in 1994 (870 respondents in east, 892 in west, mean age around 40 years) approximately 55 % in both areas reported musculo-skeletal pain during the last four weeks [ 13 ]. In another Norwegian survey from 1991, only 15 % reported no muscular pain during the previous year, 58 % had experienced pain the last week, and 15 % reported pain every day during the previous year [ 9 ]. Periodic muscular pain or stiffness in one or more body regions should probably be considered a normal phenomenon among adults. Only when symptoms are strong, they imply disease and demand of health care [ 14 ]. It is thus important that the proportion reporting to be very troubled was significantly higher in east regarding all body regions. This might be due to a higher prevalence in east of specific musculo-skeletal diseases, like rheumatoid arthritis, fibromyalgia, etc. The Oslo Health Study asked about fibromyalgia and osteoporosis: In east 49 persons reported fibromyalgia and nine osteoporosis, compared to 19 and five in west. In previous studies we found no difference in prevalence of rheumatoid arthritis [ 15 ] or osteoarthrosis [ 13 ] between the two areas. The higher level of extensive pain in east corresponds, however, with our earlier results, as higher pain intensity, more widespread pain and higher disability scores were found among residents in east compared to west [ 13 ]. Blank and Diderichsen found social inequities in both frequency and intensity of a variety of common symptoms in a Swedish population [ 16 ]. Their results led to the hypothesis of "double suffering" also promoted by Eachus [ 7 ]: That lower classes both have more illnesses and experience these illnesses with greater intensity. Their lesser resources to cope with the consequences of disease also contribute to the suffering. Our present study lends support to this hypothesis: Physical and mental ill-health are more frequently reported in east, musculo-skeletal disorders are more common, the proportion reporting to be very troubled by pain is higher, and fewer respondents believe that they can continue their daily activities and fewer express satisfaction with health care. The response rate in our material is low (39 % in east, 43.9 % in west). Total response rate in The Oslo Health Study was 46.5 % for 45- year olds, 43.7 % for 40- year olds and 46 % for all age groups. Non-attendance does not occur randomly. Analyses of the impact of self-selection on the Oslo Health Study have shown that the following sub-groups were under-represented among the attendees: unmarried or divorced, males, persons with low education, low income groups, receivers of disability benefit, inner city dwellers and those not born in Norway [ 17 ]. But when response rate is low, it also turns out that some healthy, highly educated and busy people have chosen not to participate [ 18 ]. We may suggest – but can not know for sure – that non-respondents in east belong mainly in the first group and in west mainly in the second. The implication would be that the differences observed between the areas would increase with increasing response rate. We consider it a strength to use geographical area as a marker of socioeconomic position, and not for example individual education or income. Residential areas are distinct and easy to handle for authorities and politicians, and the majority of health care resources are allocated at area level. That inhabitants in affluent areas are healthier than in less attractive areas, is hardly a surprise, but which are the mechanisms behind the differences? There may be a certain amount of selection: The financial disadvantage of disabled people make it more likely that they live in poorer areas. In our material, far more people of non-western origin lived in east compared to west. As being of non-Western origin showed a strong independent correlation with severe muscular pain, this selection contributed significantly to the between area difference observed. A less healthy physical environment, less healthy lifestyle, and the psychological impact of being poorer than other people, are also possible explanatory factors [ 19 ]. Some authors have found that geographical variations in self-reported illness persist even after allowing for socio-structural individual characteristics [ 20 , 21 ]. This was not the case in our study, as area of living did not show any independent correlation with musculo-skeletal pain after adjustment for individual explanatory variables. As our study is cross-sectional, causal interpretations cannot be made, we can only describe associations between socioeconomic measures and the health inequities observed. Several studies have shown that education and income can not explain the difference in self-reported health between socioeconomic contrasting areas [ 20 - 22 ]. Our results support this, and support the theory that psychological factors are important [ 23 ]. According to Wilkinson, socioeconomic inequality influences health through perception of place in the social hierarchy [ 24 ]. Such perceptions produce negative emotions like shame and distrust that are translated inside the body into poorer health via psycho-neuro-endocrine mechanisms [ 25 ]. Conclusion The present study shows that even in Norway today the perception and impact of a health problem (musculo-skeletal pain) is related to a person's socioeconomic situation. Self-reported health status is known to correlate with mortality, and it is a person's perceived health problems which influence the demand for health care. Significantly more persons living in a non-affluent area of Oslo reported extensive pain, compared to persons in an affluent area. Inactivity, poor mental health, and being a non-Western immigrant implied increase risk of severe symptoms. Competing interests The authors declare that they have no competing interests. Authors' contributions The two authors planned and carried out the data collection together. MB carried out the data analyses and drafted the manuscript. Boyh authors approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529299.xml |
539263 | Androgen-induced cerebral venous sinus thrombosis in a young body builder: case report | Background Cerebral venous sinus thrombosis is an infrequent disease with a variety of causes. Pregnancy, puerperium, contraceptive pills and intracranial infections are the most common causes. The patient may present with headache, focal neurological deficits and seizures. The clinical outcome is highly variable and treatment with heparin is advised. Case presentation The patient is a 22 year old male who presented with headache, repeated vomiting and papilledema. He was a bodybuilder doing exercise since 5 years ago, who had used nandrolone decaonoate 25 milligrams intramuscularly during the previous 5 months. Brain MRI and MRV showed superior sagital and transverse sinus thrombosis and extensive investigations did not reveal any known cause. Conclusions We suggested that androgen was the predisposing factor in our patient. Androgens may increase coagulation factors or platelet activity and cause arterial or venous thrombosis. As athletes may hide using androgens it should be considered as a predisposing factor for thrombotic events in such patients. | Background Cerebral venous sinus thrombosis (CVST) is a disease with a wide spectrum of non specific clinical signs and symptoms, including headache, focal neurological deficits, seizures and coma. The clinical outcome is highly variable; patients may recover completely or may develop severe and lasting neurological deficits [ 1 ]. There are many causes for this disease but the most common predisposing factors are pregnancy, puerperium, contraceptive pills, coagulopathies and intracranial infections [ 2 ]. There are few reports of patients with CVST after androgen therapy. We present a young bodybuilder man who developed CVST with abusing androgens for increasing muscle mass. Case presentation In May 2004 a 22 year old male was admitted to our department with chief complaints of headache and vomiting. The patient was well till 10 days prior to admission that developed progressive, intense bitemporal headache exacerbated with bending. The patient also had history of malaise, nausea and several episodes of vomiting from 3 days before admission. The only objective finding on physical examination was bilateral papilledema. The patient was a body builder doing exercise from 5 years ago who had used nandrolone decaonoate 25 mg once or twice a week during the last 5 months. He had injected 20 ampoules in this period. Brain computed tomography without contrast was done for the patient which showed cord sign, emergency MR imaging including T1 – T2, weighted and MRV showed prominent superior sagital and transverse sinus thrombosis. The C.S.F opening pressure was 480 mm/H2o without any other abnormality. Heparin 80 IU/kg started as loading dose then continued 1000 IU/ hr for 10 days. On the 5 th day of treatment headache resolved and warfarin added to heparin. Laboratory tests including antithrombin III activity, protein C, S factor V leiden, Plasma hemocystein and anticardiolipin were all within normal limits. The patient was discharged in a good condition and was maintained on 6 months warfarin protcol. Discussion There are few reports of CVST following androgen therapy [ 3 ], but there is just one reported case of CVST in androgen using young body builder [ 4 ]. The anabolic activity of testosterone and its derivatives is primarily manifested in its myotrophic actions which result in greater muscle mass and strength. This has led to widespread use of androgenic anabolic steroids by athletes at all levels. Nandrolone decaonoate is a synthetic anabolic steroid. In focus on homeostasis system the most important factors under testosterone regulation are fibrinogen, Plasminogen activator inhibitor-1 (PAI – 1) and platelet aggregability. The current data indicate that testosterone lowers fibrinogen and PAI – 1, however these anticoagulatory and profibrinolytic may be opposed by proaggregatory effects on platelets because high dosages of androgens were found to decrease cycloxygenase activity and thereby increase platelet functions [ 5 ]. Proaggregatory effect of testosterone and other synthetic androgens become more reliable theory for CVST, according to recent publication [ 6 ]. Conclusions This case report presents a patient with CVST following exogenous androgen usage with a mechanism which is not completely understood, but it may be related to platelet activation or an increase in coagulation factors. As androgen use may be frequent and hidden in athletes, it may be an underestimated cause of cerebral venous thrombosis in young adults and careful history should be taken in these groups of patients. Lists of abbreviations CVST: cerebral venous sinus thrombosis PAI-1: Plasminogen activator inhibitor-1 MRI: magnetic resonance imaging MRV: magnetic resonance venography Competing interests The author(s) declare that they have no competing interests. Authors' contributions M.A.S: Admitting and treating the case, preparing the article. M.M : Revising the article. A.R.A : searching previous articles, preparing case presentation part. B.M : searching previous articles, preparing background. Figure 1 T1 weighted horizontal nonGd Images show transverse sinus thrombosis. Figure 2 T1 weighted sagital Images with Gd show thrombosis in sagital sinus. Figure 3 2D MR venogram shows sagital sinus thrombosis. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539263.xml |
387265 | Activation of Arp2/3 Complex: Addition of the First Subunit of the New Filament by a WASP Protein Triggers Rapid ATP Hydrolysis on Arp2 | In response to activation by WASP-family proteins, the Arp2/3 complex nucleates new actin filaments from the sides of preexisting filaments. The Arp2/3-activating (VCA) region of WASP-family proteins binds both the Arp2/3 complex and an actin monomer and the Arp2 and Arp3 subunits of the Arp2/3 complex bind ATP. We show that Arp2 hydrolyzes ATP rapidly—with no detectable lag—upon nucleation of a new actin filament. Filamentous actin and VCA together do not stimulate ATP hydrolysis on the Arp2/3 complex, nor do monomeric and filamentous actin in the absence of VCA. Actin monomers bound to the marine macrolide Latrunculin B do not polymerize, but in the presence of phalloidin-stabilized actin filaments and VCA, they stimulate rapid ATP hydrolysis on Arp2. These data suggest that ATP hydrolysis on the Arp2/3 complex is stimulated by interaction with a single actin monomer and that the interaction is coordinated by VCA. We show that capping of filament pointed ends by the Arp2/3 complex (which occurs even in the absence of VCA) also stimulates rapid ATP hydrolysis on Arp2, identifying the actin monomer that stimulates ATP hydrolysis as the first monomer at the pointed end of the daughter filament. We conclude that WASP-family VCA domains activate the Arp2/3 complex by driving its interaction with a single conventional actin monomer to form an Arp2–Arp3–actin nucleus. This actin monomer becomes the first monomer of the new daughter filament. | Introduction The actin cytoskeleton determines the shape, mechanical properties, and motility of most eukaryotic cells. To change shape and to move, cells precisely control the location and timing of actin filament assembly by regulating the number of fast-growing (barbed) filament ends ( Pollard et al. 2000 ). The actin-related protein (Arp) 2/3 complex, a seven-subunit protein complex that contains two actin-related proteins, generates these new barbed ends in response to cellular signals ( Welch et al. 1998 ; Machesky et al. 1999 ; Rohatgi et al. 1999 ). In a process called “dendritic nucleation,” the Arp2/3 complex nucleates new actin filaments from the sides of preexisting filaments to produce a rigid and highly crosslinked filament array ( Mullins et al. 1998 ; Machesky et al. 1999 ; Blanchoin et al. 2000a ). Such crosslinked arrays form the core of many motile cellular structures, including the leading edges of amoeboid cells and the actin comet tails that propel endosomes and bacterial pathogens through eukaryotic cytoplasm. To understand the construction, function, and regulation of these structures, it is important to understand the molecular mechanism of Arp2/3 activation. The Arp2/3 complex must be activated by both a member of the Wiskott–Aldrich syndrome protein (WASP) family and a preexisting actin filament before it will nucleate a new actin filament ( Machesky et al. 1999 ; Blanchoin et al. 2001 ; Zalevsky et al. 2001 ). The structure and the orientation of the Arp2 and Arp3 subunits within the crystal structure of the complex suggest that these subunits may nucleate a new filament by forming an actin-like heterodimer that mimics the barbed end of an actin filament ( Robinson et al. 2001 ). In the crystal structure of the unactivated complex, however, Arp2 and Arp3 are separated by 40 Å so that formation of an actin-like dimer would require a conformational change ( Robinson et al. 2001 ). Binding of the Arp2/3 complex to both a preformed filament and a WASP-family protein is thought to drive at least part of this conformational change ( Blanchoin et al. 2001 ; Marchand et al. 2001 ; Panchal et al. 2003 ). The Arp2/3-activating region of WASP-family proteins, also known as the VCA domain, is composed of three sequences arranged in tandem: (1) an actin-binding verprolin-homology (or V) domain (also known as a WASP-homology 2 [WH2] domain), (2) a conserved “connecting” (or C) region that interacts with both the Arp2/3 complex and monomeric actin ( Marchand et al. 2001 ), and (3) an acidic (or A) region that binds the Arp2/3 complex. This VCA domain is both necessary and sufficient for efficient Arp2/3 activation. We and others have previously suggested that an actin monomer provided by the VCA domain to the Arp2/3 complex may drive the formation of an Arp2–Arp3–actin heterotrimer and form a nucleus for actin polymerization ( Dayel et al. 2001 ; Marchand et al. 2001 ). Both the Arp2 and Arp3 subunits of the complex bind ATP ( Dayel et al. 2001 ). Hydrolysis of this ATP could be used to perform work, to provide a signal, or, like the guanine triphosphate (GTP) bound to the α subunit of tubulin heterodimers, may simply stabilize a protein fold. On conventional actin, ATP hydrolysis is a timing mechanism that promotes construction of dynamic and polarized filament networks. Actin rapidly hydrolyzes ATP upon polymerization ( Blanchoin and Pollard 2002 ) and releases bound phosphate several hundred seconds later ( Melki et al. 1996 ). ATP hydrolysis and phosphate dissociation do not cause immediate filament disassembly, but enable interaction with depolymerizing factors such as cofilin ( Blanchoin and Pollard 1999 ). ATP hydrolysis by actin thereby determines the overall rate of filament turnover. We show here that the Arp2/3 complex rapidly hydrolyzes ATP on the Arp2 subunit upon filament nucleation. There are several events in the Arp2/3 nucleation reaction that might trigger ATP hydrolysis on Arp2: (1) binding of VCA to the Arp2/3 complex, (2) binding of VCA-Arp2/3 to the side of a preformed filament, (3) binding of a VCA-tethered actin monomer to the Arp2/3 complex, or (4) binding of a second or third actin monomer to form a stable daughter filament. We find that ATPase activity requires the combination of a preformed actin filament, a VCA domain, and an actin monomer, but does not require actin polymerization. This indicates that hydrolysis is triggered relatively early in the nucleation reaction—before completion of a stable daughter filament. Capping the pointed ends of actin filaments also stimulates Arp2 to rapidly hydrolyze ATP in the absence of monomeric actin and VCA and without branch formation. Thus, ATP hydrolysis on Arp2 is stimulated directly by interaction with conventional actin, presented to the complex either as a monomer attached to the VC domain of the WASP-family protein or as one of the subunits making up the pointed end of a preformed filament. To our knowledge this is the first direct evidence that the monomer supplied by the VCA domain is the first monomer of the new daughter filament. From these observations we propose a model for the mechanism of Arp2/3 complex activation by WASP-family proteins. Results γ- 32 P-AzidoATP Can Be Covalently Crosslinked to Arp2 and Arp3 with Approximately Equal Efficiency Previously we used sodium dodecyl sulphate polyacrylamide gel electrophoresis (SDS-PAGE) to show that UV irradiation covalently crosslinks α- 32 P-8-AzidoATP to the Arp2 and Arp3 subunits of the Arp2/3 complex ( Dayel et al. 2001 ). Here we crosslink γ- 32 P-AzidoATP instead of α- 32 P to Arp2 to measure ATPase activity. Using SDS-PAGE, we can separate the subunits and simultaneously monitor cleavage of the labeled γ-phosphate from ATP bound to both Arp2 and Arp3. This technique allows us to measure ATP hydrolysis specifically on the Arp2/3 complex in spite of a 100-fold molar excess of actin, which also binds and hydrolyzes ATP. We crosslinked γ- 32 P-AzidoATP to the Arp2/3 complex by brief (9 s) exposure to UV light. In the presence of γ- 32 P-AzidoATP at concentrations above the K D for ATP ( Dayel et al. 2001 ), γ- 32 P-AzidoATP crosslinks to both Arp2 and Arp3 with approximately equal efficiency ( Figure 1 A). Addition of large amounts of monomeric actin to the labeled Arp2/3 distorts the shape of the Arp2 band, but the 32 P signal from Arp2 remains separately quantifiable, and the magnitude is unaffected ( Figure 1 A). The efficiency of crosslinking for both Arp2 and Arp3 is approximately 10% (unpublished data); therefore, only 1% of the Arp2/3 complex has γ- 32 P-AzidoATP crosslinked to both Arp2 and Arp3. For simplicity, we refer to this partially crosslinked Arp2/3 complex as γ- 32 P-AzidoATP-Arp2/3. Reactions using γ- 32 P-AzidoATP-Arp2/3 are performed in the presence of 100 μM ATP, to occupy the noncrosslinked sites and ensure 100% of the Arp2/3 complex is active. Figure 1 Arp2 Hydrolyzes ATP Rapidly upon Filament Nucleation (A) Arp2/3 (2 μM) was covalently crosslinked to γ- 32 P-AzidoATP by exposure to UV light. Both Arp2 and Arp3 crosslink with approximately equal efficiency (lane 1). Addition of 100-fold excess monomeric actin (lane 2) distorts the shape of the Arp2 band, but the Arp2 signal remains separate and quantifiable. (B–E) γ- 32 P-AzidoATP-Arp2/3 (20 nM) was mixed with 2 μM monomeric actin in polymerization buffer. Samples were taken before and at indicated times after the addition of 750 nM VCA, which initiates rapid actin-filament nucleation by the Arp2/3 complex. (B) Subunits were separated by SDS-PAGE and stained with Coomassie. (C) 32 P signal shows remaining uncleaved γ- 32 P on Arp2 and Arp3 subunits. Arp2 rapidly loses γ- 32 P after addition of VCA. (D) Cleaved γ- 32 P was separated from free 32 P-ATP and protein- 32 P-ATP by TLC. (E) Quantitation of (B) to (D): Protein-ATP (closed circle), Cleaved Pi (closed square), Free ATP (closed diamonds), and Arp2-ATP from SDS-PAGE (open circle, normalized separately). (F) Arp2 releases phosphate soon after ATP hydrolysis. Reaction conditions were the same as (B)–(E), but with the addition of 2 mM maltose and 2 U/ml maltose phosphorylase. Timepoints were quenched into formic acid and assayed by TLC. Hydrolyzed 32 P-ATP was quantified from the decrease in protein conjugated 32 P, and released 32 P was quantified from the 32 P-glucose phosphate produced. Arp2 Hydrolyzes ATP Rapidly upon Actin Filament Nucleation We mixed 20 nM γ- 32 P-AzidoATP-Arp2/3 with 2 μM monomeric actin in polymerization buffer and initiated polymerization by adding 750 nM VCA, which activates rapid actin filament nucleation by the Arp2/3 complex (t ½ actin polymerization ≈ 20 s; unpublished observations). (Unless otherwise stated, VCA refers to 6-histidine [6His]-N-WASP-VCA [398-502]. Cleavage of the 6His tag did not affect the kinetics of Arp2/3-mediated actin polymerization [unpublished data].) We assayed timepoints both by SDS-PAGE and thin-layer chromatagraphy (TLC) during the same reaction to monitor remaining and cleaved 32 P, respectively ( Figure 1 B– 1 D; quantified in Figure 1 E). ATP is hydrolyzed by the Arp2/3 complex at the earliest timepoints after the addition of VCA (monitored by 32 P cleavage) and cleavage has ceased by 90 s ( Figure 1 D). SDS-PAGE analysis separates the subunits and shows that the γ- 32 P is cleaved rapidly from Arp2 upon addition of VCA, but not significantly from Arp3 ( Figure 1 C). The kinetics of ATP hydrolysis assayed by SDS-PAGE match the kinetics of phosphate cleavage by TLC ( Figure 1 E). Since the nucleation reaction is autocatalytic, the rate increases over time, and therefore it is not possible to derive an exact ATPase rate constant from our data. However, we can define a conservative lower bound: k hyd > 0.1 s –1 , noting that the true rate constant is probably much higher. Isolated Arp2/3 complex in polymerization buffer shows very slow spontaneous cleavage of γ- 32 P from both Arp2 and Arp3 (<1 × 10 –4 s –1 ) (unpublished data). As a control, 32 P-ATP hydrolysis is only seen when the Azido-ATP is covalently crosslinked to the Arp2/3 complex ( Figure 3 D, compare open and closed circles) indicating that the signal is due only to hydrolysis of ATP covalently bound to the Arp2/3 complex and not due to ATP hydrolysis by polymerizing actin. This is further confirmed by observations of ATP hydrolysis on the Arp2/3 complex under conditions where no actin polymerization takes place ( Figure 3 E and 3 F; Figure 4 ). Figure 3 A Single Actin Monomer, in the Presence of Actin Filaments and VCA, Stimulates ATP Hydrolysis on Arp2, without Requiring Actin Polymerization (A–C) Remaining unhydrolyzed γ- 32 P-AzidoATP on Arp2 (closed circle) and Arp3 (open circle) was quantified to assay ATP hydrolysis (same conditions as Figure 1 B– 1 D). γ- 32 P-AzidoATP-labeled Arp2/3 (20 nM) was mixed at indicated times with either 750 nM VCA then 2 μM G-actin (A), 2 μM G-actin then 750 nM VCA (B), or 2 μM F-actin then 750 nM VCA (C). (D) Latrunculin B (open square) inhibits the ability of VCA plus monomeric actin (open circle) to stimulate ATP hydrolysis on the Arp2/3 complex in the absence of actin filaments. Also, 32 P ATP hydrolysis signal requires covalent crosslinking to Arp2/3. Arp2/3 was mixed with 6 μM γ- 32 P-AzidoATP and exposed to UV either before (closed circle) or after (open circle) the addition of excess (2 mM) unlabeled ATP. Excess ATP added before the UV exposure prevents crosslinking and abolishes the ATP hydrolysis signal, indicating that all the 32 P ATP hydrolysis signals measured are due to ATP hydrolysis on Arp2/3 and not from ATP hydrolysis on actin. (E and F) In the presence of phalloidin-stabilized actin filaments, actin monomers are prevented from polymerizing by Latrunculin B, but still stimulate ATP hydrolysis on the Arp2/3 complex. 20 nM γ- 32 P-AzidoATP–labeled Arp2/3 was premixed with 1 μM phalloidin-stabilized actin filaments. The reaction was initiated by mixing with 750 nM N-WASP VCA, 1 μM G-actin and 4 μM Latrunculin B as indicated, cleaved γ- 32 P was assayed by phosphomolybdate extraction (E), and separately, actin polymerization was monitored by pyrene–actin fluorescence (F). Figure 4 Pointed-End Filament Capping Is Sufficient to Stimulate ATP Hydrolysis on Arp2 in the Absence of VCA (A) The Arp2/3 complex prevents actin filament reannealing by capping the pointed ends. The length distribution of 2 μM Alexa-488 phalloidin-stabilized actin filaments is unaffected in the absence (i) or presence (iii) of 20 nM Arp2/3 complex. (ii) 5 min after shearing the filaments, filaments have begun to reanneal in the absence of the Arp2/3 complex, but 20 nM Arp2/3 complex (iv) maintains short filaments, preventing reannealing by capping filament pointed ends. (B) ATP hydrolysis on Arp2 is stimulated by pointed-end capping. Crosslinked γ- 32 P-AzidoATP-Arp2/3 (20 nM) was mixed with 2 μM phalloidin-stabilized actin filaments. The mixture was split in two and one sample was sheared. Timepoints were taken as shown. (C) Uncleaved 32 P on Arp2 (unsheared [closed circle] and sheared [closed square]) and Arp3 (unsheared [open circle] and sheared [open square]) were quantified from (B). Arp2 rapidly hydrolyzes bound ATP upon filament pointed-end capping. Phosphate Release by Arp2 Lags Hydrolysis by Approximately 40 s To investigate the kinetics of phosphate release fromArp2/3 during the polymerization reaction, we added maltose and maltose phosphorylase to the reaction. In the presence of 32 P-labeled Arp2/3 complex, maltose phosphorylase conjugates the 32 P-orthophosphate released from Arp2 to a hydrolyzed maltose molecule to make 32 P-glucose phosphate. The phosphate from adenosine diphosphate-inorganic phosphate (ADP-Pi)-bound Arp2 is inaccessible to the enzyme and remains unconjugated orthophosphate. We quantified hydrolyzed 32 P-ATP and released phosphate by TLC ( Figure 1 F). Phosphate release from Arp2 lags behind ATP hydrolysis by approximately 40 s. The Rate of Filament Nucleation Matches the Rate of ATP Hydrolysis by Arp2 To determine whether ATP hydrolysis on Arp2 is coupled to filament nucleation, we varied the rate of nucleation and looked to see whether the rate of ATP hydrolysis by Arp2 varied accordingly. We varied the nucleation rate by using N-WASP and Scar1 VCA domains, which stimulate different rates of Arp2/3 complex-dependent actin nucleation ( Zalevsky et al. 2001 ). To slow the nucleation reaction and allow more accurate kinetic measurements, we used only 1 μM monomeric actin. We used pyrene–actin polymerization data ( Figure 2 A) to calculate the concentration of barbed ends produced over time ( Figure 2 B, open symbols) (see Methods and Materials ; Zalevsky et al. 2001 ). Note that this calculation is model-independent and simply uses the established kinetic parameters for actin polymerization and the change in the amount of monomeric and filamentous actin over time measured from the pyrene–actin curves. The same reagents were used to monitor ATP hydrolysis by Arp2 under the same conditions. We used loss of γ- 32 P labeling as a probe for ATP hydrolysis and scaled the initial labeling intensity to the Arp2/3 concentration used in the reaction (20 nM) to calibrate the stoichiometry of ATP hydrolyzed by Arp2 ( Figure 2 B). Using Scar1 VCA instead of N-WASP VCA halves both the rate of nucleation of actin filaments and the rate of ATP hydrolysis on Arp2. Figure 2 ATP Hydrolysis by Arp2 Coincides with Nucleation of New Actin Filaments, and Not Filament Debranching (A, B) The kinetics of nucleation were slowed by using only 1 μM monomeric actin (compared to 2 μM for Figure 1 ). γ- 32 P-AzidoATP-Arp2/3 (20 nM) was mixed with either 750 nM N-WASP WWA (closed circle) or Scar1 WA (closed square) and 1 μM 7% pyrene-labeled monomeric actin. (A) Actin polymerization measured by pyrene fluorescence. (B) The concentration of new filament ends (open symbols) was calculated from the polymerization data in a model-independent way (see Methods and Materials ), and Arp2-ATP hydrolysis (closed symbols) was measured under the same reaction conditions for both N-WASP WWA (open and closed circles) and Scar1 WA (open and closed squares). (C) ATP hydrolysis on Arp2 does not accompany filament debranching. Using a large excess (100 nM) of γ- 32 P-AzidoATP-Arp2/3 creates a slow hydrolysis phase that follows the rapid nucleation phase. The slow phase of ATP hydrolysis can be inhibited by excess (1.5 μM) uncrosslinked Arp2/3 added at t = 200 s, showing that the slow phase of ATP hydrolysis is from Arp2/3 being recruited from solution and not from that already incorporated in branches. We note that the total amount of Arp2 that hydrolyzes ATP in the polymerization reaction is 30% less for Scar1 VCA than for N-WASP VCA, which we interpret as 30% fewer filaments produced. Although it is possible to calculate the rate of end production from the pyrene–actin polymerization curve in a model-independent way, it is not possible to calculate the total number of barbed ends produced, since once polymerization reaches equilibrium, the pyrene–actin curve will not change even if new barbed ends continue to be produced. From the ATP hydrolysis data, therefore, the Arp2/3 complex produces filament ends more slowly when activated by Scar1, and under our conditions, the reaction ends when monomeric actin is depleted by incorporation into the new filaments. Fewer total filaments are therefore produced by the less active VCA domain. ATP Hydrolysis on Arp2 Does Not Accompany Filament Debranching A previous study claimed that ATP hydrolysis on Arp2 occurs very slowly (t 1/2 ≈ 800 s), coincident with filament debranching ( Le Clainche et al. 2003 ). Le Clainche et al. (2003 ) used a much higher concentration of Arp2/3 complex (100 nM) in their assays than the 5 nM Arp2/3 complex that they estimate was used up during their polymerization reaction. Using these conditions, we find that Arp2/3 complex hydrolyses ATP in two discrete phases: a fast (nucleation) phase, followed by a slow, approximately linear phase ( Figure 2 C, open symbols). This slow phase does not plateau within 6000 s and is similar to the data presented in Le Clainche et al. (2003 ). To demonstrate that this slow ATP hydrolysis is not due to the Arp2/3 complex hydrolyzing ATP upon debranching, we added an excess of unlabeled Arp2/3 complex into solution at t = 200 s, after the polymerization phase is complete. This unlabeled Arp2/3 complex competes for nucleating factors with γ- 32 P-AzidoATP-Arp2/3 in solution, but it does not compete with γ- 32 P-AzidoATP-Arp2/3 already incorporated in branches. Addition of excess unlabeled Arp2/3 complex at t = 200 s inhibits the slow phase of ATP hydrolysis ( Figure 2 C, closed symbols), indicating that the slow phase is due to ATP hydrolysis on Arp2/3 complex being recruited from solution and not due to ATP hydrolysis on Arp2/3 complex already in branches. This slow ATP hydrolysis probably represents a low rate of filament nucleation by the excess unused Arp2/3 complex, the rate of nucleation being limited by the low monomeric actin concentration that remains after most of the actin has polymerized. Both VCA and Monomeric Actin Are Required to Stimulate ATP Hydrolysis by Arp2 during the Polymerization Reaction Although the kinetics of ATP hydrolysis on Arp2 match the kinetics of actin polymerization, these data do not rule out the possibilities that VCA alone or the filamentous actin created during the polymerization reaction stimulates the ATPase activity independent of nucleation. To more specifically determine what stimulates ATP hydrolysis on Arp2, we varied the order of addition of the components that initiate the polymerization reaction. Incubation of the Arp2/3 complex with VCA does not induce ATP hydrolysis by the complex until monomeric actin is added to the reaction ( Figure 3 A), showing that VCA alone does not stimulate the ATPase activity. Similarly, monomeric actin alone does not stimulate the Arp2/3 complex to hydrolyze ATP until the addition of VCA ( Figure 3 B). To test whether actin filaments themselves stimulate Arp2/3 ATP hydrolysis, we used phalloidin-stabilized actin filaments to ensure that no monomeric actin would be present and took care not to shear the filaments in order to reduce the number of free pointed ends. ATP hydrolysis is not stimulated on the Arp2/3 complex by filamentous actin, even in presence of VCA ( Figure 3 C). As controls, we found that neither 5 μM phalloidin nor 20 mM phosphate inhibit the kinetics of ATP hydrolysis by Arp2 during the polymerization reaction (unpublished data). When Arp2/3 concentration is low (20 nM), and nucleation is rapid (using N-WASP VCA), initiation of the polymerization reaction causes striking and near-complete ATP hydrolysis on Arp2 (approximately 80%, i.e., approximately 16 nM; Figure 3 B and 3 C). We detect a small amount of ATP hydrolysis on Arp3 with similar kinetics but much lower stoichiometry (10%–20%). The decrease is not caused by the dilution effect of adding the second component (approximately 4%), which is already compensated for in the data presented. In the Presence of Both VCA and Actin Filaments, a Nonpolymerizable Actin Monomer Is Sufficient to Trigger Rapid ATP Hydrolysis on Arp2 The timing and stoichiometry of ATP hydrolysis and the combination of factors required to stimulate it suggest that Arp2 hydrolyzes ATP during the filament nucleation reaction. Kinetic and light-microscopy data indicate that most or all Arp2/3-dependent filament nucleation occurs from Arp2/3 complex bound to the sides of filaments produced earlier in the polymerization reaction ( Blanchoin et al. 2000a , 2001 ; Zalevsky et al. 2001 ). To test whether filament side-binding is necessary for ATP hydrolysis on Arp2, we blocked filament formation with the actin-monomer binding toxin, Latrunculin B. Latrunculin B binds to monomeric actin and prevents it polymerizing, but does not affect its binding to VCA (R. D. Mullins and A. E. Kelly, unpublished data). The combination of VCA and Latrunculin B–actin monomers does not stimulate ATP hydrolysis on Arp2/3 complex ( Figure 3 D, open squares), nor do preformed, phalloidin-stabilized actin filaments and Latrunculin B–actin monomers without VCA ( Figure 3 E, filled squares). In the presence of preformed actin filaments and VCA, however, Latrunculin B–actin monomers stimulate rapid ATP hydrolysis on Arp2/3 ( Figure 3 E, filled circles) without actin polymerization ( Figure 3 F). Table 1 summarizes the requirements for stimulation of ATP hydrolysis on Arp2. These data indicate that during the nucleation reaction, actin filament side-binding by Arp2/3 complex is a prerequisite for VCA and monomeric actin to stimulate ATP hydrolysis on Arp2. The observation that polymerization of the daughter filament is unnecessary implies that the VCA-mediated interaction of a single actin monomer with the Arp2/3 complex is the trigger for ATP hydrolysis on Arp2. Table 1 Requirements to Stimulate ATP Hydrolysis on the Arp2 Subunit of Arp2/3 Complex Abbreviations: G-actin, monomeric actin; F-actin, actin filaments; LatB, Latrunculin B Pointed-End Capping by the Arp2/3 Complex Stimulates Rapid ATP Hydrolysis by Arp2 in the Absence of Either Branch Formation or a WASP-Family VCA Domain The Arp2/3 complex is known to cap the pointed ends of preformed actin filaments in vitro, inhibiting both polymerization and depolymerization from the pointed ends of gelsolin-capped filaments ( Mullins et al. 1998 ). The Arp2/3 complex does not cap the barbed ends of actin filaments and does not affect the rate of addition of monomers from the barbed ends of spectrin-capped filaments (unpublished data). We speculated that the way the Arp2/3 complex caps a free-filament pointed end in solution might mimic the way the Arp2/3 complex anchors the pointed end of the new daughter filament in a branch. If the actin monomer that triggers ATP hydrolysis during nucleation is the first monomer of the daughter filament, pointed-end capping, like nucleation, should drive interaction with this monomer and trigger ATP hydrolysis on Arp2. To test this, we sheared preformed, phalloidin-stabilized actin filaments in the presence of the Arp2/3 complex. Mechanical shearing fragments long actin filaments into many short filaments, creating many new filament ends that rapidly reanneal to produce long filaments again ( Murphy et al. 1988 ). This reannealing process is blocked by proteins that cap filament ends ( Andrianantoandro et al. 2001 ). Without shearing, the addition of 20 nM Arp2/3 complex does not alter the length distribution of phalloidin-stabilized actin filaments ( Figure 4 A, compare [i] and [iii]). After shearing in the presence of 20 nM Arp2/3 complex, pointed-end capping by the Arp2/3 complex blocks reannealing and results in significantly shorter filaments ( Figure 4 A, compare [ii] and [iv]). No branches form within this time—it takes several hours for even a few branches to assemble under these conditions (unpublished data). To assay for ATP hydrolysis by the complex, we incubated γ- 32 P-AzidoATP-Arp2/3 complex with actin filaments under the same conditions as the microscopy experiment. We split the mixture into two parts, sheared one half, and took timepoints to assay for ATP hydrolysis from both samples ( Figure 4 B; quantified in Figure 4 C). No ATP hydrolysis occurs in the unsheared condition, confirming that binding to the sides of actin filaments is not sufficient to stimulate ATP hydrolysis. ATP hydrolysis occurs rapidly in the sheared condition and occurs only on Arp2 ( Figure 4 C). Since this occurs well before any branches form, pointed-end capping by the Arp2/3 complex is sufficient to stimulate ATP hydrolysis on Arp2 not only in the absence VCA, but also in the absence of filament side-binding. Discussion Conventional actin and all actin-related proteins share a conserved nucleotide binding pocket. Actin monomers bind ATP but do not hydrolyze it until they are induced to polymerize. Actin polymerization triggers rapid ATP hydrolysis, followed by a slow release of cleaved phosphate from the filament ( Blanchoin and Pollard 2002 ). Arp2 also hydrolyzes its bound ATP, and we find that the conditions that promote ATP hydrolysis and the kinetics of the reaction are remarkably similar to those of conventional actin. In the presence of VCA and actin filaments, monomeric actin stimulates ATP hydrolysis on Arp2 ( Table 1 ). We also find that binding of the Arp2/3 complex to the pointed end of a preformed actin filament is sufficient to trigger Arp2 ATP hydrolysis, even in the absence of VCA. The stimulation of Arp2 ATPase activity by both filament pointed ends and by actin monomers under nucleating conditions suggests that the geometry of the Arp2/3–actin interaction is the same in both cases. Interaction between the Arp2/3 complex and conventional actin can occur in three distinct ways: (1) the Arp2/3 complex binds the sides of preformed actin filaments; (2) the Arp2/3 complex binds to the pointed ends of filaments, either by remaining associated with the daughter filament following nucleation or by capping preformed pointed ends; and (3) the Arp2/3 complex may interact with an actin monomer bound to the VCA domain of a WASP-family protein. There is abundant experimental evidence for filament side- and pointed-end binding by the complex ( Mullins et al. 1998 ; Blanchoin et al. 2000a , 2001 ; Amann and Pollard 2001a , 2001b ). Evidence that a VCA-bound actin monomer interacts with the Arp2/3 complex is more circumstantial and is supported by four observations: (1) VCA domains can simultaneously bind both the Arp2/3 complex and monomeric actin ( Marchand et al. 2001 ; Panchal et al. 2003 ); (2) removal of the actin monomer-binding WH2 (V) domain from a WASP-family protein severely decreases the efficiency of Arp2/3 activation ( Marchand et al. 2001 ); (3) kinetic modeling suggests that the Arp2/3 complex requires monomeric actin to form a filament nucleus ( Zalevsky et al. 2001 ); and (4) Arp2/3-dependent nucleation is not limited to the end of the mother filament ( Amann and Pollard 2001a ), indicating that the VCA-bound actin monomer does not incorporate into the mother filament. Two of the three interactions between the Arp2/3 complex and conventional actin—nucleation and pointed-end capping—are thought to be mediated by the actin-related subunits, analogous to actin–actin interactions in a filament. Both interactions stimulate rapid ATP hydrolysis by Arp2. Based on sequence conservation and biochemical similarities, ATP hydrolysis on Arp2 is probably driven by a mechanism similar to that which stimulates ATP hydrolysis on actin. The molecular details of how polymerization activates ATP hydrolysis on conventional actin, however, are not well understood. A leading hypothesis is that a “hydrophobic plug”—a loop between subdomains 3 and 4 of actin (residues 262–274 in yeast; Kuang and Rubenstein 1997 )—undocks from the monomer surface and binds to a hydrophobic cleft formed by adjacent monomers in the opposite strand of the two-start filament helix ( Lorenz et al. 1993 ; Kuang and Rubenstein 1997 ). Our data are consistent with stimulation of ATP hydrolysis by docking of a hydrophobic plug sequence on Arp2 into a hydrophobic cleft created by Arp3 and the first actin monomer of the daughter filament ( Figure 5 ). In the crystal structure of the inactive Arp2/3 complex, Arp2 and Arp3 are oriented like a pair of actin monomers in opposite strands of the two-start filament helix ( Robinson et al. 2001 ), but they are separated by a 40 Å cleft. Our data support a model in which activation of the complex involves closure of the cleft, allowing actin to polymerize from an Arp2–Arp3 heterodimer ( Kelleher et al. 1995 ; Robinson et al. 2001 ), which then remains attached to the pointed end of the new daughter filament, anchoring it to the branch ( Figure 5 B [iv]). Based on the geometry of the subunits in the crystal structure and the hydrophobic plug model, we expect that the Arp3–actin contact creates a pocket to bind the hydrophobic plug of Arp2 (residues 265–277 in yeast Arp2). The geometry of the interaction would stimulate the ATPase activity of Arp2, but not Arp3 ( Figure 5 A). Figure 5 Model for Activation of ATP Hydrolysis on the Arp2/3 Complex and Mechanism by which WASP-Family Proteins Activate the Arp2/3 Complex to Nucleate New Actin Filaments (A) Filament pointed-end capping stimulates ATP hydrolysis on Arp2 without branch formation. (i) Arp2 and Arp3 are separated when the Arp2/3 complex is free in solution. (ii) Upon pointed-end capping, the binding energy of the actin-Arp2/3 interface drives Arp2 and Arp3 together and (iii) a conformational change on Arp2 (shown by the red the subdomain 3/4 loop flipping out) triggers ATP hydrolysis by Arp2 (filament pointed-end capping is probably not a significant function of the Arp2/3 complex in vivo). (b) A VCA-bound actin monomer drives the activation of the Arp2/3 complex and stimulates ATP hydrolysis on Arp2. (i) The Arp2/3 complex must first be bound to the side of an actin filament, and an actin monomer is bound to the VC domain of the WASP-family protein. (ii) The VC domain of the WASP-family protein docks the first monomer of the daughter filament onto the Arp2/3 complex, stabilizing the Arp2–Arp3–actin interaction and promoting the active conformation of the complex. (cf. Aii). (iii) The active conformation of the Arp2–Arp3–actin monomer triggers a conformational change on Arp2 and ATP hydrolysis by the subunit. (iv) Actin polymerizes from the activated Arp2/3 complex. ATP hydrolysis by Arp2 may promote dissociation of the CA domain of the WASP-family protein from the Arp2/3 complex, aided by actin polymerization, which competes its WH2 domain from the first actin monomer. Monomeric actin does not interact directly with the Arp2/3 complex in the absence of VCA, but under conditions that promote nucleation, a single actin monomer triggers VCA-dependent ATP hydrolysis on Arp2. By analogy with capping-induced ATP hydrolysis, the monomer that triggers ATPase activity is therefore the first monomer of the new daughter filament ( Figure 5 B [i]–[iii]). The hydrophobic pocket formed between Arp2, Arp3, and the actin monomer would therefore promote a similar conformational change in Arp2 and stimulate ATP hydrolysis ( Figure 5 B [iv]). Interaction of the Arp2/3 complex with the sides of filaments is not sufficient to trigger Arp2 ATPase activity, even in the presence of VCA. Binding of Arp2/3 to the sides of filaments is, however, required for ATP hydrolysis on Arp2 stimulated by VCA and monomeric actin. These data suggest that binding the side of an actin filament induces a conformational change in the Arp2/3 complex that enables it to interact with the actin monomer bound to VCA. The filament side-binding activity of Arp2/3 does not require the presence of the Arp2 or Arp3 subunits and can be reconstituted by a combination of the Arc2 (p34) and Arc4 (p20) subunits ( Gournier et al. 2001 ). The Arc2 and Arc4 subunits contact both Arp2 and Arp3, and therefore filament side-binding might favor association of Arp2 and Arp3. The fact that Arp2-ATP hydrolysis induced by VCA and an actin monomer requires filament side-binding strongly suggests that all Arp2/3-generated actin filaments are born on the side of preformed filaments. Our results disagree with a recent paper that claims that ATP hydrolysis on Arp2 is slow and accompanies filament debranching ( Le Clainche et al. 2003 ). Using experimental conditions similar to the previous study, we observe similar slow ATP hydrolysis kinetics ( Figure 2 C) and show that this ATP hydrolysis occurs on Arp2/3 complex recruited slowly from solution. The slow hydrolysis does not reflect delayed ATP hydrolysis on Arp2/3 complex that had been rapidly incorporated into branches early in the experiment. ATP hydrolysis on Arp2, therefore, cannot be associated with debranching. Le Clainche et al. (2003 ) claim that ATP hydrolysis does not occur during nucleation and present data with a lag of several hundred seconds between computer-simulated nucleation kinetics and measured ATP hydrolysis kinetics ( Figure 1 B in Le Clainche et al. 2003 ). In this experiment, Le Clainche et al. (2003 ) initiate polymerization in the absence of free ATP. These conditions would deactivate up to 97% of the Arp2/3 complex (the fraction that is not crosslinked to ATP on both subunits). In our experience, removal of free ATP introduces an artificial lag in polymerization that lasts until tightly bound ATP is released from monomeric actin (1/k ATP release = 330 s; Selden et al. 1999 ) and is free to interact with the Arp2/3 complex (unpublished data). The claim by Le Clainche et al. (2003 ) that the absence of free ATP does not affect ATP hydrolysis kinetics is contradicted by their observation that the 32 P signal is unchanged by the addition of free ATP. The 32 P signal is only equivalent to hydrolyzed ATP in the absence of free ATP. The addition of free ATP should cause the excess of uncrosslinked Arp2/3 complex to compete with the small fraction of crosslinked 32 P-ATP-Arp2/3 complex and thereby significantly reduce the 32 P signal. The observation that the 32 P signal is not reduced, rather than confirming that removal of free ATP has no effect, instead confirms that contaminating ATP is present for the latter part of the “ATP-free” condition, presumably released slowly from monomeric actin. The lag in the polymerization created by the initial absence of ATP would be present in the experimental ATP hydrolysis measurement, but may not have been present in the nucleation data presented because this was generated by a model-dependent computer simulation ( Le Clainche et al. 2003 ). We find that ATP hydrolysis and phosphate release from Arp2 (approximately 40 s) are more than an order of magnitude faster than debranching of Arp2/3-generated dendritic networks (approximately 1000 s) ( Blanchoin et al. 2000b ). The kinetics of phosphate release from Arp2 are also about an order of magnitude faster than phosphate release from actin (1/k Pi release = 384 s for skeletal muscle actin; Melki et al. 1996 ), suggesting that, if phosphate release controls debranching, it is the phosphate release from the daughter actin filament that is important, not the phosphate release from Arp2. This is supported by the observation that phalloidin, which slows phosphate release from actin, slows filament debranching, and cofilin, which accelerates phosphate release from actin, accelerates filament debranching ( Blanchoin et al. 2000b ). Le Clainche et al. (2003 ) show that chromium-ATP Arp2/3 debranches more slowly than magnesium-ATP Arp2/3 and claim (but do not demonstrate) that chromium-ATP Arp2/3 releases phosphate more slowly. If chromium does slow the phosphate release from Arp2/3, in light of our data, this suggests that phosphate release from Arp2 may be a prerequisite for filament debranching—but is not a direct cause, since it occurs much too rapidly. We previously showed that the Arp2/3 complex requires hydrolyzable ATP for nucleation activity ( Dayel et al. 2001 ), and the current study adds weight to the hypothesis that ATP hydrolysis has a direct role in nucleation by showing that ATP is hydrolyzed by Arp2 upon nucleation. The separation of the Arps in the crystal structure and the very low nucleation rate of the unactivated complex probably reflect the tendency of Arp2 and Arp3 to remain separated in the absence of all the required nucleation promoting factors. This suggests that there is a large free energy barrier to the formation of an Arp2–Arp3 heterodimer. Our data indicate that there are two ways to overcome this energy barrier, both using the binding energy of actin: one using the combined binding energy of the two actin monomers at the pointed end of an actin filament during pointed-end capping, and the other the combined binding energy of the side of the mother filament, the VCA domain, and a single actin monomer. The surface area of the filament pointed end that would be buried by interaction with an Arp2–Arp3 dimer would be large (approximately 6800 Å 2 ). This is consistent with the fact that in vitro the binding energy of this interface is sufficient to drive the interaction and promote the active conformation of the complex directly, even in the absence of VCA or a mother filament ( Mullins et al. 1998 ). The binding of monomeric actin alone is insufficient to overcome the free-energy barrier, which ensures that the inactive conformation of the Arp2/3 complex is robust despite high cellular concentrations of actin. Because of the free energy of all the binding partners involved in nucleation, however, the energy of ATP hydrolysis may not be needed to stabilize the nucleus. Regardless, it is very likely that ATP hydrolysis on Arp2, like actin, provides a timing signal to the system. ATP hydrolysis on Arp2/3 would promote release of VCA from the complex and allow a new actin branch to move away from the site of its creation ( Dayel et al. 2001 ). ATP hydrolysis may also regulate the timing of the interaction of the Arp2/3 complex with other binding partners such as cortactin and cofilin. Temporal regulation of these interactions is likely to be essential to construction of functional motile structures. The Arp2/3 ATP hydrolysis assay presented here provides a novel assay for activation of the Arp2/3 complex that does not rely, as all previous assays have done, solely on actin polymerization. Pyrene–actin polymerization is only useful over a limited range of actin concentrations because at high concentrations, spontaneous assembly obscures Arp2/3-mediated nucleation. The pyrene–actin assay also has temporal limits since it rapidly uses up one of the factors necessary for Arp2/3 activation–monomeric actin. Our observation that ATP is hydrolyzed by Arp2 rapidly during, or soon after, the nucleation reaction means that we can use ATP hydrolysis on Arp2 as an assay to study the factors required to promote activation of the Arp2/3 complex. The fact that nonpolymerizable actin monomers are competent to stimulate hydrolysis enables us to investigate the conditions for Arp2/3 complex activation under a wider range of conditions. This system will be useful for further studies of the biophysics of Arp2/3-mediated actin assembly. Materials and Methods Purification of proteins We purified Arp2/3 from Acanthamoeba castellini by a combination of conventional and affinity chromatography ( Dayel et al. 2001 ). We flash-froze Arp2/3 complex in aliquots of approximately 40 μM in 10% glycerol, 0.5 μM TCEP, and 2 mM Tris (pH 8.0), and stored them at –80°C for later use. We purified actin from Acanthamoeba by the method of MacLean-Fletcher and Pollard (1980 ). Actin was stored in fresh G-buffer (0.5 μM TCEP, 0.1 μM CaCl 2 , 0.2 μM ATP, 2 mM Tris [pH 8.0]) and gel-filtered before use. Rat N-WASP VCA (398–502) and Human Scar1-VCA (489–559) with N-terminal 6His tags and TEV cleavage sites were bacterially expressed and purified by nickel affinity chromatography. We prepared phalloidin-stabilized actin filaments by adding 1/10 volume of 10× KMEI to monomeric actin at room temperature for 20 min to initiate polymerization, then added twice the concentration of phalloidin and incubated for a further hour at room temperature (1× KMEI buffer: 50 mM KCl, 1 mM MgCl 2 , 1 mM EGTA, 10 mM Imidazole [pH 7.0]). We took care not to unintentionally shear the phalloidin-stabilized actin filaments by using wide-bore pipette tips. Arp2/3 ATPase assay We diluted freshly thawed aliquots of Arp2/3 to 2.0 μM in 1 mM MgCl 2 , 50 mM KCl, 10 mM Imidazole (pH 7.0) and added 6 μM γ- 32 P-labeled 8-AzidoATP (Affinity Labeling Technologies, Lexington, Kentucky, United States). After a 2-min incubation to allow nucleotide exchange, we crosslinked for 9 s using a UV hand lamp (312 nm; Fisher Scientific, Hampton, New Hampshire, United States), added 1 mM ATP and 1 mM DTT to quench the reaction and buffer exchanged into 1× KMEI plus 100 μM ATP, 1 mM DTT using a NAP5 column (Amersham Pharmacia Biotech, Little Chalfont, United Kingdom). We used the Arp2/3 for assays within 10 min of crosslinking. The same actin (including 7% pyrene–actin) was used for both ATP hydrolysis assays and correlative pyrene–fluorescence polymerization assays. We took ATPase time points by mixing 400 μl of the reaction mixture with premixed 400 μl of methanol and 100 μl of chloroform. We ran the precipitated protein on SDS-PAGE gel to separate the subunits and quantified 32 P-labeling using a phosphoimager (Storm 840; Molecular Dynamics, Sunnyvale, California, United States). For phosphate cleavage assays, we quenched timepoints into 1/10 volume 26 M formic acid, spotted on cellulose TLC plates, and separated the components in 0.4 M KH 2 PO 4 (pH 3.4). We separately ran 32 P-ATP and 32 P-ATP treated with apyrase as standards to confirm the separation of 32 P-ATP and cleaved 32 P, respectively (unpublished data). As an alternative method of quantifying cleaved 32 P, phosphomolybdate was extracted as in Shacter (1984 ) and quantified using a scintillation counter. To distinguish the ADP-Pi state of Arp2 from the ADP state, the kinetics of phosphate release were measured by performing the reaction in the presence of 2 mM maltose and 2 U/ml maltose phosphorylase (Sigma-Aldrich, St. Louis, Missouri, United States), which uses only the released Pi to form glucose phosphate. Glucose phosphate was separated from free ATP, protein-ATP, and Pi using TLC. Actin polymerization assays We doped Acanthamoeba actin with 7% pyrene–actin to monitor actin polymerization by fluorescence (λ ex = 365 nm, λ em = 407 nm, 25°C) ( Mullins and Machesky 2000 ). We calculated the number of ends produced over time from [ENDS] = (d[F-actin]/dt)/([free G-actin]*10 μM s –1 ) (cf. Zalevsky et al. 2001 ). Polymerization reactions were performed in G-buffer plus 1/10 volume 10× KMEI. The Ca 2+ cation on monomeric actin was preexchanged with Mg 2+ 30 s before use. Microscopy We prepared filamentous actin as above and stabilized filaments with stoichiometric Alexa-488 phalloidin (Molecular Probes, Eugene, Oregon, United States). We mixed 2 μM Alexa-488 phalloidin–F-actin with 20 nM Arp2/3, passed twice through a 30-gauge needle to shear the filaments, and incubated at room temperature. Timepoints were taken by diluting 500-fold and rapidly applying to poly-L-lysine–coated coverslips for visualization. Filament images were quantified for length distribution and branch frequency by a custom MATLAB (MathWorks Inc., Natick, Massachusetts, United States) routine. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC387265.xml |
548129 | Non-linear mapping for exploratory data analysis in functional genomics | Background Several supervised and unsupervised learning tools are available to classify functional genomics data. However, relatively less attention has been given to exploratory, visualisation-driven approaches. Such approaches should satisfy the following factors: Support for intuitive cluster visualisation, user-friendly and robust application, computational efficiency and generation of biologically meaningful outcomes. This research assesses a relaxation method for non-linear mapping that addresses these concerns. Its applications to gene expression and protein-protein interaction data analyses are investigated Results Publicly available expression data originating from leukaemia, round blue-cell tumours and Parkinson disease studies were analysed. The method distinguished relevant clusters and critical analysis areas. The system does not require assumptions about the inherent class structure of the data, its mapping process is controlled by only one parameter and the resulting transformations offer intuitive, meaningful visual displays. Comparisons with traditional mapping models are presented. As a way of promoting potential, alternative applications of the methodology presented, an example of exploratory data analysis of interactome networks is illustrated. Data from the C. elegans interactome were analysed. Results suggest that this method might represent an effective solution for detecting key network hubs and for clustering biologically meaningful groups of proteins. Conclusion A relaxation method for non-linear mapping provided the basis for visualisation-driven analyses using different types of data. This study indicates that such a system may represent a user-friendly and robust approach to exploratory data analysis. It may allow users to gain better insights into the underlying data structure, detect potential outliers and assess assumptions about the cluster composition of the data. | Background Systems biology is a data- and knowledge-driven discipline, which heavily relies on automated tools to support the generation and validation of hypotheses. Such tasks aim to provide novel and meaningful views of the functional relationships between biological components at different complexity levels. Over the past seven years hundreds of methods have been reported to analyse these data, with an emphasis on gene expression data classification [ 1 , 2 ]. More recently, the analysis of gene regulatory and protein-protein networks has started to attract contributions from computer and physical sciences [ 3 - 5 ]. All of these tasks are linked by a need for comparing, classifying and visualising information. The ever-increasing number and sophistication of techniques may represent an obstacle to achieve more meaningful and rigorous data analysis and discovery tasks. One important problem is that users may not have the time and knowledge required to adequately understand the dynamics and operation of several tools. These deficiencies have been reflected, for example, in a lack of sound practices for assessing the statistical significance of results and for selecting the most suitable data sets and classification models [ 6 , 7 ]. On the other hand, the emergence of multiple data sets and prediction models represents an opportunity for developing an integrative data mining paradigm, which is already significantly improving several predictive tasks in systems biology [ 5 , 8 , 9 ]. The problem domains mentioned above have mainly concentrated on the application of statistical and machine learning models for classification tasks. Emphasis has been placed on the development of supervised and unsupervised classification methods [ 2 , 10 , 11 ], as well as on the application of statistical tools for assessing the quality of classification results [ 12 , 13 ]. Relatively fewer efforts have been reported on data visualisation techniques to support exploratory analysis . It has been shown that information visualisation techniques may support predictive data mining applications, including data clustering [ 14 - 16 ]. These tasks should complement each other in order to achieve higher levels of knowledge integration and understanding. Furthermore, visualisation-based exploratory methods may support: a) the identification of key patterns in the data, and b) the selection of the most adequate models for data pre-processing and/or classification. The former task refers to the recognition of key groups of data, outliers and features based on computationally-inexpensive, user-friendly and robust analyses. Its outcomes may offer guidance to conduct the latter task by gaining a better insight into the high-level structure and relationships found in the data. Such an exploratory, visualisation-based approach may generate useful alternative views for supporting a more intelligent and meaningful application of classification models. One of the traditional approaches to functional genomics information visualisation has been the application of clustering-based visualisation techniques. Such an approach mainly consists of two steps: a) the implementation of a clustering algorithm, and b) the display of the obtained clusters. The resulting clusters may be visualised by generating, for example, dendrograms [ 16 ], other hierarchical structures [ 17 ] and maps [ 14 , 18 , 19 ], which highlight or summarise similarity relationships between groups of data. Clustering-based visualisation has become a fundamental tool for analysing gene and protein expression data. Different variations of hierarchical clustering, Kohonen Self Organising Maps (SOM) and Self-Adaptive Neural Networks (SANN) are relevant examples of techniques belonging to this approach. Their capabilities and applications have been widely reported [ 2 , 15 , 20 ]. They have been successfully tested on several classification and decision support problems. However, its application to visualisation-driven exploratory analysis is limited by several problems: Many of these techniques are not capable of explicitly and automatically detecting cluster boundaries; some of them are critically sensitive to several learning parameters that need to be selected by the user; some of these solutions were not originally designed to tackle cluster-based visualisation tasks of massive collections of data described by several thousands attributes; and they traditionally require assumptions about the inherent structure of the data, which may not be always possible in exploratory data analyses. Clustering visualisation has also become an important task for the analysis of protein-protein interaction networks. Clustering is a fundamental mathematical property of networks, which allows the identification of key connectivity patterns. Such patterns may be associated with significant functional behaviours and modularity [ 3 ]. Moreover, it allows researchers to identify relevant areas for further statistical or experimental analyses. For instance, hierarchical clustering of network nodes (proteins) has been applied to detect functional modules in S. cerevisiae [ 3 ]. Each node may be represented by a vector of connectivity values that reflects the node's interactions with other network members. Graph theoretic approaches have also been applied to detect significant clusters of interconnected proteins [ 21 ]. Another important data visualisation approach, which may be applied to clustering-based analysis, comprises the application of non-linear mapping techniques . They are based on the idea of transforming the original, n -dimensional input space into a reduced, m -dimensional one, where m < n . These methods are also known as non-linear projection methods or multidimensional scaling (MDS) methods [ 22 ]. They mainly aim to optimize a function, M , which reflects key aspects of the distance structure of the original n -dimensional space. Thus, these methods aim to preserve such global properties in the transformed, m -dimensional space. This principle has been followed by several models including those proposed by Kruskal [ 23 , 24 ] and Sammon [ 25 ]. Principal component analysis (PCA) [ 26 ] is another relevant technique for reducing n -dimensional data. In this case the resulting transformation accounts for the greatest variation of the original space, but without preserving the distances observed between the points in the n -dimensional space. MDS, including Sammon's mapping, applications to gene expression analysis have been reported, which highlight their advantages for supporting the detection of clusters [ 27 ]. PCA may not be directly used to visualise clusters. But it may be applied as a pre-processing procedure, and its resulting components may be used as inputs to clustering and supervised classification models [ 28 ]. Although widely investigated techniques, such as SOM and Sammon's mapping, are usually useful to visualise clusters of high-dimensional data, they present several limitations. For example, the SOM may be highly sensitive to its training parameters, which have to be defined by the user. It also requires the user to define map topologies, and it does not provide automated mechanisms for cluster boundary detection. Its limitations for data exploratory analysis, particularly in relation to data topology preservation, have been stressed in [ 29 ]. These and other limitations, as well as adapted solutions, have been discussed in [ 14 , 15 ]. Sammon's method may include several data overlaps when the n -dimensional input space contains noisy or weakly discriminatory information [ 30 ]. Depending of the size of the input data (number of points), the number of learning iterations and computational facilities available, Sammon's mapping might be computationally expensive. Empirical analyses have shown that Sammon's mapping may easily get stuck at local optima [ 30 ]. Moreover, it has been demonstrated that this method may be sensitive to the initialisation scheme applied [ 19 , 31 ]. This paper assesses an alternative, non-linear mapping technique which aims to address key limitations exhibited by traditional methods. Its application to clustering-driven exploratory analysis of gene expression data is investigated. Furthermore, it provides the basis for an interactome network clustering visualisation system. The following section summarises relevant results. Results A relaxation method for non-linear mapping was implemented to visualise relevant similarity relationships in data originating from gene expression and interactome data. Such a method was designed by Chang and Lee [ 32 ] to address key limitations observed in methods such as those proposed by Sammon [ 25 ] and Kruskal [ 23 , 24 ]. These techniques are related because they aim to achieve a space reduction by preserving the structure of local distances in the data. However, unlike those traditional mapping techniques, the method assessed in this paper adapts a pair of points in the transformed m -dimensional space at every processing step, instead of adapting all points at every step. Thus, the term "relaxation" is taken from the relaxation method for linear equalities [ 32 ]. Chang's and Lee's method showed to outperform Sammon's mapping both in terms of cluster detection effectiveness and computational efficiency. A mapping iteration is defined as a complete sequence of adaptation steps involving pairs of points, p ( i , j ), for each i ≠ j (see Methods for a more detailed description). In this study a point may encode a biological sample described by a gene expression profile (e.g. tumour sample), or a protein described by its interaction profile. Figure 1 summarises the mapping mechanism of this approach. Analysis of gene expression data Analyses were performed on three publicly available expression data sets. The first one includes 38 samples from a known leukaemia study [ 33 ], which are represented by 50 expression values. The samples are categorised into two classes: Acute myeloid leukemia (AML) and acute lymphoblastic leukemia (ALL). This data set has been previously validated by several experimental and in silico methods. It may be considered as an adequate example for illustrating basic capabilities of clustering algorithms. The second data set includes samples originating from small, round blue-cell tumours (SRBCT) [ 28 ]. These data consisted of 63 samples categorised into four classes: Ewing family of tumors (EWS), rhabdomyosarcoma (RMS), Burkitt lymphomas (BL) and neuroblastomas (NB), which are represented by the expression values of 2308 genes with suspected roles in processes relevant to these tumours. The third data set offers another example of gene expression analysis complexity: A data set consisting of a relatively small number of samples described by thousands of gene expression values, without incorporating pre-processing by feature selection or transformation procedures. This application comprises 20 samples described by 9504 gene expression values for both normal brain and a pharmacological model of Parkinson's disease [ 34 ]. The reader is referred to the section of Methods for a more detailed description of the data and their prediction tasks. Several experiments were performed on each data set to assess possible, key advantages and limitations of the mapping approach introduced above. Each experiment requires an input file storing a matrix, A , where each entry, a ij , represents an expression value, j , for a sample, i . The user needs to define only one learning parameter, the number of mapping iterations, as defined above. The output of the algorithm is the projection of the K samples represented in A on the reduced m -dimensional space. Mappings were analysed for m = 2 and m = 3, which allowed the generation of 2D and 3D visual displays. Experiments were also conducted for several numbers of mapping iterations. The results facilitated a high-level understanding of fundamental similarity relationships between the samples, which are also consistent with previous research. Three important exploratory data analysis tasks were accomplished: The automated, unsupervised detection of clusters relevant to the natural class structure of the data sets; the visualisation of a coherent preservation of local similarity (distance) relationships between samples; and the identification of potential outliers. Figure 2 depicts mapping results for the leukaemia data. Panels (a) to (d) show results for 0, 1, 10 and 100 mapping iterations respectively. Circles are used to represent the samples. Labels '0' and '1' refer to ALL and AML samples respectively. In the initialisation of the mapping process (0 iterations) the samples are randomly assigned to positions in the 2D space. After 10 iterations there is a clear indication of separation of samples belonging to different classes. With 100 iterations the ALL samples are clustered on the upper left side of the map, and the AML samples are clustered at the bottom of this map. Such clusters were also distinguished in different experiments. As a consequence of the random initialisation process, clusters may occupy different areas on the resulting maps for different experiments using the same number of iterations. Nevertheless, the system correctly separated samples in different experiments with more than 20 iterations. Figure 3 shows resulting maps for the SRBCT data with 100 mapping iterations. These data were pre-processed as explained in the section of Methods. EWS, RMS, BL and NB samples are represented by symbols '1', '2', '3' and '4' respectively. It suggests that the mapping process was able to identify key similarity relationships. Samples belonging to the same class tend to cluster together. For example, EWS samples are mainly located at the bottom of the map. RMS samples are clustered on its left side. The sample '1' ( x : 0.54, y : -1.79) that was located on the left side of Figure 3 far from the EWS cluster (its natural class) was consistently mapped in this fashion by different experiments. Moreover, this sample was also displayed closer to RMS samples for different experiments, which might suggest a significant relationship between such a sample and the RMS class. Furthermore, a previous study using more sophisticated, supervised learning models showed that this sample may be difficult to correctly classify [ 35 ]. This suggests that the map shown in Figure 3 highlights a potential outlier in the EWS class data. Additional experiments using a smaller number of SRBCT classes (only EWS, RMS, BL) were performed. This was mainly done to explore the possibility of obtaining alternative graphical views of the data. Results are in general consistent with the results produced with 4 classes: Samples belonging to the same class were clustered together. However, these experiments allow a clearer graphical differentiation of classes on the resulting maps. Figure 4 depicts an example obtained with 100 mapping iterations. For this and other experiments, it was also possible to detect the outlier that was observed in Figure 3 . Such an EWS sample is located at the top of the map shown in Figure 4 . SOM-based analyses using the SOM Toolbox [ 36 , 37 ] were also implemented to establish comparisons. Figure 5 shows the unified distance matrix ( U-matrix ) and the label map (panel on the right side) for a representative result. The section of Methods describes the construction of these maps. The SOM was able to correctly cluster the SRBCT samples. However, these standard SOM visualisation techniques do not provide clear information on sample-to-sample similarity relationships. Moreover, they do not adequately facilitate a direct visualisation of the distribution of samples assigned to each node and their associations. Additional file 1 includes the frequency map, which is another standard SOM visualisation technique, for these results. Figure 6 shows results obtained by applying Sammon's mapping. Symbols '1', '2' and '3' represent classes EWS, RMS and BL respectively. This approach is able to detect class differences between samples. Sammon's mapping also isolated the EWS sample suggested above as a possible class outlier (right side of the map). Figure 3 (near x : 1.2, y : -1.5) suggests another sample '1' as a potential outlier. However, the Sammon's mapping did not clearly depict it as a potential outlier because in this case, unlike Figure 3 , this sample is located closer to class '1' samples (Figure 6 , near x : 0, y : 0.4). 3D Mapping analyses were also implemented for this SRBCT application. Additional files 2 and 3 compare the results originating from the relaxation non-linear and Sammon's mapping methods. Both methods displayed coherent partitions of the data, which are consistent with the 2D mapping results presented above. Moreover, 3D mappings also suggest relatively strong similarities between EWS and RMS samples because of their proximity on the maps. Figure 7 displays a relaxation non-linear map of the Parkinson's disease model data. Parkinson's disease and Normal samples are identified by symbols '1' and '2' respectively. After 100 mapping iterations, results suggest that the method is able to differentiate between these classes. Parkinson's disease samples mainly fall on the upper region of the map ( y > 0), and 7 (out of 10) Normal samples are located below that area. Figure 8 shows SOM-based results, which adequately distinguish between classes. It also indicates that a few Normal samples may be closer to Parkinson's samples than to the main group of its own class. Figure 8 may offer a clearer graphical discrimination between classes. However, the SOM-based results are more dependent on the user's selection of an optimum set of learning parameters (see Methods). Additional file 4 shows the frequency map for these results. Although the separation of clusters is less clear, Sammon's mapping (Figure 9 ) was in general capable of grouping same class samples. In this figure symbols '1' and '2' represent Parkinson's disease and Normal samples respectively. 3D maps originating from relaxation non-linear and Sammon's mapping methods are depicted in Additional files 5 and 6 respectively. Both techniques offer alternative, but consistent exploratory views of the cluster structure of this data set. Analysis of interactome networks As a way of promoting alternative, potential applications of the methodology presented, this section illustrates an example of exploratory data analysis of interactome networks. The approach proposed encodes a network as a graph of interconnected nodes. For a network consisting of N nodes, the mapping tool (from now on referred to as interClust ) requires an N × N matrix, B , as the input data. In this symmetrical matrix each element, b ij , represents the connection strength between nodes i and j in the graph. Such values may also be interpreted as weights representing the relevance of the interaction between a pair of proteins, e.g.: number of hits observed in interaction experiments or a path distance between two nodes on the graph (indirect interactions). Thus, each row in B may be seen as the connectivity profile for a node, i . That is, the connectivity profile of a node becomes this node's coordinates in the n -dimensional, input space. An accompanying tool, inBuilder , automatically generates such a network representation from a list of pairwise protein-protein interactions and their respective connection strengths predefined by the user. The section of Methods provides more information about design and operation aspects of this approach. Before testing this approach on a real interactome data set, a simple example of a network (Figure 10 ) is used to illustrate its application. In this figure the length of the links does not reflect distances or connection strengths. All of the direct connections are considered equally. This network comprises 25 nodes forming 4 main clusters. In this case a cluster is defined as a compact group of interconnected nodes. Figure 11 shows the results obtained by applying interClust to this network with 20 iterations. This map clearly distinguishes the clusters of the network. Moreover, it preserves local interaction relationships: i.e. if two nodes have similar connectivity profiles in the original space (Figure 10 ), then they are also close to each other in the resulting map. Adequate cluster visualisations were also obtained in experiments with more than 5 iterations. This algorithm was tested on the chromatin interactome in C. elegans . It includes 303 proteins and 349 interactions. This data set regroups interactions of proteins involved in transcriptional regulation at the chromatin level. This functional module is of major interest because it is at the crossroads of many biological processes such as development, sex determination, cellular differentiation and proliferation. Its misregulation may have strong consequences such as tumorigenesis or developmental defects [ 38 ]. This data set includes both retested [ 39 ] and non-retested protein interactions obtained by high-throughput two-hybrid screens (unpublished data). Figure 12 displays resulting relaxation maps, at different regions and levels of detail, obtained with 100 iterations. Figure 12 indicates a separation of proteins according to their connectivity patterns. The network encoding scheme and the non-linear mapping algorithm applied distinguished highly-connected proteins ( hubs ) from the other components of the network. Hubs are located in the outer regions of the map. The farther a node is from the centre, the more connections it has. Figure 13 displays a partial view of a region enriched by hubs, which is well-separated from the main group of proteins. The hubs are also relatively well-separated between them. It reflects the fact that such proteins share very few direct, interacting proteins in common. Additional file 7 depicts some of the hubs automatically isolated by the mapping algorithm, as well as a few nodes located near the centre of the map (F15A2.6, C34E10.5, C14B9.6). These diagrams were drawn using the InterViewer tool [ 40 ]. A closer examination of a group of proteins located in the outer regions of the map (Figures 12 and 13 ), for example, reveals that they are involved in key processes such as mitosis and meiosis. This differentiation indicates that such hubs may act as connection components between different biological processes. This is not a surprising finding, but it highlights the capacity of the algorithm to automatically detect some of the most biologically-relevant proteins only on the basis of their connectivity profiles. Seven of these hubs (Y2H9A.1, F56C9.1, F11A10.2, T12D8.7, Y113G7B.23, Y37D8A.9, C53A5.3), for instance, show a significant enrichment of phenotypes obtained by depletion of the transcripts by RNA interference (5.6-fold enrichment compared to genome-wide levels [ 41 ]). Moreover, four of these hubs (F56C9.1, F11A10.2, T12D8.7, C53A5.3) are strongly linked to embryonic lethality ( Emb ) by exhibiting a 6-fold enrichment of this property in relation to genome-wide levels [ 41 ]. Further analyses on the map suggest that neighboring nodes may reflect functional similarity relationships between the corresponding proteins. One example is illustrated in Figure 14 (upper area), which focuses on the region surrounding T20B12.2, also known as Tbp-1, a key component of the Polymerase II holoenzyme. In this region it is possible to identify several components of the core Polymerase II enzyme, as well as several related families (histones acetylases and deacetylases, nucleosomes positioning enzymes of the Swi/snf family) that share similar phenotypes. There is a strong enrichment of phenotypes such as embryonic lethality or problems in growth. These phenotypes (like embryonic lethality) are often attributed to proteins with high connectivity. Only eleven (out of twenty seven) proteins examined in this region exhibit wild type phenotype when they are depleted. This cluster shows a 5.4-fold enrichment of phenotypes, a 5.1-fold enrichment of Emb phenotype, and a 33.3-fold enrichment of sterile progeny phenotype in comparison to genome-wide levels [ 41 ]. This cluster also exhibits significant associations with sterility, growth defect and problems in larval development. Additional file 8 describes the composition of this representative cluster. Discussion With regard to gene expression data, the relaxation non-linear mapping method was capable to support an automated, unsupervised detection of relevant clusters of samples. Results demonstrated that it may also be useful for the visualisation of local similarity relationships between samples and the identification of potential outliers. In general its performance was comparable to Sammon's mapping. One cannot of course expect that a single method would always be able to accurately map different types of high-dimensional data. However, the application of both techniques is recommended as a reliable approach to data exploratory analysis. In this study they offered consistent views of the problems under analysis. SOM, as well as many other cluster analysis techniques, may be more suitable for application after first gaining an adequate insight into the structure and organisation of the data. Such a global understanding may be facilitated through the application of different non-linear mapping methods. Even though a comparison with existing network clustering methods was not implemented, preliminary results suggest that the approach proposed might represent a useful tool for interactome network visualisation and clustering. It distinguished key hubs and facilitated the identification of functionally relevant clusters. It showed that, not surprisingly, most of the hubs detected are essential for the normal development, behaviour and reproduction of C. elegans by exhibiting an enrichment of phenotypes obtained from RNA interference. This can be explained by the fact that the partners connected to these hubs are involved in numerous fundamental processes such as mitosis or meiosis. Hence, the modification of the network caused by the absence of a hub may have strong consequences on the topology of the network and the organisation of the cellular processes. Isolation of functional clusters (e.g. chromosome condensation and segregation of the transcriptional core process) is essential to investigate relationships between groups of proteins or modules. Modules playing a role in the same process (e.g. chromosome condensation and segregation during the mitosis) also tended to be interrelated in the clustering analysis. This underlies the fact that these modules are also functionally interconnected and are interdependent to stringently regulate the cellular processes. Without these connections the process may be misregulated and generate aberrant behaviour (i.e. oncogenesis if the mitosis is not well regulated). In this way, interactions that connect these modules are likely to be at the intersection of several biological processes and to regulate the correct succession of events in a cellular process (e.g. condensation of chromosomes before segregation). We do not claim that the tool reported can be used as an interaction prediction technique. The example analysed aims to illustrate the application of exploratory data analysis for detecting regions, which may be biologically meaningful and relevant for assessing the outcomes from protein-protein interaction prediction techniques. Such patterns may be useful for guiding future computational or experimental analyses to validate interaction hypotheses. Meaningful patterns may be associated, for example, with functionally enriched regions, as shown in this paper. Moreover, because of the limitations regarding predictive accuracy and coverage exhibited by existing single-source techniques, it is important to offer user-friendly tools that may help scientists to detect possible, spurious associations. Future research should include a more exhaustive statistical description of all possible hub candidates detected by the mapping process for this and other network examples. Statistical and functional attributes represented by this method should be compared with previous findings from large-scale, comprehensive studies. Such an investigation was not implemented here because it is outside the main scope of this paper. The preservation of local distance structures is an important property to interpret the non-linear mapping techniques studied here. This is the main goal of their data transformation mechanisms. It basically means that the distance between two points, dm ij , in the transformed m -dimensional space should be very similar to dn ij (their distance in the original n -dimensional data) if dn ij is small. However, if dn ij is relatively large, dm ij is not required to be similar to dn ij . A fundamental difference between the relaxation non-linear and the original Sammon's mapping methods is that the former adapts a point-to-point distance at every processing step, instead of adapting all of the distances at every step. For relatively small data sets the computing times required by the Java-based implementation of the relaxation non-linear mapping method were comparable to those obtained from Matlab ® implementations of SOM and Sammon's techniques, i.e. in the order of seconds. But for larger data sets, e.g. Parkinson's disease data, the non-linear mapping method may run in the order of minutes. This of course depends on the number of mapping iterations and memory resources available. This concern may be addressed by implementing an optimised version of the software, perhaps using another programming language. Another important solution is the implementation of a frame method [ 32 ], which has demonstrated to improve the computational efficiency of the algorithm without significantly compromising its data structure preservation capabilities. We also intend to expand this and related research within an open-source data analysis and visualisation platform, such as the TIGR Multiexperiment Viewer [ 42 ]. An important aspect of future research is the adaptation of the relaxation non-linear mapping method to perform tasks beyond exploratory data analysis. A desirable property would be its capacity to generalise solutions for new samples in an incremental fashion. That is, the system should be able to add new samples to a map without having to re-generate it. One possible solution is the application of an artificial neural network to interpolate and extrapolate the mapping as illustrated by Mao and Jain [ 43 ]. It is also important to develop hybrid systems to combine the strengths and advantages demonstrated by various mapping techniques [ 44 ]. A two- stage approach, for example, represents a feasible solution. In this approach a data set may be firstly partitioned into a set of Voronoi spaces using clustering techniques such as k -means and SOM, and then independent mapping projections may be performed on each area. It has been suggested that such a hybrid model may be advantageous especially when dealing with massive data sets [ 45 ]. Other investigations will involve the assessment and comparison of related techniques [ 46 , 47 ] The mapping algorithm successfully recognised key topological properties and functional relationships in an interactome network based on a graph encoding scheme, which only considers direct interactions. Moreover, it considered all graph connections as equals. Nevertheless, it would be important to implement applications in which the network encoding values, b ij , may also reflect non-direct, shared interactions. This may be done, for example, by defining a graph distance function between network nodes. Another input representation scheme may exploit information relevant to the significance or confidence assigned to the interactions based on experimental evidence. Since cellular networks are organized in a modular fashion, the identification of these modules is crucial to understand relationships between biological processes and offer a higher-order, more accessible representation of the interactomes. The clustering approach proposed in this paper provides a meaningful, simplified representation of complex interactomes. This representation may significantly facilitate exploratory analysis of networks for non-specialists in bioinformatics. This type of analysis is fundamental to detect key network components, such as hubs, which are implicated in many physiological disorders. Identifying these hubs and their associated clusters is also an important step toward the functional annotation of these proteins, as well as for obtaining possible explanations of their involvement in a specific disease. The cluster visualisation tool, interClust , may represent a useful technique to analyse other protein-protein interaction networks including a future human interactome. For instance, it may be applied to isolate proteins linked to a human pathology and to associate them with a cluster or functional modules (e.g. the transcriptional core complex of the Polymerase II enzyme). Another important component of future research is the adaptation of network clustering methods to take into account spatio-temporal aspects of interactions based on, for example, microarrays, in-situ hybridization or protein localization data. Non-linear mapping methods may also be applied to support the annotation of unknown proteins. This may be done by assigning a protein to a functional role that is significantly associated with the cluster under consideration. Furthermore, it is of course fundamental to compare this network clustering methodology with existing techniques. Thus, such approaches may aid researchers in the design of further experiments and the selection of more sophisticated bioinformatics analyses. Conclusions This research studied a user-friendly cluster visualisation approach that is able to support the generation of biologically meaningful outcomes. It represents an effective and robust exploratory data analysis technique. Comparisons indicate that applying more than one mapping approach may improve the confidence of results. Moreover, this may facilitate the generation of alternative, meaningful views of the data. Relaxation non-linear and Sammon's mapping techniques may be more suitable for exploratory data analysis tasks than SOM. This study did not aim to add another algorithm to the existing collection of supervised and unsupervised classification tools. This methodology is not reported as a competing solution to clustering algorithms. Our study shows how an exploratory data analysis approach based on non-linear mapping can support the identification of relevant, biologically-meaningful patterns. We do not argue that the methodology proposed should necessarily offer more accurate results in relation to existing classification solutions. We recommend this methodology as a first step towards understanding complex data mining problems in functional genomics. Such an exploratory approach may also facilitate the selection of more sophisticated methods and highlight possible, critical features for successfully implementing clustering-based studies. This research indicates that the outcomes originating from an exploratory, pattern visualisation method may be as meaningful as those produced by more sophisticated classification approaches, i.e. SOM. Moreover, the methodology proposed does not require the user to define multiple learning parameters. Exploratory analysis frameworks may facilitate a better insight into a data set before applying more sophisticated, problem-specific classification or predictive models. Such an insight may be achieved by helping users to recognise key features of the underlying structure of the data, detect potential outliers or anomalies and test assumptions about the cluster composition of the data. An adaptation of the relaxation non-linear mapping technique, interClust , represents a promising solution to aid researchers to recognise key connectivity and functional patterns in interactome networks. Further research is underway to continue assessing its application to this area. Methods Data The leukaemia data set includes 38 samples originating from [ 33 ]. Each sample is represented by 50 expression values. The samples are categorised into two classes: Acute myeloid leukemia (AML) and acute lymphoblastic leukemia (ALL). The original data sets and experimental protocols can be found at the Broad Institute Web site [ 48 ]. For each feature standardisation was applied by subtracting each value from the mean and dividing it by the standard deviation. The SRBCT data consisted of 63 samples categorised into four classes: Ewing family of tumors (EWS), rhabdomyosarcoma (RMS), Burkitt lymphomas (BL) and neuroblastomas (NB), which were represented by the expression values of 2308 genes. The dimensionality of the SRBCT expression samples was reduced by applying PCA. It has been shown that the application of PCA is important to facilitate an adequate discrimination of samples in this data set. The 10 dominant PCA components for each case were used as the input to the analysis techniques as suggested by [ 28 ], who applied a supervised learning approach to classify the samples after reduction by PCA. Using the raw data (without PCA) the relaxation non-linear mapping was not able to adequately depict differences between the samples. The original data sets and experimental protocols can be found at the National Human Genome Research Institute Web site [ 49 ]. The Parkinson's disease data include 20 samples described by 9504 gene expression values for both normal brain (10 samples) and a pharmacological model of Parkinson's disease [ 34 ]. M. musculus was the organism studied in this disease model. The data are available at The Gene Expression Omnibus [ 50 ] (accession number GDS22). Additional files 9 to 11 include, respectively, the leukaemia, SRBCT and Parkinson's disease data analysed in this paper. The network of interactions was derived from the chromatin interactome in C. elegans . It contains 303 proteins and 349 interactions. It comprises components of the Polymerase II holoenzyme, histones modifying enzymes, nucleosomes positioning proteins and several proteins containing domains known to be essential to this process such as the chromodomain, the bromodomain or the SET domain. This is an early version of the chromatin interactome, which includes a number of retested [ 39 ] as well as non-retested interactions determined by a stringent high-throughput two-hybrid screen. It also contains several interologs [ 38 , 51 ]. A combination of experimental and bioinformatic factors (reporter genes used for the phenotypic tests, number of hits per interactions, a blast e-value less than 1E-10, a PHRED score >20 for 15% of the ISTs (interaction sequence tags) and the frame verification method) were used to provide optimal accuracy. It is known that the two-hybrid approach has the tendency to generate more false positives than the pull-down/Mass spectrometry approach, for example. However, in the data set analysed the rate of false positives is reduced by using more reporter genes (up to 4 genes, unlike traditional large scale two-hybrid screens which commonly use 2 genes). Using 4 reporter genes can reduce the rate of false positives up to 50%. Additional file 12 contains this data set. Algorithms and tools The relaxation non-linear mapping algorithm is summarised in Figure 1 and details on its design are reported in [ 32 ]. The adaptation of a pair points, i and j , in the transformed, m -dimensional map is implemented as follows. Given two points, Pm i and Pm j , in the m -dimensional map, the adjusted new values, Pm new , i and Pm new , j , are calculated using: where dn ij and dm ij represent the Euclidean distances between the points, i and j , in the n - and m -dimensional spaces respectively. The SOM results were obtained using the SOM Toolbox , which is a Matlab ® implementation [ 36 , 37 ]. Each training process consists of two phases. The following parameters were used. Initial learning rates equal to 0.5 (first phase) and 0.05 (second phase). Learning rates were controlled by an inverse-of-time function. The SOM neighborhood radius starts covering one fourth of the map size. The number of training epochs was equal to 10 times the number of map nodes (first phase). For the second phase it was equal to 4 times the number of training epochs in the first phase. A U-matrix depicts the distances between neighbouring map units by displaying a grey scale. In a label map a SOM node represents a class based on a majority voting strategy for the samples associated with this node. In case of a draw, the first class encountered is used. Empty nodes are not labeled. The Sammon's mapping analyses were implemented using the SOM Toolbox with 100 mapping iterations, iteration step size equal to 0.2 and the Euclidian distance. For the interaction data set, the tool inBuilder was used to transform it into interClust input format. The cross-platform tools interClust and inBuilder are available for academic researchers on request from the authors. Graphical outputs for the relaxation maps were obtained with the proprietary software Statistica © . Additional file 7 was created using InterViewer [ 40 ], which is freely available at [ 52 ]. Analyses were performed on a PC with a Pentium ® 4 CPU. Authors' contributions FA designed the study, implemented the relaxation non-linear mapping algorithm, performed analyses and wrote the manuscript. HW did the SOM and Sammon's mapping experiments and helped with the preparation of the manuscript. AC selected the interactome data, interpreted results and helped with the preparation of the manuscript. Supplementary Material Additional File 1 SOM frequency map for SRBCT data It shows the distribution of samples, X ( Y ), over each node in Figure 5 , where X represents the class label and Y stands for the number of Class X samples assigned to the corresponding node. Click here for file Additional File 2 3D visual display originating from relaxation non-linear mapping – SRBCT data EWS, RMS and BL samples are represented by symbols '1', '2', '3' respectively. Click here for file Additional File 3 3D Sammon's mapping results – SRBCT data Symbols "1", "2" and "3" represent classes EWS, RMS and BL respectively. Click here for file Additional File 4 SOM frequency map for Parkinson's disease data It shows the distribution of samples, X ( Y ), over each node in Figure 8 , where X represents the class label and Y stands for the number of Class X samples assigned to the corresponding node. Click here for file Additional File 5 3D Relaxation non-linear mapping of the Parkinson's disease model data Parkinson's disease and Normal samples are identified by symbols '1' and '2' respectively. Click here for file Additional File 6 3D Sammon's mapping of the Parkinson's disease model data Symbols "1" and "2" represent Parkinson's disease and Normal samples respectively. Click here for file Additional File 7 Examples of key hubs in the interactome It depicts some of the hubs automatically isolated by the mapping algorithm, as well as a few nodes located near the centre of the map (F15A2.6, C34E10.5, C14B9.6). Click here for file Additional File 8 Description of protein cluster obtained from Figure 14 Click here for file Additional File 9 Leukaemia data set analysed in this paper Log ratios are used to represent the expression levels. The last column shows the class labels. Click here for file Additional File 10 SRBCT data set analysed in this paper Log ratios are used to represent the expression levels. The last column shows the class labels. Click here for file Additional File 11 Parkinson's disease data set analysed in this paper Log ratios are used to represent the expression levels. The last column shows the class labels. Click here for file Additional File 12 Chromatin interaction network in C. elegans The last column shows the connection strength in the graph. All connections are considered equally. Click here for file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548129.xml |
549551 | An automated system for measuring parameters of nematode sinusoidal movement | Background Nematode sinusoidal movement has been used as a phenotype in many studies of C. elegans development, behavior and physiology. A thorough understanding of the ways in which genes control these aspects of biology depends, in part, on the accuracy of phenotypic analysis. While worms that move poorly are relatively easy to describe, description of hyperactive movement and movement modulation presents more of a challenge. An enhanced capability to analyze all the complexities of nematode movement will thus help our understanding of how genes control behavior. Results We have developed a user-friendly system to analyze nematode movement in an automated and quantitative manner. In this system nematodes are automatically recognized and a computer-controlled microscope stage ensures that the nematode is kept within the camera field of view while video images from the camera are stored on videotape. In a second step, the images from the videotapes are processed to recognize the worm and to extract its changing position and posture over time. From this information, a variety of movement parameters are calculated. These parameters include the velocity of the worm's centroid, the velocity of the worm along its track, the extent and frequency of body bending, the amplitude and wavelength of the sinusoidal movement, and the propagation of the contraction wave along the body. The length of the worm is also determined and used to normalize the amplitude and wavelength measurements. To demonstrate the utility of this system, we report here a comparison of movement parameters for a small set of mutants affecting the Go/Gq mediated signaling network that controls acetylcholine release at the neuromuscular junction. The system allows comparison of distinct genotypes that affect movement similarly (activation of Gq-alpha versus loss of Go-alpha function), as well as of different mutant alleles at a single locus (null and dominant negative alleles of the goa-1 gene, which encodes Go-alpha). We also demonstrate the use of this system for analyzing the effects of toxic agents. Concentration-response curves for the toxicants arsenite and aldicarb, both of which affect motility, were determined for wild-type and several mutant strains, identifying P-glycoprotein mutants as not significantly more sensitive to either compound, while cat-4 mutants are more sensitive to arsenite but not aldicarb. Conclusions Automated analysis of nematode movement facilitates a broad spectrum of experiments. Detailed genetic analysis of multiple alleles and of distinct genes in a regulatory network is now possible. These studies will facilitate quantitative modeling of C. elegans movement, as well as a comparison of gene function. Concentration-response curves will allow rigorous analysis of toxic agents as well as of pharmacological agents. This type of system thus represents a powerful analytical tool that can be readily coupled with the molecular genetics of nematodes. | Background A major motivation for establishing C. elegans as an experimental molecular genetic system was to understand how genes control behavior, especially locomotion, since the uncoordinated (Unc) mutants were discovered early in the history of C. elegans genetics [ 1 , 2 ]. While studies of hundreds of genes involved in this behavior have led to many insights into processes such as axonal guidance ( unc-5 , unc-6 and unc-40 ; [ 3 , 4 ]), synaptic transmission ( unc-13 , unc-18 ; [ 5 - 7 ]), myosin assembly ( unc-54 [ 8 ]), regulation of G protein signaling [ 9 - 16 ]), neuropeptide function [ 17 ] among many others, there has been no general understanding of how C. elegans moves. Starting with Brenner [ 1 ] C. elegans researchers have identified several hundred genes with effects on movement. Recently, RNAi screens have identified 1371 of 27,574 experiments (approximately 800 genes) that confer abnormal movement [ 18 ]. Normal nematode movement is indicative of a toxicant-free environment, and of youthful vigorous worms. Drugs and toxins affect worm movement [ 19 - 21 ], and locomotory defects are a hallmark of aging worms [ 22 ]. Models for C. elegans movement (e.g. [ 23 - 25 ]) would be enhanced by additional quantitative data. Descriptions of movement phenotypes, particularly hyperactive locomotion, have been partial and to some extent anecdotal. For example, mutations in a number of genes result in activation of the EGL-30 (Gαq) signaling pathway and cause an increase in the frequency of body bends [ 7 , 9 , 10 , 12 , 15 , 17 , 26 - 30 ]. Overexpression of or gain-of-function mutations in egl-30 also cause animals to move with exaggerated body bends [ 28 , 29 ], but the presence or absence of this phenotype has only been reported for a few of the Gαq pathway activators [ 15 , 17 , 26 ]. The rate of locomotion has often been determined by manually counting body bends per minute (e.g., [ 30 - 32 ]). Amplitude of body waves has also been determined manually [ 17 , 26 ]. Keating et al. [ 33 ] used visual inspection or manual quantification to screen for movement defects caused by RNAi depletion of neuropeptide receptors. These approaches, while useful, are labor intensive and provide only a partial description of the movement of a particular genotype. We therefore developed a system to analyze the body posture of C. elegans hermaphrodites over time and extract quantitative information concerning their movement. Here we describe a functional system, metrics, analysis tools, and example applications for distinguishing closely related C. elegans mutants and establishing concentration-response relationships for toxic agents. There have been other developments of automated systems. For example, Williams and Dusenbery [ 34 ] tracked the centroids of multiple worms simultaneously. Several other studies have also used automatic tracking of centroids to analyze velocity, dispersal and turning rate [ 35 - 41 ]. Hirose et al., [ 42 ] have recently described a system to automatically measure body length, using image processing similar to that described here. At the time we developed the prototype of the system described here (1999–2000), there were no systems available. Here we describe a set of metrics that allow intuitive use of an automated system, and show the utility of these metrics for genetic studies and studies of toxic agents. During preparation of this paper, we have implemented a combined system that uses the metrics and some algorithms described here with components of a related system developed by Schafer and colleagues [ 43 , 44 ]. The hardware and software for the hybrid system is described by [ 45 ]; the metrics and applications are described here in relation to our system. Results The movement analysis system Our system analyzes motion in two phases. First, video and worm posture data are acquired using the system hardware. Second, measures of behavior are extracted by software. Data acquisition is a two-step process: videotaping and data extraction. Videotaping We assembled a videotaping apparatus comprising a personal computer with the Tracker software package that we developed, a Matrox Meteor-II/Standard video frame grabber, a Wild M5A stereo dissecting microscope with camera mount, a Dage-MTI CCD72 video camera and controller, a Sony video monitor, a Ludl Electronic Products BioPoint motorized inverted-microscope stage and controller with a joystick for manually moving the stage, a stage-mounted custom Petri dish carrier, and a Panasonic model AG-5710P VHS video cassette recorder (which has an RS-232 connection for computer control and feedback of VCR operation) (Figure 1A ). Figure 1 Tracker and Recognizer Schematic . A. Tracker. A Petri plate with worm is placed on a computer-controlled motorized stage. A joystick is used to center the worm in the field of view. The worm is recorded on VCR. B. Recognizer. The video tape is played into a computer by a computer-controlled VCR to recognize the worm and record its body posture and position as a function of time. The behavior of individual worms is examined on Petri plates with fresh, uniform bacterial lawns (see Methods), conditions that favor continued forward movement of the worms. With the videotaping apparatus powered and the Tracker program started on the desktop computer, a single young hermaphrodite is placed at the center of a prepared Petri dish without transferring excess food, and the dish is placed onto the carrier on the motorized microscope stage. The operator uses the joystick and/or Tracker on-screen controls to position the moving worm within a bounding box on the Tracker program's graphical user interface (GUI) and starts the tracking function of the program (Figure 2A ). Figure 2 User Interfaces . A. Tracker User Interface. A simple GUI controls the tracking and recording. B. Recognizer2.1 User Interface. A simple user interface illustrates the progress of the recognition process. The image of the worm is shown with the spine and points superimposed. C. Wormproc User Interface. The interface for processing body posture shows a reference image of worm on plate to assist with worm orientation (left) and the main data processing control window (right) depicting an abstraction of worm during processing. If Recognizer2.1 inappropriately flips head and tail, it can be overridden with the Flip function. The program automatically rejects frames in which the worm length is outside of a calculated normal range, but these can be overridden with the Accpt/Rejct button, or all frames can be scored manually by hitting the Un-Reject All button. Our Tracker program grabs images (frames) from the video stream from the camera looking for differences between successive frames that indicate a moving worm. Tracker identifies the changed regions as either newly occupied or newly vacated. If a changed region falls outside of a 240 × 160 pixel bounding box (within the 320 × 240 pixel image) Tracker sends a command to the motorized microscope stage controller to shift the stage half of a screen width and/or height to relocate the itinerant worm back to the middle of the camera's view field. With the computer re-positioning the worm as necessary, the video stream from the camera is recorded onto VHS tape for use in the data extraction step. Tracker was designed to move the microscope stage in rapid, discrete shifts, allowing the worm to crawl to the edge of the view field before being reined back to the center of our camera's view field. We chose this protocol to eliminate the need for position feedback sensors and stage position data storage. Data extraction Our Data Extraction apparatus consists of a personal computer with our Recognizer2.1 software package, a Matrox Meteor-II/Standard video frame grabber, a Panasonic model AG-5710P VHS video cassette recorder (which has an RS-232 connection for computer control and feedback of VCR operation), and an optional Sony video monitor (Figure 1B ). We use our Recognizer2.1 program to extract worm position and posture data from the video recording made in the previous step. Recognizer2.1 commands the VCR to play back segments of the worm videotape made previously and, using a double-buffering paradigm, grabs and processes images (frames) from the video stream at the rate of ~6 Hz. (Processing rate is a function of available computing bandwidth.) During extraction, Recognizer2.1 displays the grabbed/annotated images in its GUI window (Figure 2B ). Recognizer2.1 locates the worm in each 640 × 480 pixel image using contrast thresholding, identifies the worm's boundary curve, and uses the boundary curve to calculate the worm's "spine." The program mathematically distributes points along the length of the "spine" and records the X-Y position of the points in an output file. ( n points define n-1 body segments, between which there are n-2 articulation points, or "bends", whose angles we calculate.) We typically set Recognizer2.1 to distribute 13 points (13 points: 12 body segments: 11 articulation points) along the worm's spine, but the program can be easily user-customized to apply as many or as few points as desired if, for example, a longer worm is analyzed. The result of running Recognizer2.1 is a set of folders saved to hard disk, each containing a file called "points" containing the table of X-Y coordinates for the 13 points on the worm spine in each grabbed image, and a set of bitmap worm images. The X-Y coordinate table's rows are the data for each image, with the first pair of rows containing the coordinate data from the first image grabbed, and the last pair of rows containing the data from the last image. The table's columns are the data for each of the points distributed along the worm's spine, but since Recognizer2.1 does not identify head versus tail, the saved coordinate data simply represents the posture and screen position of the worm in each processed image without regard to head-tail orientation. Data "orientation" is performed as part of the data processing phase. We assume the time between successive grabbed and processed images is consistent within a data set, and calculate the effective grab rate as the number of grabbed frames in the data set divided by the length of the data set (in seconds). Measurements of the distribution of inter-grab intervals were made by saving computer-generated timestamps corresponding to the X-Y coordinate data normally saved. The distribution of intervals indicates that there is less than 10% variability and thus this only accounts for a small fraction of the observed variability within observations of each animal (see Figure 4 below). Videotape stretch before or during playback could also be a source of variability but we assumed this to be negligible. Figure 3 Sample Attributes . The key attributes that are extracted by Wormproc program are shown schematically. Centroid velocity is the translation of the mean position of the rear two-thirds of the animal. Point velocity is the velocity of each point along the animal's track; velocity is the mean of the point velocities for points 5–13. Track amplitude is the maximum width of a box around the worm. Track wavelength is the length of the sine wave that fits the worm's posture. Bending frequency is the frequency of oscillations between adjacent segments. Flex is the maximum difference in angle between the ventral- and dorsal-most flexion at each articulation point. Time delay is the time required to propagate flexion between adjacent articulation points. Figure 4 Variability of wild-type movement . For each metric, the aggregate statistics are shown along with individual days' experiments. Each daily group is designated by date. The 'other' group comprises 25 individuals that were tested on 16 different days in groups of 1–3. For each metric, a bar graph of the means is displayed (A, C, E, G, I, K) as well as histograms (B, D, F, H, J, L). For bar charts: Blue, mean; green, forward; red, backwards. For histograms: Blue, all 48 N2 animals (n = 48); Green, 2-18-03 dataset (n = 5); Red, 7-18-03 dataset (n = 8); light blue, 9-05-03 dataset (n = 4); magenta, 10-17-03 dataset (n = 6); yellow, other dataset (n = 25). n = number of individuals tested; n is the same for all panels. A. Mean velocity. B. Velocity histogram. C. Mean centroid velocity. D. Centroid velocity histogram. E. Mean frequency at bend 5. F. Frequency at bend 5 histogram. G. Mean flex at bend 5. H. Flex at bend 5 histogram. I. Mean length-normalized track amplitude. J. Length- normalized track amplitude histogram. K. Mean length-normalized wavelength. L. Length-normalized wavelength histogram. Each histogram curve represents the distribution of 632 to 1485 individual measurements per worm. Data processing and analysis Data processing and analysis is performed using a suite of programs we developed in Matlab (from The MathWorks) comprised of three applications: " Wormproc " ( worm proc essing), " Metrics ", and " Histograms ." Data processing proceeds in several steps. Wormproc The researcher runs the "Wormproc" program for each worm to convert the X-Y data into a usable format. First, Wormproc loads into RAM the X-Y coordinate data and images captured and saved to disk by Recognizer2.1. Since Recognizer2.1 does not distinguish between head and tail, Wormproc orients spines by selecting the spine orientations (either oriented "as recorded" or "reversed") that are minimally different from each preceding spine. Further, Wormproc flags spines with lengths that are outside of a calculated "normal" range (asserting that they are invalid or missing data), and identifies the end of the worm that moves the most as the head-end (based on typical C. elegans foraging behavior). Disjointed data segments, for example before and after an omega bend, are treated as separate data segments; Wormproc identifies head position before and after such breaks. Next, Wormproc mathematically removes the microscope stage shifts from the X-Y data; the program recognizes stage shifts by a velocity spike (a very large displacement between two consecutive frames) uncharacteristic of a nematode. The program offsets the X-Y coordinate data after each stage shift to continue the worm's path of locomotion, interpolating over any single missing frames. Occasionally Recognizer2.1 will have grabbed a worm image while the microscope stage is moving. In these instances Recognizer2.1 either will not be able to identify any worm in the image or, because of the interlaced video and contrast, will only be able to recognize a tiny area of the smeared image as worm. In either case the X-Y data for these frames will have been automatically rejected for being outside of the normal length range for the worm. Finally, Wormproc provides a GUI (Figure 2C ) that allows the user to verify (and modify, if necessary) the computer's assertions on valid/invalid data, and worm head/tail orientations via an animation of the subject worm's movements with still images presented in a second window for reference. When the user is satisfied, the program saves the oriented and verified data to hard disk. Metrics We developed a software application called Metrics which extracts useful measures of nematode locomotion and morphology from the processed X-Y coordinates for each worm. For each worm we extract eleven attributes (Figure 3 ). One set of attributes concerns the speed of worm movement. We calculate the instantaneous speed of the animal's centroid (its 'centroid velocity' or VELC). We define the centroid as the mean position of points 5–13 (approximately the posterior two-thirds of the body), and the instantaneous centroid velocity as the change in centroid position over time. Likewise, we calculate the instantaneous velocities of all 13 points along the spine as they move over time (the point velocity, or PTVEL). We define the means of the point velocities for points 5–13 as the worm's 'velocity' (VEL). Instantaneous velocity, point velocity and centroid velocity are identified as 'forward' (positive) or 'backward' (negative) reflecting the direction the animal is moving. The MODE lists the instantaneous movement direction with 1's (forward) or -1's (backward). MODE is determined automatically in Metrics by evaluating whether the majority of points 5–12 are moving closer to their anterior or to their posterior neighbors through successive frames. Due to signal noise, MODE cannot at present be set to "no movement". THETA is the instantaneous velocity (VEL) vector direction. A second set of attributes concerns propagation of the contractile wave. The flex (FLEX) is the difference between maximum positive and negative bend angles in a sliding time window for each of the articulation points. The bending frequencies (FRE) are the time-windowed bending frequencies at each of the worm's articulation points. Time delay (PHS) is a matrix containing the time delay required for an articulation point to reach the same angle as its next anterior neighbor; this metric describes the rate of wave propagation along the worm. A third set of attributes describes the worm's waveform. The track amplitude (AMPT) is the instantaneous worm track waveform amplitude, specifically the width of a best-fit bounding box aligned with the worm's instantaneous velocity vector. Wavelength (WAVELNTH) is a measure of the instantaneous physical wavelength of the worm's sinusoidal body posture. Another attribute describes morphology. The worm's length (LEN) is the sum of the distances between the points along the worm's "spine." Histograms and other data visualization routines We developed several data visualization routines to display and compare the movement attributes of worms. The most common program we use is "Histograms" which displays a set of histograms for comparing locomotory parameters for populations of nematodes. The standard attributes displayed are: Centroid Velocity, (Mean Point) Velocity (velocity of the worm along its sinusoidal track), Flex (for several articulation points), Bending Frequency (for several articulation points), Time Delay (for several articulation points), Track Wavelength, and Track Amplitude (both in millimeters and normalized as a percent of mean worm body length). In addition to comparing populations of worms, it is often useful to compare individual worms within a population, for which we developed "iHistograms." This application produces the same charts as "Histograms," but displays the data for individual worms instead of populations. Using the flexibility of the Matlab programming environment we have developed a multitude of specialty analysis tools, ranging from toxicant concentration-response curves, to speed decay as a function of time, to animation routines to visualize wave propagation, to reversal frequency. With a bit of creativity, output can be customized to a broad range of experiments. To demonstrate the general applicability of this type of system to nematode biology, we provide a few salient examples: genetics and toxicology. Reproducibility of data One common problem in behavioral studies on C. elegans is day-to-day variability. To test whether data obtained on different days could be pooled, we analyzed the movement of small numbers of wild-type individuals on different days. We then compared the means for each movement parameter of each daily group to those for the pooled total. Specifically, we compared seven groups comprising four-twelve individuals to the total data set of 58 individuals. Two groups, with five animals each, had means for more than one parameter that were significantly different from the pooled total and were eliminated from our analysis. For the remaining groups, the p values for all parameters except FLEX ranged from 0.07 to 0.99. For two of the included groups, mean values for FLEX for the more posterior articulation points were significantly different from the pooled total. We therefore report FLEX measurements for a more anterior articulation point (bend 5) only. Comparison of the daily groups to the pooled total is displayed in Figure 4 , which shows types of graphical representation available in our system. Genetic analysis We tested whether the system was useful for comparing alleles of the same gene, and alleles of different genes that result in qualitatively similar movement phenotypes. goa-1 encodes the only Go-alpha subunit in C. elegans and is involved in locomotion [ 18 , 26 , 27 , 46 ]. goa-1(n1134) is a reduction-of-function allele, defective in the consensus sequence for myristoylation at the amino-terminus [ 27 ], and is protein-negative on a Western blot [ 47 ]. goa-1(sy192) is an antimorphic allele [ 48 ]. Although n1134 and sy192 homozygous mutant animals look very similar by visual examination, quantification of their movement revealed that s y192 affects certain movement parameters more severely than n1134 (Figure 5 ). The mean forward point velocity of sy192 is 0.37 (+/- .04) mm/sec versus 0.29 (+/- .06) mm/sec for n1134 and 0.20 (+/- .04) mm/sec for wild type (Figure 5 ). These values are significantly different (p < 0.001 for each pair). The centroid velocities show the same relationship sy192 > n1134 > wild type. The flex is also significantly different: sy192 has a flex at articulation point (bend) 5 of 1.4 (+/- .05) radians, n1134 1.3 (+/- .09) radians and N2 1.0 (+/- .09) radians (p < 0.0005 for all pairs). The frequencies of n1134 and sy192 hermaphrodites, however, are similar: 0.58 (+/- .06) Hz at bend 5 for sy192 versus 0.53 (+/- .10) Hz for n1134 (p = 1.08) versus 0.36 (+/- .08) Hz for wild type (p < 0.0005). The track amplitudes of all three genotypes are significantly different. When normalized for body length, sy192 has a track amplitude of 25.53 (+/- 1.14) % body length, n1134 22.00 (+/- 1.64) % body length, and N2 19.27 (+/- 2.34) % body length (p < 0.0005 for all pairs). The wavelengths, however, are not different for all genotypes. sy192 has a wavelength significantly different from wild type (64.19 +/- 1.40 % body length versus 62.03 +/- 2.28 % body length, p = 0.004). However, the wavelength of n1134 (63.39 +/- 2.3 % body length) is not different from N2 (p = 0.07) or from sy192 (p = 0.34). In summary, mutations in goa-1 that cause hyperactive movement increase both the point and centroid velocities, increase the flex of articulation points, increase the frequency of body bends, and increase the track amplitude compared to wild type. With the exception of frequency, the antimorph sy192 has a more profound effect on these parameters than the null mutation n1134 . sy192 also decreases the wavelength compared to wild type, but n1134 has only a mild effect on wavelength. Figure 5 Comparison of two alleles of goa-1 . Blue, wild-type (n = 48); green, goa-1 (n1134) , a null allele (n = 12); red, goa-1 (sy192) , an antimorphic allele (n = 11). n = number of individuals tested; n is the same for all panels. A. Distribution of velocity. B. Distribution of centroid velocity. C. Flex at bend 5. D. Frequency at bend 5. E. Length-normalized track amplitude. F. Length-normalized track wavelength. Each curve represents the distribution of 632 to 1485 individual measurements per worm. The population mean values reported in the text reflect only forward moving worms and are based on 550 to 1454 individual measurements per worm. For movement, goa-1 acts antagonistically to egl-30 , thus increased egl-30 activity is similar to loss of goa-1 activity [ 28 , 29 ]. We therefore compared the movement of worms bearing a strong egl-30 gain-of-function allele, tg26 [ 49 , 50 ], to those lacking goa-1 activity ( goa-1(n1134) ). Visually, tg26 mutants move with more exaggerated body bends than both wild type and n1134 . This difference was detected by our movement analysis system (Figure 6 ). The mean forward point velocity of both n1134 and tg26 are similar (0.28 +/- .03 mm/sec for tg26 and 0.29 +/- .06 mm/sec for n1134 , p = 0.76), and faster than wild type (0.20 +/- .04 mm/sec, p < 0.0005). Although not statistically significant, the forward centroid velocity for tg26 (0.21 +/- .02 mm/sec) is more similar to that of N2 (0.18 +/- .03 mm/sec, p = 0.02) and less similar to that of n1134 (0.24 +/- .05 mm/sec, p = 0.05) than the mean forward point velocities. This difference reflects the increased path length any point on the spine of the tg26 mutant must travel to displace the centroid. Both mutations cause increased flex and frequency compared to wild type. The flex of tg26 at bend 5 was 1.77 (+/- .04) radians versus 1.27 (+/- .09) radians for n1134 (p < 0.0005), and 1.00 (+/- .09) radians for wild type (p < 0.0005). The frequency of tg26 at bend 5 was 0.56 (+/- .05) Hz versus 0.53 (+/- .10) Hz for n1134 (p = 0.37), and 0.36 (+/- .08) Hz for wild type (p < 0.0005). While both alleles affect frequency similarly, tg26 has a more profound effect on flex than does n1134 . tg26 also has a more profound effect on amplitude than does n1134 . When normalized for body length, tg26 has an amplitude of 26.08 (+/- .67) % body length compared to 22.00 (+/- 1.63) % body length for n1134 (p < 0.0005) and 19.27 (+/- 2.36) % body length for wild type (p < 0.0005). The track wavelength for tg26 is shorter than that of n1134 and N2: 54.04 (+/- 1.13) % body length for tg26 , 63.4 (+/- 2.33) % body length for n1134 (p < 0.0005), and 62.3 (+/- 2.28) % body length for wild type N2 (p < 0.0005). Figure 6 Comparison of goa-1 loss-of-function and egl-30 gain-of-function mutations . Blue, wild-type (n = 48); green, goa-1 (n1134) , a null allele (n = 12); red, egl-30 (tg26) , a gain-of-function allele of egl-30 Gq (n = 8). n = number of individuals tested; n is the same for all panels. A. Distribution of point velocity. B. Distribution of centroid velocity. C. Flex at bend 5. D. Frequency at bend 5. E. Length-normalized track amplitude. F. Length-normalized track wavelength. Each curve represents the distribution of 632 to 1485 individual measurements per worm. The population mean values reported in the text reflect only forward moving worms and are based on 550 to 1454 individual measurements per worm. Our movement analysis system is thus able to discriminate between the effects of different mutations that affect the same parameters of movement. Toxicology To test the utility of our system for analyzing the effects of toxicants, we focused on the neurotoxin aldicarb and the metabolic inhibitor arsenite. We first established baseline conditions for these toxins. We then tested whether existing mutations would increase the sensitivity to these compounds. We focused on mutations that affect cuticle and P-glycoprotein transporters. We analyzed C. elegans movement in the presence of increasing concentrations of aldicarb (2-methyl-2-(methylthio)propionaldehyde O-methylcarbamoyloxime) and sodium-arsenite (NaAsO 2 ). We first determined that a 30-minute exposure to 6.4 mM aldicarb induced near paralysis in wild-type C. elegans . We then recorded movement of individual hermaphrodites following a 30-minute exposure to aldicarb concentrations from 0 to 6.4 mM. We tested 16–17 individual wild-type worms for each concentration of aldicarb and found that the concentration reducing wild-type mean forward point velocity by 50% (EC50) is 0.39 mM aldicarb (Figure 7A ). We similarly determined that a 3-hour exposure to 80 mM sodium-arsenite induced paralysis in wild-type worms. We have determined an EC50 of 9.7 mM NaAsO 2 for wild-type C. elegans (Figure 7B ). Of the movement parameters tested, mean point velocity, centroid velocity, track amplitude and track wavelength were equally sensitive to aldicarb, with differences apparent at the lowest concentration tested (0.1 mM). A reduction of frequency was seen at 0.2 mM, but flex was resistant to the effects of aldicarb and alterations were only apparent at the highest concentrations (data not shown). For sodium arsenite, most parameters were affected after exposure to 2.5 mM, except for flex which was only affected at concentrations above 20 mM (data not shown). Figure 7 Toxicant sensitivity of wild-type and cat-4 . A. Sensitivity to aldicarb. For N2, n = 16 animals for 0 mM aldicarb and 17 for all other concentrations. For cat-4 , n = 6 for all concentrations. B. Sensitivity to arsenite. For N2, n = 17 animals for 2.5 mM sodium arsenite and 18 for all other concentrations. For cat-4 , n = 4 for all concentrations. We tested three candidate hypersensitive mutant C. elegans strains for movement in response to increasing doses of aldicarb and sodium-arsenite. The cat-4 gene encodes GTP cyclohydrolase I (C. Loer, personal communication; see also [ 51 ]) necessary for biosynthesis of biogenic amines; it is hypersensitive to several disparate agents such as the neurotransmitter serotonin and the detergent SDS, suggesting a weaker or more porous cuticle (C. Loer, pers. comm.). cat-4 mutants are 2.2-fold more sensitive to aldicarb (EC50 = 0.18 mM vs. 0.39 mM; Figure 7A ) and 8.1 fold -fold more sensitive to arsenite (EC50 = 1.2 mM vs. 9.7 mM; Figure 7B ) than wild-type C. elegans . Only the EC50s of wild type and cat-4 on arsenite are significantly different. P-glycoproteins are membrane transporters in certain cells that protect the cell against environmental toxins by moving such agents out of the cell. The strain NL130 carries deletions for two P-glycoproteins encoded by pgp-1 and pgp-2 [ 52 ]. Mutants lacking these two P-glycoproteins are hypersensitive to colchicine and chloroquinone. We have shown that NL130 is 3.5-fold more sensitive to aldicarb (EC50 = 0.11 mM vs. 0.39 mM; Figure 8A ) and 1.2-fold more sensitive to arsenite (EC50 = 8.2 mM vs. 9.7 mM; Figure 8B ) than wild-type C. elegans . The strain NL152 carries deletions for mrp-1 as well as those for pgp-1 and pgp-3 [ 53 ]. The mrp-1 gene encodes a C. elegans homolog of the mammalian multidrug resistance-associated protein (MRP), another transporter that protects cells from toxins. NL152 also showed increased sensitivity (9.7-fold) to aldicarb (EC50 = 0.04 mM vs. 0.39 mM; Figure 8A ), and only slightly increased sensitivity to arsenite (EC50 = 5.8 mM vs. 9.7 mM; Figure 8B ). However, NL152 is also somewhat movement impaired in the absence of toxic agents. Figure 8 Toxicant sensitivity of NL130 and NL152 . A. Sensitivity to aldicarb. For N2, n = 16 animals for 0 mM aldicarb and 17 for all other concentrations. For NL130 and NL152, n = 6 for all concentrations. B. Sensitivity to arsenite. For N2, n = 17 animals for 2.5 mM sodium arsenite and 18 for all other concentrations. For NL130 and NL152, n = 4 for all concentrations. Our system is thus useful for analyzing the effects of toxicants on nematode movement and for examining the effects of genetic background on toxicant sensitivity. Discussion We describe a usable automated system to record and analyze C. elegans locomotion and to display quantitative data. We describe several useful parameters of locomotory behavior. These include automatic determination of the velocity of the worm's centroid, the velocity along its track, the degree of flex of body at various positions along the body axis, the bending frequency between adjacent segments, and the body-length normalized track amplitude and wavelength (Figure 3 ). Many of these measurements match what C. elegans geneticists have typically observed in describing movement variations such as velocity and body bends per minute. Our system thus provides facile quantification of standard phenotypes. Additionally, our system provides measurements that describe many aspects of movement that relate to underlying neural and mechanical mechanisms of locomotion such as propagation of the contraction wave. We envision that the ability to describe in a quantitative and automatic manner nematode movement will facilitate modeling of C. elegans movement. We have demonstrated that our system can be used to distinguish between different alleles of the same gene, different genes, and different environmental conditions. It thus provides a rich dataset for analysis of gene function and toxicology. In addition, this system can be adapted to score other phenotypes. For example, the frequency of spontaneous reversals can be easily extracted from the data set. This system gives worm length as well, and thus we can normalize the track wavelength. Measurement of body length has been used to screen for suppressor mutations and to analyze mutations that affect gene expression [ 47 , 54 ], and thus this system can be used for genetic studies besides behavior. The basic platform described here might be readily extensible to analysis of other nematode behaviors such as male mating, as well as the behavior of other organisms such as the crawling of insect larvae. Although the system is not high throughput, recording of each individual hermaphrodite takes only seven minutes of largely hands-off time on the part of the researcher. Including processing of the data, a single data set comprising 10–12 individuals can be generated in less than three hours. There are a number of limitations to our system. We use all relative coordinates, and thus tracking worms with respect to specific locations on the Petri plate, for example a gradient of chemoattractant, is not possible. In addition, the noise inherent in our tracking system, most likely arising from its use of analog video, precludes our detecting a worm that is definitively not moving. Moreover, the throughput of our system is limited by its inability to follow multiple worms simultaneously. While preparing this manuscript, we compared our system to that of W. Schafer and colleagues [ 43 , 44 ], and identified useful features of each system. We have begun to develop a joint system taking advantage of the best features of each system to allow further software development to proceed in an efficient manner. A prototype has been described [ 45 ], but the system described here has been used in a number of ongoing studies in our laboratory and is still in continual use. Methods Strains and media C. elegans N2 [ 1 ]. NL131 pgp-3(pk18)X [ 52 ]; goa-1(sy192)I [ 26 ]; goa-1(n1134)I [ 27 ]. CB1141 cat-4(e1141 ) V [ 55 ]. NL130 pgp-1(pk17)IV; pgp-3(pk18)X [ 52 ]. NL152 pgp-1(pk17) IV; pgp-3(pk18) X; mrp-1(pk89)X [ 53 ]. Nematode Growth Medium (NGM) is from Brenner [ 1 ]. Preparation of plates for observation 10 cm NGM recording plates are equilibrated to 20°C for 18–20 hours prior to being spread with bacteria. Approximately one hour before beginning our recordings, 600 μl of fresh OP50 overnight culture is spread onto each plate, rapidly swirling and shaking to achieve a thin, featureless lawn of food across the entire surface. Excess solution is drawn from the edge with a Pipetman. Each food-spread plate is covered with a tissue (Kimwipe), to ensure that the plate remains dust-free as it dries. The food is allowed to dry onto the NGM surface just until the surface exhibits a matte finish (about 45 minutes), at which time the tissues are replaced by the Petri dish lids and the plates are ready for use. The time required for drying is monitored as a crude measure of the relative moisture content of the plates. Plates requiring more than 60 minutes to dry are discarded. Each plate is used within three hours of drying. Assay conditions L4 hermaphrodites are selected 18–20 hours prior to recording to control for age. Individuals are placed on assay plates and the plate is placed in the holder on the microscope stage. After two minutes the worm is located and recording begins. Each worm is recorded for five minutes and the central four minutes of data are analyzed. Incubations and recordings are done in a constant temperature room at 20°C. Toxicant treatment Aldicarb and sodium-arsenite stocks were prepared in H 2 O (at 55°C for aldicarb), and the appropriate volume added to cooled NGM media prior to pouring plates. The volume of added solution was kept constant. The pH of media containing sodium arsenite was adjusted to 5.8–6.0 (to match NGM without toxicant) with concentrated HCl. This was not necessary for plates containing aldicarb. 10 cm assay plates and 5 cm pre-incubation plates were prepared similarly and stored at 4°C until needed. To insure an ample source of food during pre-incubation, 5 cm plates were seeded with fresh OP50 18–20 hours before use and stored at 37°C until 2 hours prior to the assay when they were placed at 20°C. 10 cm assay plates were equilibrated to 20°C and spread with a thin lawn of OP50 as described above. Hermaphrodites were placed on pre-incubation plates and incubated for 30 minutes for aldicarb and 3 hours for sodium-arsenite prior to recording. Following pre-incubation, individuals were transferred to assay plates and recorded after a 2-minute rest on the microscope stage as described above. All preincubations and assays were performed at 20°C. Microscope Our recordings were made using a Wild M5A stereo dissecting microscope with a 25× objective lens and a 1.25× camera mount. Hardware and software Tracker and Recognizer2.1 are written in Microsoft Visual C++ (6.0), using Matrox ActiveMIL-Lite libraries for image manipulation from the Matrox Meteor-II frame grabbers. Tracker is programmed to work with the optics present on our Wild microscope but can be customized to work with other microscopes with minor software changes reflecting the appropriate magnification levels. For our microscope, each stage shift commanded by Tracker to re-center a worm is between 0.362 mm and 0.904 mm depending on the orientation and speed of the worm. Our BioPoint controller is programmed to move the stage with a starting speed of 6.3 mm/sec (= 10,000 pulses/sec with a 0.628 μm/pulse step size), an acceleration rate of 44.8 mm/sec 2 (71,400 pulses/sec 2 ), and a maximum run speed of 31.4 mm/sec (50,000 pulses/sec). However, because of the small distances traveled, we expect the stage to reach a maximum velocity of 8.5 mm/sec (13,500 pulses/sec) during a stage shift. Recognizer2.1 calculates the location of the user-defined number of points (typically 13) along each worm's spine as follows: Recognizer2.1 identifies the center of the darkest portion of the image as its focus of attention and extracts a portion of the image surrounding the center point for further processing. Next, the extracted gray-scale image is turned into a binary image using a segmentation algorithm. The segmentation routine compares the ratio of pixel values resulting after applying two smoothing filters (sum of squared pixel values and square of summed pixel values, both in a typically 15 × 15 pixel neighborhood) against a user defined threshold. Pixels with pixel value ratios less than the threshold are "worm" while the others are "background." Connected regions in the binary image are labeled and Recognizer2.1 then selects the largest-area connected region as the worm in the image. Recognizer2.1 calculates the worm's boundary polygon with a user-defined number of vertices – typically 50 – by interpolating equidistant vertices along the chain of pixels on the perimeter of the worm region. Boundary polygon vertices are passed to Triangle [ 56 ], which generates constrained Delaunay triangulations across the worm boundary polygon. Finally, Recognizer2.1 connects the circumcenters of the resulting triangles to form segments of the worm's spine curve, along which Recognizer2.1 interpolates the 13 equally spaced "spine" points. Feature extraction tools are written in Matlab and C++. Attributes are calculated as follows, described based on a typical analysis distributing 13 points along a worm's spine: Speed attributes Centroid Velocity (VELC) Centroid velocity is a series of speed values, one for each pair of successive frames, and is calculated as the distance the worm's centroid moves between successive frames divided by the time between successive frames. We define the worm's centroid as the mean position of points 5–13 (approximately the rear two-thirds of the worm). VELC = (Δ Centroid Position) / (Δ Time) Sign of VELC describes forward (positive) or backward (negative) movement. Movement of the anterior one-third of the worm is ignored from this calculation (and the VEL calculation below) to minimize the effect of foraging behavior on speed. Point Velocity (PTVEL) Point velocity is a matrix of speed values, one for each of the 13 points for each pair of successive frames, calculated as the distance each point along a worm's spine moves between successive frames divided by the time between successive frames. PTVEL = (Δ Point Positions [all 13 points]) / (Δ Time) Signs of PTVEL describe each point's forward (positive) or backward (negative) movement. PTVEL describes the speed with which each point travels along the worm's serpentine path. Velocity (VEL) Velocity is a series of speed values, one for each pair of successive frames, calculated as the distance each point along a worm's spine moves between successive frames divided by the time between successive frames. VEL = Mean of PTVEL's 5–13 for each pair of successive frames Sign of VEL describes forward (positive) or backward (negative) movement. VEL describes mean speed of the worm's body along its sinusoidal path. Forward or Backward Direction (MODE) MODE is a series of flags indicating the signs of a worm's instantaneous speed attributes (VEL and VELC), that is, whether the worm is moving forward or backward. Mode is calculated automatically at the same time as speed attributes by comparing: D1: mean of the distances between points 4–11 at time τ and their posterior neighbors, points 5–12 at time τ + 1 with D2: mean of the distances between points 6–13 at time τ and their anterior neighbors, points 5–12 at time τ + 1 Forward movement is indicated by D1<D2; backward by D2<D1. The MODE flags can be either 1 (forward) or -1 (backward). (Note that the points distributed along a worm's spine are equidistant, so a non-moving worm would have no difference in the distance between neighboring points over time.) Instantaneous Velocity Vector Direction (THETA) The instantaneous velocity vector direction is calculated as the direction the worm's centroid moves between successive grabbed frames. (Calculated as: THETA = arc tangent of the worm's X-Y displacement.) Wave propagation attributes Flex (FLEX) We calculate the matrix of angles between each segment (that is, at each articulation point) for each frame by: angle = acos [(V1x*V2x + V1y*V2y) / (|V1|*|V2|)] where V1 = first segment vector and V2 = second segment vector We define the FLEX at each articulation point as the maximum angle difference during each possible 32 frame (~6 second) time window; that is, the most positive angle minus the most negative angle. Bending Frequency (FRE) We apply the spectrogram function ("specgram") from Matlab's Signal Processing Toolbox to the matrix of bend angles calculated for FLEX. Specgram calculates a windowed discrete-time Fourier transform (short-time Fourier transform) for the changing angles for each articulation point using a sliding window 32 frames (~6 seconds) wide. We quote from Matlab's documentation for specgram: " specgram calculates the spectrogram for a given signal as follows: 1. It splits the signal into overlapping sections and applies the window specified by the window parameter to each section. 2. It computes the discrete-time Fourier transform of each section with a length nfft FFT to produce an estimate of the short-term frequency content of the signal; these transforms make up the columns of B. The quantity (length(window) – numoverlap) specifies by how many samples specgram shifts the window. 3. For real input, specgram truncates the spectrogram to the first nfft/2 + 1 points for nfft even and (nfft + 1)/2 for nfft odd ." The magnitude of the function indicates the relative energies of the signal's component frequencies. We take the highest magnitude (non-constant) component frequency as the characteristic frequency of that time-window of angles. (If two or more frequencies share the highest magnitude, the lower frequency is identified as the characteristic frequency.) Time Delay (PHS) For the time delay calculation, the program correlates anterior bend angles with posterior bend angles occurring at later time using a Dynamic Time Warping function. The program then uses these correlations to calculate the time required for the posterior bend to reach the same angle as its anterior neighbor. Track waveform attributes Track Amplitude (AMPT) The program aligns the major axis of a best-fit bounding box with the worm's instantaneous velocity vector. The width of the bounding box (its minor axis) is taken as the instantaneous wormtrack waveform amplitude. Track Wavelength (WAVELNTH) We apply a rotation and translation transform to the spine of the worm from every frame to mathematically align each worm with y = 0 using the instantaneous velocity vector as the worm's centerline: We create a matrix w containing the XY coordinates of the 13 points for a worm's spine and create a translation transform that we will use to center the worm's midpoint at the XY origin: We also create a rotation transform to align the instantaneous velocity vector (with angle theta), to y = 0: We multiply the two matrices to create a convenient combined transform C = B*A; and finally multiply our matrix by the combined transform matrix ww = C*w; which yields a matrix ww with the rotated and aligned coordinates for the 13 points. We perform a spatial Fast Fourier Transform (FFT, using Matlab's built-in fft function) on the rotated "spine" of each worm, using the varying y-values as the "signal" with their corresponding x-position values defining the signal to be in a spatial (rather than temporal) domain. By working in the spatial domain, the result from our FFT is in cycles per mm. The inverse of this result is the track wavelength, in mm per cycle. Morphology attribute LEN (spine length) Sum of the distances between the points distributed along the worm's spine with each point-to-point distance calculated by (Δ x 2 + Δ y 2 ) 0.5 Data Analysis and Comparison tools are written in Matlab. Our most common tool is called Histograms, which produces a set of charts displaying distributions of measures of behavior: Velocity, Centroid Velocity, Bending Frequency, Flex, Time Delay, Track Amplitude, Track Wavelength, Length-Normalized Track Amplitude, and Length-Normalized Track Wavelength. To create a histogram curve for a given metric, for example a velocity histogram curve, we sort (or 'bin') each worm's velocity data into discrete ranges (or 'bins'). For velocity we use 'bins' that are 0.03333 mm/sec wide, i.e. bins would represent 0 to 0.03333 mm/sec, another 0.03333 to 0.06666 mm/sec, and so on. Bin sizes for each parameter are as follows: Point and Centroid Velocity, 0.03333 mm/sec; Flex, 0.1 radian; Frequency, 0.16665 Hz; Time Delay, 0.075 sec; Track Amplitude, 0.01 mm; Track Wavelength, 0.05 mm; Length-Normalized Track Amplitude, 1% mean body length; Length-Normalized Track Wavelength, 5% mean body length. Again using our velocity example, the frequency of occurrence for the number of velocity values in each bin is normalized to a percent of velocities observed for that worm, and the normalized velocity distribution is added to a list of the other normalized velocity distributions for that population. (Normalizing each worm's data affords each worm equal mathematical significance.) The final histogram curve for the population is generated by plotting the mean value of each data bin's normalized frequency of occurrence (on the y-axis) versus the bin value (on the x-axis) which shows us the proportion of worms that exhibited each velocity. The same method is used for creating the histogram curves for each metric, naturally selecting bin sizes appropriate for the data in question: Statistical analysis Standard statistical tests were performed using Matlab functions. Each p-value reported was from a one-way analysis of variance (ANOVA) comparing the distribution of mean values (one per individual) from each population of worms against that of each other population; each ANOVA tested the null hypothesis that the mean of the mean values from each populations were the same. Unless otherwise noted, statistical tests were performed on data from worms only when moving forward. For the toxicant concentration-response data, curves were fit by non-linear regression using Prism (GraphPad Software) sigmoidal dose-response equation with variable slope. Availability of source code Source code is available through a GPL at Documentation of this software is available as a pdf from Contributions of authors SM, JB, JM and PS conceived and designed the original automated system for tracking and movement measurement. SM developed a working prototype system and brought in the key algorithms used. CC wrote most of the code in the current release. JM developed the protocol for the plate assay, and devised the dose-response and genetic experiments. RS conceived of applying automated analysis of C. elegans movement for toxicology studies, YK performed most of the toxicology experiments. JM, CC, RS and PS analyzed the data. CC, JM, and PS wrote the paper. All authors read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549551.xml |
517834 | We Move in Mysterious Ways | null | A man in a suit and bowler hat walks awkwardly down the street, each convoluted step a labored movement. He lifts up one knee, then briefly stoops. Stepping forward, he swings the other leg out to the side then kicks high in the air. In this old Monty Python skit, the man works for the Ministry of Silly Walks. It's his job to walk this way. The rest of us, however, tend to stroll along—or throw baseballs, or lift coffee mugs—in a much more efficient manner. There's a nearly infinite number of silly walks, throws, and lifts, but somehow people tend to settle on one best way of doing these things. However, scientists studying motor control have been hard pressed to figure out what exactly we're doing when we move. People may be striking a balance between sloth and speed: too slow and our throws lack oomph; too fast, and instead of dunking our donuts in our coffee, we dunk our whole fist. Or people might be minimizing some version of jerk—physicists' and engineers' term for changes in acceleration. (Roller coaster engineers, for example, balance jerk against speed and g's to keep the ride smooth and safe, but also fun.) But so far, such models that start by assuming people minimize error or jerk haven't allowed researchers to deduce what dictates how people move. To help solve this recalcitrant problem, Konrad Körding and colleagues, as reported in PLoS Biology, took a page from economists, who have long used equations called utility functions that incorporate the costs and benefits of a situation. Say you like oranges better than apples, but oranges cost more. Given a certain budget for fruit, the utility function says how many of each you should buy. Similarly, Körding and colleagues observed people's preferred movements, then inferred an underlying utility function that presumably describes bias in the nervous system for different movements. To see which movements people preferred, the researchers engaged people in a simple virtual reality system. The subjects moved a joystick that fought back: it was connected to a set of motors that produced varying forces—with a strong force for a short time, say, or a mild force for a longer time. Over and over, the subjects moved their cursors from one spot to another. After each pair of moves, the subjects then chose which of the two movements they found easier. In this way, the researchers were able to rank a large set of different movements relative to each other by individuals' preferences. They found a surprising amount of agreement among the subjects on which movements were preferable. They also got a counterintuitive result: as the duration of the resistance got longer, people actually preferred stronger resistance. The researchers speculate that subjects didn't mind larger resistance when it acted over a longer period because the force takes longer to ramp up to its maximum value. Subjects would have more time to adjust—just as when someone gradually pushes into you, you can stay standing by leaning into them, whereas if they shove you with the same force it can knock you off balance. By showing that utility functions can be of use not only in explaining the marketplace but also motor control, Körding and colleagues have added a new tool to biologists' repertoire. Though their approach hasn't closed the case on the mysteries of movement, it could help explain why we settle for a particular, non-silly walk. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517834.xml |
545800 | Hidden localization motifs: naturally occurring peroxisomal targeting signals in non-peroxisomal proteins | Functional but silent peroxisomal targeting signals have been found in non- peroxisomal proteins. This discovery has important implications for sequence-based signal prediction and for evolution. | Background For an increasing number of otherwise uncharacterized protein sequences from genome-sequencing projects, function assignment is attempted solely with in silico prediction methods, as reliable and cost-effective large-scale experimental methods are not available. In addition to sequence homology and annotation transfer considerations [ 1 ], these function assignments increasingly rely on algorithms that recognize protein-sequence features responsible for posttranslational modifications, subcellular localization and interactions with specific domains of other proteins. Although considerable effort has been invested in achieving low false-positive prediction rates, our experience with tools for recognizing glycosyl phosphatidylinositol (GPI) lipid [ 2 , 3 ] and myristoyl [ 4 - 6 ] anchor attachment sites and for predicting potential targets for PTS1-dependent translocation to peroxisomes [ 7 ] shows that a small but noticeable number of proteins without appropriate biological context (for example with contradictory subcellular localization or in taxa without the modifying enzyme or receptor) are systematically hit by these tools. For example, we found more than a dozen metazoan lysozymes [ 7 , 8 ], known extracellular proteins, that are predicted to have carboxyl termini with a functional peroxisomal targeting signal 1 (PTS1) region. Are these false-positive predictions? All three of the sequence-analysis tools mentioned above check query sequences for a recognition pattern that is explicitly described in terms of its physical properties and it is possible to check the concordance between pattern descriptions and query sequence individually. Nevertheless, this visual inspection is frequently unable to rationalize the findings as false-positive predictions, as all known components of the pattern appear to be present. Even in the case of high accuracy of the prediction tool, an erroneous prediction cannot be excluded. Alternatively, these predicted sequence motifs may occur by chance and be functional in an appropriate test system, but still have no biological meaning because the necessary cellular context is absent in vivo . Only experimental tests can resolve this contradiction. As a case study, we report the results of an experimental analysis that demonstrates the existence of naturally occurring peroxisomal targeting signals in several known non-peroxisomal proteins. We also discuss the evolutionary perspective of functional localization signals in unrelated proteins as well as the consequences for experimental localization determination and function prediction from sequence. The major mechanism for targeting proteins to the matrix of peroxisomes, which are membrane-bounded organelles [ 9 ] of eukaryotic cells, is initiated in the cytoplasm by interaction of the receptor protein peroxin 5 (PEX5) with the carboxy-terminal signal PTS1 on the target protein [ 10 , 11 ]. This signal consists of three regions of sequence comprising approximately 12 residues [ 12 , 13 ]. It is composed of the most carboxy-terminal tripeptide (classically, the -SKL terminus), preceded by a region of around four residues (which interact with the surface at the mouth of the PEX5 binding cavity), and a solvent-accessible (or easily unfoldable) stretch of around five residues further upstream. The PTS1-prediction program 'PTS1' [ 14 ] identifies PTS1 signals in query protein sequences by evaluating their carboxy-terminal ends with respect to features necessary for interaction with the tetratricopeptide repeats of PEX5. The predictor's scoring function searching for this motif within the 12 carboxy-terminal residues achieves an estimated sensitivity of 90% and a selectivity above 99% [ 7 ]. Results The carboxyl termini of several non-peroxisomal proteins interact with PEX5 Screening of SWISS-PROT [ 15 ] entries with the PTS1 predictor identified proteins from several families that are clearly not peroxisomal but score highly and are predicted as PEX5 targets [ 7 , 8 ]. We were not able to rationalize these results as false predictions as the proteins' carboxyl termini did not deviate from the generalized PTS1 sequence pattern [ 13 ]. To verify whether these proteins could indeed interact with PEX5, we tested the carboxyl termini of seven representative proteins in a yeast two-hybrid system: hen egg-white lysozyme (P00698, secreted); dog lysozyme C from milk (P81708); tyrosinase from human (P14679, a melanosomal type I membrane protein); frog tyrosinase (Q04604); Drosophila sevenless (P13368, a large transmembrane protein required for photoreceptor development); precursor of lysosomal bovine cathepsin D (P80209); and a mitochondrial ribosomal protein from yeast (P12687). We also examined the carboxyl terminus of a mouse dihydrofolate reductase construct with an added SKL peptide, which has been shown not to be imported into yeast peroxisomes [ 16 , 17 ]. Depending on their taxonomic origin, the carboxyl termini of the eukaryotic sequences were assayed for interaction with the tetratricopeptide repeat domains of either human or yeast PEX5 using published methodologies [ 12 ]. The query sequences, along with prediction scores and measured β-galactosidase activities, are summarized in Table 1 . The results show that all peptide sequences interact with the PTS1-receptor PEX5 in the two-hybrid system. Hence, the carboxy-terminal sequences of these assayed non-peroxisomal proteins fulfill the requirements to function as PTS1 signals. The accessibility of the PTS1-like carboxyl terminus is critical The fact that the peroxisomal translocation machinery fails to import naturally occurring mature proteins carrying PTS1 signals into peroxisomes in vivo could be explained by the non-accessibility of their carboxyl termini. These could either be hidden in the native structure of the mature protein or of its functional complexes, or competing translocation machineries could lead to a removal of the respective proteins from the cytosol before their recognition by PEX5. The first possibility is exemplified by DHFR-SKL. The carboxy-terminal 16 residues of the DHFR-SKL construct (EKGIKYKFEVYEK SKL , sequences appended to DHFR are in bold type, see results in Table 1 ) interact with yeast PEX5 in the two-hybrid test but in vivo the complete construct is not imported into peroxisomes, thus confirming the prediction [ 16 , 17 ]. For comparison, it should be noted that two other DHFR-derived constructs with slightly longer carboxyl termini (IKYKFEVYEK GGKSKL and IKYKFEVYEK KNIESKL ) are predicted to be peroxisomally targeted. Their scores calculated with the PTS1 predictor [ 7 ] are 13.2 and 9.9, respectively (compare with data in Table 1 ). They were experimentally shown [ 17 ] to be translocated to peroxisomes. In the native three-dimensional structure of DHFR [ 18 ], the carboxyl terminus is part of a β-sheet that is buried in the fold, deprived of flexibility and accessibility. Seemingly, this structure prevents the carboxy-terminal appended residues SKL in the construct from entering the PEX5 binding cavity, whereas slightly longer carboxyl termini may do. In our two-hybrid test system, the carboxy-terminal 16-mers are always considered exposed as, in the non-native sequence environment of the carboxyl terminus of the GAL4 activation domain, they are free from interfering or blocking structural features. Thus, DHFR-SKL fails to be imported into peroxisomes because its carboxyl terminus is sequestered in the structure of the mature protein. Competing targeting signals prevent translocation into peroxisomes despite the presence of PTS1-like carboxyl termini Alternatively, functional PTS1 signals can be overruled by other localization signals [ 7 ]. For instance, distribution of the mammalian alanine-glyoxylate amino transferase (AGT) between peroxisomes and mitochondria is regulated by the variable occurrence of an amino-terminal mitochondrial targeting signal in the mature protein (depending on the usage of two alternative transcription initiation sites) [ 19 , 20 ]. Does a naturally occurring PTS1-like carboxyl terminus of a clearly non-peroxisomal protein that is capable of interacting with PEX5 indeed lead to in vivo import of the respective protein, provided that a potentially overruling sequence signal is eliminated? A set of three target proteins with amino-terminal leader sequences was chosen from Table 1 . Chicken lysozyme (SWISS-PROT id P00698), a secreted enzyme, is one of the best characterized proteins and has an apparently accessible carboxyl terminus as deduced from its three-dimensional structure (Protein Data Bank (PDB) number 1H6M [ 21 ]). The corresponding carboxy-terminal 16-mer produces moderate β-galactosidase activity in the yeast two-hybrid assay (most of the other proteins in Table 1 appear to interact even more strongly with PEX5). Human tyrosinase (P14679) is a melanosomal marker protein that functions in the formation of pigments such as melanins. Yeast 60S ribosomal protein L2 (P12687), or MRP7, is a component of the large subunit of the mitochondrial ribosome. Green fluorescent protein (GFP) was appended to the amino terminus of each of the selected proteins. It can be assumed that translocation into the endoplasmic reticulum (ER) or mitochondria is disrupted by the resulting shift of the signal peptide from the amino terminus to the center of the protein. The resulting molecules are expected to be redirected into peroxisomes if their carboxyl termini can act as PTS1 signals. Targeting of the GFP-constructs in vivo was indeed confirmed by co-localization with a peroxisomal DsRed2-SKL construct in COS7 cells for the metazoan enzymes (Figure 1 ) and with DsRed-SKL in yeast cells for the Saccharomyces cerevisiae protein (Figure 2 ). Thus, the PTS1 signals at the carboxyl termini of the assayed proteins are normally suppressed by alternative amino-terminal targeting sequences. A similar mechanism can be inferred for other eukaryotic SWISS-PROT proteins listed in Table 1 , although steric carboxy-terminal accessibility or other factors might also play a role. Functional PTS1 sequences can occur in organisms without peroxisomes The occurrence of silent PTS1s without a targeting role raises the question of whether such signals can also evolve in organisms that do not carry peroxisomes. To test this hypothesis, we extended Table 1 with a set of four predicted carboxyl termini from prokaryotic enzymes: Escherichia coli glutamate-1-semialdehyde 2,1-aminomutase (P23893), E. coli transaldolase A (P78258), Methanopyrus kandleri riboflavin synthase (NCBI-Refseq accession NP_613646) and Archaeoglobus fulgidus 2-nitropropane dioxygenase (NCBI-Refseq accession NP_070998). Indeed, these proteins harbor carboxyl termini that qualify as PTS1 signals (lower part of table 1 ). As confirmation, for the bacterial protein glutamate-1-semialdehyde 2,1-aminomutase (GSA) we used the same methodology for subcellular localization determination as for yeast MRP7. The resulting GFP-GSA construct is also imported into peroxisomes (Figure 2 ), demonstrating that its PTS1-like carboxyl terminus is functional in the mature protein. Discussion In families of orthologous proteins, peroxisomal location and its targeting signal in the amino-acid sequence are not necessarily conserved. For example, in plants the five enzymes of the glyoxylate cycle are localized to peroxisomes, but in S. cerevisiae three of the five (aconitase, isocitrate lyase, and the respective malate dehydrogenase isoform) could not be found in peroxisomes [ 22 ]. Thus, it is not surprising to find sporadically occurring PTS1 signals in protein families (see some examples in Table 1 ). In dually localized proteins such as AGT [ 23 ], the PTS1 signal has a biological role as a targeting signal. However, the carboxyl termini of the proteins from Table 1 do not seem to fulfill any specific targeting function. We suggest that these PTS1 signals occur as a result of neutral mutation. The presence of a functional PTS1 signal would not lead to evolutionary pressure in this context because mislocalization is prevented by overriding the function of these sequences either by alternative exposure of amino-terminal signals or by steric carboxy-terminal inaccessibility. The case of lysozyme is particularly noteworthy because a large number of homologous proteins were systematically hit when performing a SWISS-PROT screen using the prediction tool (30 cases with putative PTS1s and 46 other lysozyme carboxyl termini are shown in Figure 3 ). Because of the close relationship of the originating species and the occurrence of several isozymes, the lysozyme sequences in the multiple alignment share a high degree of similarity. The PTS1 carboxyl termini seem to be a mimicry of the sequence needed to support structural features of the protein. The cysteine at the antepenultimate position, which is present as part of a disulfide bridge [ 21 ] in the final secreted form of lysozyme, happens to fulfill the need for a small residue at the respective PTS1 location. The PTS1 is mostly functional, with a positively charged or amidic penultimate amino acid and the correct hydrophobic carboxy-terminal residue, which is the case for a large proportion of the lysozymes. Note that the disulfide bridge will not be formed in our GFP-lysozyme test case because translocation of the fusion protein into the endoplasmic reticulum is prevented. We conclude that a PEX5-interacting sequence can evolve simply by mutational alterations in the carboxy-terminal region of a protein. Although shuffling of a carboxy-terminal exon cannot be excluded for other examples, the fact that the open reading frames (ORFs) of the carboxy-terminal exons for human tyrosinase (GenBank accession AP000720.4), fly sevenless (GenBank accession AE003484.2) and chicken lysozyme (GenBank accession AF410481.1) reach far into the functional domains of their proteins, rather supports an evolutionary mechanism of several point substitutions. The occurrence of functional PTS1 sequences in non-eukaryotic species further supports a stochastic model for the evolution of PEX5-interacting protein carboxyl termini. In non-globular regions of proteins, sequences that code for targeting to other subcellular compartments, or for posttranslational modifications, might appear in similar ways during evolution. For example, the sequence motif coding for amino-terminal N -myristoylation of glycines behaves as an exchangeable functional module, as protein families do exist where it has been substituted by alternative sequence determinants that facilitate membrane association [ 6 ]. This is exemplified by the Arabidopsis thaliana Rab5 ortholog Ara7 and its paralog Ara6. Ara7 is geranylgeranylated on carboxy-terminal cysteines just as Rab5 is in other species. However, the closely related paralog Ara6 lacks the carboxy-terminal cysteines and has an experimentally verified amino-terminal myristoylation motif [ 24 ]. Many of these signals seem to remain silent under normal physiological conditions (as is the case for the PTS1 signal in some metazoan lysozymes) but have the potential to become important in some future evolutionary scenarios or in pathological situations. Alternatively, the PTS1 signal might have become obsolete and the corresponding sequence segment is now subject to evolutionary alterations. Apparently, the cell exploits only a fraction of the potential molecular capabilities of its proteins. Futhermore, subcellular targeting is organized in a hierarchy of cellular recognition mechanisms. The co-translational sorting into the ER serves as a first decision node. Posttranslational processes such as interaction with chaperones, folding, and covalent modifications are concomitant with the appropriate exposure of targeting signals. The amino-terminal signals are made first and are therefore favored when it comes to recognition by receptors. PEX5 needs only to categorize the remaining unsorted proteins with accessible carboxyl termini into 'stay here' or 'let's go into peroxisomes'. This might also explain why the PTS1 signal is comparatively short and permissive for a wide range of residues. Clearly, the fact that functional sequences for subcellular targeting occur in unrelated proteins needs to be considered for prediction-tool development. The construction of a negative learning set (sequences without the specific localization signal) on the basis of proteins with differing cellular localization is problematic. For example, a set of non-peroxisomal but organellar localized [ 25 ], viral [ 26 ] or bacterial sequences might contain a considerable number of proteins that potentially interact with PEX5. Thus, such a set does not directly qualify for automated learning procedures or the assessment of false-positive prediction [ 27 , 28 ]. Surprisingly, when Maurer-Stroh and Eisenhaber applied their myristoylation site predictor for eukaryotic proteins to bacterial proteomes [ 5 ], systematic hits were found despite the absence of known amino-terminal N -myristoyltransferases (NMT) in bacteria. Are these false-positive predictions? A literature search revealed that myristoylation by host NMTs has physiological relevance for several secreted proteins of intracellular bacterial parasites [ 5 ]. Thus, the sequence motif coding for amino-terminal N -myristoylation is typical for eukaryotes but occurs also in bacteria. In many cases, it remains without phenotypic effect for bacteria but may become evolutionarily important in the case of host-parasite interactions. In the case of the endothelin-converting enzyme 1 and the neprilysin-like zinc metallopeptidase family, the carboxy-terminal CXAW motif is a valid prenylation motif. This carboxy-terminus is functionally hidden because the protein is exported to the extracellular side of the cytomembrane and the carboxy-terminal residues are apparently involved in folding and enzyme function [ 29 ]. Clearly, the accessibility of the recognition motif in the substrate protein to the respective receptor or protein-modifying enzyme is a major issue. For PTS1 signal prediction from the amino-acid sequence, carboxy-terminal exposure needs to be assessed both from the steric point of view as well as in the context of competing translocation mechanisms. Analyzing only the carboxy-terminal dodecamer peptide [ 7 , 13 ] might not suffice for reliable prediction of accessibility to the receptor, but a full solution would require sufficiently accurate three-dimensional structure prediction. In databases, it should also be routine to flag proteins that contain several competing targeting signals with differing priority. Finally, silent localization signals might become active in mutant protein constructs and lead to non-native localizations, an issue that needs to be assessed especially in localization screens of proteins with uniformly incorporated fluorescent dyes such as GFP. It cannot be excluded that the subcellular location of a considerable number of proteins has not been correctly determined in published large-scale studies that rely on this methodology [ 30 , 31 ]. To conclude, sequence segments coding for subcellular targeting or for posttranslational modifications can occur in proteins that are not substrates in either of these processes. Accurate prediction techniques reveal candidate proteins carrying hidden sequence signals. Many of these can be experimentally confirmed. In the case of the PTS1 predictor program, there is no reasonable argument to assume a difference in prediction accuracies for real and hidden PTS1s as, in both cases, productive interaction of the carboxyl terminus with PEX5 is the criterion for a functional PTS1. Materials and methods Cloning procedures Oligonucleotides were purchased from MWG Biotech (Munich, Germany). The E. coli strain DH5α, Bethesda Research Laboratories) was used for all transformations and plasmid isolations. For the yeast two-hybrid-assay, the hybridized oligonucleotide pairs coded for the carboxy-terminal 16-mers of the selected proteins flanked by Bam HI (5') and Eco RI (3') restriction sites. Each oligonucleotide pair was introduced into a Bam HI- Eco RI-digested pGAD.GH fragment, generating plasmids containing the Gal4p activation domain in addition to the desired carboxy-terminal 16-mer extension (Gal4pAD-16mer). All pGAD.GH constructs were sequenced (VBC Genomics, Vienna, Austria). The plasmids pAH987 and hP87 contain the binding domain of Gal4p fused to the TPR domain of S. cerevisiae or Homo sapiens PEX5, respectively (Gal4pBD-TPR) [ 12 ]. Chicken cDNA for the amplification of lysozyme was generated from chicken oviduct using Tripure (Invitrogen) according to the manufacturer's instructions. Reverse transcription was performed using RNA-PCR Core Kit (Applied Biosystems) following the manufacturer's instructions. For the amplification of tyrosinase, we used cDNA from the melanoma cell line 29 WUBI (generous gift of Walter Berger, Vienna). The coding regions of lysozyme and tyrosinase were gained by PCR (for oligonucleotide primers see Table 2 ) using the Advantage cDNA Polymerase Mix kit from Clontech and the GeneAmp PCR-system from Perkin Elmer. The PCR-fragments were cloned into the pCR2.1 vector (Invitrogen) by T/A cloning and sequenced as control (VBC Genomics). The fragments containing the lysozyme or tyrosinase coding regions were excised with Eco RI/ Bam HI and ligated into pEGFP-C1 (Clontech). The DsRed2-SKL construct was obtained by PCR using Pfu-polymerase (Promega) and the plasmid pDsRed2-C1 (Clontech) as template (for oligonucleotides, see Table 2 ). The PCR fragment and the plasmid were both cut with Eco 47-3/ Xho I and the PCR fragment encoding the carboxy-terminal SKL was introduced to replace the original DsRed2 end sequence. The final plasmid encodes the DsRed2-SKL protein under the control of the cytomegalovirus promoter. Standard procedures were used for cloning of the GFP-MRP7 and GFP-GSA constructs including control sequencing (VBC Genomics). The plasmids expressing GFP and GFP-SKL under control of the MLS1 promoter were described previously [ 32 ]. The DNA fragment coding for DsRed-SKL was obtained by PCR (for oligonucleotides, see Table 2 ; template pDsRed, Clontech) and cloned ( Bam HI-and partially with Pst I) after the MLS1 promoter in the vector YEplac181. DNA fragments coding for MRP7 and GSA were obtained by PCR (see Table 2 for oligonucleotide sequences) and cloned ( Bam HI- Sph I) in-frame with GFP to give rise to the expression of GFP-MRP7 and GFP-GSA, respectively, all of them under the control of the MLS1 promoter. Yeast two-hybrid assay According to the Matchmaker two-hybrid protocol, yeast strain PCY3 ( MAT α, his3 Δ200, ade2 -101, trp1 Δ63, leu2 , gal4 Δ, gal80 Δ, lys2 :: GAL1-HIS3 , ura3 :: GAL1-lacZ ) [ 12 ] was transformed with the Gal4pAD-16mer constructs (plasmid pGAD.GH) together with either pAH987 or hP87. Yeast transformants were selected and grown on minimal medium containing 2% glucose and supplemented with bases and amino acids as required (SC-leu-trp). For quantitative measurement of β-galactosidase activity in accordance with published techniques [ 12 ], yeast cells were grown in selective medium (SC-leu-trp) overnight at 30°C, diluted to A 600 = 0.3 into the same medium and finally harvested at absorptions of A 600 between 0.9 and 1.1. In vivo localization study in COS7 cells COS7 cells were transfected with the pEGFP-C1-constructs and DsRed2-SKL by electroporation using 920 μF and 220 mV (Gene pulser II, Bio-Rad), grown on coverslips for 36 h, washed, fixed with 0.5% formaldehyde in PBS for 15 min and covered with geltol. Cells were analyzed using the Olympus BX51 fluorescence microscope (60 × enlargement). In vivo localization study in yeast cells The yeast strain used in this study is S. cerevisiae CB80 ( MATa , ura3-52 , leu2-1 , trp1-63 , his3-200 ). Yeast transformants were selected and grown on minimum medium containing 0.67% yeast nitrogen bases without amino acids (Difco Laboratories), 2% glucose and amino acids (20-150 μg/ml) as required (SC-leu-ura). For fluorescence microscopy, yeast cells were grown at 30°C with shaking in selective media with 0.5% glucose as sole carbon source until the glucose concentration was very low (0.05%, usually 16 h), harvested by centrifugation and resuspended in the original volume of induction medium containing 0.67% yeast nitrogen bases without amino acids, 0.1% yeast extract, 30 mM potassium phosphate pH 6.0, 0.125% oleate, 0.2% Tween-80 and amino acids as required. Cells were grown for 16 h in induction medium and observed live for fluorescence. Briefly, cells were collected by centrifugation and washed twice in water. Cell pellets were resuspended in induction medium without oleate and aliquots were spotted onto multitest slides (ICN Biochemicals) previously coated with concanavalin A (6 mg/ml, Sigma). Cells were allowed to attach for 5 min at room temperature and the slides were washed twice with induction medium and a coverslip applied for observation. Fluorescence was viewed with a Zeiss Axioplan 2 fluorescence microscope using a 63 × (1.4 NA) lens. Digital images were captured with a Quantix CCD camera using Lightview software without further modification. The pictures were mounted and false-color overlays were made in Adobe Photoshop. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545800.xml |
517820 | Sterol-Derived Hormone(s) Controls Entry into Diapause in Caenorhabditis elegans by Consecutive Activation of DAF-12 and DAF-16 | Upon starvation or overcrowding, Caenorhabditis elegans interrupts its reproductive cycle and forms a specialised larva called dauer (enduring). This process is regulated by TGF-β and insulin-signalling pathways and is connected with the control of life span through the insulin pathway components DAF-2 and DAF-16. We found that replacing cholesterol with its methylated metabolite lophenol induced worms to form dauer larvae in the presence of food and low population density. Our data indicate that methylated sterols do not actively induce the dauer formation but rather that the reproductive growth requires a cholesterol-derived hormone that cannot be produced from methylated sterols. Using the effect of lophenol on growth, we have partially purified activity, named gamravali, which promotes the reproduction. In addition, the effect of lophenol allowed us to determine the role of sterols during dauer larva formation and longevity. In the absence of gamravali, the nuclear hormone receptor DAF-12 is activated and thereby initiates the dauer formation program. Active DAF-12 triggers in neurons the nuclear import of DAF-16, a forkhead domain transcription factor that contributes to dauer differentiation. This hormonal control of DAF-16 activation is, however, independent of insulin signalling and has no influence on life span. | Introduction Sterols are essential in most eukaryotic cells and play a structural role in the architecture of their membranes. They influence the physicochemical properties of membranes, including fluidity and permeability for ions ( Haines 2001 ). Cholesterol, together with glycosphingolipids, is also proposed to organise membrane microdomains (also called “rafts”), which provide platforms for protein sorting or signal transduction ( Simons and Toomre 2000 ). In addition to this structural role in the membrane, cholesterol is essential for a variety of signalling processes. It is a precursor of important classes of physiologically active compounds such as steroid hormones in mammals or ecdysones in insects. The nematode Caenorhabditis elegans provides a valuable model system to study the orchestration of cholesterol metabolism and function at the level of a whole organism. C. elegans, like other nematodes, cannot synthesise sterols de novo ( Hieb and Rothstein 1968 ; Chitwood 1999 ). Thus, it requires an exogenous source of sterols, which enables (i) analysis of sterol metabolism using labelled precursors and (ii) analysis of sterol functions by feeding normal and mutant worms with cholesterol derivatives and related sterols. Although worms require exogenous cholesterol for survival, the effects of its depletion are still controversial ( Kurzchalia and Ward 2003 ). Worms are routinely grown in the laboratory on agar plates seeded with bacteria and supplemented with 5 μg/ml of cholesterol (Brenner conditions) ( Brenner 1974 ). Omitting sterols from agar has a weak effect on development and growth: Worms can still propagate for many generations, although some larvae fail to shed the old cuticles properly during molting, gonad development is aberrant, and movement is uncoordinated ( Yochem et al. 1999 ; Shim et al. 2002 ). Under these conditions, the amounts of sterols in both the agar and the bacteria grown on yeast extracts seem to be sufficient to support growth. A stronger phenotype is obtained by using bacteria grown on defined or sterol-extracted media ( Crowder et al. 2001 ; Merris et al. 2003 ). Results of depletion experiments indicate that although absolutely necessary, sterols are required only in very low amounts. This makes it less likely that they are structural components in worm membranes, and thus the primary role in worms should reside in signalling ( Kurzchalia and Ward 2003 ). However, no specific signalling molecules derived from cholesterol, steroid hormones, or ecdysones have been identified yet. It has been suggested that in worms cholesterol plays a role in the processes of molting and dauer formation. Involvement in molting is based on the roles of the worm homologues of mammalian megalin and insect DHR3. A worm mutant of lrp-1, a homologue of mammalian gp330/megalin protein, had a phenotype of defect in shedding of the cuticle, and this phenotype became more apparent upon partial cholesterol depletion ( Yochem et al. 1999 ). Among other functions, megalin in mammals is involved in the uptake of a cholesterol derivative, vitamin D, by kidney absorptive cells ( Willnow et al. 1999 ). Molting in insects is regulated by ecdysones, polyhydroxylated sterols derived from cholesterol, which act via nuclear hormone receptors. The analysis of the C. elegans genome did not reveal a homologue of the ecdysone receptor itself. However, disruption by RNAi of CHR3 (nhr-23), a C. elegans homologue of Drosophila orphan nuclear receptor (DHR3) that is induced by ecdysone, leads to defective shedding of the old cuticle ( Kostrouchova et al. 1998 , 2001 ). Another process that might involve cholesterol or its derivatives is dauer larva formation. Many genes can mutate to cause constitutive formation of dauer larvae (Daf-c mutants) or to prevent their formation (Daf-d mutants) ( Riddle and Albert 1997 ). Genetic studies have revealed that three pathways (TGF-β, cyclic GMP, and insulin-like IGF-1) control the formation of dauer larvae ( Riddle and Albert 1997 ). DAF-2 (insulin-like receptor, IGF-1) signals to inhibit the activity of DAF-16, a forkhead domain (FOXO) transcription factor ( Kenyon et al. 1993 ; Morris et al. 1996 ) that also influences the prolongation of adult life span ( Lin et al. 1997 ; Ogg et al. 1997 ). Under dauer formation conditions, DAF-16 is activated and translocated into the nucleus, where it may integrate insulin-like and TGF-β signalling pathways ( Henderson and Johnson 2001 ; Lee et al. 2001 ; Lin et al. 2001 ). Genetic epistasis analysis suggests that daf-16 acts upstream of two other daf genes, daf-9 and daf-12 ( Gerisch et al. 2001 ; Jia et al. 2002 ). It was proposed that DAF-16 inhibits the activity of DAF-9 when the dauer formation process is initiated. The integration of all three pathways downstream from daf-9 occurs at the level of DAF-12, a putative nuclear hormone receptor ( Antebi et al. 1998 , 2000 ), suggesting a possible hormonal regulation of dauer larva formation. In addition, daf-9 has a strong homology to several cytochrome P450s that are involved in steroid metabolism in mammals ( Gerisch et al. 2001 ; Jia et al. 2002 ). The daf-9 null mutation leads to constitutive dauer formation, consistent with the scenario where DAF-9 is an enzyme that produces a steroid hormone regulating DAF-12, which in turn ultimately triggers dauer formation. As a starting point for our investigations on the role of sterols in C. elegans, we developed a protocol for strict elimination of sterols in the medium and food. Under sterol-free conditions, the first generation of worms developed from eggs to adults without external cholesterol. In the second generation they become dauer-like larvae but molting was incomplete. We found that replacing cholesterol with its natural metabolite lophenol, a methylated sterol, induced all worms to form regular dauer larvae. Using the effect of lophenol on growth, we could partially purify activity supporting the reproduction and determine the role of sterols during dauer larva formation and longevity. In the absence of this hormone, the nuclear hormone receptor DAF-12 is derepressed and thereby activates the dauer formation program. Active DAF-12 triggers in neurons the nuclear import of DAF-16 that contributes to dauer differentiation. Thus, the effect of lophenol allowed us to reveal a novel function of DAF-16 downstream of DAF-12 that is required for the execution of the dauer program but has no effect on life span. Results Worms Grown without Cholesterol for Two Generations Become Dauer-Like Larvae with Incomplete Molting In order to establish sterol-free growth conditions, we extracted traces of sterols from agarose and used defined medium for propagation of Escherichia coli to be fed to worms (see Materials and Methods ). When eggs derived from mothers grown on normal plates (5 μg/ml of cholesterol; approximately 13 μM) hatched at low population densities on sterol-free plates seeded with bacteria, the first generation of worms developed from eggs to adults without external cholesterol. These adults, however, laid only about 60% as many eggs as normal, and 17% of the total eggs laid did not hatch ( Figure 1 A). The second-generation larvae completed their L1-to-L2 molt but then all arrested their development ( Figure 1 A, no cholesterol). Previous studies showing a weaker effect of cholesterol depletion in the second generation ( Yochem et al. 1999 ; Crowder et al. 2001 ; Shim et al. 2002 ; Merris et al. 2003 ) might be due to contaminating sterols. Figure 1 Depletion of Cholesterol Leads to Formation of Dauer-Like Larvae (A) For worms grown on plates without cholesterol, the first-generation worms (F1) laid fewer eggs than normal (133 ± 10 versus 210 ± 12) and more eggs failed to hatch. Filled ovals depict unhatched eggs; 17% of eggs laid by cholesterol-depleted worms failed to hatch in comparison to 0.02% of those laid by cholesterol-fed worms. The second generation of worms (F2) arrested after the completion of the L1-to-L2 molt. (B) Light micrograph of an arrested L2 larva. (C) Electron micrograph of the lateral cuticle of an arrested L2 larva 5 d after arrest, showing two cuticles. The outer cuticle resembles that of an L2, which has no alae, and the inner cuticle resembles the dauer cuticle with its distinctive striated layer (bracket) and an incomplete dauer ala. (D) Electron micrograph of an arrested L2 daf-12 mutant grown without cholesterol. Arrows indicate vesicles beneath the cuticle which are not present in normal larvae. (E) Electron micrograph of a wild-type L2 larva grown with normal cholesterol. Figure 1 B shows an F2 larva grown on a cholesterol-depleted plate. The arrested larvae are similar in size and appearance to L2 larvae grown on cholesterol. The number of cells in their gonads varied between five and 25 with an average of ten, similar to normally grown L2 larvae ( Kimble and Hirsh 1979 ). These larvae stopped pharyngeal pumping after 3–5 d and became immobile after 7 d. If they were transferred to cholesterol-containing plates within the first 2–3 d, larvae reversed their arrest and matured to fertile adults. The reversal required as low as 20 nM cholesterol. The arrested larvae had a double cuticle ( Figure 1 C). The outer cuticle looks like a normal L2 cuticle and the inner cuticle has the characteristics of the normal dauer larva cuticle, including partially developed alae (for comparison, see Figure 3 D) and the distinctive striated layer found only in dauer larvae ( Cassada and Russell 1975 ) ( Figure 1 C). Other dauer-like features of these arrested larvae include constriction of the gut and unstained gut granules (unpublished data). Occasionally, animals were found with partially shed cuticles (<5%; unpublished data). In contrast to normal dauer larvae ( Swanson and Riddle 1981 ), these arrested larvae were sensitive to treatment with 1% sodium dodecyl sulphate (SDS), perhaps because this shedding defect prevents complete dauer cuticle maturation. Figure 3 Wild-Type Worms Form Regular Dauer Larvae When Grown with Lophenol Replacing Cholesterol (A and B) Light microscopy of the second generation of worms grown on lophenol. Note low population density and ample bacteria on the plates (bacteria get swept into piles resembling worm tracks on agarose plates). (C–E) Electron micrographs of lophenol-grown and daf-2 dauer larvae. The alae and the striated layer (bracket) are indistinguishable from those of regular dauer larvae, with extended outer projections. We then asked whether the dauer features of the arrested larvae depend on daf-12. In the absence of cholesterol the second generation of daf-12 worms arrested with only one cuticle similar to that of normal L2 ( Figure 1 D and 1 E). These results show that on noncrowded plates with ample food, the absence of cholesterol causes L2 larvae to enter the normal dauer pathway utilising DAF-12. In summary, our results imply that cholesterol, or cholesterol derivatives, are essential either for the development of reproductive adults or for the prevention of dauer larva formation. In addition, cholesterol derivatives are needed to shed the L2 cuticle. Cholesterol Depletion Leads to Reduced Levels of Nonmethylated Sterols and Accumulation of Methylated Sterols We investigated the metabolism of cholesterol in worms under conditions of cholesterol depletion. Eggs derived from mothers fed with radioactive cholesterol were put on cholesterol-depleted plates, where they grew for another generation to finally produce arrested L2/dauer larvae. The metabolism of sterols was followed by thin-layer chromatography (TLC). It has been previously established that C. elegans metabolises exogenously added cholesterol by methylating the A-ring at the fourth position and rearranging the double bond to form lophenol as the major product ( Figure 2 A) ( Chitwood et al. 1983 ). In eggs and L1 larvae of the first generation ( Figure 2 B, lanes 1 and 2, respectively) or under conditions where cholesterol is present in the food permanently (unpublished data), methylated sterols (ms; lophenol and 4α-methyl-Δ8,14-cholestenol) are found in much lower amounts than nonmethylated sterols (nms; cholesterol, 7-dehydrocholesterol, and lathosterol). Quantification of radiographs revealed that under these conditions ms represent only 1%–3% of the total radioactivity. Cholesterol is also metabolised to a number of more hydrophilic derivatives (indicated by a vertical line), the identities of which are still unknown. In eggs derived from the second generation we observed two major changes: (i) The total radioactivity decreased and (ii) the fraction of nms showed a stronger decrease than that of ms ( Figure 2 B). The proportion between ms and nms changed even more dramatically in the L1 larvae of the second generation ( Figure 2 B, lanes 3–6). Here, about 95% of radioactivity was found as ms (lane 6). Note that the total radioactivity is on the limit of detection. Thus, upon cholesterol depletion the relationship between amounts of nms versus ms is altered. Figure 2 Depletion of Cholesterol Is Associated with a Decrease of Nonmethylated Sterols (A) Nematode-specific biosynthesis of 4-methylated sterols from exogenously added cholesterol. Open arrow shows the methylation at the fourth position. A vertical line indicates hydrophilic metabolites of cholesterol. (B) Cholesterol metabolism in the first (lanes 1 and 2) and the second (lanes 3–6) generations of worms derived from mothers fed with radioactive cholesterol. CE, cholesteryl esters; mS, methylated sterols (lophenol, 4-methylcholestenol); nmS, nonmethylated sterols (cholesterol, 7-dehydrocholesterol, lathosterol). The position of these compounds on TLC was determined by chromatography of cholesteryl stearate, lophenol, and cholesterol. E, eggs; L1, L1 larvae. Substitution of Cholesterol by a Methylated Sterol, Lophenol, Leads to Dauer Larva Formation in the Second Generation To test how nms influence dauer larva formation, we grew worms on plates with all cholesterol replaced by the methylated sterol lophenol ( Figure 2 A). When eggs from normally grown hermaphrodites were placed on lophenol plates, the first generation of worms was indistinguishable from that grown on cholesterol. They had normal brood size and normal morphology. In the second generation, however, the entire population completed two molts and became dauer larvae despite sufficient food and low population density ( Figure 3 A and 3 B). Even on plates with a single worm, the individual developed into a dauer larva. These dauer larvae formed on lophenol plates had the distinct skinny shape, their pharynx was constricted, and they had very rare pharyngeal contractions, as detected earlier for dauer larvae formed by starvation ( Keane and Avery 2003 ). Electron microscopy showed that they had the characteristic alae ( Figure 3 C) and striations of the normal dauer cuticle (compare Figure 3 D and 3 E). They were also resistant to SDS treatment like normal dauer larvae. These results show that, like cholesterol starvation, growth on lophenol leads to dauer larva formation. Unlike cholesterol starvation, however, lophenol allows shedding of the L2 cuticle to form normal dauer larvae. The formation of dauer larvae by growth on 13 μM lophenol was prevented completely by adding cholesterol or its immediate precursor, lathosterol (see Figure 2 A), in amounts as low as 20 nM. Under these conditions all the worms matured to fertile adults. The presence of contaminating nms in plates could be the reason why others did not find dauer larvae when worms were grown on lophenol ( Merris et al. 2003 ). The dauer larvae formed on lophenol plates resumed normal growth when transferred to cholesterol plates, although it required 3–4 d for them to reinitiate development, in contrast to the typical 15 h for normal dauer larvae. Methylation of the Fourth Position of Sterols Is Not Obligatory for Dauer Larva Formation The observation that small amounts of nms can prevent dauer larva formation on lophenol suggests that the lack of a cholesterol derivative which cannot be produced from lophenol is causing dauer larva formation. However, it is possible that lophenol itself actively induces dauer larva formation and methylation of sterols in the 4α position is necessary for this process. In order to distinguish between these alternatives we synthesised 5α-cholestan-3β-ols with a methyl group or fluorine substituted in 4α position ( Figure 4 ) and fed them to worms. We decided to use saturated sterols because they are much more easily accessible for chemical synthesis (for details of synthesis, see Protocol S1 ). Cholestanol ( Figure 4 A) and lophanol ( Figure 4 B) have similar effects on growth as their homologues cholesterol and lophenol, respectively. The former supports reproductive growth, whereas the latter induces dauer formation. Remarkably, when fed with 4αF-cholestanol ( Figure 4 C), worms in the second generation produced dauer larvae. Fluorinated compounds, except in very rare cases, are not susceptible to chemical modifications by living organisms and therefore 4αF-cholestanol cannot be methylated. The fluorine atom is less bulky than the methyl group ( Figure 4 , space-filling models) and differs from the latter in its chemical properties. Thus, it is not the methylation of a sterol in the fourth position per se that is required for the formation of a dauer larva, but rather its accessibility is necessary to prevent this process. We suggest that cholesterol is normally metabolised in two distinct pathways: a pathway forming lophenol, and a pathway forming a steroid hormone. This hormone is required for maintaining reproductive growth and cannot be produced from ms. Figure 4 Methylation of the Fourth Position of Cholestanol Is Not Required for Dauer Larva Formation Structural formulae and space-filling models of (A) cholestanol, (B) lophanol, and (C) 4αF-cholestanol. Abilities to support reproductive growth or dauer formation in the second generation are indicated. R, reproduction; D, dauer larva. Partial Purification of Gamravali, an Activity That Promotes Reproduction The effect of lophenol on growth gave us a unique opportunity to purify the hormone (activity) required for reproductive growth. The rationale of our approach was to rescue the formation of dauer larvae induced in the presence of lophenol by a substance derived from a lipidic extract of worms. Obviously, this substance should differ from cholesterol and its direct metabolites such as 7-dehydrocholesterol and lathosterol. The lipidic extract of worms (see Materials and Methods ) was fractionated using high-performance liquid chromatography (HPLC) on a reverse-phase C 18 column ( Figure 5 A), and fractions mixed with lophenol were fed to L1 larvae of the second generation that were grown on lophenol. As seen in Figure 5 B, two major peaks of activity rescuing dauer larva formation were detected. The major peak, according to retention times (23–30 min), should contain cholesterol, lathosterol, and 7-dehydrocholesterol. Another peak at the beginning of the gradient, however, is much more hydrophilic than major metabolic sterols. Two observations argue that this fraction is not contaminated by dietary cholesterol: (i) This region of the gradient never displayed activity even if the column was overloaded with cholesterol, and (ii) in contrast to cholesterol, active fraction #2 did not support reproductive growth alone, and instead many worms engulfed by the old cuticle were observed. This may be because another cholesterol-derived substance responsible for molting was missing. Figure 5 Partial Purification of Gamravali (A) Lipidic extract of worms was separated by HPLC using a C 18 reverse-phase column. Retention times of (1) 7-dehydrocholesterol, (2) cholesterol/lathosterol, and (3) ecdysone/estradiol/testosterone are indicated with arrows. (B) Fractions of 2 min from the chromatography were assayed for the activity to rescue the formation of dauer larvae induced in the presence of lophenol. We name this activity gamravali (from gamravleba, which means “reproduction” in Georgian; gamravali means “something supporting the reproduction”) because it is required for reproduction in worms. Currently we are attempting to determine the molecular formula of gamravali using mass spectroscopy. This task, however, is very demanding because of the tiny amounts of the substance in worms. Even more demanding will be the final identification of the structure by nuclear magnetic resonance or X-ray analysis. We estimated that the latter might require scaling of the preparation (see Materials and Methods ) up to more than two orders of magnitude. Our data indicate that gamravali is much more hydrophilic than sterols. Remarkably, retention times on the column of many mammalian steroids tested (pregnenolone, β-estradiol, testosterone, etc.) and the insect molting hormone ecdysone are very similar ( Figure 5 A). Thus, gamravali could be a polyhydroxylated sterol such as ecdysone, lack the hydrophobic side chain as in mammalian steroid hormones, or even contain a charged group. However, none of the compounds mentioned above or other commercially available steroids could rescue dauer formation in the presence of lophenol (see a list of tested compounds in Materials and Methods ). A Mutant of daf-12 Can Grow and Reproduce Normally on Lophenol for Many Generations, Whereas Several Daf-d Mutants Produce Dauer Larvae The effect of lophenol on growth also made it possible to identify steps of the dauer formation pathway at which gamravali is required. For this we examined the phenotype of several dauer formation-defective (Daf-d) mutants when grown on lophenol. We assumed that mutants that are defective in metabolism of gamravali and thus act upstream of the hormone receptor should produce dauer larvae on lophenol. Mutants in genes acting downstream of the gamravali action should reproduce normally. We first investigated the growth of a daf-12 null mutant with lophenol as the sole source of sterols. DAF-12 as a putative nuclear hormone receptor is a good candidate to be a receptor for gamravali. In contrast to wild-type worms, mutants of daf-12 grown on lophenol produced no dauer larvae and developed normally for more than seven generations. daf-12 could also reproduce normally on lophanol and 4αF-cholestanol (see Figure 4 ). Therefore, daf-12 acts downstream of gamravali depletion to promote dauer formation, leaving open the possibility that gamravali could be a ligand that inhibits the DAF-12. Our data also show that lophenol can substitute for all cholesterol functions except for the promotion of reproductive development. In contrast to daf-12, other Daf-d mutants, such as daf-22, daf-6, daf-10, daf-3, and daf-5 developed into dauer larvae when grown on lophenol. According to genetic studies all these genes are upstream of daf-12 in the pathway. daf-22 and daf-6 cannot produce or sense the dauer-inducing pheromone, respectively ( Golden and Riddle 1985 ; Perkins et al. 1986 ). DAF-3 and DAF-5 are SMAD transcription factor and its regulator Ski, which antagonise TGF-β action ( Patterson et al. 1997 ; Da Graca et al. 2004 ). The functions of these genes could be to inhibit gamravali production when the dauer pathway is initiated by starvation or overcrowding. Growth on lophenol alone would mimic this situation and result in dauer formation since gamravali cannot be made from lophenol. Mutant daf-16 Produces Defective Dauer Larvae on Lophenol and the Latter Induces Entry of DAF-16 into Nuclei of Neurons in a DAF-12–Dependent Manner Somewhat different results were obtained with null mutants of daf-16 grown on lophenol. In the second generation, neither reproductive adults nor regular dauers were observed. The larvae were fully susceptible to the SDS treatment and their morphology displayed several abnormalities in comparison to regular dauer larvae ( Figure 6 ). They did not have alae of normal morphology (compare Figure 6 A and 3 E), although a striated layer characteristic of the dauer state was visible. The gut was not constricted as in regular dauer larvae ( Figure 6 A). Remarkably, the cuticle displayed annular structures ( Figure 6 B, arrowhead) characteristic of adults but never detected in dauer larvae (compare Figure 6 B and 6 C). One in about 400 worms would occasionally mature and produce a few eggs that never hatched. Thus, in the absence of DAF-16 only a partial, defective dauer larva can be produced on lophenol. Similar defective dauers (although with very low efficiency) were produced by pheromone-treated daf-16 ( Vowels and Thomas 1992 ). Our data indicate that in the absence of cholesterol or its derivatives other than lophenol, daf-16 is still needed for normal differentiation of dauer larvae. Thus, DAF-16 should have activity downstream of the sterol requirement. Figure 6 Mutant daf-16 Worms Grown on Lophenol Form Defective Dauer Larvae (A) Low-magnification electron micrograph of lophenol-grown daf-16 . The alae are defective although the striated layer (bracket) is visible. Note that the gut is not constricted and contains remnants of food. (B and C) High-magnification electron micrographs of lophenol-grown daf-16 and wild-type dauer larvae. Arrowhead indicates an annular structure. Genetic epistasis analysis suggested that daf-16 functions upstream of daf-9, which in turn acts upstream of daf-12 ( Gerisch et al. 2001 ; Jia et al. 2002 ). Moreover, it has been proposed that DAF-16 inhibits DAF-9, a cytochrome P450 that could be involved in the synthesis and/or degradation of a gamravali-like ligand for DAF-12. Consequently, gamravali should be required downstream of DAF-16 function. However, our data ( Figure 6 ) imply that, in addition to the regulation of sterol biosynthesis, DAF-16 acts downstream of DAF-12 and is involved in the differentiation of dauer larvae. Under reproductive conditions, DAF-16 is found in both the cytoplasm and the nucleus, whereas upon activation by the IGF-1 or the TGF-β pathways the protein is accumulated in the nucleus ( Henderson and Johnson 2001 ; Lee et al. 2001 ; Lin et al. 2001 ). We asked whether the growth on lophenol had a similar effect on the cellular distribution of DAF-16. In order to answer this question, we made use of a transgenic line expressing a DAF-16::GFP fusion protein ( Lin et al. 2001 ). In L3 larvae grown on cholesterol, DAF-16::GFP showed diffuse fluorescence throughout many cells, as reported previously ( Figure 7 A). In contrast, in the second generation of worms grown on lophenol, DAF-16::GFP is localised in nuclei of neurons of the pharynx, ventral cord, and tail ( Figure 7 C). Lophenol had a very weak effect on the accumulation of DAF-16 in the nuclei of other cells (e.g., gut or muscles). Figure 7 Growth on Lophenol Induces the Accumulation of DAF-16 in the Nuclei of Neurons in a DAF-12–Dependent Manner (A) When grown on cholesterol, the transgenic line DAF-16a::GFP/b KO displays a diffuse staining in the cytoplasm and nuclei of many cells (only the pharynx region of an L3 larva is shown). (B) Staining of a larva of similar age by Hoechst. Note many nuclei in the pharynx. (C) The DAF-16a::GFP/b KO line grown on lophenol shows strong staining of nuclei in neurons of the pharynx, tail, and ventral cord of a dauer larva. (D) An L3 larva of DAF-16a::GFP/b KO in a daf-12 null background grown on lophenol. Note the diffuse fluorescence in the pharynx cell similar to that shown in (A). Is the nuclear accumulation of DAF-16 upon growth on lophenol dependent on the activation of DAF-12? DAF-16::GFP showed diffuse staining in a daf-12 null mutant grown on lophenol (compare Figure 7 D and 7 A). Our data therefore indicate that the activation of DAF-12 induced by the absence of gamravali leads to accumulation of DAF-16 in the nuclei of neurons. Since the double daf-16;daf-12 mutant grown on lophenol did not produce dauer larvae and could grow on this sterol for many generations, the phenotype of daf-16 observed in the absence of hormone (see Figure 6 A and 6 B) depends on the activity of daf-12. These results imply that the dauer formation process is initiated by DAF-12 but needs nuclear import of DAF-16, which in turn contributes to dauer differentiation, presumably through transcriptional regulation in the nucleus. Growth on Lophenol Does Not Extend the Life Span of Worms daf-2 mutants have a life span that is approximately twice as long as that of the wild-type worms ( Kenyon et al. 1993 ), and in addition mutants display strong intrinsic thermotolerance ( Gems et al. 1998 ). This effect is attributed to the activation of DAF-16 in a daf-2 mutant, since a double daf-16;daf-2 mutant suppresses this phenotype. Does the nuclear accumulation of DAF-16 in neurons when grown on lophenol have a similar effect on life span and thermotolerance? In wild-type worms of the first generation grown on cholesterol or lophenol we could not detect significant differences in the mean life span (21.0 ± 1.8 d and 20.3 ± 1.6 d for cholesterol and lophenol, respectively). It must be noted, however, that worms grown in the absence of cholesterol and the presence of lophenol do not have a developmental phenotype in the first generation and therefore may have some maternal rescue of adult life span. Because the second generation does not grow to adulthood (forms dauer larvae), the definitive experiment cannot be performed. Growth on lophenol also had no influence on the intrinsic thermotolerance of worms at 39 °C. Thus, the activation of DAF-16 induced by the absence of gamravali might have different physiological consequences than its activation by diminished IGF-1 signalling. Discussion Worms Need Tiny Amounts of Sterols for Survival In our attempts to understand the role of cholesterol in nematodes we developed strict sterol-free conditions for growth by combination of the extraction of agarose with organic solvents and the growing of bacteria on defined media. Under these conditions, the first generation of worms grew relatively normally and only the second generation arrested as dauer-like larvae. Thus, the amount of sterols deployed by mothers into embryos is sufficient not only for the survival of the first generation but even for the embryonic development of about 130 embryos that reach the L2 stage in the second generation. This makes cholesterol unlikely to be an indispensable structural component in most worm membranes, although it could play a structural role in cell types where it is concentrated ( Matyash et al. 2001 ). These results are difficult to reconcile with the widespread role of cholesterol and other nms as essential structural components of the plasma membrane. Presumably, C. elegans can regulate membrane properties in response to temperature changes by altering fatty acid composition of phospholipids ( Tanaka et al. 1996 ). Future investigations should clarify what components of nematode membranes substitute for structural functions of cholesterol and whether mechanisms exploited by nematodes to control membrane properties are also serving an analogous purpose in higher eukaryotes. Hormonal Regulation of Dauer Larva Formation: Gamravali versus Lophenol Our results obtained by growing wild-type and mutant worms without cholesterol, with cholesterol replaced by lophenol, and with lophenol supplemented by gamravali demonstrate unequivocally that the decision to enter diapause is regulated by a sterol-derived hormone(s). We propose the following model to explain these results ( Figure 8 ). Gamravali derived from cholesterol acts to promote reproduction and prevent dauer larva formation by inhibiting the nuclear hormone receptor DAF-12. The effect of external signals that induce dauer formation, starvation, and overcrowding is to prevent gamravali production, thus preventing reproduction and promoting entry into diapause. According to this model, growth on lophenol resembles the absence of gamravali. Although it is formally possible that lophenol or ms derived from it could induce dauer formation, this is unlikely for three reasons: (i) Dauer larvae are not induced in the first generation on lophenol, (ii) cholesterol and gamravali at concentrations less than 1/600 that of lophenol prevent dauer larva formation, and (iii) the 4α-fluoro derivative can substitute for lophenol. It is much more likely that lophenol supports all the functions of cholesterol, structural and hormonal, except promoting reproductive growth, since daf-12 null mutants grow and reproduce normally on lophenol. The daf-9 gene, which encodes a cytochrome P450 ( Gerisch et al. 2001 ; Jia et al. 2002 ), could be involved in the production of gamravali. Remarkably, expression of DAF-9 in daf-7 and daf-2 could rescue their Daf-c phenotype ( Gerisch and Antebi 2004 ; Mak and Ruvkun 2004 ). We did not detect any gross-level changes of cholesterol metabolism in the double null daf-9 daf-12 strain ( Figure S1 ). However, since DAF-9 is expressed predominantly in only a small subset of cells ( Gerisch et al. 2001 ; Jia et al. 2002 ), differences in overall cholesterol metabolism might be small and require more sensitive assays. Figure 8 Cross Talk between Two Signalling Pathways in the Process of Dauer Larva Formation Pheromone accumulated under the conditions of overcrowding or starvation induces the inhibition of gamravali production via the TGF-β pathway. Genes, mutants of which produced dauer larvae on lophenol, are shown in blue. Activated by the absence of gamravali, DAF-12 initiates the process of dauer larva production. One of its activities is to recruit DAF-16 into nuclei of neurons (shown in red). The insulin-like pathway has several physiological functions, among them the regulation of longevity and thermotolerance, and could be involved in the process of dauer formation by regulating the levels of gamravali via DAF-16. The Place and Role of daf-16 in the Dauer Formation Pathway Our data uncover a dual role for DAF-16, first during the reproductive/dauer decision and second during dauer differentiation. It has been established that DAF-16 acts via the insulin-dependent pathway and is involved in the inhibition of hormone production ( Gerisch et al. 2001 ; Jia et al. 2002 ), thereby controlling the reproductive/dauer decision ( Figure 8 , right branch). Our results show that, in addition, DAF-16 functions downstream of DAF-12 so that activation of the latter recruits it into neuronal nuclei ( Figure 8 , shown in red). The process of dauer formation, thus, is initiated by DAF-12 but needs DAF-16. A direct physical interaction between nuclear hormone receptors and forkhead domain (FOXO) transcription factors has recently been reported ( Schuur et al. 2001 ; Zhao et al. 2001 ; Dowell et al. 2003 ; Li et al. 2003 ). It is tempting to speculate that DAF-12 and DAF-16 can interact physically and that the activated DAF-12 can retain DAF-16 in the nucleus. Consistent with this, DAF-12 and DAF-16 have been coimmunoprecipitated in a recent in vitro study ( Dowell et al. 2003 ). Sterols and Longevity According to a current view, daf-16 is a major regulator of the longevity process ( Lin et al. 1997 ; Ogg et al. 1997 ). Reduction of DAF-2/IGF-1 signalling leads to activation of DAF-16 and to near-doubling of the life span of worms ( Kenyon et al. 1993 ; Morris et al. 1996 ). The inhibition of insulin receptor activity leads to the redistribution of FOXO transcription factors from cytoplasm into the nucleus and thus is a prerequisite for their activity ( Henderson and Johnson 2001 ; Lee et al. 2001 ; Lin et al. 2001 ). Our data show that in worms grown on lophenol, DAF-16 accumulates strongly in neuronal nuclei. The growth on lophenol, however, has no consequences on the length of the life span. A plausible explanation for this observation is that the activity of daf-16 influencing life span is tissue specific. In a recent study, Libina et al. (2003) have expressed DAF-16 in a daf-16;daf-2 double mutant under different tissue-specific promoters. Whereas expression of DAF-16 in the intestine led to the extension of the life span, expression in the neurons had no effect on longevity. This is consistent with our data showing that the DAF-12–dependent nuclear import of DAF-16 in neurons activates a different program from that in the intestine. Materials and Methods Materials Lophenol was purchased from Research Plus (Manasquan, New Jersey, United States). Electrophoresis-grade ultraPURE agarose was the product of Life Technologies (Paisley, Scotland, United Kingdom). Dulbecco's medium (DMEM) was from Invitrogen (Karlsruhe, Germany). Sterols, steroids, and antioxidant BTH were from Sigma (Sigma-Aldrich Chemie, Taufkirchen, Germany). All mutants except daf-12 (rh61rh411) and daf-9 (dh6) daf-12 (rh61rh411) were obtained from the Caenorhabditis Genetics Center. daf-12 and the double mutant daf-9 daf-12 were a kind gift of Adam Antebi (Max Planck Institute for Molecular Genetics, Berlin, Germany). The following mutant strains were used throughout the study: daf-22 (m130) II, daf-6 (e1377) X, daf-3 (mgDf90) X, daf-5 (e1386) II, daf-10 (e1387) IV, daf-12 (rh61rh411) X, daf-2 (e1370ts) III, and daf-16 (mgDf50) I. For imaging studies strains daf-16 (mu86) I; muIs71 [ pKL99 ( daf-16a::GFP/bKO )+ pRF4(rol-6) ]X and daf-16 (mu86) I; muIs61 (daf-16::GFP (pKL78)+rol-6(pRF4) were tested. Since the former, daf-16a::GFP/bKO, gave a brighter signal, as previously reported ( Henderson and Johnson 2001 ; Lee et al. 2001 ; Lin et al. 2001 ), results obtained with this strain are presented throughout the study. Preparation of sterol-depleted and sterol-containing plates for the propagation of worms Wild-type N2 Bristol and mutant strains were routinely propagated on NGM-agar plates as described in Brenner (1974) . To obtain cholesterol-free conditions, agar was replaced by agarose (extracted three times with chloroform) and peptone was omitted from plates. An overnight culture of the NA22 strain of E. coli was grown on a sterol-free culture medium DMEM. Bacteria were rinsed with M9 medium before use. For preparation of sterol-containing plates, cholesterol or lophenol was dissolved in methanol to the concentration of 5 mM and mixed 1:1 (v/v) with cholesterol-free bacterial suspension in M9. After evaporation of methanol in SpeedVac, the suspension was mixed with fresh bacteria to obtain the desired end concentration of tested substance. Bacterial suspensions were spread on cholesterol-free agarose plates. Light, fluorescence, and electron microscopy Light and confocal fluorescence microscopy were done using Zeiss Axioplan and Axiovert LSM 510 microscopes, respectively. For nuclear labelling, larvae were grown on medium containing 5 μg/ml of Hoechst (Molecular Probes, Eugene, Oregon, United States). One hour before imaging, larvae were transferred to medium without Hoechst, washed briefly with M9 medium, anaesthetised with 40 mM sodium azide in M9, and mounted on agarose pads. For electron-microscopic studies arrested larvae were washed two times with M9, harvested by centrifugation, and mixed with an equal volume of 2×Fixative (5% glutaraldehyde, 2% paraformaldehyde in M9). Worms were cut at room temperature with a razor blade on a microscopic slide, transferred to a centrifuge tube, and incubated for 2 h in a refrigerator. Afterwards worms were centrifuged and embedded in EmBed-812 (EMS, Ft. Washington, Pennsylvania, United States). Images were acquired by Tecnai 12 (FEI, Eindhoven, The Netherlands) or Phillips 400 electron microscopes. TLC of cholesterol metabolites from C. elegans To investigate cholesterol metabolism in C. elegans, 10-cm NGM agar plates were prepared without cholesterol. A quantity of 300 μl of bacterial suspension containing 13 μM cholesterol was supplemented with 4 μCi of [ 3 H]-cholesterol (70 Ci/mmol; Amersham Biosciences Europe, Freiburg, Germany). Worms were harvested from the plates with M9 medium and subjected to three cycles of freezing-thawing, and lipids were extracted by the Bligh and Dyer method ( Bligh and Dyer 1957 ). Eggs derived from mothers fed with radioactivity were put on sterol-free plates and propagated for two generations. Equal amounts of eggs and L1 larvae estimated by counting of aliquots were extracted and analysed by TLC as described above. TLC was performed on glass-backed plates of silica gel 60 (Merk, Darmstadt, Germany). Solvents used for the separation of cholesterol metabolites were chloroform-methanol (24:1). After chromatography, plates were sprayed with a scintillator (Lumasafe, Lumac LSC B.V., Groningen, The Netherlands) and exposed to a film (Hyperfilm MP, Amersham Biosciences Europe, Freiburg, Germany). We quantified relative amounts of ms and nms by scanning films exposed to radioactivity for short time and using Adobe Photoshop software. Regio- and stereospecific synthesis of 4α-substituted 5α-cholestan-3β-ols The synthesis of 4α-substituted 5α-cholestan-3β-ols is described in the Supporting Information . Preparation and HPLC fractionation of a lipidic extract from worms Worms of mixed population from 150 15-cm plates were collected by rinsing with ice-cold water and left overnight at 4 °C to sediment. The final volume of the sediment was about 150 ml. After decantation, aliquots of the worm suspension were transferred into 50-ml Falcon tubes and subjected to three cycles of freezing in liquid N 2 and thawing by sonication in an ultrasound bath at 37 °C. Worm suspension was then transferred into a glass bottle, 19 volumes of methanol containing 10 μg/ml of antioxidant BHT was added, and extraction was performed overnight at room temperature under continuous agitation. Extract was separated from worm remnants by filtration through a Whatman GF/A glass filter and remnants were reextracted with a fresh portion of methanol. Methanol extracts were combined and extracted two times with one volume of hexane. The obtained hexane extract was washed twice with a methanol-water mixture (9:1), dried under N 2 flow, and dissolved in 7 ml of hexane. In order to dispose very hydrophobic substances, the extract was subjected to a solid-phase separation. A quantity of 200 μl of the hexane fraction was applied to a 20-ml LC-18-SPE cartridge (SUPELCO, Bellefonte, Pennsylvania, United States) equilibrated with methanol. Twenty millilitres of flow-through methanol was collected, dried under N 2 flow, and dissolved in 200 μl of methanol. Two preps (400 μl) were subjected to reverse-phase HPLC chromatography on an Alliance 2695 solvent module (Waters GmbH, Eschborn, Germany) linked to a Waters 996 photodiode array detector using an XTerra Prep MS C 18 10 μm 10 × 250-mm column (Waters). The elution protocol was as follows: 15% solvent A (20% methanol in water) and 85% solvent B (methanol) for 11 min, a gradient from 85% B to 100% B in 11 min, and 100% B for 18 min. The flow rate was 5 ml/min. Fractions of 2 min were collected, dried, dissolved in 400 μl of isopropanol, and stored at −80 °C until use. Assay for gamravali Testing of the biological activity of HPLC fractions was performed in 12-well cell culture plates (Nunc, Roskilde, Denmark). Each well contained 1 ml of sterol-free agarose mixed with 0.1% tergitol. A quantity of 100 μl of HPLC fractions was added per well and dried in the laminar flow cabinet. Before seeding worms, 30 μl of sterol-free bacteria containing 10 μM lophenol was added to plates and left to stay overnight at room temperature. Worms for the bioassay were prepared as follows. The first generation of adult worms grown on lophenol (see above) was bleached, and eggs were placed on sterol-free plates without food and were kept for 3 d to obtain synchronised L1 larvae. About ten starved L1 larvae were placed in each well with HPLC fractions at room temperature. After 4 d worms were scored and the activity of fractions was represented as the percentage of worms that reached L4 or adult stages. Quadruplicates of each fraction per experiment were analysed. Compounds tested to rescue the dauer larva formation in the presence of lophenol Pregnenolone, testosterone, estrone, β-estradiol, progesterone, androstenol, vitamin D 3 , ecdysone, 20-hydroxyecdysone, 7α-hydroxycholesterol, 7β-hydroxycholesterol, 19-hydroxycholesterol, 20-hydroxycholesterol, 22-hydroxycholesterol, 24-hydroxycholesterol, 26-hydroxycholesterol, cholic acid, dehydrocholic acid, deoxycholic acid, litocholic acid, taurodeoxycholic acid, and chenodeoxycholic acid were tested. 7α-hydroxycholesterol, 19-hydroxycholesterol, and 26-hydroxycholesterol were from Steraloids (Newport, Rhode Island, United States); all others were from Sigma. Generating a double null mutant for daf-12 and daf-16 and a transgenic line expressing DAF-16::GFP in daf-12 null background The double mutant daf-16(mu86) I; daf-12 (rh61rh411) X was generated by crossing daf-16 (mu86) I; muIs71 [ pKL99(daf-16a::GFP/bKO) + pRF4(rol-6) ]X hermaphrodites with daf-12 (rh61rh411) males. Progeny displaying no Roller phenotype and fluorescence and able to grow on lophenol were selected. The daf-16(mu86) mutation was identified by PCR. Consequently, the mutations were verified by sequencing. The double mutant daf-16(mu86) I; daf-1 2 (rh61rh411) X was then used to generate daf-16 (mu86) I; daf-12 (rh61rh411) muIs71 [ pKL99(daf-16a::GFP/bKO) + pRF4(rol-6) ]X worms by backcrossing to the original daf-16 (mu86) I; muIs71 [ pKL99(daf-16a::GFP/bKO) + pRF4(rol-6) ]X. The mutations were identified and verified in a way similar to that used for the double mutant. Life span and thermotolerance The life span and thermotolerance of worms were investigated according to the method of Gems et al. (1998) . Studies with N2 animals were performed on plates containing cholesterol or lophenol at 20 °C. Day 0 corresponded to L4 stage. The life spans of about 300 worms per condition were investigated. Supporting Information Figure S1 Comparison of Cholesterol Metabolism in Wild-Type, daf-12, and Double Mutant daf-9 daf-12 Worms (3.8 MB PDF). Click here for additional data file. Protocol S1 Regio- and Stereospecific Synthesis of 4α-Substituted 5α-Cholestan-3β-ols (114 KB DOC). Click here for additional data file. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517820.xml |
511074 | Biochemical enrichment and biophysical characterization of a taste receptor for L-arginine from the catfish, Ictalurus puntatus | Background The channel catfish, Ictalurus punctatus , is invested with a high density of cutaneous taste receptors, particularly on the barbel appendages. Many of these receptors are sensitive to selected amino acids, one of these being a receptor for L-arginine (L-Arg). Previous neurophysiological and biophysical studies suggested that this taste receptor is coupled directly to a cation channel and behaves as a ligand-gated ion channel receptor (LGICR). Earlier studies demonstrated that two lectins, Ricinus communis agglutinin I (RCA-I) and Phaseolus vulgaris Erythroagglutinin (PHA-E), inhibited the binding of L-Arg to its presumed receptor sites, and that PHA-E inhibited the L-Arg-stimulated ion conductance of barbel membranes reconstituted into lipid bilayers. Results Both PHA-E and RCA-I almost exclusively labeled an 82–84 kDa protein band of an SDS-PAGE of solubilized barbel taste epithelial membranes. Further, both rhodamine-conjugated RCA-I and polyclonal antibodies raised to the 82–84 kDa electroeluted peptides labeled the apical region of catfish taste buds. Because of the specificity shown by RCA-I, lectin affinity was chosen as the first of a three-step procedure designed to enrich the presumed LGICR for L-Arg. Purified and CHAPS-solubilized taste epithelial membrane proteins were subjected successively to (1), lectin (RCA-I) affinity; (2), gel filtration (Sephacryl S-300HR); and (3), ion exchange chromatography. All fractions from each chromatography step were evaluated for L-Arg-induced ion channel activity by reconstituting each fraction into a lipid bilayer. Active fractions demonstrated L-Arg-induced channel activity that was inhibited by D-arginine (D-Arg) with kinetics nearly identical to those reported earlier for L-Arg-stimulated ion channels of native barbel membranes reconstituted into lipid bilayers. After the final enrichment step, SDS-PAGE of the active ion channel protein fraction revealed a single band at 82–84 kDa which may be interpreted as a component of a multimeric receptor/channel complex. Conclusions The data are consistent with the supposition that the L-Arg receptor is a LGICR. This taste receptor remains active during biochemical enrichment procedures. This is the first report of enrichment of an active LGICR from the taste system of vertebrata. | Background The initial event in taste transduction involves recognition of taste stimuli by plasma membrane-associated receptor proteins. These proteins are concentrated at the apical end of specialized neuro-epithelial cells (taste cells) found within multicellular end-organs known as taste buds [ 1 , 2 ]. The recognition binding sites for most taste stimuli face the exterior environment. The interaction of a taste stimulus with this recognition site triggers a chain of metabolic and ionic events in the taste cell, leading to alterations in membrane conductance, release of neurotransmitter, and a change in the firing rate of the afferent sensory nerve fibers with which taste cells synapse [ 2 ]. Receptor recognition is, therefore, largely responsible for maintaining the specificity of the taste transduction process. To date, 7-transmembrane G protein coupled receptors (7TM-GPCR's) for three taste modalities have been identified by both molecular cloning and through searches of the human and mouse genome. Sweet taste stimuli appear to be recognized by at least one heterodimer (T1R2/T1R3) of the three member family of 7TM-GPCR's, the T1R's [ 3 - 7 ]. The taste receptors for sweetness are coupled to changes in intracellular levels of either cyclic nucleotides or polyphosphoinositols [ 5 , 8 - 10 ]. Two GPCR receptor types have been implicated in the basic taste of umami (glutamate taste). One is the heterodimer of T1R1/T1R3, of the same 7TM-GPCR family as the sweet taste receptor dimer [ 11 ]. Another GPCR umami receptor is an N-terminal truncated metabotropic-type 4 glutamate receptor (taste/mGluR4) presumably coupled to an inhibition of adenylyl cyclase [ 12 ]. A third proposed, non-GPCR umami receptor is an NMDA-type ionotropic glutamate receptor [ 13 ]. Finally, a family (~40 members) of 7TM-GPCR's recognizes many bitter taste stimuli [ 14 , 15 ]. These bitter taste receptors are coupled through a gustducin-containing G protein [ 16 ] to changes in intracellular levels of cyclic nucleotides and polyphosphoinositide metabolites [ 17 - 19 ]. While these recent discoveries have markedly improved the understanding of taste transduction, it is apparent from neurophysiological, biophysical and biochemical studies that receptors and transduction processes other than the GPCR/second messenger systems are utilized by the sense of taste [ 2 , 20 ]. For example, several taste transduction processes make use of ion channels as the receptor recognition step [ 21 ]. Salty taste is likely transduced by an epithelial sodium channel (ENaC), and sour taste may also make use of channels such as acid sensing ion channels (ASICs) [ 22 ] and the hyperpolarization-activated, cyclic nucleotide-gated cation channel (HCN) (reviewed by [ 2 ]). Certain stimuli, such as quinine and perhaps denatonium co-opt potassium channels to alter membrane conductance of taste receptor cells [ 23 - 25 ]. Finally, in a variety of species, ligand-gated ion channels have been implicated as taste receptors for a number of stimuli, including sugars in the dog [ 26 ], glutamate in mouse [ 13 , 27 ], nicotinamide in crayfish [ 28 ], sugars and amino acids in fleshfly [ 29 ], bitter compounds in frog [ 30 ], and apparently for amino acids in the channel catfish, Ictalurus punctatus [ 31 , 32 ]. Little is known about the structure and function of these ligand-gated ion channel receptors (LGICR) in the taste system nor the extent to which they serve as taste receptors in other species. To evaluate the role of LGICRs in taste transduction, receptors of this class need to be identified and fully described. To date, a well characterized example of a likely LGICR class of taste receptors is found on the common channel catfish, I. punctatus . The catfish is an advantageous model system for studying taste transduction [ 33 ] because it possesses a large number of densely arrayed taste buds across its body surface, particularly on its barbel appendages and gill rakers [ 33 - 36 ], and shows high specificity and sensitivity to selected amino acids. Several taste transduction pathways for amino acids have been identified both biochemically and neurophysiologically, including those recognizing (1) L-alanine and other small neutral amino acids, (2) L-proline, and (3) L-arginine (L-Arg) [ 33 , 37 - 40 ]. Of these three receptor systems, the one tuned to the amino acid, L-Arg, appears to be of particular high specificity and affinity [ 38 , 41 ]. Calcium imaging studies on isolated catfish taste receptor cells suggest the presence of at least two subtypes of L-Arg-stimulated transduction pathways. In one, L-Arg induces a change in intracellular calcium that is independent of extracellular calcium activity. In the other, L-Arg induces an increase in intracellular calcium that is dependent upon extracellular calcium. This second type of L-Arg induced response is blocked by D-Arg, whereas the first type of L-Arg induced response is less sensitive to the D-isomer [ 42 , 43 ]. Intracellular and patch clamp studies on catfish taste cells are also consistent with there being two subtypes of responses to L-Arg [ 42 ]. The L-Arg induced increase in intracellular calcium independent of extracellular calcium and not blocked by D-Arg may utilize a mechanism such as a GPCR-polyphosphoinositol linked pathway with IP3 releasing calcium from intracellular stores. The other L-Arg-stimulated pathway, showing intracellular changes in calcium dependent upon extracellular calcium and blocked by D-Arg, is consistent with an LGICR mechanism. The LGICR mechanism was given additional credence through studies demonstrating that membranes from barbel epithelium (that contains taste buds), when reconstituted into lipid bilayers, show L-Arg-stimulated ion channel activity inhibited by D-Arg [ 31 - 33 ]. Using membrane homogenates from catfish barbel epithelium, biochemical binding studies revealed high affinity sites for L-Arg that were inhibited by D-Arg, L-arginine methyl ester, and to a lesser extent by L-lysine and L-α-amino-β-guanidino propionic acid [ 38 ]. Previous studies had also demonstrated that both of the lectins, Ricinus communis agglutinin I (RCA-I) and Phaseolus vulgaris Erythroagglutinin (PHA-E), inhibited the binding of L-Arg to its presumed receptor sites, and both lectins labelled identical SDS-PAGE-separated bands of a barbel membrane preparation [ 44 ]. L-Arg-stimulated ion channel activity of solubilized barbel (taste) membranes reconstituted into lipid bilayers was inhibited by both PHA-E [ 35 ] and RCA-I (Teeter & Brand, unpublished observations). In addition, PHA-E also labelled the apical membrane of selected taste bud cells and solitary chemoreceptor cells of catfish barbel [ 35 ]. When reacted against the exterior epithelium of fixed, unpermeabilized catfish barbel, polyclonal antibodies developed against these lectin reactive peptides immuno-labelled the apical membrane of a subset of barbel taste receptor cells [ 45 ]. While these prior studies are consistent with the hypothesis that one of the receptors for L-Arg is an LGICR, both the neurophysiological studies and the reconstitution experiments could not definitively test this hypothesis. Without actually isolating an LGICR for L-Arg, it remains possible that a GPCR moiety for L-Arg could conceivably be tightly coupled to a separate ion channel. The purpose of the studies reported here was to investigate the possibility that an LGICR exists for L-Arg, by: 1. biochemically enriching this putative LGICR to near homogeneity, and 2. biophysically characterizing the L-Arg-stimulated ion channel activity of the presumed LGICR at each step of enrichment to • demonstrate that the same LGICR is being enriched with each step, and • demonstrate that the ion channel properties and kinetics of the enriched LGICR are similar to those of the presumed LGICR in situ . This enrichment and characterization are necessary steps towards eventually cloning this putative LGICR. The data are consistent with the interpretation that a receptor for L-Arg can be solubilized in an active state and can be enriched to a point where, upon denaturation and SDS-PAGE, material containing receptor-like activity elutes as a single band. Results Lectin specificity: lectin blots and lectin histochemistry An SDS-PAGE of a detergent-solubilized epithelial membrane fraction from barbel of I. punctatus , called "Sp" (defined in methods), revealed numerous proteins labelled by silver stain (Fig. 1 , Lane "Sp"). Yet the lectins, PHA-E and RCA-I, both clearly labelled only one major glycoprotein band, in the range of 82 – 84 kDa (Fig. 1 , Lane "PHA" and Lane "RCA"). A few other protein bands were more lightly labelled by these two lectins, including ones near 88 – 90 kDa and ~120 kDa. This recognition specificity is notable because previous work had shown that both of these lectins inhibited binding of L-Arg to a membrane suspension of barbel epithelium with respective specificity confirmed by control studies using the hapten sugar [ 44 ]. Other lectins that did not inhibit L-Arg binding labelled other glycoproteins of barbel epithelium [ 44 ]. Previous studies showed that the lectin, PHA-E, labelled primarily exterior-facing (presumably) glycoprotein motifs of the taste buds of catfish barbel [ 35 ]. However, no comparable labelling studies were carried out using RCA-I. Since RCA-I was used in this current study as an affinity reagent, it was important to establish the labelling specificity of RCA-I – conjugated lectin to catfish barbel taste buds. Figure 2A shows that RCA-I labels primarily the taste buds on the surface of the barbel, with Figure 2B showing labelling of two taste buds in a transverse section at higher magnification. Figure 2C demonstrates that the lectin labelled primarily the apical region of the taste bud. Some of the spotty labelling scattered among the taste bud field (Fig. 2A ) may due to the RCA-I recognizing an epitope on solitary chemoreceptor cells (SCC). The SCCs are a dispersed chemoreception system found in aquatic vertebrates, and are possibly related to the taste system [ 35 , 46 , 47 ]. Apparent labelling of the catfish SCC was reported with PHA-E as well [ 35 ]. The specificity shown by these lectins in binding inhibition, ion channel conductance inhibition, lectin blots, and lectin histochemistry, all give credence to use of lectin affinity chromatography as the first step in enrichment of L-Arg-stimulated channel activity. Enrichment of L-Arg-stimulated ion channel activity Step 1 – lectin chromatography Since the lectins, RCA-I and PHA-E, inhibited the binding of L-Arg but not of L-alanine [ 44 ], the assumption was made that an L-arginine receptor (L-ArgR), or a fragment thereof, was among the glycoproteins labelled by these two lectins, and lectin affinity chromatography was, therefore, chosen as the first step in enrichment of this L-ArgR. Verification of the presence of active putative L-ArgR ion channel in each eluted fraction was assessed by a bilayers – incorporation assay. This assay proved to be more readily performed and provided more reproducible results than soluble binding assays that yielded high non-specific binding and required much more material. A quantitative protein assessment of the eluted materials from the RCA-I column indicated that over 99% of the protein from fraction Sp passed through the column unbound. Only occasionally did this protein material contain minimal L-Arg-stimulated ion channel activity. In contrast, the remaining protein eluted from the affinity column by galactose and reconstituted into lipid bilayers (See below) consistently contained ion channels activated by L-Arg and inhibited by D-Arg. Silver staining of SDS-PAGE separated proteins before (Fig. 3A ) and after (Fig. 3B ) the RCA affinity enrichment step revealed numerous proteins from total Sp (Fig. 3A ) with fewer and more heavily stained bands of protein present in galactose-eluted fractions (Fig. 3B ). These included primarily protein of molecular weight ~82–84 kDa, with apparently lower abundance proteins near 115–120 kDa, 60–70 kDa, and 40–45 kDa (Fig. 3B ). The 82–84 kDa band matched the general position of the principal protein labelled in the lectin blots (Fig. 1 ). The protein eluting at 115–120 kDa in Figure 3B may correspond to the lightly labeled glycoproteins at ~120 kDa seen in the lectin blots of Figure 1 . Protein of molecular weight 60 – 70 kDa and 40 – 45 kDa observed in the galactose-eluted protein fraction of Figure 3B have no apparent match in the lectin blots of Figure 1 . These may, possibly, be degradation products of the other protein fractions that are labelled by the lectins or they may be proteins of low abundance, visible here due to the enrichment of protein resulting from the affinity procedure. On the assumption that the 82–84 kDa protein was at least a subunit of this L-ArgR, the SDS-PAGE band at 82–84 kDa was electroeluted and used to develop polyclonal antibodies from three guinea pigs (See Methods). The affinity purified and pre-treated polyclonal antibodies from two of these guinea pigs proved most specific and these were labelled, "GP1" and "GP2," respectively. In Western blots of fraction Sp (Fig 4A ), both of the GP antibodies labelled a wide band between 74 and 84 kDa (containing its antigen) and occasionally a second higher molecular weight band near 110 kDa. In Western blots of RCA lectin-galactose eluted proteins, a narrower band of 82–84 kDa was labelled (Fig. 4B ). Little GP labelling was seen in Western blots from SDS-PAGE of the protein not retained by the RCA lectin column, but those bands that were labelled were near 84 kDa and 110 kDa (data not shown). The GP polyclonal antibodies were used as a confirmatory marker of the 82–84 kDa peptide during subsequent enrichment steps. Both the GP1 and GP2 antibodies labelled the 82–84 kDa protein band in Western blots of an SDS solubilized partial membrane preparation from catfish barbel and both immuno-labelled taste cells of the catfish (see ahead). Step 2 – gel filtration The CHAPS-solubilized, dialyzed, non-denatured protein eluted from the galactose wash of the lectin column was applied to a Sephacryl S-300 HR column and eluted with Tris/CHAPS. Each protein-containing peak of the elution was assayed for L-Arg-stimulated channel activity and subjected to SDS-PAGE with silver staining and Western blotting against the GP1 antibody. Only protein from the first peak, in fractions 1 and 2, (eluting at an equivalent molecular weight of > 670 kDa.) contained L-Arg-stimulated ion channel activity (see ahead). Silver staining of material in these first two fractions run on SDS-PAGE revealed a prominent band at 82–84 kDa (Fig 3C ). The broad band near 110–115 kDa seen after lectin chromatography (Fig. 3B ) was not seen while protein at 60–70 kDa and 40–45 kDa remained. The corresponding Western blot (Fig. 4C ) demonstrated that the antigen to which GP1 was developed was still present in the active fraction. Step 3 – ion exchange chromatography As a final enrichment step eluted material from fractions 1 and 2 of the Sephacryl column were lyophilised and resuspended into Start Buffer (see methods) and loaded on a Hitrap Q anion exchange column. Both the pH 9 and pH 8 elution resulted in protein being released from the column, with the majority of protein eluting at pH9. However, only the pH 9 fraction contained L-Arg-stimulated ion channel activity when reconstituted into lipid bilayers. A silver stain of an SDS-PAGE of the pH 9 eluent showed a deeply staining band at 82–84 kDa along with weak staining above 200 kDa and in the 35 kDa range (Fig 3D ). There was also an almost complete loss of other stained bands observed in SDS-PAGE of protein from the previous enrichment steps (compare Fig. 3D with Figs. 3C and 3B ). The corresponding Western blot of 0.2 μg of the pH 9 eluent showed very strong reactivity at 82–84 kDa (Fig. 4D ). The SDS-PAGE of the pH 8 eluent showed a faint band at 82–84 kDa with other less intense bands at lower molecular weight. This pH 8 eluent may contain an inactive or partially denatured form of the L-ArgR. Immunohistochemistry of GP1 and GP2 on catfish barbel The GP antibodies were developed against the 82–84 kDa fraction of solubilized catfish barbel membranes, since it was a band of this molecular weight that was labelled by the lectins, PHA-E and RCA-I. While it is expected that GP1 and GP2 would label an 82–84 kDa band by Western blots, the fact that the GP1 and GP2 antibodies faithfully marked each enriched fraction that exhibited L-Arg-stimulated ion channel activity (see below) suggests that they are immuno labels of the receptor. As such, their localization within the barbel may be a marker for this presumed L-ArgR. Figure 5 shows GP1 and GP2 immuno labelling of paraformaldehyde fixed, sectioned catfish barbel. Figure 5A , a low power image of the barbel sectioned length-wise, demonstrates that it is primarily the taste buds that are labelled by GP1 (1/8000 dilution). Figure 5B shows labelling by GP2 (1/12000 dilution) of the apical area of three taste buds from a surface viewpoint. Figures 5C and 5D show taste bud labelling by GP1 (1/16000 dilution) and GP2 (1/8000 dilution), respectively of taste buds in tangential sections. Figure 5E shows a negative control where the primary antibodies were omitted. These immunohistochemical studies suggest that the antigen epitopes for GP1 and GP2 are concentrated at the apical portion of taste buds. Characterization of the RCA lectin-, Sephacryl S 300 gel-, and ion exchange-protein reconstituted into lipid bilayers LGICR enrichment was followed and verified by measuring the L-Arg-stimulated conductance of lipid bilayers (equimolar mixture of DOPS:DOPE) into which the protein fractions derived from each purification step were fused. Nearly identical L-Arg-stimulated single channel activity was observed from active fractions of all three steps: the galactose-eluted protein from the RCA lectin column, the protein of the first peak, fractions 1 & 2, of the material eluted from the Sephacryl S-300 column, and the protein of the pH 9 elution from the ion exchange column. The fact that consistent and nearly identical L-Arg-stimulated activity was observed with material from each subsequent enrichment step indicates that the enrichment procedures were sufficiently benign so as to permit the retention of LGICR-type activity and, presumably, native receptor conformation. General agonist/antagonist channel properties. Figure 6 illustrates single channel activities observed during the enrichment steps. Since active material from each of the three steps yielded almost identical channel properties (See Table 1 ), only that activity seen with fractions 1 and 2 (combined) off of the Sephacryl S-300 column is shown here. The data of Figure 6 are from the same experiment. In lipid bilayers into which active material had been fused, but in the absence of added L-Arg, no spontaneous channel activity was observed (Fig. 6 top panel ). Addition of 10 μM L-Arg to the cis side buffer solution induced appearance of ion channels (Fig. 6 middle panel ). L-Arg-stimulated channel activity in positive fractions from all columns was readily blocked by the addition of 100 μM D-Arg to the same side of the chamber wherein L-Arg had been added (Fig. 6 bottom panel ). (In control experiments with bilayers into which no protein was incorporated, neither L-Arg nor D-Arg alone (in a range 10 – 1000 μM) induced channel activity.) While 10 μM L-Arg was generally used in the screening and assay procedures, we estimate that the threshold for L-Arg-induced channel activity of the solubilized putative L-ArgR is about 1 μM L-Arg. In addition to inhibition by D-Arg, the lectins, RCA-I and PHA-E (not shown here, but see [ 35 ]), also inhibited the L-Arg-induced ion channel activity. In contrast, none of the antibodies developed to the denatured 82–84 kDa proteins inhibited L-Arg stimulated ion channel activity. Neither L-alanine nor L-proline (up to 200 μM) activated the ion channels stimulated by L-Arg. Amplitude histogram Figure 7 illustrates an all points amplitude histogram of L-Arg activated channels from material contained in fractions 1 & 2 from the Sephacryl column elution step measured at a fixed transmembrane potential of -100 mV. These data imply that there is one major peak of current amplitude (with some fluctuation) giving a unitary current of -6.6 pA. Similar unitary currents were measured for material from both the lectin column and the ion exchange procedure. Cation/anion selectivity & current/voltage relationship The cation-anion selectivity of L-Arg activated channels found in material contained in fractions 1 & 2 off the Sephacryl column was determined for Na + and Cl - . A potential of zero current (reversal potential) was measured after formation of a 4-fold transmembrane concentration gradient of electrolyte (100 mM NaCl at cis side and 25 mM NaCl at trans side) across the bilayer containing a few ion channels. The average reversal potential (n = 5 bilayers, ± S.D.) elicited by voltage ramps (see Fig. 8 ) was -14 ± 3 mV. This value corresponds to weak cation selectivity (P Na /P Cl = 2.2). The current-voltage relationships for L-Arg activated channels formed by active protein fractions from all three enrichment steps are illustrated in Figure 9 . The data are well fit by a linear regression (r = 0.99) with slopes of 58, 67, and 73 pS for channels formed by protein from the RCA-I lectin column (open circles), protein from Sephacryl S-300 HR column (solid boxes) and protein from the pH 9 fraction of the ion-exchange column (triangles), respectively. From these current-voltage relationships it follows that the conductance of the channels from protein at any stage in the enrichment process are nearly identical, and that L-Arg activated ion channels are essentially potential-independent. Comparison with biophysical properties of native channels The electrophysiological properties of ion channels formed by the protein fractions throughout the enrichment scheme described here closely resemble those measured for the native channels [ 31 ] (see Table I ). Both native channels and channels formed by proteins after these enrichment steps 1. are activated by L-Arg and inhibited by D-Arg over the same concentration ranges; 2. display nearly the same unitary conductance (The membrane-associated channels show two conductance states, one at 40 – 60 pS, the other at 75 – 100 pS (Table I and [ 31 ])); 3. are cation selective; and 4. are potential independent. Discussion Ligand-gated ion channel receptors (LGICR) may be used for selected stimuli of several taste modalities. In spite of their likely role in taste, little is known about these LGICRs [ 21 ]. The best characterised apparent taste LGICRs that recognise non-ionic stimuli are those of the catfish, I. punctatus . This animal possesses apparent LGICRs of low affinity for L-proline and of high affinity for L-Arg [ 31 , 33 ]. The observation that L-Arg acts as a stimulus for the taste system of the channel catfish was first reported by Caprio [ 48 ]. Subsequent neurophysiological and biochemical binding studies demonstrated that, unlike most other vertebrate taste receptors, the catfish taste receptor(s) for L-Arg is of both high specificity and high sensitivity [ 32 , 38 , 41 ]. Contemporaneous neurophysiological cross-adaptation and single unit studies indicated that L-Arg stimulates unique sites independent of those for other amino acids such as L-alanine or L-proline [ 49 , 50 , 40 ]. The receptor sites for L-Arg have narrow structural requirements, with only a few structural analogs of L-Arg acting as cross-adapting stimuli [ 41 ]. The receptor binding studies found a high affinity site for L-Arg, with Kd of 20–50 nM, and demonstrated inhibition of L-Arg binding by D-Arg, L-arginine methyl ester, and to a lesser extent by L-lysine and L-α-amino-β-guanidino propionic acid [ 38 ]. Other amino acids were without effect at reasonable levels. Interestingly, L-Arg and D-Arg are non-reciprocal cross adaptors, where neural adaptation to D-Arg eliminates responses to L-Arg, while adaptation to L-Arg still leaves some response to D-Arg [ 40 ]. This non-reciprocal cross-adaptation predicts the presence of a receptor site for D-Arg and suggests that any receptor for L-Arg – be it a GPCR or an LGICR – should be sensitive to D-Arg. The fact that no responses to D-Arg were ever observed in the bilayer experiments suggests that the major receptor for D-Arg is a GPCR. Consistent with these receptor specificities is the behavioural observation that at micromolar levels, L-Arg induces oropharyngeal motor behavior in I. punctatus [ 33 , 51 , 52 ] with D-arginine acting as a partial antagonist of this behavior [ 33 ]. More recently, whole cell patch clamp and calcium imaging of isolated catfish taste receptor cells, along with earlier in situ intracellular electrophysiological recordings, indicated that the majority of L-Arg-induced depolarizations are generated by inward currents [ 42 , 43 ]. As predicted by neurophysiological cross-adaptation studies [ 40 ] and consistent with biochemical binding experiments [ 38 ] the increases in intracellular Ca 2+ activity observed in taste cells stimulated by L-Arg could be blocked by D-Arg [ 43 ]. Studies in our laboratory demonstrated that plasma membrane vesicles from barbel epithelium incorporated into lipid bilayers displayed L-Arg (μM)-stimulated ion channel activity that was inhibited by D-Arg [ 31 , 32 , 35 ]. No channel activity was observed toward L-alanine, but another, apparently less abundant, channel was stimulated by mM levels of L-proline, with the L-proline response being inhibited by D-proline. The L-Arg-stimulated responses were not inhibited by D-proline, nor were the L-proline responses inhibited by D-Arg [ 31 ]. The L-Arg-stimulated channels were found to be 50–80 pS in size, cation selective, but of low ion specificity. In contrast to the taste system, the olfactory system of the catfish transduces the stimulus, L-Arg and some other basic amino acids apparently through a GPCR [ 53 ] as does the olfactory system of goldfish [ 54 ]. In addition, L-Arg is an appetitive stimulus for the leech, Hirudo medicinalis, where the transduction process for L-Arg can be influenced by bitter stimuli, suggesting an integration at the receptor cell level [ 55 ] Preliminary reports on the localization of an L-ArgR showed that the antibodies, GP1, and the lectin, PHA-E, when incubated with intact, unfixed barbels, labelled exterior-facing epitopes on catfish barbel taste buds [ 35 , 45 ] and SCCs scattered in the epithelium among taste buds [ 35 ]. Immunoelectron microscopy using GP1 revealed labelling primarily on those cells of the taste bud containing large microvilli [ 35 ]. Because these data were of surface labelling only, the labelling specificity towards other areas of the taste bud and the barbel epidermis for GP1 and the lectins was not known. This current report demonstrates primarily apical labelling of taste buds by both conjugated RCA-I, GP1 and GP2 (Figs 2 and 5 ). The scattered punctuate labelling seen with both RCA-I and the GP antibodies may represent epitope recognition on solitary chemoreceptor cells (particularly within the epidermis), the apical processes of which were previously shown to label with PHA-E [ 35 ]. As the secondary antibody controls suggest (Fig. 5E ) very little of this punctuate labelling can be attributed to a second antibody effect. Considering collectively the neurophysiological, biochemical, behavioural, biophysical and localization data, a putative receptor for L-Arg emerges as one of high structural specificity, with D-Arg acting as an antagonist, one of high sensitivity, and one expressed in the apical membrane of a specific sub-class of taste receptor cells. Yet, because the previous biophysical studies were carried out with intact cells, epithelial homogenates or reconstituted membrane vesicles, the data are insufficient to permit a distinction between the L-ArgR as a single LGICR macromolecular complex and the receptor as two separate entities, a recognition molecule coupled to ion channel activity. To help make this distinction, solubilization and enrichment of the receptor were required. We assumed that if receptor activity survived solubilization and increasing enrichment, and if this receptor appeared to purify as a unitary entity by SDS-PAGE, then it is likely that the major receptor for L-Arg is indeed a LGICR. This enrichment procedure was also a necessary first step in cloning the receptor, since partial amino acid sequences may be obtained from the purified product. Enrichment of the putative L-ArgR The initial enrichment step of lectin affinity was predicated upon the observation that the lectins, RCA-I and PHA-E, inhibited the specific binding of L-Arg to its presumed receptor sites [ 44 ], and that PHA-E and RCA-I inhibited L-Arg-stimulated conductance activity of catfish barbel (taste) membranes reconstituted into lipid bilayers [ 35 ]. In addition, both lectins labelled only a few protein bands of an SDS-PAGE of barbel Sp, and only one major band, that near 82–84 kDa, was common between the two (Fig. 1 ). Both conjugated RCA-I and PHA-E labelled cells within the taste buds (Fig. 2 and [ 35 ]). Lectin affinity chromatography employing agarose bound RCA-I was therefore used to achieve an initial partial enrichment of the putative L-ArgR. RCA-I was chosen for lectin affinity because in our hands it was more stable and easier to work with than the agarose bound PHA-E. In pilot studies, lectin affinity chromatography with PHA-E led to results similar to those obtained with RCA-I. The two additional enrichment procedures using Sephacryl-S300 and Hitrap Q ion exchange were chosen to further enrich the putative L-ArgR because both could be carried out using solubilization buffers that were less likely to destroy the activity of the receptor and both are standard biochemical purification techniques. Each chromatography step that retained L-Arg-stimulated ion channel activity resulted in increasingly concentrated and increasingly purified protein of molecular weight near 82–84 kDa. This enrichment is readily seen in SDS-PAGE of Figure 3 , where the protein profile changes dramatically over the course of each chromatography step. The Western blots of material from each step (Figs. 4A,4B,4C,4D ) suggest a substantial enrichment of the 82–84 kDa protein. Throughout enrichment, the 82–84 kDa band was consistently recognized by the GP1 and GP2 antibodies in Western blots (Fig. 4 ). These blots suggest that the same entity(ies) labelled by the lectins was retained and enriched through the course of each chromatography step. The Western blots also speak to the specificity of the antibodies, GP1 and GP2, in that both antibodies labelled almost exclusively the 82–84 kDa band. The subsequent observation that GP1 and GP2 recognized epitopes at the apical region of the taste bud (Fig. 5 ) indicates that at least a portion of the proteins in the 82–84 kDa region are taste cell-related, membrane associated and, given the biophysical characteristics, likely receptors. While the 82–84 kDa protein(s) may be a major constituent of the active ion channel complex, the actual molecular weight of the active receptor, and therefore some estimate of its quarternary composition, was difficult to determine. Attempts at running native gels led to inconsistent findings. Of the procedures used for enrichment, the Sephacryl column was the one that could, theoretically, at least, give an estimate of the size of the complex. Using this column, all of the L-Arg-stimulated ion channel activity was located within the initial eluted peak (fractions 1 & 2). Calibration of the column suggested a molecular weight of > 640 kDa for eluted material at this initial peak. However, this high apparent molecular weight may not represent the actual weight of the unitary LGICR, since, like many other LGICRs, the L-ArgR may form clusters [ 31 , 56 - 58 ]. Theoretically, use of Sephacryl 400HR should be able to resolve such high molecular weights. However, when Sephacryl 400HR was used, ion channel activity was spread across many eluted fractions, making estimates of unitary molecular weight impossible. After these enrichment procedures, the L-Arg-stimulated ion channel activity was retained and the 82–84 kDa protein fraction was greatly purified. The correlation of these two observations suggests that protein in the 82–84 kDa range is at least part of the L-ArgR. Biophysical characteristics of the putative L-ArgR The biophysical characteristics of the L-Arg-stimulated channels reconstituted from each step in the enrichment scheme remained nearly unchanged from those described for the channel observed in native membrane fragments [ 31 , 32 ] : 1. both channels were activated by the same range of concentration of L-Arg and blocked by the same concentration range of D-Arg (Fig. 6 ); 2. neither were activated by L-alanine nor by L-proline; 3. both displayed similar amplitude histograms (Fig. 7 and [ 31 ]); 4. both had similar unitary conductance (see Table 1 ), with the enriched channel displaying unitary conductance of 73 +/- 7 pS (Fig. 7 ), and the channel in situ displaying two conductance ranges, 40 – 60 and 75 – 100 pS. This difference is likely due to the difference in buffers, where the bilayer studies were performed in NaCl/CaCl2/MOPS, while the reconstituted membrane in situ studies were performed with a complex ringer buffer (Table 1 , Fig. 8 ); 5. both channel preparations when stimulated by L-Arg exhibit linear current voltage relationship (Fig. 9 ). The results of this study demonstrate that the isolated channel protein shows recognition-specificity for L-Arg and acts as a non-specific ion channel upon binding L-Arg, properties consistent with the in situ activity of the putative L-ArgR and consistent with expected taste receptor criteria. Conclusions Collectively, the data presented here suggest that one major taste receptor for L-Arg in the catfish, I. punctatus, is a ligand-gated ion channel receptor. The active receptor was biochemically enriched from taste bud-containing epithelium and biophysically validated. Immunohistochemical studies using an antibody raised against peptides labelled by lectins that inhibited the binding of L-Arg to a likely receptor revealed specific labelling at the apical region of the taste bud. Analogous with other LGICRs, taste receptor cell depolarization to L-Arg is suggested through L-Arg binding to a receptor site that is associated with and activates an ion channel of low ion selectivity. In the animal, activation of this channel by L-Arg or other select agonists will open the channel and allow influx of Na + and Ca 2+ , present in the mucus covering the taste epithelium [ 32 ], into the receptor cell. Alternatively, given the relatively low cation/anion ratio, at least part of this charge could be carried by efflux of Cl - . This flow of charge will result in cellular depolarization, release of neurotransmitter to the innervating sensory nerve, and transmission of the taste signal to the central nervous system. This enrichment procedure can be used to generate sufficient material for obtaining partial peptide sequences necessary for eventual cloning of this LGICR. Methods Animals Use of the channel catfish for these studies was approved by the Institutional Animal Care and Use Committee of the Monell Chemical Senses Center. Channel catfish, Ictalurus punctatus , purchased from local suppliers were usually euthanized on the day of delivery, but if not, were held less than 4 days in 250 gallon aquaria under dim light and fed commercial catfish chow. Chemicals All electrolytes, buffers and other chemicals were reagent grade from Sigma (St. Louis, MO). Water was deionized and further purified through a Milli-Q Plus PF system (Bedford, MA). 3-[(3-Cholamidopropyl)-dimethylammonio]-1-propanesulfonate (98%) (CHAPS), polyoxyethylenesorbitan monolaurate (TWEEN 20), phenylmethanesulfonyl fluoride (PMSF) and pepstatin A were purchased from Sigma. n-Octyl-β-D-gluco-pyranoside (n-octylglucoside) was purchased from Calbiochem (LaJolla, CA). The absolute enantiomer, L-arginine HCl, was purchased from Sigma, and the absolute enantiomer, D-arginine HCl, was a gift of the Ajinomoto Co., Tokyo, Japan. Sephacryl S-300 HR, High Range Gel Filtration Calibration Kits, and 1.0 ml Hitrap Q columns were purchased from Pharmacia Biotech (Piscataway, NJ). Agarose-bound lectins, Ricinus communis agglutinin I (RCA-I), Phaseolus vulgaris Erythroagglutinin (PHA-E), in their biotinylated forms, the ABC kits, and rhodamine-conjugated RCA-I were purchased from Vector Lab (Burlingame, CA). The second antibody, Cy3-conjugated goat anti-guinea pig IgG, was obtained from Jackson ImmunoResearch Labs. (West Grove, PA). The 4 CN Membrane Peroxidase Substrate System was purchased from Kirkegaard & Perry Laboratories (Gaithersburg, MD). Gels and ampholytes were purchased from BioRad (Hercules, CA). Protein was quantitated using a BioRad DC Protein Assay. Synthetic 1,2-dioleoyl- sn -glycero-3-phosphoserine (DOPS), 1,2-dioleoyl- sn -glycero-3-phosphoethanolamine (DOPE) and 1,2-dioleoyl- sn -glycero-3-phosphocholine (DOPC) were purchased from Avanti Polar Lipids, Inc. (Pelham, AL). Tissue homogenate preparation and solubilization Maxillary and mandibular barbels from approximately 50 euthanized channel catfish, ~25–40 cm long, were removed and placed in a beaker containing 100 ml of 50 mM Tris-HCl buffer (pH = 7.8), 100 mM NaCl, 1 mM EDTA ("TRIS Wash"). After one exchange of buffer, barbels were removed to a 50 ml polypropylene screw cap tube filled with TRIS Wash buffer and stored at -80°C until used. The preparation of a solubilized, plasma membrane-enriched fraction from catfish barbels was adapted from Kalinoski [ 59 ]. A typical homogenate/solubilization run used barbel tissue from 150 fish, prepared in aliquots of 50 fish at a time. The entire procedure was carried out at 4°C. The epithelium of thawed barbels from 50 fish at a time in 75 ml of TRIS Wash was stripped from the supporting pseudo-cartilage with two 10 second bursts (with a 10 second interval) from a hand-held Toastmaster Hand Blender, Model 1738 (Boonville, MO). The suspension was allowed to settle, and the supernatant decanted into a 600 ml beaker through two layers of USP Type VII gauze (Kendall Co., Boston, MA). Fifty ml of TRIS Wash (4°C) was added to the remaining settled barbels, and the suspension subjected to a third 10 second burst from the Blender. This entire second suspension was rapidly poured over the gauze layers into the same beaker. A second and third tube of barbels from 50 fish were treated identically, and the filtrate from all three combined, and the volume brought to 470 ml. This homogenate, divided into two aliquots, was centrifuged at 4000 × g for 15 min. The supernatants were recovered and centrifuged at 21,500 × g for 45 min. The pellets were retained. To solubilize the pellets from the 21,500 × g spin each pellet was recovered by two, 1 ml rinses of 50 mM TrisHCl (pH 7.8), 50 mM NaCl ("Low Osmolar Buffer") (total, 4 ml) and transferred to a 15 ml Teflon/glass homogenizer. Twenty milligrams of CHAPS and 40 μl of protease inhibitor-mix (0.275 mM pepstatin A, 57.5 mM PMSF, in ethanol) were added to the ~4 ml suspension in the homogenizer. The suspension was homogenized by ten slow strokes of the Teflon pestle using a rotating motor drive at moderate speed. The homogenate was transferred to a 15 ml capped tube, diluted to 10 ml with Low Osmolar Buffer, and placed on a Clay Adams Nutator rocker/shaker overnight at 4°C to solubilize the proteins. After the overnight agitation, the suspension was centrifuged for 1 h at 100,000 × g. The supernatant was recovered and is referred to as "Sp." This Sp was then used in the subsequent lectin affinity chromatography step. Lectin affinity chromatography All steps in the lectin affinity procedure were performed at 4°C. Agarose-bound RCA I affinity resin was pre-washed by adding 8 ml of gel slurry to a 125 ml conical glass tube and bringing the final volume to 14 ml with TRIS Wash. The tube was inverted several times, then centrifuged at 500 × g for 1 min. The supernatant was discarded and the wash step repeated three additional times with TRIS Wash buffer, then two more times with Low Osmolar buffer. The resulting agarose-bound RCA I gel, with a bed volume of 4 ml, was used for affinity chromatography. The 4 ml agarose-RCA gel in the 15 ml tube was combined with ~7 ml of Sp, and the mixture equilibrated on the Nutator tube rocker for 30 min. The mixture was poured into a 1 cm × 10 cm column and the effluent collected as 1 ml fractions at a flow rate of 6.0 ml/h. Effluent from the column was monitored by absorbance at 230 nm and/or 280 nm. To remove unbound protein, the column was washed with a sufficient volume (~12 ml) of Low Osmolar Buffer until no detectable protein eluted from the column. Proteins bound to the RCA resin were then eluted from the column with 10 ml of 50 mM Tris-HCl (pH = 7.8), 200 mM NaCl, 1 mM EDTA, 20 mM D-galactose. Protein from both the Low Osmolar elution step and the D-galactose elution step was reconstituted into lipid bilayers and assayed for L-Arg-stimulated ion channel activity (See ahead). The galactose-eluted protein fractions showing L-Arg-stimulated ion channel activity were pooled and dialyzed over night against 2000 ml Milli-Q water containing 0.05% CHAPS. Dialyzed samples were lyophilized and stored at -80°C. Gel filtration chromatography All steps in the gel filtration procedure were performed at 4°C. Lyophilized proteins from the galactose elution of the lectin column were prepared for further enrichment using size exclusion chromatography by dissolution in 300 μl of TRIS Wash with the addition of 0.2% CHAPS and 10% sucrose. This preparation was loaded onto a Sephacryl S-300 HR column (1 cm × 40 cm). Fractions of 0.4 ml were eluted from the column with a buffer composed of TRIS Wash plus 0.2% CHAPS. The column was run at 2.4 ml/hr, and the effluent absorbance monitored at 230 nm and 280 nm. Eluted fractions containing measurable protein were evaluated for L-Arg-stimulated ion channel activity by incorporation into a lipid bilayer (See ahead). In addition, SDS-PAGE (See ahead) was performed on each eluted fraction. Those fractions exhibiting L-Arg-stimulated ion channel activity were dialyzed overnight against 2000 ml of Milli-Q water containing 0.05% CHAPS, lyophilized and stored at -80°C. The column was calibrated with a Gel Filtration Calibration Kit with protein standards from 158 to 669 kDa. Ion exchange chromatography All steps in the ion exchange procedure were performed at 4°C. Further enrichment of the fractions exhibiting L-Arg-stimulated ion activity was achieved using an anion exchange column (Hitrap Q). The column was prepared by washing with 5.0 ml of Start Buffer (25 mM Tris-HCl, pH = 9.0), followed by 2.0 ml of Regeneration Buffer (Start Buffer plus 1.0 M NaCl) and then 5.0 ml of Start Buffer. Lyophilized proteins of the active fraction from the gel filtration column were dissolved in 300 μl of Start Buffer and applied to the column. Prior to elution, the loaded column was washed with 2.0 ml Start Buffer. Elution was at 12 ml/h in 400 μl fractions. First, 1.0 ml of First Elution Buffer (25 mM Tris-HCl, pH = 9.0, 500 mM NaCl) was applied to the column followed by (second), 2.0 ml of Start Buffer (no NaCl), followed by (third) 4.0 ml of Second Elution Buffer (25 mM Tris-HCl, pH = 8.0, 500 mM NaCl). Effluent from the column was monitored with in line UV (230 and 280 nm) and conductivity detectors. Each fraction was assayed for L-Arg-stimulated activity (See ahead.). The active fractions were pooled, dialyzed overnight against 2000 ml of Milli-Q water containing 0.05% CHAPS, lyophilized, and stored at -80°C. Development of polyclonal antibodies The procedure for development of antibodies against the catfish barbel peptides labelled by the lectins, RCA-I and PHAE has been described previously [ 45 ]. Briefly, electroeluted material from that area of a gel congruent with an identical gel labelled by the lectins was injected into three female guinea pigs using a schedule and procedure previously found to raise high titer polyclonal antibodies [ 60 ]. Antisera were aliquoted into 500 μl lots in 1.5 ml Eppendorf tubes and kept frozen at -80°C until used. Antisera from animal #1 (GP1) and animal &2 (GP2) were found to be the most specific in that they reacted primarily with their antigen within the 82–84 kDa band in Western blots of SDS-PAGE of catfish barbel membranes. The IgG fraction of GP1 and GP2 antisera was purified using an E-Z-SEP antibody purification kit (Pharmacia). To reduce non-specific binding in the immunohistochemical studies and Western blots, the GP1 and GP2 antibodies were incubated with powder derived from an acetone precipitation of a catfish brain homogenate. One ml of E-Z-SEP-purified antibody was incubated with 10 mg of powder for 40 min at 4°C. The powder was removed by centrifugation and the procedure was repeated once with fresh powder. The resulting pretreated antibodies are called simply, "GP1" and "GP2." Gel electrophoresis, lectin blots and Western blots Tris-glycine gels (4–20%, Bio-Rad) were used for SDS-PAGE. Protein of fractions before any chromatography (i.e., Sp), as well as those from each chromatography step, were denatured by mixing 10 μl from each fraction, 1:1, with sample buffer containing 125 mM Tris-HCl (pH = 8.0), 20% glycerol, 4% SDS, 4% β-mercaptoethanol and 50 μg/ml bromophenol blue, and placing the mix in a boiling water bath for 5 min. Gels were run at a constant 20–25 mA for about 1 h, using prestained broad range molecular weight markers (Bio-Rad) in 1 or more lanes. Proteins were stained with Bio-Rad Silver Stain Plus kit. For lectin blots, proteins were electrophoretically transferred to nitrocellulose sheets (Bio-Rad Mini Trans-Blot, Hercules, CA). The sheets were incubated in a blocking solution (2% gelatine, phosphate buffered saline [(PBS) (150 mM NaCl, 100 mM sodium phosphate, pH = 7.4)] and 0.05% Tween 20) for 2 h at room temperature, and then incubated with biotinylated RCA-I or PHA-E (10 μg/ml) overnight at 4°C. Following exposure to biotinylated lectin, nitrocellulose sheets were washed extensively with blocking solution. Lectin-bound protein bands were visualized using a Vectastain peroxidase ABC kit with a 4 CN Membrane Peroxidase Substrate system. Development was stopped after one hour using 5% glacial acetic acid. For Western blots, proteins were transferred to nitrocellulose and incubated in blocking solution (PBS, pH = 7.4, 5% non-fat dry milk, 1% goat serum and 0.05% TWEEN 20) for 2 h at ambient temperature with constant slow rocking (Nutator). The nitrocellulose was incubated overnight at 4°C (rocking) with primary antibody (GP1, 1/500) in blocking solution. The nitrocellulose was washed with blocking solution and incubated with biotinylated secondary antibody (1:250) for 1 h with slow rocking. Bands were visualized as above. Lectin histochemistry and immunohistochemistry Rhodamine-conjugated RCA-I (Vector Labs) was used to estimate the specificity of lectin interaction with glycoproteins of catfish barbel. To assess the localization of the antigen contained in the 82–84 kDa band from the SDS-PAGE of a membrane fraction of catfish barbel (and thereby the likely localization of the putative taste receptor for L-Arg), immunohistochemistry was performed using the GP1 and GP2 antibodies pretreated as described above. Barbels were removed from euthanized albino channel catfish ( I. punctatus ) (the fish being 5 – 7 cm in length) and immediately placed in 4% buffered paraformaldehyde (PFA) (0.1 M sodium phosphate buffer, pH 7.2–7.4) for eight hours at 4°C. After washing out the PFA with several rinses of excess buffer, the barbels were placed successively in 10%, 20%, and 30% sucrose (in buffer) for 24 hr, all at 4°C. After the final cryoprotect sucrose step, barbels were cut into pieces of less than 1 cm and mounted with M-1 Embedding Matrix (Thermo Shandon, Pittsburgh, PA). The tissue was sectioned at 10 microns on a Microm HM500OM cryostat. For lectin histochemistry, ten micron sections of fixed barbel were incubated in the dark with rhodamine-conjugated RCA-I lectin (Vector Labs., Burlingame, CA), diluted 1/200 for 2 to 3 hr at ambient temperature. The sections were then washed quickly with Dulbecco's PBS (GIBCO/Invitrogen Corp), followed by three incubation washes of 10 min each. For immunohistochemistry, 10 micron barbel sections, pre-washed 3 times for 10 min each in Dulbecco's PBS, were first incubated at ambient temperature for 3 to 5 hr in blocking buffer consisting of 3% bovine serum albumin, 2% goat serum, 0.3% TritonX100, and 0.1% sodium azide in Dulbecco's PBS at pH 7.1. The sections were then incubated with primary antibody, GP1 or GP2, in blocking buffer for 18 hr at 4°C. The primary antibody solution was removed and the sections were then washed once quickly with Dulbecco's PBS, followed by three incubation washes with PBS of 10 min each. The sections were then incubated in the dark with second antibody, Cy3-conjugated goat anti-guinea pig IgG (Jackson ImmunoResearch Labs., West Grove, PA) at 1:1000 dilution for 60 min. at ambient temperature. The secondary antibody was removed and the sections washed once quickly with Dulbecco's PBS, followed by three incubation washes with PBS of 10 min each. Excess fluid was removed from the slides and the sections mounted under cover slips with VectaShield (Vector Labs). Sections were observed with a Nikon Microphot FXA fluorescence microscope, photographed, and images sized and enhanced using the GNU Image Manipulation Program software [ 61 ]. Immuno-specificity was verified by running negative controls where the primary antibody was omitted from the procedure. In all cases, this control step showed no taste bud labelling (Fig. 5E ). The scattered, spotty background labelling seen with the primary antibodies is likely due to both an unknown factor in pre-immune serum and to labelling of solitary chemoreceptor cells in the barbel epithelium, as was previously documented [ 35 ]. Lipid bilayer reconstitution Reconstitution of protein fractions containing likely L-Arg-stimulated channel activity was carried out with material from the low osmolar and galactose-eluted fractions of the lectin affinity procedure, from the protein-containing fractions off the Sephacryl S-300 gel filtration column and from each fraction of the ion exchange column. Lipid vesicles were prepared by sonication of 5 mg of DOPE:DOPC (2:1) and 0.5 ml of 5 mM Tris-HCl (pH = 7.2), 300 mM NaCl and 500 mM sucrose, to which 10 μg n-octylglucoside was added. To prepare the liposome-detergent mixture for incorporation into a lipid bilayer, approximately 0.2 to 0.5 μg of presumed receptor protein was added to the liposome in a cassette dialysis unit, and the mixture dialyzed overnight against 2000 ml of the Tris/NaCl/sucrose buffer at 4°C. Virtually solvent-free lipid bilayer membranes were prepared as described [ 62 ]. The membrane-forming solution was an equimolar mixture of DOPS:DOPE in hexane. The bilayer chamber consisted of two symmetrical halves of a Teflon chamber, each with solution volumes of 1 ml divided by a 15 μm thick Teflon partition containing a round aperture of about 150 μm diameter. Hexadecane in n-hexane (1:10, v/v) was used for aperture pre-treatment. A pair of Ag-AgCl electrodes was connected to the solution in the chamber via 3 M KCl-4% agar bridges. "Virtual ground" was maintained at the trans side of the bilayer. The bilayer was bathed symmetrically with 5 mM MOPS (pH 7.2), 1 mM CaCl 2 , 100 mM NaCl (unless otherwise stated). Fifty to 100 μl of the dialyzed liposome vesicles containing presumed receptor was added to the cis -side of the membrane. Fusion of the vesicles was initiated mechanically by gently mixing the membrane bathing solution from the cis -side using a micro-pipette. L-Arg was added to the cis side approximately 20 min after addition of vesicles. Unless otherwise stated all other additions of reagents also were made from the cis side. Channel sidedness was determined by sensitivity of the bilayer to L-Arg. The orientation of the channels was such that the L-Arg sensitive side was normally in the cis compartment. All bilayer experiments were performed at room temperature. The current was amplified by a Dagan 3900 integrating patch-clamp amplifier (Dagan Corp., Minneapolis, MN) in the voltage clamp mode. Single channel data were digitized at 15 kHz (Digidata 1200, Axon Instruments, Foster City, CA) and analyzed using pClamp6 (Axon Instruments) and Origin 5.1 (Microcal Software, North Hampton, MA) software on an IBM compatible computer. The calculated success rate of incorporation of vesicular proteoliposomes into lipid bilayers was about 25%. Success rate is defined as the ratio of the number of successful incorporation attempts to the total number of incorporation attempts (an "incorporation attempt" refers here to the formation of a new bilayer and application of putative channel protein). Over the course of these studies, the number of incorporation attempts for each fraction tested as indicated above was about ten. Authors' contributions WG developed the enrichment procedures and drafted the manuscript. YK developed protein stabilization procedures and performed and interpreted the bilayer studies. DB performed the lectin- and immunohistochemistry and refined many of the enrichment procedures. AS developed the antibodies and performed pilot chromatography procedures. DLK initially designed the lectin procedure and carried out preliminary work on solubilization that made subsequent procedures possible. JT performed initial bilayer studies and advised YK on a continuing basis. JGB, the corresponding author and head of the laboratory, conceived the study, developed tissue preparations, designed and coordinated all phases of the study, and finalized the manuscript. All authors read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC511074.xml |
549545 | Functional complementation of RNA interference mutants in trypanosomes | Background In many eukaryotic cells, double-stranded RNA (dsRNA) triggers RNA interference (RNAi), the specific degradation of RNA of homologous sequence. RNAi is now a major tool for reverse-genetics projects, including large-scale high-throughput screens. Recent reports have questioned the specificity of RNAi, raising problems in interpretation of RNAi-based experiments. Results Using the protozoan Trypanosoma brucei as a model, we designed a functional complementation assay to ascertain that phenotypic effect(s) observed upon RNAi were due to specific silencing of the targeted gene. This was applied to a cytoskeletal gene encoding the paraflagellar rod protein 2 ( TbPFR2 ), whose product is essential for flagellar motility. We demonstrate the complementation of TbPFR2 , silenced via dsRNA targeting its UTRs, through the expression of a tagged RNAi-resistant TbPFR2 encoding a protein that could be immunolocalized in the flagellum. Next, we performed a functional complementation of TbPFR2 , silenced via dsRNA targeting its coding sequence, through heterologous expression of the TbPFR2 orthologue gene from Trypanosoma cruzi : the flagellum regained its motility. Conclusions This work shows that functional complementation experiments can be readily performed in order to ascertain that phenotypic effects observed upon RNAi experiments are indeed due to the specific silencing of the targetted gene. Further, the results described here are of particular interest when reverse genetics studies cannot be easily achieved in organisms not amenable to RNAi. In addition, our strategy should constitute a firm basis to elaborate functional-dissection studies of genes from other organisms. | Background RNA interference (RNAi) can be triggered by introduction of long double-stranded RNA molecules (dsRNAs) in cells [ 1 ], and proceeds in a number of sequential steps, starting with the cleavage of long dsRNAs into shorter ≈ 21–23 nucleotide-long dsRNAs called short interfering RNAs (siRNAs; these were initially discovered in plants [ 2 ]). The enzyme responsible for this chopping (DICER; [ 3 , 4 ]) displays RNase III activity, producing characteristic siRNAs with a phosphorylated 5' end and a two nucleotide-overhanging 3'OH end. These siRNAs enter an RNA-induced silencing complex, or RISC [ 5 , 6 ]. A helicase activity unwinds the two strands of the siRNA, and RISC scans the mRNAs in the cytoplasm and cleaves the molecules that are found complementary to the RISC-contained siRNA [ 5 ]. RNA-silencing processes have been described in a variety of organisms: post-transcriptional gene silencing in plants [ 7 , 8 ], quelling in fungi [ 9 ], homology-dependent gene silencing in ciliates [ 10 ], or RNA interference in worms [ 1 ], flies [ 11 , 12 ], trypanosomes [ 13 , 14 ] and mammals [ 15 , 16 ]. It is thought that this machinery has evolved to protect cells against undesirable RNAs, like RNA viruses in plants [ 17 , 18 ], or to limit the mobility of transposable elements in animals [ 19 - 21 ]. While RNAi and associated phenomena constitute exceptional recent basic science findings, they also provided a basis for the elaboration of powerful research tools. RNAi methodologies have been set up to perform reverse-genetics studies in a number of organisms. RNAi potency and flexibility have allowed to perform high-throughput genetic screens in several organisms [ 22 - 26 ]. In mammalian cells, the presence of long dsRNA ( > 50 base pairs) triggers the activation of sequence-unspecific interferon-related pathways [ 27 - 29 ]. To circumvent this difficulty, researchers resorted to the transfection of small interfering RNAs [ 16 ] or in vivo synthesis of small hairpin RNAs, which were demonstrated to produce gene-specific silencing [ 27 , 30 , 31 ]; reviewed in [ 32 , 33 ]. However, an siRNA might trigger a number of potential unspecific events such as the degradation of partially complementary mRNA due to cross-hybridization, leading to unspecific RNAi, or the translational arrest due to a micro RNA-like effect where an siRNA hybridizes to a mRNA with one or few mismatches. It is thus of paramount importance to ensure that the phenotypic effects observed as a result of siRNA presence in cells are due to silencing of the target gene only. Two large-scale studies show that siRNA-induced gene silencing of transiently- or stably-expressed mRNA is highly gene-specific and does not produce secondary effects detectable by genome-wide expression profiling [ 34 , 35 ]. In contrast, other works provided evidence that siRNAs can be target-unspecific, with the observation of silencing of genes that had limited sequence homology with the siRNA [ 36 , 37 ]. These reports should prompt scientists to assess the specificity of RNAi-silencing in any experiment. A solution to that problem, that we devised in trypanosomes and which is described in this report, is based on the rescueing of the RNAi-mediated loss-of-function phenotype by expressing an RNAi-resistant version of the target gene. Trypanosomes are protozoan parasites belonging to the Kinetoplastida order. These unicellular flagellated organisms diverged very early in eukaryotic evolution, and exhibit a number of original features [ 38 - 40 ]. Trypanosomes were amongst the first organisms where RNAi was discovered [ 13 , 14 ], and a number of strategies have been devised to either transiently or permanently induce gene-specific RNAi-silencing in these cells [ 14 , 41 - 43 ]. Examples of successful RNAi in trypanosomes used flagellar genes as targets which yielded easily monitored phenotypes [ 44 ]. From a structural point of view, the most conserved morphological feature of eukaryotic flagella is the axoneme, which is made of nine doublets of outer microtubules plus 2 central microtubules (so-called 9+2 axonemal structure). In trypanosomes, the flagellum not only has that axone-mal structure, but it also has a lattice-like structure called the paraflagellar rod (PFR) that is positioned along the axoneme. The two main components of the PFR are TbPFR2 and TbPFR1, that share 60% primary sequence identity [ 45 , 46 ]. TbPFR2 silencing leads to flagellar paralysis and trypanosomes do not swim anymore [ 13 , 47 ]. During the cell cycle, the cell first replicates its mitochondrial DNA (kinetoplast) and starts to grow a new flagellum whilst maintaining the old flagellum in place. Hence, a trypanosome which has two kinetoplasts and two nuclei will be close to completion of its cell cycle and will possess an old and a new flagellum [ 48 ]. This aspect is an interesting feature for RNAi-based studies of flagellar morphogenesis, because bi-flagellated cells have an "internal control" flagellum (the old one), while the new one has a phenotype corresponding to the RNAi-based gene knock-down. The presence of both the old and the new flagella in the same cell gives an indication of the time course of events when RNAi is induced in trypanosomes, leading to the appearance of a visible phenotype in the new flagellum while the older one is unchanged because it is not affected by gene silencing. We previously established the degree of identity between the gene sequences capable of leading to cross-RNAi [ 47 , 49 ]. However, as mentioned earlier, each time a phenotype is observed in RNAi experiments, it is necessary to ensure that it is indeed due to the specific silencing of the targeted gene(s). Inspired by the procedure with which gene knock-out is usually performed (the control experiment is done by re-introducing the knocked-out gene to ensure that the lost function gets complemented), we devised a functional complementation strategy aimed at assessing that RNAi indeed targets the intended gene. This strategy, elaborated using the TbPFR2 gene as a model system, involved the silencing of the TbPFR2 target via its UTRs and the expression of a RNAi-resistant copy of the targetted gene. The RNAi-resistant gene was either a copy of TbPFR2 with different UTRs or its Trypanosoma cruzi orthologue: TcPFR2 . We found that inter-species complementation experiments were straight forward. This strategy opens a venue for functional gene dissection experiments where modified gene sequences can be tested for their ability to encode functional protein that can complement the RNAi-based loss-of-function phenotype. Results and discussion Multiple RNAi on trypanosomes We wanted to establish if the co-transfection of two distinct dsRNAs, targeting two different genes, could trigger their simultaneous silencing. The genes selected were TbPFR2 and FLA1 ; TbPFR2 encodes one of the two major components of the paraflagellar rod and is necessary for flagellum motility [ 13 ]; FLA1 encodes a protein required for flagellum attachment to the cell body [ 50 ]. These dsRNAs were transfected simultaneously in wild-type trypanosomes. As a control experiment, we used GFP dsRNA. Cells were monitored for their acquired phenotype 15 h and 22 h after transfection (Table 1 ). The extinction of TbPFR2 was followed by immunofluorescence microscopy using the L8C4 anti-TbPFR2 monoclonal antibody. FLA1 gene silencing was analyzed by differential interference contrast microscopy, as it results in the visible detachement of the flagellum from the cell body (Figure 2 ). The transfection of TbPFR2 dsRNA yielded potent silencing, as more than 60 % of the cells showed no staining for L8C4 15 h later. Since old flagella pre-exist in cells which were affected by RNAi at the beginning of cell replication, the real percentage of silenced cells is probably higher than 60 %, which is confirmed by the fact that it built up to more than 74 % at time point 22 h. The transfection of FLA1 dsRNA produced a phenotype in which the flagellum was detached from the cell body in more than 50 % of the cells. When both dsRNAs were co-transfected, both phenotypes were indeed observed, with similar frequencies to experiments where only one dsRNA was transfected. All the transfected cell populations did show a comparable growth rate (data not shown). The trypanosomes shown in Figure 2 had been transfected with both dsRNAs and the cell on the right is starting cytokinesis. The old flagellum of that cell is detached, while the new flagellum is attached along the cell body. The new flagellum exhibits a dilation of its distal tip, probably corresponding to the accumulation of TbPFR1, that is not assembled but still transported to the distal tip of the flagellum in the absence of TbPFR2 [ 13 ]. This observation demonstrates the usefulness of double-transfection experiments also for kinetics analysis. In our case, 22 h after dsRNAs transient transfection, the phenotype due to the FLA1 silencing is no longer visible in the new flagellum while that same flagellum still exhibits the phenotype due to the TbPFR2 silencing, clearly indicating different turn-over for TbPFR2 and FLA1 proteins. The RNAi machinery could cope with two different dsRNA populations, without – in our conditions – any visible saturation effect. These results show the feasibility of experiments involving the use of multiple dsRNAs, thus allowing studies on complex processes in the cell physiology. However, such complex experiments can only be envisaged after ensuring that the phenotypes resulting from RNAi are specifically due to silencing of the target gene. In order to address that specific problem, we elaborated a method that involves RNAi experiments on trypanosomes that were engineered to possess an extra RNAi-resistant copy of the targeted gene, leading to functional complementation. Gene silencing by dsRNA targeting UTRs As a model for this study, we chose the TbPFR2 gene, which is present in four copies in the WT trypanosome genome (Figure 1A ), all transcribed as a single long polycistronic mRNA. All these gene copies are separated by three identical intergenic UTRs (igUTR), while the first copy has a unique 5' UTR and the last copy has a unique 3'UTR. Three types of dsRNA populations were used in our experiments, and termed as follows. dsRNA homologous to the GFP sequence was labelled "GFP dsRNA"; dsRNA homologous to coding sequence of the TbPFR2 gene was labelled "CDS dsRNA"; finally, the mixture of three dsRNAs homologous to the 5' UTR, igUTR and 3'UTR of the TbPFR2 gene was termed "UTRs MIX dsRNAs" (Figure 1A ). These dsRNAs were transfected into three cell lines: WT, TbPFR2tag and TbPFR2tag-ΔHLA (see Methods). For each experiment, the presence or absence of TbPFR2 in the new flagellum of bi-nucleated/bi-flagellated cells was monitored by immunofluorescence 14 h after the transfection (Table 2 ). Reports in [ 14 , 24 ] showed that RNAi silencing of a gene can be accomplished by targeting transcribed non-coding sequences. Here, we wanted to make sure that this kind of experiment was still feasible with a more complex system such as the TbPFR2 multigene locus, where multiple and distinct UTRs regulate the expression of four TbPFR2 isogenes. We first transfected WT trypanosomes with GFP dsRNA as a negative control and did not detect any TbPFR2 silencing (Figure 3A ). Second, WT trypanosomes were transfected with the CDS dsRNA: 88 % of cells showed typical TbPFR2 silencing with an anti-TbPFR2 immunofluorescence showing that the protein was missing from the new flagellum (Figure 3B ). Finally, the WT trypanosomes were transfected with the UTRs MIX dsRNAs, yielding the same phenotype as for the CDS dsRNA, although the silencing appeared less pronounced (Figure 3C and Table 2 ). Overall, these results demonstrate that RNAi could efficiently silence all of the TbPFR2 gene copies by targeting non-coding sequences present at the mRNA level. When WT trypanosomes were transfected with TbPFR2 dsRNA complementary to only one UTR, the cells did not display any specific phenotype (data not shown). This observation is probably explained by the organization of the TbPFR2 locus: the polycistronic transcript is rapidly spliced into three different types of mRNA, each encoding one of the four copies of TbPFR2 [ 51 - 53 ]. Thus, even if one type of TbPFR2 RNA is destroyed, the three remaining ones would likely provide enough RNA to synthesize TbPFR2 levels compatible with normal PFR formation. To demonstrate that the silencing observed upon transfection of WT trypanosomes with the UTRs MIX dsRNAs was due to the actual targeting of TbPFR2 , we used two cell lines expressing a supplementary tagged TbPFR2 gene copy. The TbPFR2tag cell line expresses the TbPFR2-TAG protein which correctly localizes to the flagellum. The TbPFR2tag-ΔHLA cell line expresses TbPFR2-TAG-ΔHLA, lacking the HLA tripeptide, which prevents its localization to the flagellum. To determine both the cellular localization of the tagged TbPFR2 proteins (TbPFR2-TAG and TbPFR2-TAG-ΔHLA) and the completeness of the PFR assembly, immunofluorescence experiments were carried out with the BB2 and ROD-1 antibodies; the former recognizes the Ty-1 epitope tag present on the two tagged TbPFR2 proteins [ 54 ], while the latter is a marker for full PFR assembly [ 55 , 56 ]. GFP dsRNA was transfected as a negative control in each cell line. As expected, this did not yield any TbPFR2 silencing: TbPFR2 was decorated in both the old and new flagella by the anti-TbPFR2 antibody, and the PFR could be assembled fully, as evidenced by its staining with the ROD-1 antibody (data not shown, Table 2 ). In TbPFR2tag cells, TbPFR2-TAG was able to localize to the PFR, as evidenced by the PFR decoration with BB2 (red color, Figure 4A ). In contrast, TbPFR2-TAG-ΔHLA failed to do so in TbPFR2tag-ΔHLA cells, and the BB2 signal was detected in the cytoplasm, as expected (red color, Figure 4D ). We next compared TbPFR2tag trypanosomes after transfection with either CDS dsRNA or UTRs MIX dsRNAs. TbPFR2tag cells transfected with CDS dsRNA had a flagellum not (or faintly) decorated with the anti-TbPFR2 antibody, demonstrating that both the WT and the recombinant TbPFR2 gene copies were effciently silenced (Table 2 ). That result was confirmed with anti-TAG immunofluorescence that showed no staining of the flagellum, demonstrating that TbPFR2-TAG was absent (no red color, Figure 4B ). This lack of both TbPFR2 and TbPFR2-TAG led to an incomplete assembly of the PFR, which was therefore not decorated with the ROD-1 antibody (no yellow color, Figure 4B ). In contrast, cells transfected with the UTRs MIX dsRNAs exhibited a WT phenotype, with only 2 % of the cells displaying TbPFR2 silencing in the flagellum (Table 2 ). In this case, the tagged protein was expressed, leading to complete assembly of the PFR (yellow color, Figure 4C ) because the protein is functional and localized to the flagellum (red color, Figure 4C ; [ 56 ]). This remarkable result indicates a complementation phenomenon that is explained by the fact that the tagged TbPFR2 gene was not silenced, as it was expressed from a coding sequence flanked by UTRs from the expression vector: from the 5' UTR of the procyclin gene and from the 3'UTR of the aldolase gene (Figure 1B ; [ 57 ]). To definitely demonstrate that the complementation described above is indeed due to the expression of functional TbPFR2-TAG, we transfected the same dsRNA into TbPFR2tag-ΔHLA trypanosomes expressing a modified TbPFR2 protein missing three amino acids (that is nine nucleotides out of 1800). TbPFR2tag-ΔHLA does not access the flagellar compartment and thus cannot be functional [ 58 ]. Transfecting either CDS dsRNA or UTRs MIX dsRNAs produced cells in which the new flagellum was not (or faintly) decorated by the anti-TbPFR2 antibody (Table 2 ). TbPFR2-TAG-ΔHLA was not decorated by the anti-TAG antibody when cells were transfected with CDS dsRNA (red color, Figure 4E ), indicating that both the WT and the tagged TbPFR2 copies were silenced, thus leading to an incomplete PFR edification (yellow color, Figure 4E ). In contrast, transfection of UTRs MIX dsRNAs did not prevent the expression of the recombinant TbPFR2-TAG-ΔHLA protein, as it appeared stained by the anti-TAG antibody (red color, Figure 4F ). However, that non-functional protein could not participate in the construction of the PFR, as shown by the absence of ROD-1 signal in the new flagellum, since it cannot access the flagellum (yellow color, Figure 4F ). RNA-directed RNA polymerase activity (RdRP) has been implicated as one possible step in the formation of siRNA in fungi [ 59 ], plants [ 17 , 18 ], and worms [ 60 ]. The fact that we could specifically silence WT TbPFR2 by targeting its UTRs, without interfering with the tagged TbPFR2 genes, suggests that spreading of silencing beyond the initial targeted sequence does not occur in trypanosomes [ 61 - 64 ]. Functional complementation with orthologue genes We next asked if an RNAi-mediated loss of function could be complemented by the expression of a gene orthologue to the silenced one. The system used to answer that question involved the TbPFR2i cell line – that expresses TbPFR2 dsRNA under the control of a tetracycline-inducible promoter [ 47 ] – into which constitutive expression of Trypanosoma cruzi TbPFR2 orthologue ( TcPFR2 ) was established using stable transfection procedures. TbPFR2 and TcPFR2 proteins share 90 % identity (both of them are recognized by the anti-TbPFR2 L8C4 antibody), but their gene sequences have diverged enough for us to envisage that the RNAi-silencing of TbPFR2 would not affect significantly the introduced TcPFR2 gene (83 % overall nucleotide identity). We thus created two new cell lines based on the previously described TbPFR2i cells [ 47 ] (see Methods). TbPFR2 expression and cell motility were analyzed. Our first experiment showed that the PCGFP cells constitutively expressed GFP, as detected by microscope observation of living cells (data not shown). Both the PCGFP and PCTcPFR2 cell lines were induced to express TbPFR2 dsRNA for 48 hours. Immunofluorescence revealed that non-induced PCGFP cells exhibit a WT-like phenotype (Figure 5A , - TET). When these cells were induced with tetracycline, expected TbPFR2 silencing occurred (Figure 5A , +TET). Non-induced PCTcPFR2 cells displayed an intense anti-TbPFR2 antibody decoration with bright dots in the cytoplasm, indicative of TcPFR2 overexpression (such overexpression by the EP procyclin promoter is frequent; Figure 5B , - TET). When these same cells were tetracycline-induced, the flagellar staining was still perfectly visible, at a level comparable to the one previously observed in the non-induced PCGFP control cells (Figure 5B , +TET). Bright dots previously observed had disappeared, probably as a result of TbPFR2 silencing. The fact that the paraflagellar rod was still neatly decorated by the anti-TbPFR2 antibody demonstrated that the structural inter-species complementation had indeed taken place in these cells, with TcPFR2 being effectively located at the flagellum. Did these structurally-complemented cells show a functional complementation, i.e. a normal flagellum motility (hence a normal cellular mobility)? To address this question, we performed a sedimentation assay [ 56 ] on non-induced and tetracycline-induced PCGFP and PCTcPFR2 trypanosomes (Figure 6 ). Non-induced PCGFP cells showed a little tendency to sediment due to the fact that expression of TbPFR2 dsRNA in TbPFR2i cells is partially leaky, producing low amounts of TbPFR2 dsRNA even in the absence of tetracycline ([ 43 ]; Durand-Dubief and Bastin, unpublished data). When expression of TbPFR2 dsRNA was fully induced, motility stopped leading to increased sedimentation (Figure 6 , left panel). In contrast, expression of TbPFR2 dsRNA in PCTcPFR2 cells did not reduce motility (Figure 6 , right panel). That result definitely demonstrates that the ortholog protein TcPFR2 fully complemented the loss of function resulting from TbPFR2 silencing. The complementation described above shows the robustness of our strategy, because TbPFR2 and TcPFR2 are highly similar (they share 82 % identity at the nucleotide level [ 65 ]) and are nonetheless correctly differenciated by the RNAi machinery. However, our complementation strategy might be more diffcult to implement when the gene studied is too similar to the T. brucei counterpart. While this unfavorable case might happen with extremely evolutionarily-related organisms, studies have shown that the overall genetic sequence identity between Trypanosoma brucei and Trypanosoma cruzi , for example (the closest evolutionarily-related organisms envisaged for these studies), is roughly 80 % ([ 65 ] and [ 66 ]). [ 47 ] showed that this identity percentage is still compatible with an RNAi-based complementation strategy. It goes without saying that when the organisms are evolutionarily-distant, gene sequences diverge more rapidly than the protein sequences, thus laying off a field where our strategy can be implemented with good confidence that complementation will occur. Conclusions In this report, we demonstrated that RNAi-mediated silencing of a gene by targeting its UTRs is useful in studies where the loss of function resulting from this silencing must be complemented with the expression of an RNAi-resistant copy of the silenced gene, in order to demonstrate that the phenotype is indeed due to silencing of that gene, and not to inactivation of another one. The results obtained in this work are of particular interest when reverse-genetics studies cannot be easily achieved in organisms not amenable to RNAi, like Leishmania [ 67 ] or Trypanosoma cruzi [ 68 ], or where genetics experiments are hardly set up, like mammals. When genes from these organisms are to be studied, a complementation experiment can be set up as a three-step procedure whereby: 1) the ortholog gene in Trypanosoma brucei is RNAi-silenced and the loss-of-function phenotype is established; 2) T. brucei cells are engineered to ensure constitutive heterologous expression of the gene of interest, still allowing RNAi-mediated silencing of the T. brucei gene; 3) function of the investigated gene is assessed by checking if the loss-of-function phenotype observed in the first place gets complemented. Additionally, one application of the strategy described herein is genetic functional dissection, which is of interest when protein domains are to be characterized with respect to their function (e.g. the HLA tripeptide sequence in TbPFR2 that localizes the protein to the flagellum). Complementation had previously been demonstrated following transformation of mammalian cells with EGFP siRNA and expression of a codon-modified, but functional, EGFP version [ 69 ]. Our strategies are increasing flexibility for complementation studies after RNAi as unmodified genes can be used for rescue. Methods Trypanosomes The procyclic T. brucei brucei strain 427 (or its derivatives) was used throughout this work. Cells were cultured at 27°C in semi-defined medium 79 (SDM 79) containing 10% foetal calf serum. PFRAi cells were described in [ 47 ]. The TbPFR2i trypanosomes can be tetracycline-induced to express TbPFR2 dsRNA, thus eliciting an RNAi response against that gene. Note that this cell line is referred to as TbPFR2i in this article because of a change in the gene nomenclature [ 70 ]. RNAi assays by transient transfection RNA was synthesized in vitro with T3 and Sp6 polymerases using PCR products as templates [ 71 ]. The following primers (incorporating T3 or Sp6 promoters) were used: for GFP (from the nucleotide coding sequence 476–691 of the EGFPN2 gene; Clontech), AATTAACCCTCACTAAAGGGAGAAG AACGGCATCAAGGTGAAC (T3 promoter italicized) and ATTTAGGTGACACTATAGAAG AGTGATCCCGGCGGCGGTCACG (Sp6 promoter italicized); for FLA1 , AATTAACCCTCACTAAAGGGAGA CCAAACCGTGGGCACCAAGG (T3 promoter italicized) and ATTTAGGTGAACTATAGAAGAG GTGGGATGATTAAAACGAGC (Sp6 promoter italicized); for the TbPFR2 5' untranslated region (5' UTR; nucleotide sequence [-545→-1] upstream of TbPFR2 ATG start codon), AATTAACCCTCACTAAAGGGAGA (T3 promoter) and ATTTAGGTGACACT-ATAGAAGAG (Sp6 promoter); for the TbPFR2 intergenic untranslated region (igUTR), AATTAACCCTCACTAAAGGGAGA CGCTGCGCTTAAATGTCTT (T3 promoter italicized) and ATTTAGGTGACACTATAGAAGA GTGATGCTTTATTGCTTTCT (Sp6 promoter italicized); for the TbPFR2 3'untranslated region (3'UTR; nucleotide sequence [1→533] downstream of the TbPFR2 TAG stop codon), AATTAACCCTCACTAAAGGGAGA (universal T3 promoter) and ATTTAGGTGACACTATAGAAGAG (universal Sp6 promoter); for the TbPFR2 coding sequence (CDS; nucleotide coding sequence [1084→1358]), ATTTAGGTGACACTATAGA GAGGTGAAGCGCCGTATTGAGGA (Sp6 promoter italicized) and AATTAACCCTCACTAAAGGGAGA GTTTTGTACAGGCGACGGAA (T3 promoter italicized); Figure 1A shows the TbPFR2 locus and the position of the two dsRNA populations that were used, and their homology to either the coding sequence (labelled "CDS dsRNA") or the different 5'UTR, igUTR and 3'UTR all together (labelled "UTRs MIX dsRNAs"). A third dsRNA, homologous to the GFP gene is labelled "GFP dsRNA" throughout this work and was used as a control dsRNA. dsRNA was introduced into trypanosomes by electroporation, as described [ 14 ]. Plasmids Plasmid pPC was generated from plasmid pSk1-GFP [ 50 ] as follows: pSk1-GFP was digested with Hin d III and Eco RI to remove the GFP gene. Oligonucleotides AGCT GTCTAGCGATATCGGATCCG (forward) and AATT CGGATCCGATATCGCTAGCA (reverse) were annealed (protruding ends italicized) and the resulting double-strand oligonucleotide was ligated into the pSk1-GFP plasmid, resulting in the insertion of a poly-linker containing restriction sites Cla I, Hin d III, Nhe I, Eco RV, Bam HI and Eco RI (Branche and Bastin; unpublished data). Plas-mid pPCTcPFR2 was generated as follows: amplification of the TcPFR2 gene was performed using Trypanosoma cruzi genomic DNA (kind gift of Cécile Gallet and Philippe Grellier, MNHN) and the two primers TcPFR2H (GAGTCTAAGCTTATGAGCTACAAGGAGGCATC) and TcPFR2ER (GCGTGGAATTCTTACTGTGTGATCTGCTGCAC). Both the amplified DNA fragment and the pPC plasmid were digested with Eco RI and Hin d III. The fragment was ligated into pPC so as to yield the plasmid pPCTcPFR2 (Figure 1B ). Cell lines The different constructs used to transform trypanosomes are shown on Figure 1B . The cell lines were established as follows. WT-derived trypanosomes constitutively expressing TbPFR2-TAG proteins The TbPFR2tag cell line was derived from the WT cell line into which the pTbPFR2TAG430 plasmid [ 72 ] was transfected. The recombinant cells constitutively expressed the TbPFR2-TAG protein, that is localized in the flagellum (Fig 4D ). Tagged TbPFR2 is known to be functional [ 56 , 72 ]. In contrast, transformation of WT cells with the pTbPFR2TAGΔHLA430 plasmid lead to the expression of slightly modified TbPFR2 protein, missing only three amino acids, that failed to enter the flagellum compartment and hence was found in the cell body cytoplasm [ 58 ] (Fig 4G ). This cell line was called TbPFR2tag-ΔHLA . After electroporation [ 73 ], cells were grown overnight and then distributed in 24-well plates in the presence of phleomycin (2 μ g/mL) for selection. TbPFR2i-derived trypanosomes constitutively expressing GFP and TcPFR2 TbPFR2i cells [ 47 ] constituted the genetic background into which we established the PCGFP and PCTcPFR2 new cell lines. The PCGFP cell line was established by transfecting TbPFR2i cells with plasmid pPCGFP after linearization with BstX I. For establishing the PCTcPFR2 cell line, the pPCTcPFR2 plasmid was linearized with BstX I and transfected into TbPFR2i cells. Recombinant cells were selected by addition of puromycin (1 μ g/mL), phleomycin (2 μ g/mL), G418 (15 μ g/mL) and hygromycin (20 μ g/mL) to the culture medium. Immunofluorescence and microscopy Three different monoclonal antibodies were used as hybridoma supernatants: L8C4, IgG recognizing T. brucei TbPFR2 and cross-reacting with T. cruzi orthologue TcPFR2 [ 74 ]; BB2, IgG recognizing the Ty-1 tag of the TbPFR2-TAG and TbPFR2-TAG-ΔHLA recombinant proteins [ 54 ]; and ROD-1, IgM recognizing a doublet of minor PFR proteins [ 55 ]. For immunofluorescence, trypanosomes were spread onto poly-L-lysine-coated slides, fixed in cold methanol and processed as described [ 75 ]. Experiments involving the use of L8C4 only were performed with an FITC-conjugated anti-mouse IgG secondary antibody. Double-staining experiments using BB2 and ROD-1 were performed with a TRITC-conjugated specific anti-mouse IgG secondary antibody and an FITC-conjugated specific anti-mouse IgM secondary antibody. DNA was systematically stained with 4',6-diamidino-2-phenylindole (DAPI). Slides were examined with a Leica DMR microscope, images were captured using a cooled CCD camera (Cool Snap HQ, Roper Scientific) and processed with the GNU image manipulation program version 2 [ 76 ]. Cell sedimentation assay The trypanosome sedimentation assay was performed as described in [ 56 ]. Briefly: trypanosomes were grown at ≈ 5.10 6 cells/mL in normal culture medium, with or without 48 hour tetracycline induction. 1 mL of these cultures was dispensed to 5 plastic spectrophotometry cuvettes, for time points 0, 2, 4, 6, 8 hours, and left still. At each time point, the optical density at 600 nm was measured twice: first without mixing (O.D. no mix ) and second after mixing the cuvette (O.D. mix ). Data were plotted as a function of time. Authors' contributions F.R. carried out most of the experiments reported and wrote the manuscript, M.D.-D. performed the double transfection reported at Table 1 & Figure 2 and P.B. conceived the study and participated in its design and coordination. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549545.xml |
545196 | Intermittent Presumptive Treatment for Malaria | A better understanding of the pharmacodynamics of intermittent presumptive treatment, says White, will guide more rational policymaking | Intermittent presumptive treatment (IPT) in pregnancy involves giving a curative treatment dose of an effective antimalarial drug at predefined intervals during pregnancy. IPT in pregnancy was first introduced in areas of high malaria transmission as a measure to reduce the adverse impact of Plasmodium falciparum malaria in pregnancy [ 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 ]. Later, based on trials showing that IPT could reduce anaemia in young children and also malaria episodes in infants, it was extended as a measure to reduce morbidity and mortality in the first year of life [ 9 , 10 , 11 , 12 ]. Antimalarial chemoprophylaxis for pregnant women living in endemic areas has been recommended for many years, but in practice has been limited to the use of chloroquine and pyrimethamine [ 13 , 14 ]. Unfortunately, there are few places left in the world where these drugs can still be relied upon to prevent P. falciparum malaria. There are insufficient safety data on the newer antimalarials to warrant their systematic use in pregnant women. IPT with sulphadoxine-pyrimethamine (SP) has been introduced as an alternative. Antimalarial chemoprophylaxis in young children has been shown to reduce the adverse impact of P. falciparum malaria [ 15 , 16 , 17 ], but this intervention never obtained the same endorsement as chemoprophylaxis in pregnancy. Five randomised trials of IPT in pregnancy in East Africa have been reported [ 1 , 2 , 3 , 4 , 5 ], all with SP, all in high-transmission settings, and all done between 1992 and 1999 ( Table S1 ). The alarming recent increase in resistance to SP in Africa confounds the cost-effectiveness assessments upon which subsequent policy recommendations for IPT in pregnancy were based [ 18 , 19 ]. There is no consensus on how IPT works, making planning difficult. This article argues that IPT provides mainly intermittent suppressive chemoprophylaxis (as opposed to treatment effect alone or some other magical effects which have never been specified). If this is correct dosing schedules should be individualised for each antimalarial depending on the drug's pharmacokinetic and pharmacodynamic properties. As increasing resistance to SP must seriously compromise IPT regimens based on this drug, the evaluation of available new effective antimalarials is needed urgently, in both high- and low-transmission areas. Pharmacokinetics After a treatment dose of SP (25 mg sulfadoxine/1.25 mg pyrimethamine per kilogram body weight), plasma concentrations of pyrimethamine (half-life, 3 days) and sulfadoxine (half-life, 7 days) decline log-linearly [ 20 , 21 ]. The antimalarial effect depends on synergy between the two components, but the effect from one treatment dose can last as long as 60 days with fully sensitive P. falciparum [ 20 , 21 ]. For slowly eliminated antimalarial drugs ( Table S2 ), the terminal elimination phase crosses the in vivo dose–response curve ( Figure 1 ). Thus, if a full treatment dose is given, concentrations at the beginning of the terminal elimination phase exceed the minimum parasiticidal concentrations (MPCs)—the lowest concentrations that give maximum effect [22] . The exceptions to this are chloroquine (and probably piperaquine), as resistance to these drugs increases, because the elimination of chloroquine is multiexponential, and the terminal elimination phase begins at concentrations that are low by comparison with the peak concentrations after treatment ( Figure 2 ). Figure 1 In Vivo Antimalarial Pharmacodynamics The parasite burden in an adult (vertical axis) is shown in green. After parasite burden expands to the point where it causes illness, treatment is given (red arrow), which causes a log-linear decline in parasite numbers until concentrations of the antimalarial drug (grey shading) fall below the MPC. As the antimalarial blood levels fall further, the decline in parasite burden slows until it reaches a multiplication rate of one (the antimalarial concentration at this point is the in vivo MIC). The parasite population then expands to cause a recrudescence six weeks later. The sigmoid concentration–effect relationship is shown in brown; it is depicted in the reverse direction to that normally drawn. PMR, parasite multiplication rate. Figure 2 Blood Concentration Profiles of Two Antimalarials with Different Elimination Profiles The examples shown here are mefloquine (orange) and chloroquine (pink). An increase in MIC has different effects on the shortening of post-treatment suppressive prophylaxis (hatched bars). MIC R , MIC for resistant parasites; MIC S , MIC for sensitive parasites. The pharmacokinetic properties of many drugs are altered in pregnancy; lower concentrations often result from an expanded volume of distribution. Strangely, despite the wide endorsement of SP IPT in pregnancy, there are no pharmacokinetic studies of sulphadoxine or pyrimethamine in pregnancy, so it is not known whether the current dosing is optimal. The absorption and disposition of many drugs are also altered in infancy, but there are very few data on antimalarial pharmacokinetics in the first year of life. For some drugs (e.g., amodiaquine) there is insufficient information for any age group. Pharmacodynamics Is the benefit of IPT gained only through clearing parasites from the placenta (“treatment effect”), or is the prevention of new infections (“prophylactic effect”) an important component? If only the treatment effect is important, then how long does the beneficial effect of eradicating an asymptomatic low-density infection persist for? If it lasts until the next infection becomes patent (i.e., detectable), then rapidly eliminated drugs will provide protection only for a few days longer than the average incubation period (about two weeks). Establishment of a new placental infection (i.e., pathologically significant placental sequestration) may take longer because the placenta selects and accumulates parasites that bind to the proteoglycans chondroitin sulphate and hyaluronic acid [23] . If only the treatment effect is important, then for sustained benefit we must hypothesise that the parasites that persist asymptomatically before IPT is given are a selected subpopulation that is more pathological than the parasites that cause subsequent reinfection. This seems implausible in infancy, and even in pregnancy it seems unlikely that it would take more than ten weeks in high-transmission settings to re-establish a significant placental infection. This suggests that the prophylactic effect is important for the efficacy of IPT. The duration of prophylactic effect is compromised particularly by resistance. For most antimalarials the duration of antimalarial effect is a simple function of the in vivo concentration–effect (dose–response) relationship and the pharmacokinetic properties of the antimalarial drug [22] . But for SP this function is more complicated, as synergy between the two components needs to be considered. The duration of synergy depends on resistance levels determined by mutations in the parasites' genes encoding dihydropteroate synthase and dihydrofolate reductase, the respective targets of sulphadoxine and pyrimethamine [ 20 , 21 ]. Information on this temporal pattern of reinfection following IPT in pregnancy or infancy is lacking. Such information is essential if the choice of drug and the dosing is to be rationalised. Resistance is defined by a right shift in the concentration–effect relationship and results in reduced effects for any concentration below the MPC for resistant parasites [24] (see Figure 1 ). As the concentration of a slowly eliminated antimalarial in the blood declines, it continues to suppress the growth of newly acquired infections as they emerge from the liver. Eventually, however, concentrations fall below the minimum inhibitory concentration (MIC) for the prevalent parasites (i.e., the concentration at which the net multiplication rate is one), and parasite expansion is possible ( Figure 3 ). It follows, then, that the duration of “post treatment prophylaxis” (PTP) (i.e., the length of time after an antimalarial treatment dose for which newly acquired infections are suppressed) is determined by the concentrations of the drugs used (determined by dose and pharmacokinetics) and the sensitivity of the prevalent parasites. The more resistant the parasites are, the shorter is the duration of PTP; for each doubling of MIC the duration of PTP is shortened by one half-life ( Protocol S1 ). The triple dihydrofolate reductase mutants now prevalent across much of Africa have an approximate 1,000-fold reduction in pyrimethamine susceptibility, which would translate into a reduction in PTP of one month ( Figure 4 ). As a further confounder, folic acid, which is prescribed widely in pregnancy, is a competitive antagonist of pyrimethamine. Figure 3 Hypothetical Parasite Burden Profiles during Pregnancy with SP IPT in a High-Transmission Setting Entomological inoculation rate is about 50 infectious bites per person per year. Note that many infections self-cure (each infection is depicted as a green line). The hatched bars represent the duration of “suppressive prophylactic activity”, and the solid bars represent the period during which parasite multiplication is suppressed (i.e., levels exceed the in vivo MIC). The horizontal dotted line at 10 8 parasites represents the level at which malaria can be detected on a blood film. (A) represents a drug-sensitive area; (B) represents a moderately resistant area. Figure 4 Relationship between MIC and PTP The proportional increase in malaria parasite MIC with resistance is plotted against the shortening of the duration of PTP, expressed as multiples of the terminal half-life. This applies only to drugs for which suppressive antimalarial prophylaxis occurs in the terminal elimination phase (i.e., most drugs). Preventing Placental Pathology In a high-transmission setting infections are acquired every few days or weeks throughout life (see Figure 3 ). Mortality is high in childhood, but by the time of adulthood and pregnancy, infections are largely asymptomatic—although they are often still patent (which requires a total burden of greater than 100 million parasites) [22] . Thus, immunity prevents life-threatening parasite burdens, and suppresses the pro-inflammatory response (which causes illness), but it does not prevent infection. In pregnancy this immune control is impaired in the placenta, which acts as a “privileged site” for parasite multiplication. The objective of IPT in pregnancy is to reduce or eliminate the adverse effects of malaria on maternal anaemia and birth weight, and, in addition, in a low-transmission setting, to prevent severe malaria in the mother [ 25 , 26 ]. How malaria produces intrauterine growth retardation is still unresolved, but in P. falciparum malaria, retardation tends to be greatest in the first pregnancy, and often occurs without maternal illness. The greater the placental parasite burden, the greater is the reduction in birth weight. “Placental malaria”—histological evidence of placental accumulation of parasitized erythrocytes or malaria pigment deposition—has often been used as an endpoint in intervention studies, although the quantitative relationship between placental malaria and reduction in birth weight remains poorly characterised. Preventing Malaria in Infancy There are fewer data on the efficacy of IPT in infancy than in pregnancy ( Table S3 ). The pharmacodynamics of IPT in infancy are probably similar to those in pregnancy, although there is no “privileged site” for parasite multiplication. Protection in the first months of life is mediated by a variety of factors, which include transplacentally acquired maternal antibody (IgG) and a relatively high haemoglobin F content in the infants' erythrocytes. After about six months of age, protection from these factors wanes, and the infant becomes much more vulnerable to malaria than the mother (because protective immunity has yet to be acquired). As delivery of antimalarials in the rural tropics is so difficult, for operational reasons IPT is currently being given to infants at the same time as the EPI immunisations (at 2, 3, and 9 months). This regimen leaves a six-month gap between the second and third administrations, which, even for fully SP-sensitive parasites, leaves four unprotected months. This is at a time when the infant is increasingly vulnerable to severe malaria. More information is needed on the duration of protection afforded by currently available antimalarial drugs when administered to healthy infants. Should IPT be Used in Low-Transmission Settings? If IPT is just a simple, albeit imperfect, way of administering chemoprophylaxis, then there are also strong arguments for evaluating this approach in low-transmission settings. The adverse impact of malaria in pregnancy is greater in low-transmission than in high-transmission settings. The reduction in the birth weight of first-borne infants is similar, but extends to the second and subsequent pregnancies; treatment failure rates are higher than in non-pregnant adults [ 27 , 28 , 29 ]; and there is a significant risk of severe malaria with attendant very high mortality. In Asia and South America, where low-transmission areas predominate, P. vivax is also an important cause of low birth weight, and so useful preventative measures must also be effective against this infection [30] . Antimalarial prophylaxis has been recommended and used in low-transmission settings, but whereas chloroquine remains generally effective against P. vivax , there are no safe and effective available drugs for P. falciparum infections. Use of IPT in these areas would provide a sterner test than in a high-transmission area because there would be little or no background immunity to assist antimalarial drug efficacy. What Is the Correct Dose and the Correct Interval Between Doses? The dose used in IPT is usually the full age- or weight-adjusted treatment dose derived either empirically or from dosefinding studies (usually in non-pregnant adults). The dose and dosing interval should be determined by the tolerability, absorption, distribution, and elimination kinetics of the drug used, and the in vivo MIC. Unfortunately, for the two drugs that have been used for IPT (SP and amodiaquine) there are no pharmacokinetic data in pregnant women or infants. The in vivo MIC is an important measure but it is parasite specific and difficult to assess [ 31 , 32 , 33 , 34 , 35 ]. Ideally, the interval between doses should not be more than one week longer than the time needed for plasma concentrations to fall from peak post-dose levels to the MIC value. This timing is a conservative choice as it assumes all infections are equally harmful, and it does not take into account either the delay in selecting a placenta-binding P. falciparum subpopulation in pregnancy or the delay in achieving full growth rates because of continued sub-MIC suppression. Although the MIC for drugs that are succumbing to resistance obviously varies considerably, for newer drugs such as lumefantrine or piperaquine the variance is considerably less and generalisations can be made. Should an Artemisinin Combination Be Used for IPT? If IPT is simply prophylaxis, then a rapidly eliminated artemisinin component provides very little direct benefit for the additional cost and risk (although the risks are thought to be very small in the second and third trimesters of pregnancy and in infancy). The addition of an artemisinin component would accelerate parasite clearance and prevent gametocyte production, but the benefits of this in an asymptomatic pregnant woman or child are uncertain. The main benefit would be in providing protection against the emergence of de novo resistance to the slowly eliminated drug, although, because parasitaemias tend to be low, the probabilities of de novo selection are much lower than in acute symptomatic infection. But there is a genuine concern that if monotherapies are made available, then they will be used and abused, and resistance may develop. Discussion: The Policy Implications Without a better understanding of the pharmacodynamic effects of IPT, it will be difficult to make rational improvements in this promising approach to malaria prevention. The most parsimonious explanation for its effectiveness is that IPT provides antimalarial prophylaxis that, if sufficiently lengthy and effective, is beneficial both to the pregnant woman and the infant. But how lengthy and how effective? For IPT in pregnancy the only drug that has been evaluated is SP, at a time when the drug was more effective than it is today. A significant improvement in birth weight was found in only two of four randomised trials. In a large prospective observational study conducted in western Kenya, IPT was associated with an odds ratio of 0.65 (95% confidence interval, 0.45 to 0.95) for low birth weight [7] . A dose–response relationship was found, with an adjusted mean increase in birth weight of 61 g for each increment in the number of SP doses (up to three doses). So, two doses of SP did not provide maximal benefit. But given the alarming decline in SP efficacy in Africa (resulting from rapid spread of “quintuple” dihydrofolate reductase/dihydropteroate synthase mutants that are about 1,000 times less sensitive to pyrimethamine than wild-type parasites), there are grave doubts about whether the efficacy observed in these various studies would still be observed today, even if the dosing was increased. SP cure rates in children in Malawi—where IPT in pregnancy is widely used—have been consistently less than 40% for the past five years [36] . It has been suggested that asymptomatic pregnant women in high-transmission settings may have sufficient immunity to complement a failing drug—i.e., treatment responses would be better than in symptomatic children. However, if the duration of PTP is the main determinant of benefit, then this benefit is shortened progressively by increasing resistance (see Figure 4 ). Alternatives to SP are needed urgently. Is IPT safe? There is no evidence to date that IPT is harmful. But the incidence of serious adverse effects when amodiaquine (agranulocytosis, 1:2,000) and SP (Stevens-Johnson syndrome,1:7,000) were used as antimalarial prophylaxis by Western travellers was so high that they are contraindicated [37] . Both drugs were associated with severe hepatitis. Single treatments are considered safer, but how much safer is not known. There are insufficient data on the safety of amodiaquine in pregnancy [38] . The closer IPT comes to continuous prophylaxis, presumably the higher the risks of serious adverse effects. The risk–benefit assessment is difficult to make, but with the current high levels of SP resistance, these important uncertainties also argue strongly for the evaluation of alternatives. Proguanil and quinine are regarded as safe, but both are eliminated very rapidly. Treatment doses of mefloquine are not well tolerated by healthy subjects, and there are safety concerns in pregnancy [39] . Serious contenders all require more than a single dose and will need urgent evaluation. These include artemether-lumefantrine, although more information is needed on safety and on the pharmacokinetics in pregnancy, and the duration of PTP provided by lumefantrine needs further assessment. It may be too short. Dihydroartemisinin-piperaquine may be the most promising candidate. It is very well tolerated, and piperaquine is slowly eliminated. Indirect evidence from the pattern of new infections following clinical trials suggests protracted suppressive activity. It has not yet been evaluated in pregnancy, so more information is needed on safety and pharmacokinetics in this context. For IPT in infancy, however, there seems every reason to evaluate this drug as soon as possible. Supporting Information Protocol S1 Calculations Showing That for Each Doubling of MIC the Duration of PTP Is Shortened by One Half-Life (27 KB DOC). Click here for additional data file. Table S1 Randomised Trials of IPT in Pregnancy (31 KB DOC). Click here for additional data file. Table S2 Terminal Elimination Half-Lives of Currently Available Antimalarial Drugs (34 KB DOC). Click here for additional data file. Table S3 Randomised Trials of IPT in Infancy (27 KB DOC). Click here for additional data file. Summary Points Intermittent presumptive treatment (IPT) with sulphadoxine-pyrimethamine (SP) in pregnancy and with amodiaquine or SP in infancy has been proposed for use in areas with high levels of malaria transmission. The duration of post treatment prophylaxis is likely to be an important determinant of the benefit of IPT. Because of rapidly increasing resistance, it is very unlikely that IPT in pregnancy with SP is as effective now in east Africa as it was 5–10 years ago, when it was evaluated. More effective antimalarial drugs such as artemether-lumefantrine and particularly dihydroartemisinin-piperaquine should be evaluated for IPT in both low- and high-transmission settings. Choice of drug, dosing, and dose spacing for IPT should be based on a better understanding of pharmacokinetics and pharmacodynamics. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545196.xml |
539277 | Altered protein dynamics of disease-associated lamin A mutants | Background Recent interest in the function of the nuclear lamina has been provoked by the discovery of lamin A/C mutations in the laminopathy diseases. However, it is not understood why mutations in lamin A give such a range of tissue-specific phenotypes. Part of the problem in rationalising genotype-phenotype correlations in the laminopathies is our lack of understanding of the function of normal and mutant lamin A. To investigate this we have used photobleaching in human cells to analyse the dynamics of wild-type and mutant lamin A protein at the nuclear periphery. Results We have found that a large proportion of wild-type lamin A at the nuclear periphery is immobile, but that there is some slow movement of lamin A within the nuclear lamina. The mobility of an R482W mutant lamin A was indistinguishable from wild-type, but increased mobility of L85R and L530P mutant proteins within the nuclear lamina was found. However, the N195K mutant shows the most enhanced protein mobility, both within the nucleoplasm and within the lamina. Conclusion The slow kinetics of lamin A movement is compatible with its incorporation into a stable polymer that only exchanges subunits very slowly. All of the myopathy-associated lamin A mutants that we have studied show increased protein movement compared with wild-type. In contrast, the dynamic behaviour of the lipodystrophy-associated lamin A mutant was indistinguishable from wild-type. This supports the hypothesis that the underlying defect in lamin A function is quite distinct in the laminopathies that affect striated muscle, compared to the diseases that affect adipose tissue. Our data are consistent with an alteration in the stability of the lamin A molecules within the higher-order polymer at the nuclear lamina in myopathies. | Background The nuclear lamina is a filamentous network of lamin proteins that underlies the inner nuclear membrane (INM). It is thought to make connections between both integral membrane proteins of the INM, and chromatin. It may therefore play a fundamental role in the functional organisation of the nucleus. Lamins are type V intermediate filament (IF) proteins, consisting of a central coiled-coil region, and globular N-terminal and C-terminal domains. The N-terminal domain has a nuclear localisation signal (NLS) and most lamins, except for lamin C, are farnesylated at their carboxy termini via a CaaX motif [ 1 ] (Figure 1A ). The mammalian genome contains two lamin B genes ( LamB1 and 2 ) and lamins A/C . The latter is alternatively spliced to produce lamins A and C, as well as other minor species. Lamin B is expressed in all cell types and is essential for cell viability. A-type lamins are expressed in more differentiated cells [ 2 ] and are non-essential for cell viability [ 3 ]. Figure 1 Structure of lamin A protein. A) Diagram of lamin A amino acid sequence showing the domains of the protein, and the position of the four laminopathy-associated missense mutations in DCM, FPLD and AD-EDMD. B) Structure of the C-terminal globular domain of Lamin A showing the relative positions of the FPLD associated R482W missense mutation and the AD-EDMD associated L530P mutation. (Adapted with permission from [25]). Lamins readily form parallel coiled-coil dimers, which then associate into larger polymers. However, whereas cytoplasmic IF proteins assemble in vitro into 10 nm filaments that resemble those formed in vivo , lamins assemble in vitro into paracrystalline arrays rather than filaments [ 4 ]. This suggests that, in vivo , assembly of correct lamin higher-order structures requires the interaction with other molecules/proteins. Lamin A certainly has the ability to interact with other proteins, and also to influence their localisation. In the absence of lamin A, emerin relocates from the INM to the endoplasmic reticulum [ 3 , 5 , 6 ]. The interaction domain with emerin is in the C-terminal domain of lamin A [ 7 , 8 ]. The coiled-coil region can interact with chromatin [ 9 , 10 ] (Figure 1A ). There is also an interaction between lamins A/C and the INM proteins LAP2β [ 11 ] and muscle-specific nesprin1 [ 12 ]. In addition to its localisation at the nuclear lamina, lamin A is also found within the nucleoplasm where it might interact with other nuclear proteins. Interaction and/or co-localisation between lamin A and; Rb, mRNA splicing factors, LAP2 , sites of early DNA replication, and specific transcription factors have been reported [ 13 - 18 ]. Recent interest in the function of the nuclear lamina has been provoked by the discovery of lamin A/C mutations in several human diseases, termed the laminopathies [reviewed in [ 19 ]]. What is striking about these diseases is that so many apparently disparate phenotypes arise from mutations in one widely expressed gene. The overt phenotypes of the laminopathies can be grouped according to the major cell types that are affected. Striated (skeletal and cardiac) muscle is affected in autosomal dominant Emery-Dreifuss muscular dystrophy (AD-EDMD), limb girdle muscular dystrophy type 1 (LGMD-1B), and dilated cardiomyopathy (DCM). Adipose and bone tissues are affected in familial partial lipodystrophy (FPLD) and mandibuloacral dysplasia (MAD). Charcot-Marie-Tooth neuropathy type 2B1 (CMT2B1) is a demyelination disease of peripheral neurons. Lastly, Hutchinson-Gilford Progeria Syndrome (HGPS) [ 20 , 21 ] and atypical Werner's Syndrome [ 22 ] affect multiple tissue types, including many of those involved in the other laminopathies (muscle, fat, bone), and also results in some premature ageing phenotypes. There are currently three main hypotheses for laminopathy disease mechanisms – nuclear weakness, altered nuclear-cytoskeletal interactions, or changes in gene expression [ 19 , 23 ]. To understand the disease pathology of the laminopathies it will be necessary to better characterise the properties of mutant lamin As. The mutations in AD-EDMD are distributed throughout the coiled-coil domain and the first half of the C-terminal globular domain of lamin A. LGMD and DCM appear to be caused mainly by mutations in the coiled-coil domain [ 19 ], although an R571S mutation at the end of the globular domain, that affects only lamin C, has been found in a mild case of DCM. [ 24 ]. In contrast, FPLD and MAD mutations cluster tightly within part of the C-terminal globular domain. An explanation for this came from structural analysis of this domain. The residues mutated in FPLD and MAD are on the surface (solvent exposed), whereas residues mutated in other laminopathies are located internally within the hydrophobic core of the domain structure [ 25 , 26 ] (Figure. 1B ). The latter mutations may therefore have more profound affects on the structure of the mutant protein, whereas FPLD and MAD mutations may leave the overall structure of the lamin A molecule largely unperturbed but might, for example, interfere with protein-protein interactions. To better understand the affects of laminopathy-associated mutations on lamin A function we have used fluorescence recovery after photobleaching (FRAP) and fluorescence loss in photobleaching (FLIP) to investigate the protein dynamics of GFP-tagged wild-type and disease-associated mutant lamin As in living cells. Results and discussion Expression of mutant lamin A in human cells To investigate the biological affect of different mutations on lamin A nuclear localisation and dynamics we expressed epitope tagged forms of the protein, carrying disease-associated missense mutations, in human HT1080 cells. The mutations chosen were; L85R (DCM), N195K (DCM), R482W (FPLD), and L530P (AD-EDMD). Although L85R and N195K are both located within the coiled-coil domain of lamin A, and associated with DCM (Figure 1A ), they have been shown to have different behaviours when transiently expressed [ 27 , 28 ]. R482W and L530P are associated with different disease phenotypes (lipodystrophy and myopathy, respectively), and although they are both within the globular domain, R482W is a surface residue, whilst L530P is internal (Figure 1B ). Since these mutations are responsible for autosomal dominant forms of disease they should still exert their molecular phenotype in the cell in the presence of wild-type (wt) protein. Both FLAG-tagged and GFP-tagged prelamin As were transiently transfected into human fibrosarcoma cells. Each protein was processed into mature lamin A [ 29 ] and incorporated into the nuclear lamina, as evident by the bright nuclear ring of staining visualised either by immunofluorescence with anti-FLAG antibody or from the GFP signal (Figure 2 ). The mutant forms of lamin A generally had a more uneven distribution at the nuclear periphery, compared to wt, as has been reported previously [ 30 ]. We saw high levels of N195K lamin A in the nucleoplasm in addition to the nuclear periphery, but we did not see much evidence for its aggregation into intra-nuclear foci, as has been reported in mouse myoblasts and embryonic fibroblasts [ 27 , 28 ],. This might reflect differences in cell-type or relative expression levels of the mutant protein. Apparently internal sites of epitope-tagged lamin As are seen, but analysis of 3D image stacks (Figure 2B ) shows that these are invaginations of the nuclear periphery and not intra-nuclear foci. Such invaginations has previously been reported in many types of cultured cells [ 31 - 33 ]. Figure 2 Sub-cellular localisation of epitope-tagged lamin As. A) Detection of FLAG-tagged wt and mutant lamin As transfected into human HT1080 fibrosarcoma cells. The FLAG tag was detected by immunofluorescence with M2 anti-FLAG (red in merge), in DAPI stained nuclei (blue in merge). Bar = 10 μm. B) Detection of GFP-tagged wt and mutant lamin As transfected into human HT1080 fibrosarcoma cells. GFP signal in images collected at 2 μm intervals from the top to the bottom of the nucleus is shown in black and white. The merged colour images (far right) show mid-plane images of the GFP signal (green) in DAPI stained nuclei (blue). Bar = 10 μm. Analysis of lamin A dynamics by FRAP The lamin A mutations that we have studied are within different domains of the protein (Figure 1A ), or within different parts of the same structural domain (Figure 1B ). Therefore they likely have different interactions, either with other molecules of lamin A, or with other proteins of the nuclear periphery or nucleoplasm. Such interactions affect the kinetic properties of a protein, and photobleaching and time-lapse imaging can probe this [ 34 ]. We therefore analysed the mobility of GFP tagged wt and mutant lamin As by FRAP in transiently transfected human cells. In each case a region at the nuclear lamina was bleached. The fluorescence within a 1.8 × 1.8 μm region of interest (ROI) of this bleach region was then followed every 5 minutes over a period of up to 65 minutes. To calculate the loss of fluorescence attributed to the imaging process alone, the sum of pixel intensities was also calculated for a control (unbleached) cell in each case. This was used to normalise the fluorescence intensity for each ROI [ 35 ]. The mean relative fluorescence intensity for each time point was then calculated for 9 cells of each of the GFP-lamin A proteins (WT, L85R, N195K, R482W and L530P). For wild-type lamin A, fluorescence at the nuclear lamina is visibly bleached (t = 0 in Figure 3A ), and only about 20% of the signal recovers over the time course of the experiment (Figure 3C ). This indicates that a large proportion (~80%) of lamin A at the nuclear periphery is immobile, at least within the time-frame of these experiments. This is similar to the reported immobility of 60% of lamin B receptor (LBR) in the INM [ 36 ]. The recovery curve shows that wt lamin A moves back into the bleach area only very slowly (Figure 3C ). The extrapolated t1/2 is ~140 minutes, similar to that reported for lamin B1 (>180 mins) [ 37 ]. GFP-tagged lamin C expressed in CHO cells has also been reported to show very little recovery after 1 hour [ 33 ]. Most nuclear proteins e.g. transcription factors, and even chromatin-associated proteins such as HP1, are very dynamic with t1/2 values in the range of a few seconds [ 38 ]. Even the INM proteins emerin, Lap2β, and Man1 have recovery halftimes of about 1 minute [ 39 ]. The slow recovery of lamin A is compatible with its incorporation into a stable polymer that only exchanges subunits very slowly. Figure 3 FRAP analysis of wild type and mutant lamin As. A and B) Single z -plane confocal images of GFP-tagged (A) wt and (B) N195K lamin A expressing cells. Images were captured before (t = -5) and immediately after (t = 0) photobleaching of an area of the nuclear periphery, and at 5 min intervals thereafter. The bleach region is boxed in red. C) Graphs of mean (± s.e.m) relative fluorescence in the bleach area during FRAP, averaged over 9 cells each. In each graph, data for wt (black) and a mutant (red) lamin A are compared. The recovery kinetics for the R482W lamin A mutant are indistinguishable from wt and the extrapolated t1/2 = 145 mins (Figure 3C ). However, the other lamin A mutants analysed show significant differences. The L85R and L530P mutant proteins appear to be more mobile than wild-type lamin A. They recover more rapidly: t1/2 L85R = 75 mins, L530P = 80 mins. Compared to wt, a higher proportion of the L85R fluorescence (35%) also recovers, suggesting that less of this mutant lamin A is in an immobile fraction. The most dramatic difference in dynamics was seen for the N195K mutant. Compared to the other lamin As it does not bleach to the same extent, and this is attributable to rapid diffusion of the high levels of nucleoplasmic protein, since at t = 0 recovery of fluorescence can be seen in the nucleoplasmic part of the bleach region, but not in the nuclear periphery itself (Figure 3B ). It is known that in early G1 cells the nucleoplasmic pools of lamin A recovery their fluorescence immediately following bleaching [ 37 ]. However, even within the nuclear periphery fluorescence recovers within the observation period (Figure 3B , t = 15) and the t1/2 = 30 mins (Figure 3C ). Therefore the N195K lamin A mutant is considerably more mobile within the nuclear lamina than wt lamin A, or indeed the other lamin A mutants studied here. Analysis of lamin A dynamics by FLIP To further analyse the movement of lamin A within the nuclear lamina, and between the lamina and the nucleoplasm, FLIP experiments were performed on wt, and N195K and L530P mutant GFP-lamin A expressing cells. After successive rounds of photobleaching at a region of the nuclear periphery, the fluorescence at a region of the nuclear periphery distant from the bleach, and at a region within the nucleoplasm were measured (Figure 4 ) in 10 cells each. As in FRAP, the data was normalised for the loss of fluorescence caused by the successive rounds of imaging. Figure 4 FLIP analysis of wild type and mutant lamin As. A) Single z -plane confocal images of a GFP-tagged wt lamin A expressing cell captured before (left) and immediately after (right) a round of photobleaching of an area of the nuclear periphery (red box). Fluorescence was also recorded for an unbleached area (blue box) of the nuclear periphery, and a region of the nucleoplasm (green box). Bar = 10 μm B ) Graphs of mean (± s.e.m) relative fluorescence in the bleach area (red) during successive rounds of FLIP, and in unbleached regions of the nuclear periphery (blue). and the nucleoplasm (green). Data are averaged over 10 cells each for wt lamin A and for the L530P and N195K mutant lamin As. For both wt and L530P lamin A there is little loss of fluorescence from either a distant region of the nuclear periphery, or the nucleoplasm after repeated rounds of photobleaching (Figure 4B ). This reflects the slow FRAP recovery kinetics of these forms of lamin A (Figure 3 ). In contrast, the nucleoplasmic fraction of the N195K mutant lamin A shows a substantial decrease (24%) in fluorescence after successive rounds of bleaching at the nuclear periphery. This may reflect diffusion into the small region of nucleoplasm contained within the bleach region, but could also be due to exchange of protein between the nucleoplasm and the lamina. A 10% decrease in fluorescence is also seen at a non-bleached part of the nuclear periphery. This suggests that there is enhanced lateral movement of mutant lamin A within the nuclear lamina compared to wild-type protein. Conclusions For GFP-tagged wild-type lamin A we have determined that a large proportion of the protein at the nuclear periphery is immobile (Figure 3 ), and that any recovery of fluorescence that does occur there is very slow (t1/2 = ~140 mins). This is consistent with the slow recovery halftimes of lamin B1 [ 37 ], and the incorporation of lamin A into a stable IF polymer at the nuclear lamina. Of the four laminopathy-associated mutant forms of lamin A studied by photobleaching all, except for R482, show altered dynamics relative to wt protein. The R482W mutation is associated with FPLD, and other lamin A mutations found in this disease are also either a loss of positive charge at R482, or K486, or the gain of a negative charge (G465D). The amino acid residues involved all map to a solvent-exposed surface in the structure of the Ig-like C-terminal domain [ 26 ] (Figure 1B ). By NMR and circular dichroism the structure and thermostability of the R482W mutant is similar to that of wt lamin A [ 26 ]. Our analysis suggests that the dynamics of the R482W mutant protein within the cell are also similar to wt. It has been suggested that FPLD-associated mutations of lamin A do not destabilise the Ig-like domain of lamin A, but may alter the interaction of the protein with other cellular components. The Ig-like lamin A domain interacts with LAP2α [ 16 ], emerin [ 28 ], DNA [ 10 ] and SREBP1 [ 18 ]. Emerin can still interact with R482W lamin A [ 16 ], though altered emerin-lamin A interactions have been reported for the R482L mutation [ 40 ]. Mutations at R482 have a 5-fold lower affinity for DNA binding in in vitro assays [ 10 ], and a slightly lower affinity for SREBP1 [ 18 ]. We suggest that if lamin A-protein or -DNA interactions are perturbed by the R482W mutation they are not sufficient to affect the dynamics of lamin A movement within the nucleus. The EDMD-associated L530P mutation is also within the Ig-like domain (Figure 1 ), but unlike R482W it is located inside of the structure and so is predicted to destabilise protein folding [ 25 , 26 ]. Compared with wild-type and R482W lamin A, we detected increased mobility of L530P lamin A within the nuclear lamina by FRAP (Figure 3 ). Expression of L530P has been reported to result in decreased emerin localisation at the INM [ 30 ]. Therefore, the stability of both emerin and lamin A at the nuclear periphery may be mutually dependent. In the absence of lamin A, emerin completely fails to localise at the INM [ 3 , 5 , 6 ]. Our analysis of protein dynamics suggests that an altered interaction between emerin and lamin A could alter the stability of the nuclear lamina, reflected in the increased mobility of lamin A. Missense mutations in the coiled-coil domain of lamin A are associated with the myopathies, not FPLD (Figure 1 ). They likely impair the dimerization and formation of higher-order filaments of lamin A. The increased mobility of the L85R mutant lamin A, as assayed by FRAP (Figure 3 ), would be consistent with this. The most dramatic change in lamin A dynamics was seen with the N195K form. FRAP indicates that it is considerably more mobile than wt lamin A (Figure 3 ). FLIP suggests that there might be exchange between the nucleoplasmic and lamina pools of this mutant protein, as well as enhanced mobility within the nuclear lamina (Figure 4 ). Like L530P, this mutation is also in the coiled-coil domain, but clearly has a more drastic affect on lamin polymerisation and intra-nuclear dynamics. Given the genetically dominant nature of many of the laminopathies, it would be interesting to determine whether the presence of a mutant lamin A has an affect on the mobility of the (GFP-tagged) wild-type protein in the same cells. FLPD is clinically distinct from AD-EDMD and DCM. Patients with FPLD do not have striated muscle pathology, conversely adipose tissue is normal in AD-EDMD and DCM. Whereas we find increased mobility of all the myopathy-associated lamin A mutants we studied, we cannot distinguish between the protein dynamics of wt and an FLPD mutant form of lamin A (R482W). We conclude that in AD-EDMD and DCM laminopathies the structure of the nuclear lamina is perturbed in such a way as to allow for more rapid exchange of lamin A molecules. In contrast, we suggest that, in this respect, the structure of the lamin polymer is normal in FPLD. Methods Generation of GFP-tagged lamin A constructs Green Fluorescent Protein (GFP)-tagged human lamin A cloned in pCDNA-3-EGFP (GFP-HLA) was obtained from L Karnitz (Mayo Clinic, Rochester). GFP-tagged mutant lamin As were then generated by transfer from N-terminal FLAG-tagged fusion constructs [ 27 ]. GFP-HLA was digested with AccI/EcoRI and fragments of 5.4 Kb (the vector backbone) and 894 bp (GFP coding sequence plus 0–175 bp of lamin A) were purified. FLAG-pre-lamin A coding sequences (carrying laminopathy mutations) were then excised from each of the pSVK vectors using EcoRI/SalI, and cloned into XhoI/EcoRI digested pUC21. A 1.8 kb AccI/SpeI fragment of the coding sequence (removing the FLAG tag and the first 175 bp of lamin A) was purified. The AccI site is upstream of each mutant codon and the SpeI site is downstream of the stop codon. A three way ligation was then performed of this fragment together with the 5.4 Kb and 894 bp AccI/EcoRI fragments from GFP-HLA. Cell transfection and immunofluorescence Human HT1080 fibrosarcoma cells were transfected with plasmids using Lipofectamine™2000 according to the manufacturer's recommendations. Cells grown on glass slides were fixed 24 hours later for immunofluorescence or GFP analysis. Cells were fixed for 10 mins in 4% paraformaldehyde and permeabilised for 10 mins in 0.05% Triton-X100. FLAG-tagged proteins were detected with 1:200 dilution of M2 anti-Flag mouse monoclonal antibody (Sigma) and 1:100 anti-mouse Texas Red Fab'2 heavy and light chains (Jackson Labs). Slides were counterstained with DAPI and analysed using a Zeiss Axioplan microscope fitted with a Xillig CCD camera and a focus motor to collect images at 0.5 μm intervals in the z plane [ 41 ]. Live cell analysis Cells were grown on DeltaT 0.17 mm culture dishes (Bioptechs Inc) and were mounted onto a heated stage (Bioptechs Inc) on a Zeiss LSM510 confocal microscope. An objective warmer (Bioptechs Inc) was also used to help to maintain a stable temperature of the medium in the culture dish. Photobleaching For FRAP, a 1.8 × 1.8 μm region at the nuclear periphery in the mid-focal plane was bleached with 100 iterations at 100% power of the argon laser running at 6.1 mA (50% power). The pinhole size for the confocal was set at 1 Airy unit. The time series software option was used to specify the appropriate time delay between rounds of 3D image stack capture. Each bleached cell was imaged with a ×100 objective, in a window that included other non-bleached cells to allow for relative fluorescence levels to be normalised. Immediately following the bleach, images in the same z-plane were captured at 1s intervals (t = 0 in Figure 3 ). Thereafter 3D z-plane stacks were captured of the cell at 5 min intervals for a further 65 mins, using 8% of laser power. Because of the length of FRAP analysis, nuclear rotation, cell movement and focus drift presented a problem in registering the bleach ROI between time points. To account for this, the best z -plane image for the bleach ROI was selected from each time point 3D stack. Each of these was then processed by an interactive rotation script (v3.6 IPLAB, Scanalytics) to correct for nuclear rotation and cell movement. This enabled all images to be superimposed with the pre-bleach image. For FLIP, an ROI at the nuclear periphery was bleached with 10 laser iterations at 100% of 50% total laser output (~6.1 mA). Following the bleach, 5 images were taken at 2 sec intervals using 8% of laser output. The bleach procedure was repeated for 16 rounds. In both FRAP and FLIP, the loss of fluorescence attributed to the imaging process alone was assessed from the sum of pixel intensities in a control (unbleached) cell, in each analysis. The relative fluorescence intensity over time was calculated for each defined ROI using a normalisation equation [ 35 ]. Authors' contributions SG constructed the GFP-tagged lamin A constructs, did the cell transfections, the fluorescence microscopy and the photobleaching analysis. NG gave assistance and advice in the photobleaching studies. PP advised and assisted in confocal microscopy and wrote the scripts for image registration over the time course of FRAP. CO and HJW constructed the FLAG-tagged lamin A mutants and provided advice. WAB conceived of the study and drafted the manuscript. All authors read and approved the final manuscript | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539277.xml |
517808 | A mechanism of airway injury in an epithelial model of mucociliary clearance | We studied the action of sodium metabisulphite on mucociliary transport in a frog palate epithelial injury model, hypothesizing that it may be useful for the study of mechanisms of airway injury. Sodium metabisulphite (MB) releases SO 2 on contact with water. SO 2 is a pollutant in automobile fumes and may play a role in the exacerbation of airway disease symptoms. We first investigated its effect on mucociliary clearance. MB 10 -1 M, increased mucociliary clearance time (MCT) by 254.5 ± 57.3% of control values, (p < 0.001, n = 7). MB 10 -4 and 10 -2 M did not interfere with mucus clearance time compared to control values. In MB-treated frog palates, MCT did not return to control values after one hour (control, 97.3 ± 6.3% vs. MB, 140.9 ± 46.3%, p < 0.001, n = 7). Scanning EM images of epithelial tissue were morphometrically analyzed and showed a 25 ± 12% loss of ciliated cells in MB palates compared to controls with an intact ciliary blanket. Intact cells or groups of ciliated cells were found in scanning EM micrographs of mucus from MB-treated palates. This was associated with increased matrix metalloproteinase (MMP-9) activity in epithelial tissue and mucus. We suggest that the loss of ciliated cells as a result of MMP-9 activation prevented full recovery of MCT after MB 10 -1 M. The mechanism of action may be on epithelial cell-cell or cell-matrix attachments leading to cell loss and a disruption of MCT. Further studies are warranted to determine whether this is an inflammatory mediated response or the result of a direct action on epithelial cells and what role this mechanism may play in the progression to chronic airway diseases with impaired mucociliary clearance. | Background Particle clearance in the airways is dependant on mucus and cilia [ 1 ]. The cilia beat frequency, mucus secretion rate and the properties of mucus are variables important in normal and effective mucociliary clearance [ 2 ]. However, the study of mucociliary clearance in intact mammalian airways in humans or small mammals is technically difficult. It is worthwhile, therefore, to develop alternate models that, by way of ease of preparation and homology to human conductive airways, can yield important knowledge in understanding the basic mechanisms involved in airway diseases. The bullfrog palate provides an excellent integrated model system for studying all the relevant variables for mucociliary clearance including mucus secretion rate, cilia beat frequency, linear velocity of mucus, the viscoelastic properties of mucus and the transepithelial potential difference, indicative of changes in epithelial ion fluxes and water transport [ 2 ]. We have extended the physiological applications of the frog palate model to study the initial events of airway injury. To create an injury model from the fresh frog palate model, a solution of sodium metabisulphite was topically applied to the palate. Sodium metabisulphite has been shown to release sulfur dioxide (SO 2 ) on contact with water and has been employed as an aerosol in other airway injury models to study hypersecretion and hyperplasia [ 2 - 6 ]. In dog studies, chronic exposure to SO 2 produced symptoms similar to chronic bronchitis in humans [ 3 ]. We hypothesize that sodium metabisulphite will interfere with mucociliary clearance on the frog palate by disrupting the action of the ciliated epithelium, vital to the process of mucociliary clearance. The objective of this study was to evaluate the effect of sodium metabisulphite on mucociliary clearance on the frog palate. A further objective was to analyze tissue and mucus samples in ultra-structural and molecular studies to characterize the nature of the injury and to assess the potential involvement of matrix metalloproteinases which have been shown to play a role in airway injury and remodeling [ 7 , 8 ] and in cell-signaling pathways [ 14 ]. Materials and Methods Development of a frog palate injury model A fresh frog palate was prepared as previously described [ 1 , 2 ]. Briefly, the upper palate of the bullfrog ( Rana catesbiana ) was excised by cutting in the coronal plane from the lateral border of the mouth on one side of the head to the other. The excised palate was placed horizontally on gauze soaked in frog Ringers (2/3 Ringers + 1/3 distilled water, 207 mosml L -1 ) in a Petri dish. The palate was placed in an enclosed chamber (20 × 20 × 30 cm) maintained at a constant temperature (22–24°C) and continuously humidified at 100% with aerosolized frog Ringers generated by a Pari Jet ® nebulizer at a airflow rate of 8 L/min. The palate was allowed to stabilize for 15–20 min before any procedures were carried out on the palate. Mucociliary clearance time (MCT) was measured by applying a droplet of mucus collected from the inferior (cut) edge of the palate that was placed at the superior edge of the palate near the midline. The action of cilia carries the mucus toward the inferior edge. The effect of various concentrations of sodium metabisulphite on MCT was measured following topical application on the palate. Frog Ringers was used as a control solution and vehicle for sodium metabisulphite. The volume of solution (either frog Ringers or sodium metabisulphite) that was applied to each palate was normalized among different sized palates according to the area of the palates surface. The area of the palate was approximated by measuring across the lateral-most borders of the jaw at the base of the palate, and calculating the area of the equivalent half-circle. The volume of solution applied was normalized to the area of each palate as shown: area = 3.5 cm 2 (volume applied = 2 μl), 4.5 (3 μl) 5.5 (4 μl) to 6.5 cm 2 (5 μl). Using bromophenol blue in frog Ringers applied to the palate, it was shown that within two minutes of application, the solution was carried from the superior edge of the palate to the inferior edge by ciliary action. Therefore, when frog Ringers was applied to the palate, two minutes was allowed for the droplet of solution to disperse on the palate. This was followed by the measurement of MCT using a drop of frog mucus collected off the inferior (cut edge) of the palate and marked with carbon particles to enhance its visibility on the palate surface. The movement of the mucus droplet down the palate by ciliary action was observed through a stereomicroscope with a reticulated eyepiece and timed over a set distance of 4 mm, once it reached a steady speed. For each solution tested, five consecutive mucus clearance times were recorded and the average was used as the time point for that particular group of recordings. After a recovery period, sodium metabisulphite 10 -4 M was applied to the palate. After two minutes, another five measurements of MCT were recorded followed by a recovery period. At this point in time (70 min, shown in Figure 1 ), frog Ringers was applied and MCT was measured again. If this value was within 10% of the first application, the palate was considered to have recovered back to the control condition. Sodium metabisulphite 10 -2 M (at 80 min) was applied followed by the measurement of MCT again. This was followed with a recovery period with the measurement of frog Ringers MCT again, which was shown to be not different from the previous controls. Figure 1 The effect of sodium metabisulphite on mucociliary clearance time (MCT). The results of seven independent experiments performed on seven different frog palates are shown in real time as displayed on the x-axis of the graph. Application of frog Ringers (FR) is shown by grey bars, while black bars indicate the application of sodium metabisulphite shown by the concentration (10 -4 , 10 -2 or 10 -1 M). Frog Ringers following each recovery period also represented a timed control prior to each dose of metabisulphite. However, in order to control for deterioration of palate over the course of the experiment, three sets of frog Ringers controls measured before the application of sodium metabisulphite 10 -1 M were plotted versus time and a line of best fit was determined (data not shown). No change in the significance of the slope of the line (equal to or close to '0') indicated that no significant deterioration of the palate had taken place over time. Within each individual experiment there was, however, some variability in controls. Therefore the control mucociliary clearance times that were used to determine a line of best fit (taken as 100%) were compared to the actual MCT measured at that particular point. Thus for each experiment the actual MCT was expressed as a percentage of the line of best fit of the control time which was extrapolated to the time of application of frog Ringers or sodium metabisulphite to the palate as shown ((actual MCT/predicted control MCT) ×100). Thus, there is variation within and between controls that are shown as a standard deviation for each time point (representing seven independent frog palate experiments). MCTs for metabisulphite were expressed as a percentage of the line of best fit for frog Ringers controls, extrapolated to the time metabisulphite was applied. An increase in MCT compared to control times, indicated a slowing of the mucociliary clearance time. A minimum of fifteen minutes was allowed after metabisulphite, for MCT to return to the normal range, i.e. within 10% of the frog Ringers MCT, measured prior to metabisulphite. If recovery of MCT had not occurred after twenty minutes to within the range specified, frog Ringers was re-applied and the recovery period was repeated. Injury to the palate A 50% increase in MCT was established a priori as indicative of a quantifiable injury to the mucociliary clearance system. Sodium metabisulphite was applied in progressively increasing concentrations from 10 -4 , 10 -2 and 10 -1 M. Each test solution was alternated with frog Ringers. A higher concentration of metabisulphite was not applied until the MCT had returned to within 10% of the previously measured frog Ringers control. In several experiments, pH was measured on the surface of the frog palate, using a solid-state micro pH electrode (Lazar Research Laboratories, Los Angeles, CA) connected to an Accumet ® pH meter (Model 925, Fisher Scientific, Nepean, ON, Canada) to continuously monitor changes on the palate surface during the application of sodium metabisulphite. Scanning electron microscope (SEM) studies Samples of frog palate epithelial tissue and mucus were placed in 2.5% glutaraldehyde solution, immediately after collection and kept in a refrigerator at 4°C until processing. Samples were prepared for the SEM by standard methodology. Briefly samples were post-fixed in 1% osmium tetroxide in Milonig's buffer at room temperature for one hour. They were then washed briefly in distilled water and dehydrated in an increasing series of ethanol (50, 80 and 100%), ten minutes at each concentration, followed by two additional periods of absolute ethanol. The samples were further dehydrated by critical point drying at 31°C for 5–10 minutes, and then mounted on a specimen holder for drying overnight in a desiccator. In the final stage of preparation before viewing, the samples were sputter coated with gold (Edwards, Model S150B Sputter Coater) and examined with a Hitachi 2500S scanning electron microscope. High-resolution digital images were acquired directly to a computer for storage and reproduction. Morphometry To quantify the area of cilia loss in fields of view in the electron microscope studies, image files were analyzed using Sigma Scan ® image analysis software to trace areas of cell loss and determine the areas of loss relative to the field of view. Fifteen fields from 3 samples exposed to sodium metabisulphite 10 -1 M were examined as well as samples from control tissue (exposed only to frog Ringers). Gelatinase zymography Samples of frog palate epithelial tissue were removed following mucus clearance studies, snap frozen in liquid nitrogen and stored at -80°C until they were prepared for zymography. At that time the tissue samples were ground with a mortar and pestle to a powder, adding liquid nitrogen to the mort to keep the tissue frozen. Homogenization buffer (KCl, ZnCl 2 , EDTA and Tris-HCl) was added to the samples that were sonicated for 30 seconds and then centrifuged at 14,000 rpm for 15 minutes. The supernatant was collected and an aliquot removed for protein assay (BCA protein assay kit, PIERCE). A 10 μl sample, normalized for protein content, was loaded on a separating gel (acryl amide and gelatin) and run at 120 volts for one hour. After electrophoresis, the gel was washed for one hour in 25% Triton-X100 at room temperature followed by incubation overnight in zymography development buffer (0.15 M NaCl, 0.5 mM CaCl 2 , 0.05% Azide NaN 3 , 50 mM Tris-Cl, 2 M Tris-HCl). The gel was then stained for 2 hours with 0.05 % Coomassie blue (R-250) in methanol: acetic acid: water (2.5:1:6.5) followed by de-staining in 20% isopropanol in 4% ethanol and 8% acetic acid. The presence of gelatinases (MMP 2 and 9) was shown by clear bands (no staining) corresponding to MMP standards (MMP 2 and 9) run in leftmost lane on the gel. Optical density was measured in a Bio-Rad Scanning densitometer. Statistical treatment of data All measurements were expressed as mean ± standard deviation. Overall significance of the MCT results were tested using a one-way analysis of variance in SPSS, with differences among groups (of more than two) evaluated using planned orthogonal comparisons. For comparisons between two groups (density comparisons between control and MB in zymograms), a Student T-test was used. The level of significance was set at p < 0.05. Results Figure 1 shows the effect of sodium metabisulphite on the MCT expressed as a percent of frog Ringers control times. MCT is shown for frog Ringers, sodium metabisulphite 10 -4 , 10 -2 and 10 -1 M and 3 consecutive recovery periods following sodium metabisulphite 10 -1 M in which frog Ringers was applied to the palate in twenty-minute intervals followed by a measurement of MCT. The average frog Ringers MCT (in 7 frogs) measured initially at 15 minutes following an initial stabilization period was 97.3 ± 6.3 %. No difference in MCT was measured after the application of sodium metabisulphite 10 -4 (30 min) and 10 -2 M (60 min); whereas 10 -1 M sodium metabisulphite (at 100 min) increased MCT by 254.5 ± 57.3% compared to Ringers control MCT (taken as ~100%). Between 10 -4 and 10 -2 M sodium metabisulphite, there was no significant difference compared to control MCTs. This is illustrated by the dotted line in Figure 1 . However, twenty minutes after sodium metabisulphite 10 -1 M, frog Ringers was applied but MCT did not recover to previous frog Ringers control times. Another twenty minutes of recovery was allowed and frog Ringers MCT was still not recovered. After an additional twenty minutes, frog Ringers MCT was measured for the third consecutive time, showing that after one hour of recovery (~170 min in the time course of the experiment), the MCT was still significantly different from the initial frog Ringers MCT (140.9 ± 46.3 vs. 97.3 ± 6.3%, p < 0.001, n = 7). MCT was significantly increased after sodium metabisulphite 10 -1 M. To determine if this acute effect was due to pH changes, possibly representing altered ion fluxes in the tissue, a micro pH electrode was placed on the palate to measure pH before and after the application of metabisulphite to the palate surface. This is shown in Figure 2 . Prior to metabisulphite, the pH on the surface was 6.8–7.0 units. The pH was not significantly altered after the application of sodium metabisulphite 10 -4 and 10 -2 M. However, after sodium metabisulphite 10 -1 M, the pH declined within seconds, reaching a nadir at ~60 seconds. After 300 seconds, there was some recovery toward normal, but the pH was still 0.3–0.5 units below the initial control value. As shown in Figure 1 , MCT recovered somewhat by 20 minutes after sodium metabisulphite 10 -1 M and showed continued (but incomplete) recovery after one hour. No corresponding pH measurements were taken at these time points. Figure 2 The effect of sodium metabisulphite on the pH, measured continuously on the surface of the palate, is shown for before and after three concentrations of sodium metabisulphite were applied (vol = 5 μl) to the palate. Scanning electron microscope studies In Figure 3 , Panel A (X400) shows the normal cilia blanket, with pores of secretory cells visible. Panel B (X400) shows regions of the palate surface devoid of cilia after 10 -1 M sodium metabisulphite was applied. The normal continuous covering of cilia is shown greater detail in Panel C (x3500) and in Panel D after 10 -1 M sodium metabisulphite, where a region of exfoliation is shown more clearly at the higher magnification. The absence of cilia and ciliated epithelial cells is visible, with only the extracellular matrix remaining. Morphometry to quantify the area of exfoliation determined in a five different field from three independent experiments, revealed that after sodium metabisulphite 10 -1 M, there was a 25 ± 11.8% loss of ciliated epithelial cells from these palates compared to none in control palates. Figure 3 Scanning electron micrographs of control and MB-treated palates at a magnification of 400× (panels A and B respectively) and at 3500× (panels C and D respectively). In panel A, the ciliated epithelium completely covers the surface of the palate except where the openings to secretory cells are seen. In panel B, it can be seen that the ciliated surface is not continuous, but punctuated with numerous spaces where ciliated cells are not present. Panel C shows the high density of cilia on the palate surface, which under normal transport conditions, beat in a metachronal pattern to move a mucus layer over them. In panel D, the continuity of the ciliated layer is interrupted by spaces where ciliated epithelial cells are no longer present. SEM of the palate surface following sodium metabisulphite 10 -4 and 10 -2 M, showed no ultra structural changes compared to control palates to which frog Ringers had been applied. Figure 4 shows a split micrograph of mucus collected from a palate after sodium metabisulphite 10 -1 M. At lower power (X400) a grouping of ciliated cells are visible in the mucus. At the higher power (x2000), intact ciliated epithelial cells are clearly shown. Figure 4 A sample of mucus taken off the palate after MB treatment showed groups of intact ciliated cells. This would suggest that the cells, which were exfoliated from the epithelial surface, were carried off the palate in the mucus layer by the process of mucociliary clearance. Gelatinase zymography Figure 5 shows two representative zymograms from tissue and mucus. In 5A from tissue, in the left lane, two bands are visible representing MMP-9 (92 kD) and MMP-2 (72 kD) standards. In sodium metabisulphite 10 -1 M treated tissue (two rightmost columns), bands representing MMP 9 activity were seen whereas only faint bands were visible in control tissue. Statistical comparison of densitometry bands showed significant activation (p < 0.05, n = 3). MMP-2 activity (in the bottom row on the zymogram) may have also increased, but since control tissue showed similar activation these results are inconclusive. A similar state of MMP activation in mucus is shown in Figure 5B . Increased activated MMP-9 was observed in the mucus from metabisulphite-treated palates (p < 0.05, n = 3) compared to mucus from frog Ringers-treated palates. To test if MMP-9 activation was related to sodium metabisulphite concentration, samples of epithelial tissue were treated with sodium metabisulphite 10 -4 , 10 -2 and 10 -1 M, and prepared for zymography (Figure 6 ). Optical density analysis showed that activation of MMP-9 after sodium metabisulphite 10 -2 M was greater than after sodium metabisulphite 10 -1 M (#, p < 0.05, n = 3) while both were greater than MMP-9 activation following application of 10 -4 M sodium metabisulphite (*, p < 0.05, n = 3). Figure 5 Representative zymograms from tissue (A) and mucus (B) shows the standards for MMP9 (top band, ~92 kD, latent size) and MMP2 (lower band, ~72 kD, latent size) in the leftmost lane. To the right of standard in each zymogram, two sets of bands are visible, corresponding to MMP-9 and MMP-2 levels of activity in duplicate samples of control tissue. In the next two lanes are duplicate sets of bands from an experiment which shows increased activated MMP-9 and possibly MMP-2 activity in sodium metabisulphite 10 -1 M treated tissue in both tissue and mucus. The bar graph only shows a comparison of the scanning density of the MMP-9 bands since the MMP-2 control and sodium metabisulphite-treated tissue showed similar activation. A significant increase in activated MMP-9 was seen in sodium metabisulphite-treated mucus and tissue (* p < 0.05, n = 3 for each). Figure 6 MMP-9 activation in palate tissue is a dose-related effect. The representative zymogram shows bands corresponding to MMP-9 activity in tissue samples treated with MB 10 -1 , 10 -2 and 10 -4 M. The MMP-2 bands have been removed from this gel as no differences were seen. Densitometry of the MMP-9 bands showed that MB 10 -1 M showed less activity than MB 10 -2 M, whereas MB 10 -4 M showed significantly less activation than either of the higher doses. The bar graph shows the average results in tissue from three separate experiments. Discussion The important findings of this study are: 1. the development of a model of airway epithelial injury that can be used for study of ultra-structural and molecular events in airway injury that are directly related to the disruption of mucus clearance; 2. that sodium metabisulphite (by releasing SO 2 on contact with water) has an acute effect on mucus clearance followed by incomplete recovery of mucus clearance time; 3. ultra-structural studies showed that areas of ciliated epithelial cells were lost from the palate surface resulting in an incomplete recovery of mucus clearance. Loss of cilia has been previously reported following exposure to SO 2 in dogs [ 3 ]. The implication is that loss of cilia may affect mucus clearance in number of airway diseases. The mechanism of this effect requires further study for a more complete understanding of the events involved in this process. Intact ciliated epithelial cells were found in the mucus from 10 -1 M sodium metabisulphite-treated palates but not from frog Ringers-treated control palates; 4. Gelatinase zymography showed increased activity of MMP-9 after sodium metabisulphite (10 -4 to 10 -1 M) and this was shown to be a dose-related effect. It is noteworthy that gelatinase zymography showed increased activity of MMP-9 at each concentration of sodium metabisulphite, whereas ultrastructural damage was only found at the highest concentration; 5. The finding that intact ciliated cells were found in the mucus suggests that the action of activated gelatinases was on cell-cell or cell-matrix attachments resulting in the exfoliation of intact ciliated epithelial cells, which may have contributed to a slowing of mucus clearance over the surface of the palate. Additional studies are underway in our laboratory to identify possible the role of inflammatory mediators in the activation of matrix metalloproteinases in this model. Sodium metabisulphite may cause the release of oxidants or other mediators by epithelial cells [ 10 , 11 ] or from typical inflammatory cells, possibly activated neutrophils resident in the tissue, although the question of a time frame, related to neutrophil recruitment and activation would need to be clarified [ 12 ]. Oxidant products may cause activation of precursor forms of collagenase or gelatinase, leading to breakdown of the extracellular matrix [ 14 ]. It has been recently shown that mechanical stress resulted in the expression and release of gelatinases from epithelial and endothelial cells in the rat lung [ 7 ]. Further studies need to be undertaken to identify the source of MMP release following sodium metabisulphite and other airway modulating agents. A high concentration of sodium metabisulphite may not be biologically relevant and represents a practical limitation to the applicability of the model. Nevertheless, a dose-response curve showed little effect on mucus clearance in the frog palate model at lower concentrations of sodium metabisulphite. Our findings suggest that this ex vivo model may be particularly useful in characterizing how an initial injury may be induced in ciliated epithelium. The ability to make functional measurements of mucociliary clearance in the ex vivo frog palate model allows for a correlation of variables in follow-up in vitro studies of tissue and mucus that may be interfering with mucociliary clearance. Sodium metabisulphite, when applied to the palate is diluted in the periciliary fluid [ 9 ]. The dilution in palate surface fluid reduces the concentration of the applied metabisulphite. By approximating the area of the palate as one-half the area of a circle (~5 cm 2 on average) and assuming a mucus plus periciliary layer of 10 μm, a volume of 5 μl would effectively be diluted by as much as 1–2 orders of magnitude (assuming it spread over at least half the area of the palate). This calculation would suggest that sodium metabisulphite 10 -1 M was effectively and rapidly diluted to 10 -2 M or less. It follows that the lower concentrations of metabisulphite would be effectively less than the stock concentrations. Since, the effective concentration was determined experimentally in a dose-response experiment as that dose that produced a 50% or greater increase in the mucus clearance time, and since only the highest concentration of metabisulphite produced this effect, this concentration became physiological relevant to the outcome of these experiments. Lower concentrations (10 -4 and 10 -2 M) were also used, even though no effect on mucociliary clearance time was observed in the dose-response experiments, to determine if there might be some quantifiable effect at the cellular level, which was not manifested as a decrement in mucociliary clearance. In several experiments the continuous pH response following the application of sodium metabisulphite 10 -1 M to the palate surface was monitored for 5 minutes. The pH measured on the palate prior to metabisulphite was 6.9 ± 1.4 units. There was a decrease in pH following sodium metabisulphite 10 -1 M, reaching a nadir of 6.4 ± 0.25 pH units after 60 seconds. Sodium metabisulphite 10 -4 and 10 -2 M did not cause any decrease in pH on the palate after application. Although the observed decrease in pH with sodium metabisulphite 10 -1 M is relatively minor (<0.5 pH units), it may have been sufficient to influence ion channels, possibly disrupting ciliary beating and causing chemical changes such as the induction of inflammatory mediators [ 5 , 15 ]. The dramatic increase in mucus clearance time seen 1–2 minutes after the application of MB 10 -1 M occurred in a similar time frame to the pH changes. After five minutes, the pH was returning toward normal, and within 20 minutes there was some recovery of mucus clearance time. An in vitro study [ 13 ] examined the effect of pH changes on ciliary beat frequency and found that the beat frequency was stable between 7.5 and 10.5 pH units. A significant decrease in beat frequency was noted at lower pH values. This report is consistent with our study that suggests that the transient decrease in pH caused a transient slowing or even cessation of ciliary beat frequency. The increase seen in mucus clearance times after 10 -1 M sodium metabisulphite (~250% compared to control, ~100%) was followed by a recovery (120 to 170 min in Figure 1 ) to ~150% compared to control (still significantly different from control) nevertheless, demonstrated recovery from the acute response. It is possible that recovery could have been attenuated by the inability of the cilia to clear sodium metabisulphite off the palate. Alternately, SE micrographs showed that, in metabisulphite-treated palates, significant areas of exfoliation were present. It was shown by morphometric analysis that areas of the palate were devoid of ciliated cells, compared to an uninterrupted "carpet" of cilia in control palates. Although mucus continued to move across the palate, the loss of a significant portion of the ciliary layer, replaced by gaps in the ciliated surface, would contribute to a sustained (non-recoverable) increase in MCT. A further finding of intact, ciliated epithelial cells in mucus, recovered from metabisulphite-treated palates, suggested that exfoliation of intact ciliated cells may involve the action of proteases on cell-cell or cell matrix attachments. Gelatinase zymography showed increased activity of MMP-9 in tissue and mucus from metabisulphite-treated palates compared to controls. Conclusion We have shown from the zymographic studies, taken together with the scanning electron microscope studies, that MMP-9 activation was associated with the loss of ciliated cells from the palate. These results suggest the sustained increase in MCT as measured directly on the frog palate may have been due to the action of sodium metabisulphite to activate MMP-9 leading to a loss of ciliated epithelial cells. How this occurs at the cellular level is a question that remains to be answered. Further studies that clarify a site of action of the MMPs and a source of MMPs in this model will be important to determine the mechanism of action of this effect. How MMPs are activated in the tissue is another important question. An understanding of this injury mechanism may lead to ways to intervene in the early stages of airway diseases with symptomatic signs of impaired of mucociliary clearance. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517808.xml |
514497 | Care-seeking patterns for fatal malaria in Tanzania | Background Once malaria occurs, deaths can be prevented by prompt treatment with relatively affordable and efficacious drugs. Yet this goal is elusive in Africa. The paradox of a continuing but easily preventable cause of high mortality raises important questions for policy makers concerning care-seeking and access to health systems. Although patterns of care-seeking during uncomplicated malaria episodes are well known, studies in cases of fatal malaria are rare. Care-seeking behaviours may differ between these groups. Methods This study documents care-seeking events in 320 children less than five years of age with fatal malaria seen between 1999 and 2001 during over 240,000 person-years of follow-up in a stable perennial malaria transmission setting in southern Tanzania. Accounts of care-seeking recorded in verbal autopsy histories were analysed to determine providers attended and the sequence of choices made as the patients' condition deteriorated. Results As first resort to care, 78.7% of malaria-attributable deaths used modern biomedical care in the form of antimalarial pharmaceuticals from shops or government or non-governmental heath facilities, 9.4% used initial traditional care at home or from traditional practitioners and 11.9% sought no care of any kind. There were no differences in patterns of choice by sex of the child, sex of the head of the household, socioeconomic status of the household or presence or absence of convulsions. In malaria deaths of all ages who sought care more than once, modern care was included in the first or second resort to care in 90.0% and 99.4% with and without convulsions respectively. Conclusions In this study of fatal malaria in southern Tanzania, biomedical care is the preferred choice of an overwhelming majority of suspected malaria cases, even those complicated by convulsions. Traditional care is no longer a significant delaying factor. To reduce mortality further will require greater emphasis on recognizing danger signs at home, prompter care-seeking, improved quality of care at health facilities and better adherence to treatment. | Background Malaria continues to be the largest single component of the burden of disease in sub-Saharan Africa, even though simple, effective and affordable treatments exist. Malaria's pervasive morbidity and high mortality persist because of failed transactions between those at risk of malaria transmission and available preventive and curative health systems. The consequence is not just an intolerable burden for individuals, their families and national health systems, but is also a devastating and continuing impediment to socio-economic development on the continent. Unlike HIV and TB, the other major fatal communicable diseases in Africa, malaria deaths can be prevented by prompt treatment with relatively affordable and efficacious drugs. Yet this goal continues to be elusive. The paradox of a continuing, but easily preventable, cause of high mortality raises important questions for policy makers and health systems in Africa. Malaria in Tanzania The United Republic of Tanzania has a population of 34.5 million, all of whom are at risk of malaria. However, endemicity and risk of transmission varies and have recently been mapped by the MARA collaboration [ 1 ](Figure 1 ). This GIS-based analysis reveals that 75% of the population is subject to stable perennial or stable seasonal malaria transmission; 8% to unstable highly seasonal transmission; and 17% to no malaria transmission in the average year, but still at risk of epidemic malaria. Tanzania has the third largest population at risk of stable malaria in Africa after Nigeria and the Democratic Republic of Congo (MARA-Lite Software 3.0.0, ). Malaria is the leading cause of out-patient and in-patient health service attendance for all the ages and the leading cause of death in both children and adults in all regions of Tanzania [ 2 ]. In Tanzania, malaria is believed to be directly or indirectly responsible for about 16 million annual malaria episodes and 100,000 to 125,000 annual deaths (70–80,000 in under-fives) [ 3 ]. Figure 1 Risk of malaria transmission. Length of malaria transmission season in Tanzania based on the MARA climate model. (Source, Ministry of Health TEHIP and MARA-Tanzania). National Responses Increasing global political commitment to malaria control in recent years stimulated by the Roll Back Malaria partnership and the Global Fund to fight AIDS, TB and Malaria, has been reflected in renewed attention to malaria in Tanzanian national level policies, and to a lesser extent, in local government practices. The National Malaria Control Program's strategic plan is built around four pillars: 1) improved malaria case management; 2) national scale use of insecticide treated nets (ITNs); 3) prevention of malaria in pregnancy; and 4) malaria epidemic prevention and control [ 3 ]. Integrated Management of Childhood Illnesses (IMCI), intermittent presumptive treatment in pregnancy (IPT) and Insecticide Treated Nets (ITNS) are all part of Tanzania's national package of essential health interventions. In late 2001 the national antimalarial drug policy ceased chloroquine as the first line drug due to high drug resistance. On average there was 52% total treatment failure in sentinel surveillance of antimalarial drug efficacy [ 4 ]. The new policy includes sulfadoxine-pyrimethamine (SP) as first line, amodiaquine as second line and quinine as third line antimalarials. In 1998 a district-scale, and later in 2000, a national-scale social marketing programme for ITNs was implemented by the Ministry of Health and its NGO and donor partners in order to develop and test processes for increasing affordable supply, demand and coverage for ITNs and to stimulate the commercial market for ITNs. As part of the health sector reforms, a sector-wide approach to financing places per capita resources under the control of local government councils at district level where they can be used to support the provision of the national package of health interventions, including malaria interventions at both public and non-governmental health facilities. Household responses Tanzanians enjoy relatively good geographic access to primary health services, with 90% of the population within one hour of a government health service [ 5 ]. Government health services for children under five years of age and for pregnant women are officially free. However, household health needs and demands are great. Prevalence of overall morbidity is high, with 28.3% of the population reporting illness or injury in the previous four weeks. Utilization of the health system is relatively good and 67.1% of these episodes were reported to attend a health provider (predominantly government). The most commonly reported complaint resulting in a health service consultation is fever or malaria – reported in 69.3% of ill children (less than 15 years of age) and 60% of ill adults (15+ years). Non-governmental health providers are also common and work in partnership with government facilities at rural level. Private-for-profit health providers are relatively new and still largely available only in urban areas and large towns. Over-the-counter drugs are increasingly available in rural settings through private pharmacies, shops and kiosks [ 6 ]. Nevertheless, the most accessible health service for the rural household, both in socio-economic as well as spatial-temporal terms, is traditional medicine and traditional healers. Economic considerations Coincident with and consequent to having one of the highest malaria burdens, Tanzania is also one of the poorest countries in the world with an annual GDP of $213 USD per capita (2000) and 36% of the population below the basic needs poverty line. Malaria is estimated to consume 3.4% of GDP or about $240 million USD dollars annually [ 5 ]. This is stifling for an already fragile economic performance [ 7 ]. Tanzania spends about USD $11.37 per person per year on health [ 8 ]. Of this, $2.14 is spent on malaria services. About 75% of malaria expenditures are borne by the household, with the government contributing 20% and partners 5% [ 9 ]. Of the household malaria expenditure, about one-third is spent on antimalarial drugs and almost half on bed nets, insecticides, coils and other preventive strategies. This burden is greatest on the poorest households and contributes to the continuing cycle of poverty. Care-seeking There have been a number of studies of care-seeking for malaria in Africa reviewed by McCombie in 1996 [ 10 ] and updated in 2002 [ 11 ] with much additional work since then [ 12 - 18 ]. Many of these studies involve qualitative and sometimes quantitative analyses of data from illness narratives for recalling episodes of recent illness. Common themes emerge which can be summarized as follows: almost every study identified local community or folk perceptions, terminology or explanations of illness that overlap with malaria disease in ways that distinguished fever, malaria and convulsions as distinct in aetiology and required treatment. Care-seeking patterns for simple fever or uncomplicated malaria were more likely managed initially at home while cases with convulsions or severe malaria were more likely to seek care from a health care practitioner. Multiple care-seeking events and switching between types of providers were common. Cases with simple fever or uncomplicated malaria were more likely to seek formal, modern biomedical care and antimalarial drugs, while cases with convulsions were more likely to be managed by traditional healers or traditional practices, as well as modern care. The hierarchy of such events is likely to affect timely access to effective care. One feature of much of this prior work is that, because severe and fatal malaria is relatively rare, nearly all studies based on illness recall ask what people would do if they/their child experienced a severe illness such as "degedege" (cerebral malaria with convulsions) rather than what they did do . Rationale Although malaria mortality rates are high, fatal malaria is still relatively infrequent when compared to the number of malaria illness episodes. It is possible that the care-seeking patterns of the majority who are ill, but survive, will potentially mask different patterns of those whose care-seeking choices fail and result in a fatal outcome. To understand how best to reduce malaria mortality through improved access to antimalarials, it will be important to examine the care-seeking of individuals who actually died from what they or the health system considered was malaria. No studies in Africa have specifically focused on short-term recall of care-seeking patterns for fatal malaria to see whether and how the general themes above prevail in this sub-group of greatest interest [ 19 ]. In this paper an analysis is reported of care-seeking events in a large series of malaria deaths recorded in the course of longitudinal demographic surveillance. Methods Study area The general context of malaria and malaria control in Tanzania has been outlined in the background. The specific setting of this study is in the stable perennial malaria transmission belt that runs along the coast of Tanzania and up the Rufiji and Kilombero River basins (Figure 1 ). This transmission risk is typical of that experienced by the majority (75%) of Tanzanians and of sub-Saharan Africa in general. There are two main rainy seasons, October-December and February-May. The specific data for the study comes from a demographic surveillance system (DSS) in the Rufiji District of Coast Region, managed by the Ministry of Health and the Tanzania Essential Health Interventions Project (TEHIP). Details of the study populations, DSS methods, life tables and results are available for the Rufiji DSS [ 20 ]. Household characteristics of the Coast Region are provided in Table 1 . These are shown to be generally representative of rural mainland Tanzania. Table 1 General household-level characteristics of Coast Region in comparison to Tanzania rural mainland Basic Indicators National Mainland Coast Region* Household and Housing Average Household Size 4.9 4.9 Percentage of female-headed households 23 18 Percentage of households with a modern roof 43 24 Percentage of households with modern floor 25 10 Percentage of households with modern walls 25 1 Percentage of households with electricity 12 6 Percentage of households using a toilet 93 98 Mean distance to firewood (km)(rural households only) 3.1 1.7 Mean distance to a shop (km)(rural households only) 1.8 1.0 Mean distance to a bank (km)(rural households only) 37.5 31.3 Education, Health and Water Percentage of adult men without any education 17 24 Percentage of adult women without any education 33 52 Percentage of adults literate 71 58 Primary net enrollment ratio 59 56 Percentage of individuals ill in 4 weeks before survey 28.3 34 Percentage of ill individuals who consulted any health provider 69 83 Percentage of above who consulted a government provider 54 69 Percentage of households within 6 km of primary health facility 75 69 Mean distance to a dispensary / health centre 4.7 3.5 Mean distance to a hospital (km) 25.6 25.9 Percentage of households with a protected water source 57 23 Percentage of households within 1 km of drinking water 55 51 Mean distance to a primary school (km) 1.8 1.7 Mean distance to a secondary school (km) 12.6 13.1 Economic Activities Percentage of adults whose primary activity is agriculture 63 62 Percentage of children age 5–14 years who are working 62 57 Mean area of land owned by rural households (acres) 6 2.9 Consumption and Poverty Consumption expenditure per capita (2000/01 TZS / month) 10,120 9,922 Percentage of consumption expenditure on food 65 71 Percentage of population below the food poverty line 19 27 Percentage of population below the basic needs poverty line 36 46 * Rural result provided where available; ** Exchange rate, January 2001: TZS/USD = 803 Source: Government of Tanzania, National Bureau of Statistics, Tanzania Household Budget Survey 2000/01 Rufiji District is 178 km south of Dar es Salaam on the Indian Ocean coast and has a population of 203,000 in 2002 in an area of 14,500 km 2 . The district is entirely rural with 94 registered villages, no urban areas or towns, and has a large area set aside as a game reserve. The economy is predominantly subsistence farming and fishing. Rufiji district is home to several ethnic groups. The largest is the Ndengereko who, according to oral tradition, are the original inhabitants of the area. Other groups include the Matumbi, Nyagatwa (concentrated in the delta area), Ngindo, Pogoro and Makonde. The majority of the people are Moslems (98%) with a few Christians (1.3%) and followers of traditional religions. In addition to local languages, Kiswahili is widely spoken; English is not commonly used in the area. The population has access to 57 formal health facilities: two hospitals (one government and one NGO), five health centres with in-patient facilities (all government) and 50 outpatient dispensaries (46 government). Over-the-counter drugs are available from many private shops and kiosks in the villages. People also obtain services from traditional healers including traditional birth attendants. Immunization coverage ranges from 85% for BCG (tuberculosis) to 66% for measles in children 12–23 months of age. Acute febrile illness and malaria are the leading causes of attendance at health facilities, and the largest cause of mortality. For malaria, the district provides Integrated Management of Childhood Illness (IMCI), Intermittent Presumptive Treatment of malaria in pregnancy (IPT), and first, second and third line antimalarial services at all formal health services, as well as social marketing of insecticide-treated nets (ITNs). Demographic Surveillance The Rufiji District hosts a sentinel DSS area that covers 1,800 km 2 north of the Rufiji River and west of the Rufiji Delta (7.470 to 8.030 south latitude and 38.620 to 39.170 east longitude). The Rufiji DSS monitors a total population of 85,000 people in 17,000 households in 32 villages. All residents are registered in the system and all births, deaths, in-migrations, out-migrations, pregnancies and other vital events are monitored and registered. Events are recorded in the Demographic Surveillance Area (DSA) by 150 village key informants and verified by DSS staff. Twenty-eight full-time enumerators update the population register every four months by household survey cycles. The field and data system is based on the Household Registration System Software [ 21 ]. The database also includes key household level information on household structure, socio-economics and assets, food-security and environmental features that are updated annually. All households and community structures have been geo-located by global positioning satellite (GPS) systems. The Rufiji DSS is part of the Ministry of Health's National Sentinel System (NSS) for monitoring health and poverty status and serves as a sentinel for rural coastal districts. Annual Burden of Disease profiles are produced by the DSS and used for district planning purposes in the NSS. Verbal Autopsy The Rufiji DSS continuously records vital events within households and among individuals over time in a systematic way. The vital events reporting system consists of key informants who notify the system of any death occurring in the DSS area. This information is passed to a DSS key informant supervisor (or DSS enumerator who informs the key informant supervisor). The key informant supervisor visits the households in which death has been reported within two weeks and contacts the DSS data centre for verification of the registry status. A verbal autopsy (VA) (post mortem interview) is then scheduled and administered to one of the deceased's relatives or the individual who is most well informed of events and details of illness of the deceased. A DSS VA supervisor, who is also a trained clinical officer or health officer, conducts the VA interview. Respondents are not aware of the health care qualifications of VA interviewers. Enumerators also ascertain death events at fixed enumeration rounds three times per year, using specific event forms that are reconciled with the mortality database. There is no population sampling. The entire population of the DSS area is in the DSS and all deaths to DSS residents are subject to VA. Population compliance in both the DSS and VA interviews was very high resulting in high completeness of death registration for registered members. Verbal autopsy was available on 97.7% of deaths, missing only those where the family out-migrated shortly after the death or declined the VA interview. The VA tool used is that of National Sentinel System [ 22 ] based on an evolution of forms developed by the Adult Morbidity and Mortality Project (AMMP) [ 23 ] and very similar to that proposed by INDEPTH . It uses individual specific standard questionnaires for: a) children under 31 days of age; b) children under five years but 31 days and older; and c) population aged five years and older. The questionnaires and responses are in Kiswahili. Information such as household ID number, name, age and sex are re-collected for confirmation. In addition, data is collected by open-ended and closed questions on history of events leading to death, together with previously diagnosed medical conditions as well as signs and symptoms before death. Questions about use of health facilities prior to death, reasons for using or not using a particular health facility and confirmatory evidence of medical care and cause of death (if available) are also asked and recorded in the questionnaires. A typical bereavement interview in the course of a VA takes 45 to 60 minutes. The tentative cause of death is established from the sequence and severity of signs and symptoms, as well as the available confirmatory evidence, by the VA supervisor and recorded on the forms. However, it is physician coding that determines the final cause of death that is subsequently entered in the database. Completed questionnaires are coded independently by two physicians according to a list of causes of death based upon the tenth revision of the International Classification of Diseases. A third physician independently codes the VA in case of discordant results from the first two physicians. Where there are three discordant codes, the cause of death is registered as undetermined (about 6% of cases). A single cause is assigned as the main cause, with contributing causes also indicated. All death coded as the following were included as suspected to be directly or indirectly due to malaria and included in the study: acute febrile illness 1–4 weeks; acute febrile illness < = 7 days; acute febrile illness including malaria; acute febrile illness with convulsions; acute febrile illness with anaemia; cerebral malaria; fever plus malnutrition; malaria; malaria confirmed; and unspecified acute febrile illness. Quantitative methods All data from the DSS and the VA were entered, cleaned and managed using FoxPro (Microsoft Corp). Databases were linked and selected data transferred to Stata 7.0 (Stata Corp) for analysis. The VA database was linked to the household registration database to obtain other indices, such as the socio-economic status. In a separate study, we determined socio-economic indices for individuals in 14,440 rural households in the Rufiji DSS area for the year 2000. The index was based on principal components analysis of the presence or absence of items from a list of 22 specific household assets and nine household characteristics dealing with household ownership, construction features, water supply, sanitation and type of fuel. Further details on the socio-economic index are provided elsewhere [ 24 ]. The household index was applied to each individual in the respective household and all deaths due to malaria were partitioned into socio-economic quintiles by this index. Univariate analyses were used to assess the affect of age, sex, socio-economic status, household headship and severity of malaria on initial choices from 13 potential categories of health care providers. Chi-square was used to identify significant factors associated with choice of care sought during the final illness. Qualitative methods The health behaviour research component of the Tanzania Essential Health Interventions Project (TEHIP) investigated the care-seeking and compliance patterns for malaria in a separate study in the Rufiji District from 1998–2001. Eight villages were purposely selected to include four villages with a local health facility and four villages far from a health facility. From these villages 80 households with children under-five years of age were selected by simple random sampling. Ethnographic approaches (semi-structured interviews, case histories and focus group discussions) were used to explore and describe households' responses to childhood illnesses including malaria. A two-step coding strategy was used. In the field, a research assistant, using a provided guide, performed initial thematic coding of the data. Field coding was supervised and consistency checked by a senior social scientist. At computer data entry level, field codes were replaced by corresponding thematic codes written in text macros by experienced data clerks. A data manager supervised the data entry and was responsible for quality and further consistency checks. All qualitative data was processed in a text editor and analysed using text analysis software, Text-Base Beta (Centre for Qualitative Research, University of Aarhus, Denmark). The codes allowed retrieval and compilation of text segments of interest for thematic analysis. Terminology The ethnographic literature on treatment seeking in Africa uses a variety of terms, none of which are wholly satisfactory in capturing the nature and complexity of available health systems. In this paper the term "modern care" is used to describe what conventionally includes biomedical, western, pharmaceutical, professional, official or formal health care and the term "traditional care" is used to describe what conventionally includes traditional medicine, traditional healers, traditional providers, lay providers, traditional practices or folk care. Ethical Considerations All household visits, surveys and questionnaires in the DSS and TEHIP surveys were administered with individual informed consent. All individual and household data are confidential. All reports are based on summary data that cannot be linked to individuals or individual households. The Ministry of Health, National Institute for Medical Research's Tanzania Medical Research Coordinating Committee has approved the research protocols of TEHIP and its Rufiji DSS. Information is fed back to the communities concerned on a semi-annual basis and provided to the local council authorities and the Ministry of Health for planning purposes on an annual basis. Results Qualitative themes: illness terminology Qualitative studies confirmed that the population refers to the signs and symptoms associated with the biomedical condition of malaria as three distinct conditions, each with its own aetiology, treatment-seeking patterns and prognosis. The three conditions are: "homa" (fever, vomiting, feeling cold, loss of appetite, limp body, red eyes, not considered life threatening); "malaria" (high fever, vomiting, loss of appetite, feeling cold, some caretakers considered life threatening) and "degedege" (high fever, loss of appetite, stiffness of body, rolling of eyes, lips twisted sideways, twitching, considered life threatening). These are well recognized by most households in the study population. " [...] you are able to recognize an episode of degedege in one day. It begins with mild fever and the next day the fever becomes more severe and results in symptoms of epilepsy. The child opens the eyes wide and the black spot cannot be seen, he begins to twist the arm and leg, and then, even if you pour cold water over the child, does not react..." (Female respondent aged 37 from Bungu – Rufiji). Although the local population distinguish between the illness "homa" and malaria the distinction is not always very clear to them. Analysis of case studies revealed that the illness term "malaria" has been obtained from modern health care. When mothers take their children to these health services with what had been diagnosed at home as "homa", they are told it is malaria. The following is illustrative of experiences reported: "I first thought it was normal homa (fever) and I could see the child had homa. Now, when I took the child to the hospital, they checked the child's blood and informed me the child had malaria. [...] the child was not playing. I touched the child and the body was like fire (mwili wake ulikuwa wa moto), the body was very hot". (Female respondent aged 29, Bungu, Rufiji). Anaemia is not often recognized, and where recognized, is not associated with malaria. Qualitative themes: aetiology Although "homa" and especially "malaria" were seen as associated with malaria and mosquitoes, in most cases the signs and symptoms of "degedege" are not attributed to malaria. Life threatening malaria with convulsions is not only perceived as a different illness from malaria through local symptom definition but is attributed to different causes than malaria. Few households mention the mosquito as a cause of the illness described as "degedege". Popular beliefs as to the cause of "degedege" were found to include: fever, evil spirits and a change in weather/wind. The following translation is typical of quotes obtained from respondents on perceived causes of "degedege": "....Evil spirits or demons cause degedege. If it happens that evil spirits or demons pass in front of the child, then the child is likely to get degedege. This may result in paralysis of the body or leg or arm or any part of the body..." (Male respondent aged 46, from Kilimani, Rufiji). Qualitative themes: Care-seeking pattern "Homa" and "malaria" are seen as conditions that can be managed at least initially at home with modern medicine available from shops and from health facilities. But "degedege" is perceived as a serious life-threatening condition for which prompt treatment-seeking is required. People reported different sources of care they used for the treatment of "degedege". These sources encompass more than the biomedical health system and fall into three broad categories: home treatment, traditional healers and biomedical. Home treatment was reported to include the use of modern medicines, such as aspirin from local shops, in the early stages the illness. If the illness reaches a severe stage (convulsions) people claim to use traditional healers in the home or outside the home. Biomedical care ranging across government hospitals, health centres, dispensaries and equivalent private facilities was used in the later stage, when convulsions had subsided. However, some respondents perceive traditional healers as not competent to deal with such illness and claim to seek care from biomedical providers at the beginning of the illness. "We use traditional remedies only to treat degedege. They ( remedies ) must have a very bad smell for this will chase away the evil spirit. It is just like telling you to stay in the latrine; surely you will have to find another place because of the bad smell. This is just the same case for the evil spirit attacking the child because of the bad smell it will have to find another place to stay...." (Male respondent aged 57 years, Kiomboni, Rufiji). "I had gone to the dispensary for treatment; my child was suffering from homa. The first day he was given panadol tablets and chloroquine injection and was asked to return the next day for chloroquine injection. The next day while I was there at the dispensary waiting for treatment my child started convulsing. This I believed to be a sign of degedege. Immediately I left the dispensary in search of a traditional healer. Degedege is never treated in the dispensary. Child may die after being injected." (Female respondent, 39 years old, Kiomboni, Rufiji). "When my child developed degedege I was at Kibiti. I had to look for transport to take the child to Songa Hospital (Mchukwi Missionary hospital). There you have reliable service because you find almost all kinds of investigations. I don't like going to traditional healers because they are not reliable and do not have equipment to investigate well your child. They end up telling you things related to superstition." (Female respondent aged 42, Bungu, Rufiji). Quantitative results: care-seeking pattern In the period January 1999 to December 2001 inclusive, the Rufiji DSS conducted 243,042 person years of follow-up. In this series, 3,023 deaths occurred to resident members and 2,953 (97.7%) verbal autopsies were conducted. Of these, 24.4% (722) had a cause of death suggestive of malaria as the direct or underlying cause, of which 44.3% (320) were in children less than five years of age. Among these child deaths, there was no difference in frequency between sexes, with 51.3% being male and 48.7% female. Of the child malaria-attributed deaths, 282 (88.1%) sought care at least once before death, while 38 (11.9%) did not, or could not, seek care. Convulsions (possible cerebral malaria) were recorded in 30 (9.4%) of these fatal cases. The verbal autopsies contained both an open-ended narrative account of the final illness and a specific chronological account of where and in what sequence care was sought. There were 13 possible sources of treatment that were collapsed for purposes of certain analyses into three sub-categories of care types (Modern Care; Traditional Care; and No Care) and into six sub-categories of provider types (Government; Home/Shops; Non-Government; Traditional Medicine at Home; Traditional Medicine at Practitioner; and No Care). Table 2 compares the level and detailed source of initial care in acute febrile illness (malaria) for children less than five years of age compared with older cases. The initial treatment-seeking choice for children less than five years of age was modern care (78.7%), whereas only 9.4% used traditional care initially. The remainder (11.9%) sought no care (Figure 2 ). Table 2 Level and source of initial care in fatal acute febrile illness / malaria by age group in the Rufiji DSS sentinel area, 1999–2001 Level of Care Provider Age <5 5+ Government VHW 0.0% 0.7% Dispensary 19.4% 11.2%** Health Centre 20.0% 14.4%* Hospital 5.3% 5.0% Home Mothers 2.5% 2.2% Family 9.4% 13.2% Drug Shops 8.1% 20.6%** Non-Government Dispensary 10.3% 5.5%* Health Centre 1.6% 2.0% Hospital 2.2% 2.5% TM at Practitioner 6.6% 6.5% TM at Home 2.8% 1.7% None None 11.9% 14.3% 100% 48% Number 320 402 Total 722 * Significant at 5% level; ** Significant at 1% level TM Traditional Medicine or Practice Figure 2 Initial care-seeking patterns. Care of first resort sought during the final illness by 320 fatal "malaria" cases in children less than five years of age in the Rufiji DSS sentinel area, 1999–2001. Within modern care, government providers were most prominent (44.7%) followed by home care with antimalarials from private shops (20%) (Table 3 ). Children were statistically more likely to be taken to government health centres and government and non-government dispensaries and less likely to be served by drug shops as the initial resort to care (p < 0.05). There were no significant differences between treatment-seeking patterns for male and female patients regarding the broad choices of modern, traditional or no care. Even though there was no difference in the proportion of males and females receiving traditional care, within the traditional care group, females were statistically more likely to be kept home to receive traditional medicine, and males were more likely to be taken out of the home to see a traditional healer (p < 0.05). There were no significant differences in specific or general care-seeking patterns by sex of the household head. There was no difference in treatment-seeking patterns when comparing choices made by households in the poorest quintile and households in the least poor quintile. Table 3 Type and provider of initial care in fatal acute febrile Illness / malaria by age group, sex, socio-economic status, and type of illness in the Rufiji DSS sentinel area, 1999–2001 Type of Care Provider Age Sex of Child Sex of HH Head Poverty Quintiles Convulsions <5 5+ Male Female Male Female Poorest Least Poor With Without Modern Care Government 44.7% 31.1%** 46.4% 42.9% 38.6% 33.3% 42.6% 51.0% 55.6% 43.3% Home / Shops 20.0% 36.1%** 21.3% 18.6% 28.2% 31.9% 22.2% 15.7% 19.4% 20.1% Non-Government 14.1% 10.0% 12.2% 16.0% 12.6% 9.8% 9.3% 9.8% 2.8% 15.5% Traditional Care TM at Practitioner 6.6% 6.5% 7.3% 5.8% 2.1% 2.9% 7.4% 5.9% 16.7% 5.3%* TM at Home 2.8% 1.7% 1.2% 4.5% 6.0% 7.8% 1.9% 3.9% 0.0% 3.2% No Care None 11.9% 14.4% 11.6% 12.2% 12.0% 14.3% 16.7% 13.7% 5.6% 12.7% 100% 33% 100.0% 100.0% 100% 100.0% 100% 100% 100% 95% Number 320 402 164 156 485 204 54 51 36 284 Total 722 320 689 105 320 * Significant at 5% level; ** Significant at 1% level TM Traditional Medicine or Practice HH Household. Note, 33 households had a change in headship during the study period and were excluded from the analysis in the sex of HH Head column. Cases with convulsions were as likely to receive initial modern care as cases without convulsions (77.8% and 78.9% respectively) (Table 3 ). However, cases with convulsions were less likely to receive no care. Therefore, although the predominant choice of care was modern, inclusion of care from traditional healers was significantly more frequent in those with convulsions than in those without convulsions (p < 0.05). All traditional care was provided by traditional healers and no case claimed to give traditional medicine at home, which is contrary to what is often described in non-fatal treatment seeking. Among children for whom care was actively sought, 82.4% of those with convulsions and 90.3% without convulsions sought modern care as the initial care (Table 3 ). Multiple episodes of care-seeking were common. More than half of cases had two or more treatment-seeking events for the same illness involving a different type of provider (Figure 3 ). There is also a difference in pattern when initial care choices and cumulative care choices are compared (Table 4 ). The latter indicates important switching between providers over time and this phenomenon is most apparent when comparing malaria without convulsions to malaria with convulsions. Multiple provider care-seeking was more common if convulsions were present. These synchronic choices (frequency of use of a particular resort to care) are shown in Figures 3 and 4 . In the multiple-care-seeking group, switching between modern care and traditional care can be a factor in the delay of effective care. Of the multiple-care-seeking group that did not have convulsions, 88.4% and 99.4% had used modern care at least once by their first or second choice respectively. In this group, of those who started with modern care, only 0.9% switched to traditional care as the second choice. Of the few who started with traditional care as their first choice, most (94%) switched to modern care for their second choice. For the group that had convulsions, 90% chose modern care as their first choice, but by the second choice, 29.6% switched to traditional as the second provider. Switching did not seem to be based on differences in likelihood of receiving treatment. All provider categories were generally able to supply the expected treatment, the poorest being government providers who were able to give treatment for 94% of cases and the best being traditional healers at 96.8% of cases. Figure 3 Frequency of care-seeking events. Distribution of frequency of care-seeking events at differing categories of provider among those who sought care during the final illness in fatal episodes of malaria in 320 children under five years of age with (dark shading) and without convulsions (light shading). Table 4 Level and source of accumulative care in fatal acute febrile illness / malaria, all ages, in the Rufiji DSS sentinel area, 1999–2001 Level of Care Provider Cumulative Events No. % Government VHW* 5 0.8% Dispensary 92 14.5% Health Centre 104 16.4% Hospital 67 10.6% Home Mothers 19 3.0% Family 64 10.1% Drug Shops 36 5.7% Non-Government Dispensary 77 12.2% Health Centre 39 6.2% Hospital 30 4.7% TM** at Practitioner 73 11.5% TM** at Home 27 4.3% Total care seeking 633 100.0% VHW* Village Health Worker; TM** Traditional Medicine or Practice Figure 4 Loyalty to first provider. Comparison of loyalty to first provider of modern or traditional care during the final illness in fatal cases (all ages) that saw two or more providers. Discussion Limitations of verbal autopsy methods for malaria deaths have long been recognized, especially with regards to specificity and sensitivity [ 25 , 26 ]. This has provoked efforts to improve and validate verbal autopsy procedures in the settings in which they are used [ 23 , 27 - 34 ]. The general consensus is that, although imperfect, verbal autopsies are reasonably reliable in determining major causes of death at population level, but may not be suitable for detecting specific impacts of interventions. However, recent work applying adjustments for sensitivity and specificity at differing prevalence levels based on validation studies shows how VA data could be used to monitor progress towards malaria-specific mortality reduction [ 35 ]. It must be emphasized that not all of the cases identified as "malaria" in this series are malaria, especially those with unspecified acute febrile illness at older ages. Undoubtedly, some malaria deaths were coded as a cause other than malaria-related. For example, severe and life threatening anaemia, likely to be due to malaria, is prevalent in young children over six months of age in the study area [ 36 ] yet VA coded deaths due to anaemia with malaria are infrequent. Despite improvements in verbal autopsy methods in recent years, any study based on verbal autopsy is subject to bias. The recall abilities of respondents can be faulty, although for major events such as a death in the family, it tends to be better than recall of less significant events [ 33 ]. In the current study of care-seeking as reported in verbal autopsy, respondents might inflate the number of care-seeking events or exaggerate the choice of modern care if they perceive the DSS to be an instrument of the modern health system or if they feel guilt regarding the care-seeking decisions they took. This would tend to bias responses in favour of more modern care. Much has been learnt in recent years concerning treatment seeking for malaria in Africa, largely through ethnographic research on illness recall narratives [ 10 - 12 , 14 , 37 - 42 ]. This literature confirms that, for the majority of cases deemed as uncomplicated malarial fevers, modern care based on antimalarial drugs is favoured over traditional medicine. Usually treatment starts at home using anti-pyretics and antimalarials obtained over-the-counter from local shops or left over from previous episodes. Knowledge of appropriate treatment regimens is lacking on the part of the public as well as on the part of private providers [ 43 , 44 ]. Under-dosing in home-based care is common. Malaria is perceived by adult care givers as a mild disease, and if it becomes serious or life threatening, then, it is generally believed that the perceived diagnosis changes from malaria to something that is more likely to be treated with traditional medicine or practices. These beliefs are not rigid. Every case is subject to a process of continuing debate and re-evaluation such that modern pharmaceuticals are also sought, albeit with delay, when convulsions fail to resolve or reoccur after traditional medicine [ 16 , 45 ]. If this is the case in studies of illness recalls, where most patients recover, the question remains whether this general and widespread pattern of treatment seeking holds in those cases where effective treatment seeking clearly failed and the patient died. Since most cases of malaria death in Africa occur at home rather than in health facilities, facility-based data and studies cannot answer this question. The increasing use of demographic surveillance field sites to monitor health at population level in Africa [ 46 ] presents an opportunity to examine large series of verbal autopsy findings. Modern verbal autopsy goes beyond cause of death data to collect additional contextual data on, for example, care-seeking events prior to death. This study confirms that the general patterns seen in illness recalls for uncomplicated malaria in Africa also apply to what people actually do in episodes of fatal malaria in a holoendemic area of Tanzania. Modern care is the first choice for children in over 78% of all child malaria deaths. Government health facilities and shopkeepers were the main source of modern antimalarial drugs. Traditional care may have caused delay in modern care in only 9.4% of fatal cases. 11.9% had no care of any kind. This general pattern held over broad age, sex and socio-economic status groups. Among children with and without the complication of convulsions for whom care was actively sought, 82.4% and 90.3% respectively sought modern care as the initial care (Table 3 ). In the case of convulsions, although the majority of initial care-seeking was modern, the use of traditional healers increased while the no-care group decreased accordingly. Among those of all ages who sought care two or more times in the course of fatal malaria, modern care was included in the first two choices in 99.4% of cases excluding convulsions and in 90% of cases with convulsions. Clearly, the perceived severity and danger signs posed by convulsions provoke polyvalent treatment seeking. Nevertheless, modern care is now more popular than previous reports and qualitative studies suggest. One other study of care-seeking patterns in a large series of verbal autopsy reports from the mid 1980's from Tanzania analysed a similar number of all-cause child deaths from Bagamoyo District, a nearby district in the Coast Region [ 47 ]. In that study, malaria deaths were not analysed separately, but government providers were the choice in only 45% of deaths. At that time government providers were often without an adequate drug supply and a preference for traditional healers was cited by 41% of mothers as the reason for not using government providers. At the time of the present study in Rufiji, all government providers had adequate drug supplies under the health reforms and offered the integrated management of childhood illness (IMCI) strategy. This could be a factor in the current popularity of government providers. A relatively small proportion (21.3%) of malaria-attributable child deaths failed to seek modern care (9.4%) or any care (11.9%). This is considerably better than was seen in the mid-eighties, when 55% of children who died had not utilized any modern care [ 47 ]. It is also better than seen for deaths in general in the same area during the same period, when 20% of all-cause deaths had no prior care-seeking events [ 48 ]. Part of these non-care groups would include those who had sudden death following apparently mild illness, including severe anaemia. This study shows that most patients now include modern care early in their treatment seeking patterns for eventually severe and fatal malaria, including malaria with convulsions. So why is malaria still the largest single component in the burden of mortality? With belief systems for malaria treatment seeking now firmly on the side of modern care, there is obviously something still failing in 1) the transaction to obtain this care; 2) the quality of the care and referral once it is reached; and/or 3) patient adherence to treatment once it is obtained. This would suggest that policies, efforts and implementation research aimed at improving early recognition of symptoms and danger signs at home, prompt treatment or treatment seeking, the quality and efficacy of the antimalarial available and compliance with the full course of treatment, are now, more than ever, highly justified. When appropriate care-seeking is as high as it is in Tanzania, continuing malaria deaths should be considered as sentinel events deserving of close scrutiny and audit to identify the best remedial strategies for the health system. There are promising developments. IMCI has recently been introduced in the study area. It places heavy emphasis on training care-givers on early recognition of danger signs and the need for prompt treatment and on improving quality of assessment and care at primary health facilities [ 36 , 49 ]. Replacement of chloroquine with directly observed treatment with sulfadoxine-pyrimethamine (SP) and its simpler single dosing schedule should result in less under-dosing while the introduction of pre-packaged doses has also been shown to be effective in improving provider and client adherence [ 50 , 51 ]. This study was conducted over the last three years of a policy period that used chloroquine as the first line antimalarial. It will be repeated for a similar time frame over the initial three year period of a new policy that uses SP to see if the care-seeking and care-getting patterns change. A qualitative analysis is also planned for the narrative portion of the verbal autopsy questionnaire to look at categories and sub categories of health care related themes in VA reports. This would focus on reasons for delay in seeking modern care (e.g. tried to treat at home without antimalarials, transport, beliefs, poor recognition of severity, lack of confidence in modern care, no power to decide, insufficient finances); delay in receiving modern care (e.g. outside of working hours, weekends, long queues, satisfaction); ineffective modern care (poor communication, no referral, drugs not available, abusive health worker). Conclusions This preliminary study examined what families of children who died from malaria in a holoendemic setting in Africa actually did in terms of treatment-seeking choices and sequence. It confirms that modern medicine in the form of antimalarial pharmaceuticals from shops or government or non-governmental heath facilities is now the preferred choice in an overwhelming majority of cases (78.7% and 97% as their first or second choice respectively). Traditional medicine could only be implicated in a possible delay of modern care in 9.4% of cases. 11.9% sought no care of any kind. There were no differences in these broad patterns of choice by sex of the child, sex of head of household, socioeconomic status of the household or presence or absence of convulsions. Contrary to what is concluded from much of the historical and qualitative work on this subject, modern care is now the care of first choice, even for those who seek care for children with malaria with convulsions (82.4%), although traditional medicine also played an important role in later choices. But despite high rates of modern care-seeking for all forms of malaria, and despite relatively high attendance and utilization of modern care as seen in Tanzania, malaria mortality remains high. This must, therefore, be due to excessive delay in seeking modern care, and/or poor quality of modern care (providers and/or drugs) once sought, and/or poor patient adherence to treatment regimens once obtained. Certain policy and practice implications arise: 1) public messages need to focus aggressively on improving early recognition of malaria and severe malaria at home and improving promptness of treatment seeking (within 24 hours of onset of malaria symptoms or immediately in the case of severe malaria); 2) quality of modern care providers and modern care must be improved in all sectors, private, NGO and Government; and 3) patient adherence with modern care at home must be simplified and reinforced. List of abbreviations DSA Demographic Surveillance Area DSS Demographic Surveillance System GIS Geographic Information System HBS Household Budget Survey HH Household IMCI Integrated Management of Childhood Illness ITN Insecticide-treated netting MARA Mapping Malaria Risk in Africa Collaboration SP Sulfadoxine-pyrimethamine TM Traditional Medicine VA Verbal Autopsy Authors' contributions DD conceived the study, participated in the design, coordination and quantitative analysis and co-wrote the article. CM conceived, conducted and analysed the qualitative studies and co-wrote the article. HM led the analysis of quantitative data. EM managed the surveillance system and participated in design and coordination. AM managed and cleaned the quantitative data. YM managed the field work. CM, HK and GR participated in the coordination and management of the study. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514497.xml |
509411 | Hormones Act in Concert to Direct Plant Growth | null | Anyone who thinks plants are passive inhabitants of their environment has never seen time-lapse footage of a seedling bursting from its protective shell or a climbing vine coiling around a tree. Such films dramatize a fundamental fact of plant life: survival depends on responding to environmental cues. Shoots grow toward light and against gravity. Stems and roots curl around obstacles that block their paths. In plants, environmental cues trigger hormonal changes that in turn regulate cells' shapes and proliferation. In this way, subtle changes in the environment affect plant growth. Auxin, the first known plant hormone, spurs growth and shapes growth patterns in nearly every plant tissue throughout a plant's lifecycle. Brassinosteroids—a class of hormones chemically similar to animal steroids like testosterone—are linked to many of the same processes as auxin. Emerging leaf tips (yellow arrow) and hypocotyl (orange arrows) of an Arabidopsis mutant Early physiological and molecular experiments gave conflicting evidence about whether auxin and brassinosteroids had similar effects. For many years, biologists believed that these hormones acted through independent signal transduction pathways—chains of molecules that relay stimuli and elicit cellular responses, such as gene expression. But in the last few years, microarray studies, which can measure the transcription of thousands of genes simultaneously, showed that auxin and brassinosteroids do regulate expression of several genes in common. In this issue of PLoS Biology , Jennifer Nemhauser et al. assay the entire genome of Arabidopsis thaliana , a favorite for plant genetics studies, for effects of auxin and brassinosteroids. The group's microarray analyses show that these hormones affect transcription of about 80 genes in common—including many known players in the hormones' signal transduction pathways. To see how this regulation could occur, the research team looked at the genes turned on by both hormones to find common promoter sequences—regions of the genome that do not code for protein but instead help regulate gene transcription. They used a new computational approach to tease out promoter regions that auxin and brassinosteroid pathways both act upon, showing how these hormones have overlapping effects on gene transcription. The group also compared the effects of auxin and brassinosteroids on seedlings' stem growth and gene expression in a variety of mutant Arabidopsis lines. They showed that auxin and brassinosteroids greatly enhance each other's effects on stem growth, demonstrating that the interaction of these hormones is important for normal plant development. Mutants with a disabled auxin pathway don't respond normally to brassinosteroids, and vice versa. Also, mutants with abnormally high levels of auxin have a reduced number of genes that respond to brassinosteroids. Thus, these hormones act through overlapping, interdependent pathways—but they don't regulate each other directly. Instead, the researchers suggest, the pathways likely converge on the promoters of a few key genes. It's still an open question why plants use these hormones with such redundant effects. Nemhauser speculates that—as is known to be the case in animals—by having dual, interdependent pathways, plants can finely tune how these ubiquitous hormones act in different cells and tissues to shape patterns of growth. By showing clearly that auxin and brassinosteroids act together and how they affect many of the same genes, Nemhauser and colleagues have set the stage for more detailed studies of how these hormones act in specific parts of plants to shape growth. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC509411.xml |
534790 | Factors involved in nurses' responses to burnout: a grounded theory study | Background Intense and long-standing problems in burn centers in Tehran have led nurses to burnout. This phenomenon has provoked serious responses and has put the nurses, patients and the organization under pressure. The challenge for managers and nurse executives is to understand the factors which would reduce or increase the nurses' responses to burnout and develop delivery systems that promote positive adaptation and facilitate quality care. This study, as a part of more extensive research, aims to explore and describe the nurses' perceptions of the factors affecting their responses to burnout. Methods Grounded theory was used as the method. Thirty- eight participants were recruited. Data were generated by unstructured interviews and 21 sessions of participant observations. Constant comparison was used for data analysis. Results Nurses' and patients' personal characteristics and social support influenced nurses' responses to burnout. Personal characteristics of the nurses and patients, especially when interacting, had a more powerful effect. They altered emotional, attitudinal, behavioral and organizational responses to burnout and determined the kind of caring behavior. Social support had a palliative effect and altered emotional responses and some aspects of attitudinal responses. Conclusions The powerful effect of positive personal characteristics and its sensitivity to long standing and intense organizational pressures suggests approaches to executing stress reduction programs and refreshing the nurses' morale by giving more importance to ethical aspects of caring. Moreover, regarding palliative effect of social support and its importance for the nurses' wellbeing, nurse executives are responsible for promoting a work environment that supports nurses and motivates them. | Background Working in a burn unit has been described as a stressful occupation [ 1 ]. Every nurse who cares for a burn victim knows that stress is a part of working in this field. Some authors have emphasized that these nurses experience dealing with self-inflicted burns, uncooperative patients, inter-staff conflicts and dying patients on a daily basis [ 2 ]. Unresolved job stress may results in emotional withdrawal and burnout [ 1 ]. Professional burnout has been defined as a syndrome manifested by emotional exhaustion, depersonalization, and reduced personal accomplishment [ 3 ]. Nurses who have worked in burn centers of Tehran have experienced burnout in comparison to nurses working in other areas. The main researcher's previous study of burnout and coping in burn centers of Tehran indicated that the majority of nurses had been experienced high levels of burnout [unpublished thesis]. The consequences of professional burnout for nurses are serious. It results in emotional withdrawal or indifference; reduces the limits of nurses' activity and their contact with patients [ 4 ]. Burnout results in a poor quality and quantity of nursing care and has negative effects on the most areas of personal, interpersonal and organizational performance [ 5 ]. While no health-care professional is immune to these pressures, there is evidence that suggests that areas of nursing particularly those areas we think of as critical care environments such as burn units, are often the most vulnerable to stress, and in need of much support [ 6 , 7 ]. Nurses in burn centers of Tehran are also vulnerable to burnout because these centers have many problems. The managers of the burn centers have not the authority for recruiting new nurses. Moreover, self-management of burn centers in Tehran, poverty of most of the burn victims and lack of supportive organizations, resulted in financial problems in burn centers. These in turn have resulted in intense staff shortages, a heavy workload, and low pay. These factors, in addition to inherent characteristics of burn centers have put nurses under a huge pressure and many times they have indicated that they haven't any motivation to work and they wish to leave burn centers as soon as possible. Lewis et al. had the same idea and concluded that the scope and intensity of problems nurses encounter in burn units indicate that they need psychiatric consultation [ 2 ]. However, regarding emotional, attitudinal, psychosomatic, behavioral and organizational responses of these nurses to burnout, it is vital to identify the factors that involve in their perception of burnout. Some authors also referred to these factors in burn centers [ 1 ] and other units or populations [ 8 , 9 ]. Nurses in burn centers of Tehran also pointed implicitly or explicitly to some factors that have played a role in their perceived stress and altered their responses to burnout. The challenge for managers and nurse executives of burn centers is to understand the intervening factors and their impacts on these burn nurses' responses to burnout. As a result they can develop and promote delivery systems that support positive adaptation to stressors in burn centers, retain nurses and facilitate quality nursing care. Methods In order to understand nurses' perceptions of factors modifying their responses to job stress and burnout, qualitative research adapted from the grounded theory method was chosen [ 10 ]. Grounded theory The value of using a qualitative research method such as grounded theory is embedded in the subjective and often emotional nature of care, stress and coping. As a descriptive study, the qualitative paradigm, with its emphasis on understanding factors modifying nurses' responses to job stress and burnout from the view point of practicing nurses themselves seemed logical. Grounded theory is a theory that is derived from data, systematically gathered and analyzed through the research process [ 10 ] The aim of grounded theory is to generate rather than verify theory [ 11 ]. The researcher's purpose in using grounded theory is to explain a phenomenon from within the social situation itself and to identify the inherent processes operating therein [ 12 ]. In effect, grounded theory is guided by simultaneous analysis. Both analysis and data collection inform each other. The analysis process is systematic and ends when new data no longer generate new insights. This has been also described as ' category saturation' [ 13 , 14 ]. Pilot study Five clinical nursing instructors participated in the pilot study. They were faculties of School of Nursing in Iran University of Medical Sciences (IUMS) and had been supervising nursing students in burn centers of Tehran for many years. Their age ranged from 40–48 years and had been working in burn centers for 7–14.5 years. The aim of pilot study was using the experiences of nursing instructors in the original study and reducing the informant and researcher bias in the interviews and participant observations [ 15 ]. The results of the study indicated that the staff of burn units felt drained, they haven't had any motivation or desire to care and they had been working purely for their pay. The findings were strongly indicative of the symptoms of burnout. It revealed that their behavior was representative of an indication of their professional dilemma. The pilot study also indicated that social support, patients' cooperation/motivation and the nurses' unique characteristics had been modified to alter the nurses' responses to burnout. Analysis of data from the original study was conducted keeping these findings in mind. Conduct of study The research proposal was approved by the ethics committee of IUMS. Then permission was granted from the managers of two burn centers and their nursing administrators. Further permission and written consent was obtained from all who participated in this study. Data collection and sample Following ethical approval, data was collected through tape- recorded, unstructured interviews. Initially data was collected in one center and analyzed. Then data gathering was initiated in the second center. There were 19 informants from the first center and 14 from the second center that participated (except 5 participants in pilot study). From this sample, 25 were nurses in different levels and positions and 8 were other members of burn team. The nurses' sample included 8 staff nurses, 8 licensed practical nurses, 2 nurses' aids, 3 head nurses, 2 supervisors, and 2 nursing administrators. Since nursing staff pointed to some issues concerning the burn team, the researcher interviewed one physician, one social worker, 2 physiotherapists, and 4 patients in the process of theoretical sampling. Criterion for recruiting nursing participants was at least one year of experience in the burn center. Patients were selected according to their desire as well as their physical and psychological stability. Selecting patients occurred by consultation with head nurses. Participants were 11 males and 22 females. The nursing staff participated were 19 females and 6 males. Twelve of the nursing staff had been working 2–3 shifts in burn centers or other hospitals as well as working in other jobs due to financial needs. Other demographic information of the nursing staff is displayed in Table 1 . Table 1 Demographic information of nursing staff. Demographic item Mean Range SD Age of participants 39 23–52 ± 9.5 Years of experience 17.5 1–29.5 ± 10.5 Length of time at study hospital 12.85 1–29.5 ± 10.9 Samples were recruited from all units of both centers. The first purposeful sample included 6 of nursing staff. Theoretical sampling was used after emerging the tentative theory. The basis for theoretical sampling was the questions which emerged during data analysis. At this stage the researcher interviewed nursing administrators and other members of the burn team. Theoretical sampling helped in verifying nursing staff's responses and credibility of categories and resulted in more conceptual density. No new data were emerged in the last two interviews; therefore data gathering by interviews were terminated. Interview process Interviews were conducted in a private place with mutual agreement of the interviewer and interviewees. All interviews were completed by the main researcher. The duration of interviews ranged between 30–165 minutes. All the interviews were tape recorded except one. Some notes were taken during dialogues. Unstructured interviews were conducted using a topic guide which has been drawn up by the researchers initial review of the literature related to the concepts of the subject of the study. This topic guide included the structure, process and outcome of care [ 16 ]. The following grand tour question guided the study:" please tell me about the nursing care in your unit". Subsequent questions were based on the participants' responses and demands of the emerging theory. Interviews were terminated when data redundancy occurred. Participant observation In each center after the termination of interviews, participant observations were performed in all wards at morning, evening and night. 14 sessions of observation in the first center and 7 in the second center occurred. For this purpose the researcher informed the nursing administrators of her program. By selecting all the wards in all shifts there was no need for theoretical sampling in this stage (place and time) [ 10 ], but theoretical sampling of different situations were made in each ward or dressing room based on the questions which emerged during the interviews and observations. Descriptive, focused and selective observations were occurred in a non- linear fashion. Theoretical sampling occurred during focused and selective observations. Some of the questions which guided the theoretical sampling were," is there any difference between nursing cares received by different patients?", "is nursing care different in a large or small ward?" Prolonged engagement of the researcher in the field reduced the obtrusiveness. The level of participation varied from complete observation to participation in some activities. Some informal interview also occurred during the observations. Immediately after each session of observation field notes were completed systematically. Analysis of the field notes helped in determining contextual conditions and explaining variations in the nurses' responses in each context. This led to proposition of several hypotheses. Data analysis Data collection, analysis and interpretation occurred simultaneously, in keeping with grounded theory methodology [ 10 ]. After each interview the transcript was manually transcribed by the main researcher onto a personal computer, providing an opportunity for identifying themes as the tape was transcribed(for the purpose of this paper, quotes from the participants were translated verbatim). Following transcription, a print- out was obtained and the tape replayed making notes onto the transcripts. Notes included comments about tone of voice, recurrent themes and the researcher's own initial thought and feelings about the nature and significance of the data. Field notes of each session of observations were also typed in double space and were analyzed. The transcripts were re- read and codes assigned to recurrent themes. This is known as "open coding", whereby the data are examined word by word and line by line [ 10 ], and codes were freely generated, often reflecting the words of the respondents themselves. For example the code "head nurse support" was given to the response:" relationship is heartfelt, perhaps I do many extra things because she is positive, she is supportive, she gives me motivation". The codes similar in meaning grouped in the same categories. Analytical tools include asking questions and making comparisons helped in finding the properties of each concept [ 10 ]. In axial coding, categories were related to their subcategories; coding was occurred around the axis of a category, linking categories at the level of properties and dimensions [ 10 ]. In this stage the structures of care were related to the processes. For example it indicated that which group of factors has contributed to the nurses' distancing from patients. The process of integrating and refining the theory occurred in selective coding [ 10 ]. In this stage the core category" emergence of negative trends: nurses' responses to burnout" was identified. Selective sampling of literature related to job stress and burnout was very helpful. The core category linked other main categories (emotional, attitudinal, psychosomatic, behavioral, and organizational responses) and their subcategories. For the purpose of this article, the main categories and their subcategories are displayed in Table 2 . Table 2 Emergence of negative trends: nurses' responses to burnout. Main categories Subcategories Emotional responses Personal desperation Professional desperation Attitudinal responses Depersonalization Negativity Psychosomatic responses Physical attrition Psychological attrition Behavioral responses Intolerance Justification Organizational responses Perfunctory care Declining performance Data trustworthiness The researchers accepted the perspective of Guba and Lincoln. They translated internal validity into credibility, external validity into transferability, reliability into dependability, and objectivity into confirm ability [ 17 ]. Credibility enhanced by the researchers' describing and interpreting her experiences. For this purpose the researcher kept a field journal in which she noted the content and the process of interactions, including reactions to various events. This journal became the record of relationships and provided material for reflection. Prolonged engagement and persistent observation helped to data credibility. In this way the process of data collection and analysis took 8 months. Data triangulation and method triangulation confirmed credibility [ 15 ]. Maximum variation sampling, participant observation and using published literature met this criterion. Furthermore, once the description of the phenomenon was complete, it was returned for verification to 4 participants of each center and they validated the descriptions. The original context described adequately, so that a judgment of transferability can be made by readers. The process of the study was audited for meeting dependability [ 18 ]. In doing so, student's supervisors and two other experts reviewed the process of the study and they arrived at a same conclusion. Confirm ability requires one to show the way in which interpretations have been arrived at via the inquiry. In this study, confirm ability was established, because credibility, transferability and dependability were achieved [ 17 ]. The signposts indicating research decisions and influences were present throughout the study and the entire study functioned as an inquiry audit. Results The findings related to factors which involved in the nurses' responses to burnout are presented in this article. Analysis and interpretation of data indicated that personal characteristics and social support have involved in the nurses' responses. These factors are presented in Table 3 . Table 3 Factors involve in nurses' responses to burnout. Personal characteristics Nurses' characteristics Patients' characteristics Social support Head nurse support Nursing administrator support Peer support Nurses' characteristics Data from interviews and participant observations indicated that special personal characteristics and personality traits have involved in the nurses' emotional, attitudinal, behavioral, and organizational responses to burnout. Personal characteristics such as conscience, religious beliefs, personal philosophy, commitment, a sense of responsibility, and altruism facilitated caring behaviors. Nurses with these characteristics were more patient and empathetic. They were more cooperative and rarely justified their faults by fatigue, workload or staff shortage. Conscience, commitment, and religious beliefs such as fear of divine requital were the most prominent traits that modified the responses to burnout. One of participants stated:" God knows. I always feel it's me there on the bed. Sometimes a patient calls and I ignore, but I tell myself, what I expected if I were on this bed? I fear god and say to myself, his authority is great and whatever I do, I will see the reflection of my doings". Some of the participants pointed to interest and love in caring of burn victims. One of participants stated:" these patients are different from other ones. I have worked more than 18 years in general hospitals, I haven't worked more than 7 years in burn centers, but I think that was a blessing in disguise, I am glad because clinical work for a burn patient means love, means everything, believe me. I don't care the managers' behaviors, workload, nursing shortage and other deficiencies, because I love burn victims. Believe me". Many of the nursing staff, distanced from patients, they had immoral beliefs and demonstrated humiliation and reproach in their behaviors. They related these attitudes to fatigue, micro and macro conditions in burn centers, and loss of motivation; but participant observation indicated that, this isn't the case for all the nurses. Nurses, who had been known as good nurses, were very calm and intimate with their patients and focused on the patients' needs. The researcher wrote in one of her field notes:" she is very calm and speaks with compassion. She makes jokes and patients are relaxed with her. She follows the principles and procedures more strictly than others". Nurses in all levels were under pressure of workload, low pay, staff shortage, environmental conditions of burn units and other structural inhibitors, but as the excerpts of interviews indicated, appraisal of these inhibitors was different in the presence of specific personal characteristics. Data indicated that in some instances, when there were a number of inhibitors and they were long stay, even positive personal characteristics couldn't work. This was often the case in infectious dressing rooms and busy wards. The worst kinds of treating patients were seen in these places. It seemed that the inhibitory factors which are persistent and too frequent interact with personal characteristics and finally overcome the positive characteristics. One of participants stated:" patients expect to receive care, expect a friendly encounter which does not happen. To tell the truth, some days I am excessively distressed and tired that I don't have the patience to answer the patient' questions and concerns. I emotionally can not do what I really want to do on a daily basis for the patients. The reason is persistent day and night problems occurring with this job". This process is displayed in Table 4 . Table 4 Interrelationships between personal characteristics, inhibitory factors and caring behavior in burnout. Frequency and intensity of inhibitory factors Positive personal characteristics Caring behavior High Defeat Deteriorates Low Overcome Improves Patients' characteristics Data strongly indicated that nurses' appraisal of the patients' characteristics have influenced some aspects of their attitudinal and behavioral responses. When the appraisal was positive, the relationships improved, and when it was negative, relationships deteriorated. Positive appraisal occurred often when the patients were cooperative and motivated for recovery and in cases where they had an advantage of high socio- economic class, cultural and educational levels, or whenever they stimulated the nurses' senses of compassion and pettiness. Negative appraisal occurred often when the patients were from lower socio- cultural levels, addicts or there was a possibility of having acquired immune deficiency syndrome (AIDS) or hepatitis. The first group was treated kindly and more respectfully. The use of humiliating words and reproach towards them diminished and as a result the aggressive behavior and physical withdrawal less occurred. One of participants stated:" ...I am more supportive and compassionate towards children, those who are very alone, who have no one to love and care for them, those who have committed self-inflicted burns, a woman whose husband caused her to burn herself and doesn't have any one to support her. In many occasions I have even paid them to buy juice from outside the hospital. I feel that these patients are needy". The second group was treated very unethical. They encountered a humiliating, reproaching, and aggressive behavior. One of participants stated:" ...he fights when I am dressing him, he pulls his hand and leg, he isn't cooperative, and he has no class. They drain all my energy to the point that I don't want to talk to them. I think they are mentally retarded. They keep still when I shout at them just the way that the children act. I tell them I'll pull your ear, and I'll beat you up. This is the behavior that has worked with them ". Moreover patients with extensive burns, whose survival was an improbable event, not only were badly treated, but also were sometimes ignored and received poor care. In other words, they received only those treatments which had been ordered by physicians to prevent being reprimanded by supervisors. One of the participants justified herself and stated:" if you want the truth, a patient with 90% burns can not benefit from tetracycline ointment, but it's there in his order, I prefer to spend my time with a patient who has a better chance of surviving. Right or wrong I don't apply the ointment, because I can spend that time for a patient who will survive". We can conclude that burnout has made nurses to modify their caring behaviors to fit the different type of patients they care for. Interaction between nurses and patients' characteristics As described later, nurse's and patient's characteristics modified the nurse's responses to burnout and altered caring behaviors. More analysis and interpreting of data indicated that interaction between these two variables resulted in a more powerful combination that alters responses to burnout and identifies the kind of caring behavior. This process is displayed in table 5 . It is important to mention that the meaning of patient's characteristics in this study is the nurse's appraisal or perception of these characteristics and nurse is clinical nursing staff in different levels. Table 5 Interrelationships between nurses and patient's characteristics and caring behavior in burnout. Nurse's characteristics Patient's characteristics + - + Naturally good behavior Relatively good behavior related to ethical aspects- sometimes non-ethical - Relatively good behavior related to inhibition of non-ethical aspects Bad behavior related to the opportunity for emerging non-ethical aspects Table 5 indicates that when both nurse's characteristics and her/ his appraisal of patient's characteristics are positive, then the nurse's caring behavior is naturally effective and efficient. In this case, patient is treated respectfully, there is an empathetic behavior, and nurses spend more time with her/his clients to value their emotional needs. When the nurse's characteristics are positive and her/ his perception or appraisal of the patient's characteristics is negative the nurse doesn't have a natural empathetic behavior. She thinks that she has to be good and behave well because of her beliefs; therefore she demonstrates a good behavior. Sometimes when the patient has been perceived as having a negative outburst, from the nurse's point of view, she/he has a very negative attitude towards the patient. This will cause ethical issues and misbehavior by the nurse. The good behavior occurs when the nurse's characteristics are negative but the appraisal of the patient's characteristics is positive. In this case, the patient's characteristics do not permit for emergence of negative characteristics of the nurse; therefore an ethical/respectful and caring behavior will result. At times when nurse's characteristics are negative and her/his perception of the patient's characteristics is also negative, non-ethical behaviors find a good opportunity to emerge. In this situation the patient encounters the worst behavior. Humiliation is intense, physical withdrawal is often seen and aggressive behavior is routine. One of participants stated: "I take care of some patients with love and conscience and take care of the other patients only with conscience and some nurses doesn't have love at all, I take care of silent, calm and lonely patients better, I perform routine care for the other". Another participant also stated:" it seems I give positive energy to the patients to whom I am more interested and take care of them with more love. I've seen that they respond better to therapeutic measures". Therefore interaction between nurses and patients' characteristics has a very powerful effect on the nurses' responses to burnout and determines their nature of caring behavior. Social support Data from interviews and participant observations strongly suggested that social support influences the nurses' responses to burnout. Supportive behavior of head nurses, nursing administrators and coworkers modified the nurses' responses. Among these, head nurse's the support was the most effective factor. Nurses believed that they do not have any motivation or desire to perform well when they are not supported well. One of participants stated:" we have a very close loving relationship with the head nurses who are supportive. In that situation I do many things for her. Her supportive attitude and caring/positive attitude helps me a lot. Do you understand?" another participant believed that he couldn't endure if the nursing administrator weren't supportive. He stated:" I have seen that she is doing a good job. I have been so stressed at times that I have thought of quitting. The nursing administrator has changed my mind during those stressful moments by being caring, loving and supportive and I have decided to stay in spite of low pay and benefits." Support from peers also modified the nurses' responses. They could tolerate more with the support of their coworkers. One of participants stated:" by god, I like every single one of them. It seems like we live together seven hours a day (major part of our day). We are so familiar with each other's character and behavior patterns. You may not believe this, these relationships has been very helpful and caused us to have a very strong unit. We care for each other in times of weakness, illness or pressure". It is worthy of mention that the effect of social support on the nurses' responses to burnout was not as powerful as the nurses' and patients' characteristics. Social support influenced the emotional and in some cases the attitudinal responses, but it didn't have enough power to alter the organizational and behavioral responses, therefore it didn't change caring behaviors significantly. Moreover, there were not any data indicative of the modifying effect of the factors proposed in this article on the psycho- somatic responses to burnout. Discussion Findings of this study indicated that nurses' and patients' characteristics and social support modified the nurses' responses to burnout. These factors altered the nurses' perceptions of the inhibitory factors; in other words they could have a more positive appraisal and this in turn modified their responses. Lazarus and Folkman (1988) proposed that the initiative of behavioral manifestations is a transactional appraisal which is important for the person's wellbeing and betrays the confrontation as noxious, useful, threatening or requiring struggle [ 19 ]. Moreover Lazarus (1976) believed that personal variables including values, beliefs, commitment and a sense of control over the environment are modifying factors that influence a person's cognitive appraisal [ 20 ]. In this study, conscience, religious beliefs and commitment were the most prominent characteristics that modified the nurses' responses to burnout. Nurses with positive characteristics had a non- threatening evaluation of their confrontations, therefore they cared better for their patients. Garrett and MC Daniel (2001) in their study concluded that a nurse's perception of the environment is more a function of personality than education or experience [ 8 ]. Conscience and commitment were of the most prominent characteristics that modified nurses' responses to burnout, and were related to the caring behavior. Focusing on caring attitudes, Roach (1987) also proposed that caring behavior in nursing is manifested through the five 'C ' attributes: compassion, competence, confidence, conscience and commitment [ 21 ]. Stuart and Sundeen (1987) referred commitment as representative of whatever important for the person. It includes decisions the person considers as necessary in his life and it can direct people to (or far from) the conditions which could be threatening, noxious or probably useful [ 22 ]. Participants pointed mostly to religious beliefs as a modifying factor. Stuart and Sundeen (1987) also concluded that spiritual beliefs can essentially reduce stress and influence on the persons' potential coping capabilities [ 22 ]. In this study, when inhibitory factors were persistent and frequent, even positive personal characteristics defeated. Selye (1976) concluded that the influence of stressors on a person depends on the number of stressors that must be confronted in a time, the duration of confrontation and existence of previous experience with the same stressors [ 23 ]. This happened more in infectious dressing rooms, where nurses had the closest longest contact with the bare bodies of burn victims. There were only one dressing room with 18–27 patients per each day for dressing and some of nurses had been there for more than 28 years. Patients with different characteristics treated differently. Nurses changed their caring behavior with different patients. It seemed that they didn't have enough emotional and physical energy and motivation for caring for all patients; but some patients stimulated their emotions and gave them the needed positive energy for caring. In other words, patients' characteristics could both reduce and intensify the nurses' responses to burnout. When these characteristics appraised as positive, caring behaviors improved, and when it was negative, the worst kind of behaviors occurred. Maslach (1982) in congruence with this conclusion believed that there is little evidence that caring is a uniform state [ 24 ], and Benner and Wrubel (1984) concluded that it's not clear that this is because the caring affect is depleted, or the nurse's personal needs for emotional protection take precedence over the human caring for others. They also believed that physical exhaustion may reduce the nurse's ability to continue to provide care [ 25 ]. Patients with extensive burns received the worst kind or care. Nurses stated that caring for these patients is futile. Data implied that caring for these patients have been incongruous with the nurses' values. Meltzer and Huckabay (2004) in their study of the relationship between critical care nurses' perceptions of futile care and its effect on burnout, concluded that feeling of emotional exhaustion in these nurses was highly influenced by the frequency with which nurses were involved in life- sustaining interventions that conflicted with the nurses' values and standards in term of what the nurses thought are ethically appropriate and could result in improvement in a patient's condition and outcome [ 26 ]. It was the interaction between nurses' and patients' characteristics that identified the caring behavior. Some authors have found according to their clinical practice that nurses have the ability to adjust their approach and their style of interaction with different patients. These authors proposed that they not only alter the nature of dialogue and the tone of voice to meet each patient's needs, but also adjust their affective response. They pointed that delineation of these behaviors would be a significant contribution, yet to date these styles of care have not been explored [ 27 ]. In this study some patients received a natural care, but others faced with an ethical and in some instances a non- ethical care. Natural care occurred spontaneously and without thinking, but ethical care happened thoughtfully by the mediation of the patients and/ or nurses positive characteristics. Typically non- ethical cares was thoughtful, but in this case both the nurses and patients characteristics were negative and the appraisal was too threatening. In her discussion of caring, Noddings (1984) also distinguished between natural caring and ethical caring. According to Noddings, natural caring comes from a remembrance of being cared for, whereas ethical caring " is an active relation between my actual self and a vision of my ideal self as one- caring and cared- for" [ 28 ]. Other authors also proposed that nurses may care naturally or they may care out of a desire to be a good nurse [ 29 ]. Social support from head nurses, nursing administrators and peers made nurses to endure and tolerate in the face of problems. Lazarus (1976) proposed that the resources reducing the potential harm could be found in the environment and essentially in others who indicated one can rely on them [ 20 ]. Social support didn't have enough power to modify behavioral and organizational responses and it could only change emotional and some aspects of attitudinal responses. Therefore caring behaviors didn't change. Garrett and MC Daniel (2001) also concluded that other people in the work setting like supervisors and peers might limit depersonalization and emotional exhaustion [ 8 ]. Conclusions This study as a part of more extensive research (PhD dissertation) identified the most important factors that intervened in the nurses' responses to burnout in burn centers of Tehran. Nurses had responded to burnout. These responses included emotional, attitudinal, psychosomatic, behavioral and organizational. This part of study indicated that the nurses and patients characteristics and interaction between these two factors had a very powerful effect on the responses and determined the kind of caring behavior. Moreover social support from managers (e.g. head nurses and nursing administrators) and peers modified some of the nurses' responses to burnout. The influence of positive personal characteristics, especially conscience, religious beliefs, philosophy, commitment, a sense of responsibility and altruism on the nurses' responses to burnout, the finding that, long lasting and persistent problems in the work setting can deteriorate even the personal characteristics, and regarding the numerous problems in burn centers of Tehran, there is an urgent need for helping the nurses. We suggest that due to the intense staff shortage in these centers, the managers try to keep these nurses. The only way they can do this is by using stress reduction programs. Data strongly indicated that these nurses need to rest periodically to preserve energy and to refresh their morale. Moreover, giving importance to moral and ethical aspects of care by managers could be helpful and motivating. Changing burn patients' inherent characteristics and their other characteristics such as poverty and socio- cultural level is not a possible alternative. Promoting nurses' morale is possible and must be done promptly if we want our burn survivors receive at least an ethical effective care. Social support made the problems tolerable. Therefore we recommend nurse executives in burn centers of Tehran promote a work environment that help to decrease the perception of pressures and increase perceptions of social support. The best solution is of course eliminating the micro and macro conditions which overshadow the nurses' responses and caring behaviors. This grounded theory created several hypotheses in this stage. Nurses with specific traits were more resistant to burnout and were more caring. It is suggestive of conducting a quantitative research to test the relationship between personality traits, burnout and caring behaviors. Differences in caring behavior for different patients were related to the influence of burnout on the nurses' personality traits and appraisal of patients' characteristics. It was a very new finding that needs to be investigated in more detail. The dramatic effect of social support on the nurse' perceptions of pressures in burn centers of Tehran is suggesting of identifying the relationship between social support and the level of burnout the nurses experience in these burn centers. However the nature of a qualitative research like this, limits it's generalizability; therefore we suggest conducting more qualitative research in other critical care units to support these findings. Application of the findings of this study and conducting the suggested studies would help the managers of burn centers to enhance an environment conductive to morale, promote natural and ethical care, refresh the nurses' positive emotions and facilitate a supportive environment for their nurses. Competing interests The author(s) declare that they have no competing interests. Authors' contributions FR initiated and designed the research, collected and analyzed the data and wrote the paper. FO was the main supervisor, helped in analysis, and revised and edited the drafts. MN was co- supervisor and revised the drafts. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC534790.xml |
493275 | Algorithms for optimizing drug therapy | Background Drug therapy has become increasingly efficient, with more drugs available for treatment of an ever-growing number of conditions. Yet, drug use is reported to be sub optimal in several aspects, such as dosage, patient's adherence and outcome of therapy. The aim of the current study was to investigate the possibility to optimize drug therapy using computer programs, available on the Internet. Methods One hundred and ten officially endorsed text documents, published between 1996 and 2004, containing guidelines for drug therapy in 246 disorders, were analyzed with regard to information about patient-, disease- and drug-related factors and relationships between these factors. This information was used to construct algorithms for identifying optimum treatment in each of the studied disorders. These algorithms were categorized in order to define as few models as possible that still could accommodate the identified factors and the relationships between them. The resulting program prototypes were implemented in HTML (user interface) and JavaScript (program logic). Results Three types of algorithms were sufficient for the intended purpose. The simplest type is a list of factors, each of which implies that the particular patient should or should not receive treatment. This is adequate in situations where only one treatment exists. The second type, a more elaborate model, is required when treatment can by provided using drugs from different pharmacological classes and the selection of drug class is dependent on patient characteristics. An easily implemented set of if-then statements was able to manage the identified information in such instances. The third type was needed in the few situations where the selection and dosage of drugs were depending on the degree to which one or more patient-specific factors were present. In these cases the implementation of an established decision model based on fuzzy sets was required. Computer programs based on one of these three models could be constructed regarding all but one of the studied disorders. The single exception was depression, where reliable relationships between patient characteristics, drug classes and outcome of therapy remain to be defined. Conclusion Algorithms for optimizing drug therapy can, with presumably rare exceptions, be developed for any disorder, using standard Internet programming methods. | Background During the last decades the possibility of computer support for optimum management of various disorders has attracted interest [ 1 - 3 ]. Regarding drug therapy the approach has usually been limited to one particular disease [ 4 - 7 ]. Diabetes has been extensively studied and thoroughly analyzed also from a systems perspective [ 8 ]. Several attempts have been made to develop a general model for medical guidelines, which could be used also for the development of computer programs [ 9 - 12 ]. Computerized physician order entry has been shown to reduce the frequency of serious medication errors. Decision support tools such as alerting functions for patient medication allergy are a key part of these applications. However, Abookire and co-workers [ 13 ] analyzed trend data obtained over a five-year period that showed decreasing compliance to allergy alert functions within computerized order entry. They concluded that optimal performance requires iterative refinement and that, as systems become increasingly complex, mechanisms to monitor their performance become increasingly critical [ 13 ]. In another study large differences were seen for all main types of medication errors: dose errors, frequency errors, route errors, substitution errors, and allergies [ 14 ]. In a university hospital setting, the reductions in transcription errors and medication turn-around times supported the view that computerized physician order entry and an electronic medication administration record can provide a good return on investment [ 15 ]. In the current study guidelines for 246 diseases were analyzed regarding the content of facts, and relationships between these facts, influencing the selection and dosage of therapeutic drugs. The aim of study was to identify one or more algorithms (descriptions of how to solve a problem in a definite number of steps), which could be used to support optimum drug therapy in any disease. Methods The studied material consisted of 110 officially endorsed text documents, published between 1996 and 2004. Forty-four of these documents were published by the Swedish Medical Products Agency [ 16 ], 13 by the Swedish Agency for Evaluation of Medical Technologies [ 17 ] and 53 (in three booklets) by the Swedish Strategic Programme for The Rational Use of Antimicrobial Agents and Surveillance of Resistance (STRAMA) [ 18 ], established in order to prevent bacterial resistance to common antibiotics. The documents contained guidelines for drug therapy in altogether 246 disorders. Information regarding contraindications, dosage etc was available from an existing pharmaceutical website [ 19 ] and was retrieved by linking to context-relevant pages. These documents were analyzed as follows: All statements in the particular document regarding patient-, disease- and drug-related factors were recorded. In a second step the statements in the texts describing the relationship between these factors were identified. After that the obtained information was rewritten in a format similar to a generic computer language, using if-then statements and logical operators as needed. These rewritten documents were grouped according to similarity in structure, based on the number of if-then statements, the need for branching or calculations in the code and the occurrence of 'left-overs', i.e. information in the text that could not be explicitly expressed using if-then statements and logical operators. In theory this procedure could result in an intractable combinatorial explosion. This was observed only in one situation, regarding side effects of anti-hypertensive medication. Results Figure 1 shows the number of facts, relevant for the selection of appropriate drug class in the 246 studied diseases. Three algorithms were sufficient for the intended purpose. The simplest model is a list, containing the data, which imply that the patient should or should not receive a particular treatment. This model is adequate in situations where only one pharmacological treatment modality exists, such as influenza, where neuramidase inhibitors are the only drug class approved in Sweden, see figure 2 . A 16 page summary of 35 pages of background documentation contained the following data items influencing drug therapy: four risk groups, time from start of symptom, two available neuramidases, one of which was approved also for use in children. Pharmaceutical properties regarding all approved drugs is available online (accessed from the program using 'deep linking'). In the actual program implementation information was added about current epidemiological state regarding influenza by linking to a government authority website, diagnostic criteria and some management aspects (in pop-up windows) (Fig. 2 ). In situations, where treatment can by provided using drugs from different pharmacological classes, a more elaborate model is required. An example is heart failure. A 25-page summary of 69 pages of background documentation described six physiological states and 11 available drug classes. This knowledge was modelled using an ordered set of it-then statements, see figure 3 . The function containing these sets of ordered rules is activated every time the user enters some information on the screen. For an example of user interface, see figure 4 . In the actual program implementation information was added about treatment in emergency situations and in different clinical stages of the disease (in pop-up windows) (Fig. 4 ). The algorithms mainly consist of sets of if-then statements, grouped in a single function, which is activated each time the user enters some information by checking a check box or radio button on the screen [ 20 ]. This information is equivalent to a fact list, and the if-then rules mimic a knowledge base. Thus in these aspects the algorithms are similar to conventional expert systems [ 21 ], but a separate inference engine was not needed. However, the if-then rules had to be ordered according to the output such that positive information, i.e. recommendations to use a particular drug, always was overwritten when factors were present that implied that this particular drug was contra-indicated or less suitable. Such sets of if-then statements were able to manage the identified information in all, but one instance. The exception was depression, where explicit knowledge regarding the relationships between patient characteristics and optimum choice of drug is lacking. In some diseases the selection and dosage of drugs depend on the degree to which one or more particular factors are present. In such cases a specific program implementation of an established decision model based on fuzzy set theory [ 22 ] could be used. An example is the adjustment of medication according the patient's description of side effects in the treatment of high blood pressure [ 6 , 23 ], see figure 5 . The data in this program were obtained from a study of 1013 treated and 125 untreated hypertensive patients [ 24 ]. Thirteen types of complaints were recorded and correlated to the ten available drug classes. The resulting matrix was analysed by four physicians, one professor of endocrinology, with special interest in hypertension, two clinical pharmacologists and one family physician, and modified according to scientific evidence and clinical experience. Clicking on radio buttons enters the type and degree of complaint and the program ranks the available drug classes according to the degree to which they can be assumed not to cause the patient's complaint(s). The algorithm is an implementation of a decision theory, based on fuzzy sets [ 22 ]. A verbal description of the implementation would be quite lengthy, but the workings of the algorithm and the code can be seen on the author's website [ 23 ]. The pharmaceutical information about all therapeutics drugs, approved in Sweden, is available on the Internet at a site managed by the association of pharmaceutical industries [ 19 ]. By linking to this site, the programs provide access to the relevant and continously updated drug information regarding all drugs, registered for use in the current disease. In addition, each program is linked to the appropriate guideline document, so that the user can verify the program's treatment advice. Discussion The main findings in the current study are that simple algorithms can support drug therapy and that such algorithms can be easily implemented using standard Internet programming techniques. Problems in the implementation of information system designed to support actual care are discussed in a study by Bates et. al. [ 25 ]. The authors concluded that 'ten commandments' have to be taken into account. The most important factors for success were speed, anticipation of needs and good fit into the user's workflow. The programs developed in the current study easily meet these criteria, provided that the users have access to the Internet. However, it should be noted that the programs require no communication of patient-related information over the Internet, other than transmission of the prescription to the pharmacy, if such a system is implemented. One disorder, depression, was identified for which relevant computer program support could not be constructed. Initially, the program model with the above-mentioned decision algorithm, based on fuzzy sets, was thought to be applicable, since the number of relevant facts is well within manageable range; four major classes of drugs, slightly more than ten diagnostic variables, some ten receptors and seven neurotransmitters. However, the information in the official consensus document [ 26 ], did not provide sufficient information, due to the lack of explicit relationships between the patient's characteristics, as described e.g. by diagnostic rating scales, and the pharmacological properties of the various types of antidepressants. Outcome studies regarding depression also give little information about which drug should be recommended for a particular patient. The exception is a recent report by Joyce et al [ 27 ], who found a markedly superior response to nortriptyline compared to fluoxetine in men above 40 years of age, and the reverse effect in women, aged 18 – 24 years. Problems in pharmacotherapy related to the simultaneous occurrence of multiple disorders in the same patient are well studied and presented in the pharmacological information as contraindications or recommendations of "caution in patients with disease ...". This information is used as output from the programs in the current study. The information in the analyzed text documents was not always sufficiently detailed and logically consistent to allow the construction of computer code. Additional information could usually be gathered from other sources, but such procedures may induce errors and should be avoided. Methods are currently available to support logical structure in text documents, e.g. the Guideline Elements Model [ 8 ]. This model relies on XML tagging, a standard for content description in documents intended for electronic processing. A formal structure would increase the usefulness of guidelines [ 24 ], since it would be possible to present them in the format of fast, easily available and "easy-to-use" programs, as in the recently published PresGuid system [ 11 ], which relies on XML tagging of text based guidelines for automatic conversion to computer programs. Experience will tell whether this approach will be more efficient than the simpler strategy of manual extraction of relevant data from the guideline document, combined with JavaScript coding, used in the current study. The introduction of a new drug for treatment of a particular disease, or the results from new scientific studies may have considerable influence on guidelines. Yet only a few documents were changed during the study period and the lifespan of pharmacotherapeutic guidelines can be assumed to be 2–5 years. However, new information may appear unexpectedly, and a reasonably complete system for pharmacotherapy requires frequent updates. The approach taken in the current study, with the programs stored on a website as scripted HTML pages, facilitates the maintenance and update of the system and ensures that all users always have access to the latest version. An additional way to increase relevance over time is to link from the programs to recently published studies, e.g. by providing automatic searches in the PubMed database [ 28 ]. Conclusion In the current study based on guidelines regarding two hundred and forty-six diseases, three basic program models were found to be sufficient for computerized decision-support in pharmacotherapy. The programs could be developed using HTML and JavaScript, i.e. simple and widespread programming techniques. No specific development tools are needed although standard programming tools such as Macromedia Dreamweaver ® are helpful. The use of Internet implies that only one version of the programs is in use, that upgrading is very convenient, and that no interference occurs with existing systems. No confidential patient data has to be communicated and the programs are always available from any Internet connected computer. In summary, algorithms for optimizing drug therapy can, with presumably rare exceptions, be developed for any disorder using standard Internet programming methods. Competing interests None declared. Authors' contributions PW conceived and designed the study, collected data, supervised data analysis, did most of the JavaScript and some of the HTML programming and contributed to the writing of the paper. LM did most of the HTML and some of the JavaScript programming and contributed to the data analysis and the writing of the paper. Both authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC493275.xml |
449784 | Subcellular Localization of Frizzled Receptors, Mediated by Their Cytoplasmic Tails, Regulates Signaling Pathway Specificity | The Frizzled (Fz; called here Fz1) and Fz2 receptors have distinct signaling specificities activating either the canonical Wnt/β-catenin pathway or Fz/planar cell polarity (PCP) signaling in Drosophila. The regulation of signaling specificity remains largely obscure. We show that Fz1 and Fz2 have different subcellular localizations in imaginal disc epithelia, with Fz1 localizing preferentially to apical junctional complexes, and Fz2 being evenly distributed basolaterally. The subcellular localization difference directly contributes to the signaling specificity outcome. Whereas apical localization favors Fz/PCP signaling, it interferes with canonical Wnt/β-catenin signaling. Receptor localization is mediated by sequences in the cytoplasmic tail of Fz2 that appear to block apical accumulation. Based on these data, we propose that subcellular Fz localization, through the association with other membrane proteins, is a critical aspect in regulating the signaling specificity within the Wnt/Fz signaling pathways. | Introduction Pattern formation in multicellular organisms relies on inductive signaling events. Several evolutionarily conserved ligand–receptor combinations and associated signal transduction pathways are used again and again during development to induce tissue- and cell-type-specific responses. Thus, context-dependent signaling specificity is an important prerequisite for patterning and differentiation. Although for many signaling pathways the flow of information is largely established, the underlying signaling specificity mechanisms remain unclear. Members of the Frizzled (Fz) family of seven-pass transmembrane proteins act as receptors for the Wnt family of secreted ligands ( Bhanot et al. 1996 ). In most cases, Wnt/Fz signal transduction leads to posttranslational stabilization of the intracellular protein β-catenin (β-cat) (β-cat or Armadillo [Arm] in Drosophila ; reviewed in Polakis 1999 , 2000 ). However, recent work has established that some Wnt ligands and Fz receptors can also signal through pathways independent of the Wnt/β-cat (Wg/Arm) cascade in certain contexts in vertebrates and invertebrates (reviewed in Mlodzik 2002 ; Veeman et al. 2003 ). In particular, the Fz/planar cell polarity (PCP) pathway has been studied extensively in both Drosophila and vertebrates ( Adler 2002 ; Keller 2002 ; Mlodzik 2002 ; Tada et al. 2002 ; Strutt 2003 ). PCP is easy to study and evident in all adult tissues in Drosophila . For example, in wing cells the PCP response is the formation of an actin spike (the wing “hair”) that points distally, and in the eye PCP is manifest in the regular ommatidial arrangement in the anteroposterior and dorsoventral axes (reviewed in Adler 2002 ; Mlodzik 2002 ). These distinct PCP manifestations are regulated by the same set of genes, the so-called primary polarity genes, of which Fz is the most prominent and best studied. Similarly, this noncanonical Fz/PCP pathway has been implicated in PCP establishment in vertebrates, with prominent examples including the polarization of the sensory epithelium in the inner ear ( Curtin et al. 2003 ; Dabdoub et al. 2003 ; Montcouquiol et al. 2003 ) and aspects of cell polarization in the convergent extension process during gastrulation (for a description of the similarities, pathway conservation, and specific readouts see reviews ( Keller 2002 ; Mlodzik 2002 ; Veeman et al. 2003 ). Despite the increasing knowledge about the distinct pathways mediated by Wnt/Fz signaling, the regulation of Fz signaling specificity remains largely obscure. Both pathways, Wnt/β-cat and Fz/PCP, signal via Disheveled (Dsh) (reviewed in Boutros and Mlodzik 1999 ). This raises the intriguing question of how structurally very similar receptors can signal through a common protein into distinct downstream effector pathways. In Drosophila, Fz (for clarity we will refer to it as Fz1) and Fz2 are functionally redundant receptors for Wg, activating the canonical Wg/Arm cascade ( Bhat 1998 ; Kennerdell and Carthew 1998 ; Bhanot et al. 1999 ; Chen and Struhl 1999 ). In addition to this redundant role in canonical signaling, Fz1 has a specific nonredundant role in the Fz/PCP pathway ( Vinson and Adler 1987 ; Vinson et al. 1989 ). Subdomains of Fz1 and Fz2 have been analyzed with respect to the functional similarities and differences of the two receptors ( Boutros et al. 2000 ; Rulifson et al. 2000 ; Strapps and Tomlinson 2001 ). These studies have suggested that signaling differences between Fz1 and Fz2 could lie in their different affinities for ligands (e.g., Wg has a 10-fold higher affinity for Fz2; Rulifson et al. 2000 ) and in additional cytoplasmic sequences which govern distinct intrinsic signaling preferences between Fz1 and Fz2 for the canonical and Fz/PCP pathways ( Boutros et al. 2000 ; Strapps and Tomlinson 2001 ). Signaling specificity could be regulated by distinct Wnt-Fz combinations that would activate either the canonical or noncanonical pathway. Although a PCP-specific Wnt ligand for Fz1 has not yet been identified in flies, in vertebrates specific Wnt(s)-Fz(s) combinations are associated with either canonical or Fz/PCP signaling. However, the specificity is not simple. For example, although Wnt5a and Wnt11 cause embryonic phenotypes associated with the Fz/PCP-like pathway ( Heisenberg et al. 2000 ; Tada and Smith 2000 ), coexpression of Wnt5a with Fz5 causes axis duplications, a canonical Wnt/β-cat phenotype ( He et al. 1997 ). Similarly, vertebrate Fz7 receptors have been shown to affect both noncanonical ( Djiane et al. 2000 ; Medina et al. 2000 ) and β-cat signaling ( Kuhl et al. 2000 ). These data suggest that signaling specificity is not necessarily associated with a particular Wnt ligand or Fz receptor. Wnt/Fz signaling specificity may be determined, in part, by the presence of distinct coreceptors. For example, the Arrow-LRP5/6 protein acts as a Wnt/Wg coreceptor and is only required for Wnt/β-cat signaling ( Tamai et al. 2000 ; Wehrli et al. 2000 ). No coreceptor of Fz1 has been reported for Fz/PCP signaling. Clearly this is a complicated issue and is likely to be context and cell-type dependent. Endogenous Fz2 is difficult to detect, but in the wing hinge region it is localized evenly in membranes along the apical–basal axis (M. Strigini, unpublished data). Similarly, overexpressed Fz2 (under dppGal4 control) is localized throughout the apical–basal axis of larval imaginal disc epithelia, and extracellular Wg binds to Fz2 predominantly at the basolateral membrane ( Strigini and Cohen 2000 ), suggesting indirectly that canonical Wg/β-cat signaling is initiated at the basolateral cell surface. The existing anti-Fz antibodies are, similarly, not sensitive enough to detect endogenous levels of Fz protein ( Krasnow and Adler 1994 ), but green fluorescent protein (GFP)–tagged Fz (Fz1-GFP) expressed under the control of a ubiquitous promoter shows apical localization in pupal wings and larval eye discs during PCP signaling ( Strutt 2001 ; Strutt et al. 2002 ). All PCP molecules analyzed (Dsh, Flamingo [Fmi; a.k.a. Starry Night], Strabismus [a.k.a. Van Gogh], Prickle, and Diego ) are also localized in the apical region of pupal wings and eye epithelia (reviewed in Strutt 2003 ). Importantly, the apical localization of many PCP genes is lost in mutants of Fz1/PCP signaling components, suggesting that Fz/PCP signaling regulates apical localization ( Axelrod 2001 ; Feiguin et al. 2001 ; Shimada et al. 2001 ; Strutt 2001 ; Bastock et al. 2003 ; Jenny et al. 2003 ). Thus, as Fz1 and Fz2 show different subcellular membrane localization within the apical–basal axis, we have here extended our analysis of Fz1 and Fz2 to determine whether the specific subcellular localization is important for signaling readout and to identify the molecular aspects responsible for the localization differences. Our data indicate that localization to apical junction complexes promotes Fz/PCP signaling and inhibits canonical Wg/β-cat signaling, and that the subcellular localization of Fz receptors is mediated through sequences in the cytoplasmic tail (C-tail). In addition, we show that the seven-pass transmembrane region contains elements that are critical for PCP signaling. Based on our data, we propose a model in which subcellular localization, possibly through the association of Fz with other membrane proteins such as coreceptors, is a critical aspect in regulating the signaling readout and specificity within the Wnt/Fz signaling pathways. Results Different Subcellular Localization of Fz and Fz2 in Imaginal Disc Epithelia To confirm that Fz1 is localized apically, we analyzed Fz1 distribution in third instar larval discs ( Figure 1 ). Similar to previous reports ( Strutt 2001 ), we found that a ubiquitously expressed Fz1-GFP is always enriched at apical junctions (although the expression in third instar larval discs is weaker than in pupal wings; Figure 1 A). Expression of a Myc-tagged Fz1 under dpp-Gal4 control also displays a strong enrichment in the apical region of the disc epithelium ( Figure 1 C–1F) and in some punctae that appear to be intracellular vesicles ( Figure 1 F). There are only low levels of Fz1 detected basolaterally ( Figure 1 F; unpublished data). Apical Fz1 largely colocalizes with DE-Cadherin (DE-Cad; a marker for adherens junctions; Figure 1 D), whereas it only slightly overlaps with Discs large (Dlg) staining ( Figure 1 E; Dlg is localized to septate junctions just basally to adherens junctions [reviewed in Tepass et al. 2001 ]). This Fz1 localization pattern is very similar to the PCP factor Strabismus/Vang, which also largely colocalizes with DE-Cad, and only slightly with Dlg ( Bellaiche et al. 2004 ). Figure 1 Subcellular Localization of the Fz1 Protein (A) Anti-GFP staining of Fz-GFP in arm-fz-GFP third instar wing disc ( arm drives ubiquitous expression); xz-section is shown. (B) Illustration of a cross section of a third instar wing disc. Wing epithelium forms several folds in the hinge region, where apical–basal localization can be visualized in a horizontal xy-section. The purple line in (B) indicates the position of the xy-optical section in such folds shown in (C–F). (C) Staining of a dpp-Gal4/UAS-fz1–1-1(myc) third instar wing disc. Localization of DE-Cad (in red), Dlg (green), and Fz1–1-1 (anti-Myc, blue) is shown. Apical region of the epithelium faces the lumen in the fold, and the basolateral regions are away from the lumen. (D) same staining as in (C) with two channels shown: DE-Cad and Fz1–1-1. DE-Cad (red) and Fz1–1-1 (blue) largely overlap. (E) Dlg (green) and Fz1–1-1 (blue) from (C) are shown. Fz1–1-1 localizes generally more apical than Dlg (with only a very slight overlap). (F) Fz1–1-1 single-channel staining. In summary, Fz1–1-1 is mainly localized in the apical adherens junctions and strong punctae inside cells (probably intracellular vesicles). Low levels of Fz1–1-1 also exist more ubiquitously in the basolateral region. (G) Schematic illustration of relative positions of DE-Cad, Dlg, and Fz1–1-1 along the apical–basal axis epithelial cells. DE-Cad marks the adherens junctions, whereas Dlg localization correlates with septate junctions. Taken together, these data suggest that Fz1 is mostly localized at adherens junctions. This is in contrast to Fz2, which is distributed throughout the cellular membrane along the apical–basal axis in the wing imaginal disc epithelium ( Strigini and Cohen 2000 ). As only Fz1 can signal effectively in the Fz/PCP pathway and other PCP proteins also show apical localization ( Strutt 2003 ), we speculate that apical Fz1 localization is an important feature of signaling specificity. The C-Tail of Fz Family Receptors Controls Subcellular Localization As Fz1 and Fz2 show different subcellular localization, we wished to determine which domains or sequences within the receptors are responsible for the specific localization. To address this question, we examined the localization of Fz1/2 chimeric receptor proteins (expressed under the control of dpp-Gal4 ) in wing imaginal discs. Fz1 and Fz2 were subdivided into three parts: (1) the N-terminal Wnt-interacting cysteine-rich domain (CRD), (2) the remaining proximal extracellular domain and 7 transmembrane region and loop region (collectively referred to as 7-TM), and (3) the intracellular C-tail. All chimeric proteins were Myc-tagged between the CRD and 7-TM region (see Materials and Methods ) and labeled with three digits (separated by dashes) corresponding to the three domains of Fz1/2, with “1” and “2” reflecting Fz1 and Fz2 origin, respectively. In all cases tested, the hybrid Fz1/2 proteins carrying the C-tail of Fz1 were enriched apically ( Figure 2 ), comparable to wild-type Fz1, and colocalized with apical junctional markers (see Figure 1 ; unpublished data). In contrast, chimeric Fz receptors carrying the Fz2 C-tail, including Fz2–2-2, were localized evenly along the apical–basal axis ( Figure 2 B, 2 C, and 2 G), comparable to wild-type Fz2 (e.g., endogenous Fz2 [M. Strigini, personal communication] or overexpressed Fz2 under dpp-Gal4 control [ Strigini and Cohen 2000 ]). In summary, these data indicate that the C-tails of Fz receptors are responsible for their specific subcellular localization. Figure 2 The Cytoplasmic Region of Fz Regulates Subcellular Localization All Fz1/2 chimeras shown are Myc-tagged (the tag being inserted right after the CRD of Fz1 or Fz2; see Materials and Methods ; Boutros et al. 2000 ). The respective Fz1/2 chimeras, with their schematic structure shown under each photomicrograph, were expressed under dpp-Gal4 (expression domain marked with UAS-EGFP in example in [A]) and analyzed by confocal microscopy xz-sections (perpendicular to the stripe of expression in the wing pouch region). (A) Subcellular localization of wild-type Fz-Myc (Fz1–1-1, in green; red channel shows coexpressed GFP to mark expressing cells). Single-channel black-and-white staining of Fz-Myc is shown on right. (B–F) Anti-Myc staining of different Fz1/2 chimeras: (B) Fz1–2-2, (C) Fz1–1-2, (D) Fz2–1-1, (E) Fz1–2-1, and (F) Fz2–2-1. (G) Fz2–2-2. Note the correlation of apical Fz localization with the presence of the Fz1 C-tail. To address whether subcellular localization correlates with specific Fz signaling events, we tested the signaling preferences of the respective chimeric receptors. This was analyzed in adult wings by scoring for either a PCP or canonical Wg-signaling gain-of-function (GOF) phenotype ( Figure 3 ; Table 1 ). Expression of Fz1–1-1 under dpp-Gal4 control in wing imaginal discs caused wing cell hairs to point away from the expression domain ( Figure 3 B). This is consistent with the notion that hairs point away from regions of higher Fz signaling levels in the PCP context ( Adler et al. 1997 ). Expression of the chimeric Fz receptors showed that the presence of the Fz1 C-tail is necessary for a strong PCP GOF phenotype ( Figure 3 ; Table 1 ), suggesting that the apical localization of Fz is important for normal PCP signaling. These experiments also indicated that, in addition to apical localization, the 7-TM region of Fz1 is necessary for effective PCP signaling ( Figure 3 ; Table 1 ). Similar results were obtained in GOF PCP assays during eye development ( Table 1 ; Boutros et al. 2000 ; unpublished data). Figure 3 GOF Planar Polarity Wing Phenotype of Fz1/2 Chimeras dpp-Gal4 was used to express the respective Fz1/2 chimeras in the wing (same as described in Figure 2 ). (A) Wild-type wing. The dpp-Gal4 expression domain is highlighted by a thick orange line. In wild-type, all wing hairs are pointing distally. (B) dpp-Gal4; UAS-EGFP/UAS-fz1–1-1 wing ( dpp>fz1–1-1 ; the expression domain is again highlighted with light orange). Wing hairs flanking the expression domain point away from it, consistent with previous observations that hair point away from higher levels of Fz1 activity ( Adler et al. 1997 ). (C) dpp>fz1–2-2 wing. Wing hairs are not pointing away from expression domain, suggesting that Fz1–2-2 is not active for PCP signaling. (D) dpp>fz1–1-2 wing. Hairs point away only very slightly (less than 45 o ; compare with Fz1–1-1, showing a 90 o reorientation next to expression domain). Several different lines of UAS-fz1–1-1 and UAS-fz1–1-2 were compared, showing identical behavior (Fz1–1-1 having a much stronger phenotype), suggesting that the C-tail is required for full PCP Fz activity. (E) dpp>fz2–1-1 wing. Most wing hairs point away from expression domain. The phenotype is weaker than Fz1–1-1. (F) dpp>fz1–2-1 wing. Wing hair orientation is hardly affected. Since Fz1–2-1 is apically localized (see Figure 2 E), this result indicates that the presence of the Fz1 7-TM region is important for PCP activity. Table 1 Behavior of Chimeric Fz Receptors a Rescue that resembles wild-type b For schematic presentation of these mutants see Figure 4 A wt, wild-type; NA, not available; ND, not determined Behavior of chimeric Fz receptors was assayed in four different ways as indicated. The eye rescue phenotype was quantified by analyzing 3–6 independent eyes for each genotype. Wild-type represents 99.5%–100% correctly oriented ommatidia, whereas in the fz −/− background only about 30% show the correct orientation. Note that for rescue both the CRD and 7-TM region of Fz1 are required in addition to apical localization Taken together, these experiments demonstrate that (1) apical Fz1 localization correlates with higher levels of Fz/PCP signaling activities and (2) the 7-TM region of Fz1 is critical for effective PCP signaling. Sequence Requirement for Apical Localization within the C-Tail Next we wished to determine which part of the C-tail of Fz1 or Fz2 is responsible for the difference in subcellular localization. The protein sequences of the Fz1 and Fz2 C-tails are homologous over the first 29 amino acids (45% identity), but Fz2 is longer by an additional 61 amino acids ( Figure 4 ). The apical localization sequence could thus be located either in the nonconserved stretches within the common 29 residues, or within the Fz2 C-tail extension. We addressed both possibilities systematically and analyzed the localization of the respective mutants and their effects in the functional GOF assay in the wing (see above). Figure 4 Effects of Fz1/2 C-Tail Mutations on Subcellular Localization and PCP Activity (A) Sequence alignment of Fz1 and Fz2 C-tails. Note high degree of conservation within the membrane proximal shared portion of the Fz1 and Fz2 C-tails. The respective mutations generated and analyzed are indicated above the sequence (see also Table 1 for complete data set). As in Figures 2 and 3 , dpp-Gal4 was used to drive expression of the respective mutants, and these were detected by anti-Myc staining in third instar wing discs. Examples for Fz1–1-1V559E (V to E substitution) are shown in (B) (localization) and (F) (function). All other mutants analyzed as shown in (A) are listed in Table 1 . (C–E, G, and H) show the effects of the Fz2 C-tail-specific sequences. The Fz2 C-tail was truncated at the position of the Fz1 stop codon (amino acid L633), yielding a short Fz2 C-tail (2S). The localization (C and D) and GOF PCP function (G and H) of the respective chimeras, Fz1–2-2S and Fz1–1-2S, is shown. Note that both chimeras localize apically (C and D), and Fz1–1-2S shows a strong PCP GOF phenotype (H), very similar to Fz1–1-1 (see Figure 3 B). Fz1–2-2S shows only a very weak PCP phenotype (G), mainly occurring at an anterior distal region of the wing (marked by arrow; the rest of the wing is wild-type). (E) Subcellular localization of Fz1–1-1C2. Fz1–1-1C2 is Fz1 with the addition of the Fz2-specific tail extension (see Materials and Methods ). Note ubiquitous protein localization within the apical–basal axis (E) and a much reduced PCP activity, as compared to wild-type Fz1–1-1, in the functional assay (I). The phenotype is much weaker than in wild-type Fz1 (compare with [F] and [H] and Figure 3 B). First, we mutated several Fz1–1-1–specific residues to those of Fz2, or deleted conserved amino acid stretches within the Fz1–1-1 C-tail (see Figure 4 A and Table 1 for specific mutations analyzed). All mutated Fz1–1-1 receptor proteins showed normal localization to apical junctions ( Figure 4 B; Table 1 ), and when analyzed for their function also showed a typical Fz GOF PCP phenotype in the wing in that the wing hairs were directed away from the source of expression ( Figure 4 F; Table 1 ). Second, we tested whether sequences within the extended Fz2 C-tail have an effect on localization or PCP signaling. We introduced a stop codon after the L633 residue of Fz2 (corresponding to the position of the stop codon in Fz1) in Fz1–1-2 and Fz1–2-2 chimeras ( Figure 4 A, blue arrowhead), thus truncating the Fz2 C-tail and generating chimeras Fz1–1-2S (“S” for “short”) and Fz1–2-2S. Whereas Fz1–1-2 and Fz1–2-2 are ubiquitously localized, both Fz1–1-2S and Fz1–2-2S localize apically to adherens junctions, in a manner indistinguishable from that of Fz1–1-1 and Fz1–2-1 (compare Figure 4 C and 4 D to Figure 2 A and 2 E). These data suggest that the Fz2 C-tail extension interferes with apical localization. These same chimeras were tested in the functional assay for PCP signaling activity. Strikingly, expression of Fz1–1-2S caused a phenotype very similar to that of Fz1–1-1 ( Figure 4 H), but different from that caused by Fz1–1-2 (see Figure 3 D). Expression of Fz1–2-2S resembled that of Fz1–2-2 or Fz1–2-1, with very weak PCP effects (compare Figure 4 G to 3 C and 3 F). In summary, these results confirm that both apical localization and sequences located within the 7-TM region are functionally important for PCP signaling. To test whether the extension within the Fz2 C-tail can more generally block apical localization, we added the Fz2 extension on to Fz1–1-1 ( Figure 4 ; see Materials and Methods for details). This Fz1–1-1C2 receptor isoform was not apically enriched ( Figure 4 E), resembling the localization of Fz1–1-2. Consistently, in the functional PCP readout assay, expression of Fz1–1-1C2 showed only very weak GOF PCP effects ( Figure 4 I). Based on the results with Fz1–1-1C2 and Fz1–1-2S, we conclude that the Fz2 C-tail extension causes Fz receptors to acquire a ubiquitous membrane distribution, preventing them from accumulating at the apical junctions and thereby affecting their ability to signal via the Fz/ PCP pathway. Apical Localization Affects Rescue Capability of the Fz Chimeras The chimeric Fz1/2 receptors (driven directly by the ubiquitous tubulin promoter [tub] ) were also tested for their ability to rescue the fz − eye and wing PCP phenotype. tub -Fz1–1-1 and tub -Fz1–1-2S (which are both apically localized) fully rescue the fz − loss-of-function ( fz P21 / fz R52 ) phenotype in both the eye and wing ( Figure 5 ; Table 1 ; unpublished data), suggesting that the shortened Fz2 C-tail is functionally equivalent to the Fz1 C-tail. In contrast, tub -Fz1–2-2S and tub -Fz1–2-1 did not rescue the fz − mutant phenotype ( Figure 5 F; Table 1 ), confirming again that the Fz1 7-TM region is important for Fz/PCP signaling. Although Fz2–1-1 has activity in GOF studies ( Figure 3 E; Table 1 ), tub -Fz2–1-1 did not rescue the fz − phenotype, suggesting that the specific extracellular CRD is required for normal receptor regulation ( Table 1 ; unpublished data). This could be due to a Fz1 requirement to interact with a ligand (or extracellular domain of another transmembrane protein) to provide regulation to Fz/PCP signaling. Figure 5 Rescue of the fz − Eye Phenotype with tub -Promoter-Driven Fz Chimeras Tangential eye sections with corresponding schematic in lower part of panel reflecting ommatidial polarity (respective genotypes are also marked below each panel). Black arrows, dorsal chiral form; red arrows, ventral chiral form; green arrows, symmetric ommatidia; black circles, ommatidia with missing photoreceptors. Anterior is to the left, dorsal is up, and an area around the equator is shown for each genotype. (A) Section of a wild-type eye (equator is indicated by yellow line). (B) fz P21 /fz R52 ( fz null). Note random orientation of ommatidia. (C) fz P21 /fz R52 ; tub-fz1–1-1 . The fz − phenotype is fully rescued (100% with respect to chirality; only a minor rotation wobble is rarely seen). (D) fz P21 /fz R52 ; tub-fz1–1-2 . Note partial rescue with respect to polarity (approximately 83%) and occasional photoreceptor loss representative of Wg/β-cat signaling. (E) fz P21 /fz R52 ; tub-fz1–1-2S . Note 100% rescue, identical to wild-type Fz1 (compare with [C]). (F) fz P21 /fz R52 ; tub-fz1–2-1 . No rescue due to the presence of the Fz2 7-TM region. This chimera actually shows a mild dominant negative behavior as apparent by the increased percentage of symmetric clusters (approximately 50% as compared to fz − [approximately 15%]). The tub- Fz1–1-2 receptor, which contains the Fz1 7-TM region, but is localized throughout the cellular membrane, is also able to rescue the fz − eye and wing phenotype. However, it does so less efficiently ( Figure 5 D; Table 1 ), and it also causes eye phenotypes reflecting the activation of Wg/β-cat signaling, such as photoreceptor loss (Wg/β-cat signaling during photoreceptor induction and differentiation blocks the development of these cells as photoreceptors; Wehrli and Tomlinson 1998 ). These effects of Fz1–1-2 suggest that proper apical enrichment is critical for a clean PCP readout, but that a ubiquitously distributed Fz1 chimera might be sufficiently present at apical adherens junctions to allow for partial rescue. In summary, our data are consistent with the notion that the C-tail provides the information for correct localization required for full and clean PCP signaling specificity, and that sequences within the Fz1 7-TM region and extracellular domain are required for PCP signaling activity or regulation (see also Discussion ). Flamingo Is Not Required for the Initial Apical Fz Localization Previous work has shown that Fz1 is not localized to apical junctions in the wings of fmi mutants 30–32 h after puparium formation (APF) ( Strutt 2001 ). Similar observations were made in the late third instar eye imaginal disc ( Strutt et al. 2002 ). These data suggest that Fmi is required for apical localization of Fz1 during PCP signaling. Similarly, Fmi depends on Fz1/PCP signaling to maintain its apical junctional localization in wings 30–36 h APF ( Usui et al. 1999 ), suggesting that Fmi and Fz1 localization are interdependent when PCP signaling is active . However, this might not reflect initial requirements for apical localization. To test whether Fmi is required for the initial apical localization of Fz1, which happens prior to the initiation of PCP signaling, we examined Fz1–1-1 localization in fmi E59 clones in larval wing imaginal discs. Fz1–1-1 is localized apically in fmi E59 mutant cells in third instar imaginal discs, indistinguishable from its localization in wild-type tissue ( Figure 6 ). These data suggest that Fmi is not required for the initial apical localization of Fz1–1-1. The difference between the early stage (larval discs) and late stage (pupal wings, late eye discs posterior to morphogenetic furrow during PCP signaling) suggests that initial apical localization is independent of the later maintenance evens regulated by PCP signaling (see also Discussion ). Figure 6 Subcellular Localization of Fz1–1-1 in fmi − Mutant Clones Fz1–1-1 (Myc-tagged; shown in green) is expressed with omb-Gal4 (in large parts of the third instar wing pouch). fmi E59 clones were labeled by the absence of anti-βGal staining (red). A projection of several horizontal sections in the apical region (A) and the corresponding xz-section (B) across the clone (as indicated by a white line in A) are shown. Fz1–1-1 is localized apically inside and outside the clone, indicating that initial apical Fz recruitment is independent of Fmi. Apically Localized Fz1/2 Chimeras Act As Dominant Negatives for Wnt/β-Cat Signaling During imaginal disc development and patterning, Wg binds to the Fz2 receptor at basolateral membranes of the wing epithelium ( Strigini and Cohen 2000 ). This result suggests that canonical Wnt signaling occurs mainly at the basolateral side of the epithelium in imaginal discs. In contrast, apically localized Fz appears to have high PCP signaling activity (as described above). These results suggest that PCP signaling and canonical Wnt/βcat signaling occur in different subcellular locations or membrane compartments. Previous work has suggested that Fz2–1-1 and Fz2–2-1, which are shown here as localized to apical junction complexes (see Figure 2 D and 2 F), act as dominant negative isoforms for canonical Wg signaling ( Boutros et al. 2000 ). We have noticed that expression of the Fz chimeras (with en-Gal4 in the posterior wing compartment) often causes wing notching and loss of wing margin bristles (in the posterior wing region; Figure 7 ), indicative of reduced Wnt/βcat signaling ( Couso et al. 1994 ). To gain insight into why chimeric Fz receptors can behave as dominant negatives, we analyzed ubiquitous Dsh-GFP localization (expressed from the endogenous promoter; Axelrod 2001 ) in en-Gal4- and dpp-Gal4- driven UAS-fz wing discs ( Figure 7 ; unpublished data). In wild-type, Dsh-GFP is mainly cytoplasmic with a mild, slightly stronger apical enrichment at membranes ( Figure 7 E– 7 H, anterior compartments). In wing epithelia with overexpressed Fz1–1-1 or Fz2–1-1, much more Dsh-GFP is recruited apically in cells expressing the Fz chimeras ( Figure 7 E– 7 H, posterior compartments). At the same time, Dsh-GFP levels are reduced in basolateral regions of these cells. These data suggest that Fz in adherens junctions (apical) is trapping Dsh there, depleting it away from Wnt/βcat signaling components located possibly more basally and thus reducing canonical Wnt signaling. Figure 7 Overexpression of Apically Localizing Fz1/2 Chimeras Has an Inhibitory Effect on Canonical Wnt Signaling (A–D) show adult wings of the respective genotypes. Anterior is up and distal to the right. (A) Adult wing of an en-Gal4/+; UAS-fz1–1-1/+ fly ( en>fz1–1-1 ). en-Gal4 drives UAS reporter genes only in the posterior compartment. Inset shows high magnification of region marked by arrowhead. Some wing margin bristles are missing (arrow) in the posterior compartment. The border between anterior (“a”) and posterior (“p”) compartments is marked with black line. (B) dsh V26 /+; en>fz1–1-1 adult wing. Note enhancement of the margin bristle phenotype: all margin bristles are missing from the area between the arrows in the posterior compartment. (C) en>fz2–1-1 wing. Most of the wing margin bristles are missing in the posterior compartment. Note also that the posterior compartment is smaller. (D) en>fz2–2-1 wing. Again the posterior compartment is smaller and most of the margin is missing. (E–G) show that Fz1–1-1 expression increases apical localization of Dsh-GFP and reduces Dsh-GFP in more basolateral areas of wing cells. (E) and (F) are xy-horizontal optical sections, and (G) is an xz-cross section. The positions of (E) and (F) sections are indicated in (G). (E) Apical xy-optical section of a third instar wing disc. Fz1–1-1 (red) is overexpressed by en-Gal4 in the posterior compartment (anterior–posterior border is labeled by white line, and the corresponding compartments are labeled “a” and “p,” respectively). Dsh-GFP (green) accumulates at higher levels apically in the posterior compartment. Single-channel Dsh-GFP staining is shown at right. In wild-type disc, Dsh-GFP is evenly distributed with no anterior–posterior bias (not shown). (F) A more basal xy-section of the same disc as in (E). Note reduction of Dsh-GFP staining in the posterior compartment, except at the apical junctions as seen in folds (arrowhead). In the anterior compartment, where Fz1–1-1 is not overexpressed, Dsh-GFP is only slightly enriched in the apical folds (arrow). (G) xz-section of the same wing disc shown in (E) and (F), with top panel showing double labeling for anti-Myc (red) and anti-Dsh-GFP (green) and bottom panel showing single channel of Dsh-GFP staining. (H) xz-section of a comparable disc expressing Fz2–1-1 in the posterior compartment . Fz2–1-1 overexpression (red) also causes accumulation of Dsh-GFP in apical junctions and reduction of Dsh-GFP along the basolateral region. To test this hypothesis, we analyzed the effect of reducing dsh gene dosage in en-Gal4/UAS-fz1–1-1 flies, where wing notching and loss of marginal hairs is mild (21% of wings have large areas of margin bristles missing; Figure 7 A; Table 2 ). Strikingly, the en-Gal4/UAS-fz1–1-1 effect is enhanced in dsh heterozygous flies ( dsh V26 /+), with 65% of wings showing large areas of margin bristles missing and severe wing notching ( Table 2 ; see Figure 7 B for example). To corroborate the dsh dosage sensitivity in this context, we generated flies with three copies of dsh (by introducing an additional dsh copy as a dsh-GFP transgene expressed under its endogenous promoter; Axelrod 2001 ). In this genetic background with three dsh copies, only 4% of the en-Gal4/UAS-fz1–1-1 wings displayed a large area of missing wing margin bristles ( Table 2 ), suggesting that the presence of extra Dsh suppresses the en-Gal4/UAS-fz1–1-1 wing phenotype. Taken together, these Dsh dosage effects support the idea that trapping Dsh into apical junctional complexes reduces its availability for Wnt/βcat signaling, and thus reduces the strength of canonical signaling. Table 2 Wing Margin Phenotypes of en-Gal4; UAS-fz1/2 Chimeras Note dsh dosage sensitivity of the Fz1–1-1-induced wing nick frequency and size by removal of one copy of dsh, and the correlation of a dominant negative effect with apical localization (e.g., presence of Fz1 or Fz2S C-tails; compare also to Table 1 ). “Large nick” is defined as an area of the wing lacking more than 20 margin bristles In further support of this explanation, en-Gal4/+; UAS-fz1–1-2/+ flies show only a very mild effect on wing margin bristles ( Table 2 ). As Fz1–1-2 is ubiquitously localized along the apical–basal axis, recruiting of Dsh by such chimeras should not have an adverse effect on canonical Wg signaling. In contrast, when Fz2–1-1 and Fz2–2-1 are expressed (with en-Gal4 ) we observe very strong wing notching effects and a general reduction of the posterior wing compartment ( Figure 7 C and 7 D; Table 2 ). This can be explained as follows. As the Fz2 ligand-binding CRD has a much higher affinity for Wg than the Fz1 CRD ( Rulifson et al. 2000 ), the strong dominant negative behavior of Fz2–1-1 and Fz2–2-1 can be explained by adverse effects on both Dsh and Wg: Fz2–1-1 and Fz2–2-1 have a high-affinity Wg-binding CRD (sequestering Wg efficiently) and can trap Dsh at junctional complexes as well ( Figure 7 E– 7 H), making large pools of Wg and Dsh unavailable for canonical signaling, and thus causing a strong dominant negative effect. In summary, the dominant negative effect of the overexpression of Fz1–1-1, Fz2–1-1, and Fz2–2-1 is caused by trapping Dsh into apical junctions, making it unavailable for canonical Wnt/βcat signaling, and, when present, the Fz2 CRD enhances this effect by also sequestering Wg to these complexes. These results suggest that a Fz-Dsh complex in the apical junctions is largely incapable of canonical β-cat signaling, suggesting that the subcellular localization of Fz receptors contributes significantly to the signaling outcome and specificity (see Discussion ). Discussion We have shown that Fz1 and Fz2 have different subcellular localizations within the wing imaginal epithelium. This difference is mediated by sequences in the cytoplasmic tail of Fz2 that appear to block apical accumulation. The subcellular localization difference directly contributes to the signaling specificity outcome. Whereas apical localization favors Fz/PCP signaling, it interferes with canonical Wnt/β-cat signaling. The Relationship between Apical Localization of Fz1 and Its PCP Signaling Activity Is the apical localization of Fz required for PCP signaling? The Fz1–1-2 chimera, which is distributed ubiquitously within the apical–basolateral membrane, only partially rescues the fz − eye phenotype, and it can also cause defects related to canonical Wg/Arm signaling (see Figure 5 D). In contrast, apically localized Fz1–1-2S fully rescues the fz − phenotype and has no additional effects. The Fz1–1-2 chimera also shows much weaker PCP phenotypes in the GOF assay (see Figure 3 and Boutros et al. [2000] ). Taken together, these results suggest that a reduction in the apical localization of Fz leads to a reduction in PCP signaling activity. However, about 80% of the chirality defects in fz − eyes are rescued by tub-fz1–1-2 , and in the wing tub-fz1–1-2 rescues the fz − mutant to a similar extent as tub-fz1–1-1 and tub-fz1–1-2S (unpublished data), suggesting that Fz1–1-2 contains substantial PCP signaling activity. Because both GOF and loss-of-function studies indicate that the Fz1 7-TM region is critical for Fz1 function, Fz1–1-2 is expected to have Fz/PCP signaling activity, although with altered subcellular distribution. Thus, the remaining PCP signaling activity of Fz1–1-2 seen is probably due to the presence of some of this protein in apical regions. It is difficult to determine how much of Fz1–1-2 is actually localized to this membrane region. Since the immunohistochemical staining indicates that it is not excluded apically, we assume that Fz1–1-2 has enough apical localization to participate when PCP signaling is initiated. It has been suggested that wing cell orientation does not depend on absolute Fz levels, but instead depends on relative Fz/PCP activity differences in a Fz activity gradient across a field ( Adler et al. 1997 ). Thus, although the absolute activity of Fz1–1-2 is reduced (based on weaker GOF phenotypes and weaker rescue of fz − in the eye), the relative difference might be sufficient for the partial rescue. In this context, it is worth noting that tub-fz1–1-2 rescues the fz − phenotype better in the wing than in the eye, whereas there is no apparent difference in rescue activity between the eye and the wing for tub-fz1–1-1 or tub-fz1–1-2S . The difference could be due to the observed nonautonomous PCP signaling effects in the wing ( Vinson and Adler 1987 ), where neighboring cells affect each other's planar polarization. Fz1–1-2 may allow some wing cells to adopt the correct orientation, which then in turn influences many of the remaining wing cells to also orient themselves correctly through nonautonomous interactions. Regulation of Fz Apical Localization It has been shown that Fz1 localization is affected in fmi mutant clones at about 30 h APF ( Strutt 2001 ), leading to the proposal that Fmi recruits Fz1 into apical junctions ( Strutt 2001 ; Bastock et al. 2003 ). However, we find that Fz1 is localized normally in fmi null mutant clones earlier in the third instar wing disc. What causes the difference between these two observations? PCP signaling in the wing is thought to act in two phases (one 6–24 h APF and the second 24–32 h APF [ Strutt and Strutt 2002 ]), and it results in the distal enrichment and maintenance of Fz1 ( Strutt 2001 ). As Fz1/PCP signaling is modulated by Fmi ( Usui et al. 1999 ), Fmi-dependent changes in Fz1 localization likely result from effects on PCP signaling activity. At the same time, Fmi localization is also dependent on Fz1 activity and becomes also less apically localized in fz − tissue at 30–36 h APF ( Usui et al. 1999 ), suggesting that the regulation of apical localization between Fz1 and Fmi is complicated and mutual at these late stages. We showed here that initial apical localization of Fz1, preceding both stages of PCP signaling, is not fmi dependent. This result suggests that Fmi and Fz1 get recruited to apical junctions independently. During later stages, Fmi and Fz1 then affect each other's localization through PCP signaling. At this point, it remains unclear which molecules initially recruit Fz1 into the apical junctional region. Fz Receptor Localization and Canonical Wnt Signaling Secreted Wg mainly binds to Fz2 at basolateral membrane regions of the wing epithelium ( Strigini and Cohen 2000 ), indirectly suggesting that canonical signaling occurs in the basolateral membrane compartment. Our experiments show that overexpression of Fz1–1-1 or Fz2–1-1 leads to a cell-autonomous loss of wing margin bristles and associated tissue, suggesting that these molecules act like dominant negatives, inhibiting Wnt/β-cat signaling. As these molecules are enriched apically and sequester Dsh there, Fz-Dsh complexes at apical junctions may be largely inactive for canonical Wnt signaling. This result suggests that canonical Wnt signaling and PCP signaling occur in different subcellular compartments. Basolateral Wnt/β-cat signaling is also suggested by the fact that (1) secreted Wg binds to Fz2 at the basolateral membrane and that (2) apical Wg secretion and signaling could lead to mis-specification in disc folds and cells in the peripodial membrane ( Strigini and Cohen 2000 ). Both Fz1 and Fz2 are capable of canonical Wnt/β-cat signaling ( Bhat 1998 ; Kennerdell and Carthew 1998 ; Bhanot et al. 1999 ; Chen and Struhl 1999 ). Consistently, different Fz1/2 chimeras, including related versions of Fz2–1-1 and Fz2–2-1, are capable of rescuing the fz , fz2 double mutant phenotype ( Strapps and Tomlinson 2001 ). However, when Fz1–1-1, Fz2–1-1, or Fz2–2-1 is expressed at high levels, Dsh accumulates at apical junctions, thus decreasing cytosolic Dsh levels. As the chimeric receptors can rescue the fz, fz2 double mutant when expressed at low levels (under the control of the tub promoter; Strapps and Tomlinson 2001 ), the relative level of each receptor together with its subcellular localization appear critical for the signaling outcome. In conclusion, we have shown that subcellular localization contributes to Fz signaling specificity. Our data indicate that the localization of Fz1 at apical junctions promotes Fz/PCP signaling, whereas this localization can inhibit canonical Wnt/β-cat signaling. The localization is mediated through sequences in the C-tail. Materials and Methods Flies and constructs The flies carrying the chimeric receptor constructs UAS-Fz1–1-1, UAS-Fz1–1-2, UAS-Fz1–2-2, UAS-Fz2–1-1, and UAS-Fz2–2-1 are described in Boutros et al. (2002). UAS-Fz1–2-1 was constructed by combining the Fz1 CRD (Fz1 residues 1–166) with the Myc tag, the Fz2 7-TM region (amino acids 220–617), and the Fz1 C-tail (amino acids 558–585). A HindIII site (generated in vitro) was used to combine Fz1 CRD with the Fz27-TM region. An XhoI site was used to link the Fz2 7-TM region with the Fz1 C-tail. C-tail mutation constructs of Fz1 were generated through PCR-based site-directed mutagenesis (Quikchange kit, Stratagene, La Jolla, California, United States). Fz1–1-2S and Fz1–2-2S were generated by introducing a stop codon after residue L633 of Fz2. Fz1–1-1C2 was generated by introducing a BsiWI site at the residues R574–T575 of Fz1 and R626–T627 of Fz2 (the RT residues remain the same by this mutagenesis). This added the Fz2 amino acids 627–694 to the Fz1 C-tail at the RT residues. The respective UAS transgenic flies were generated by standard procedures. The Gal4/UAS system was used to express the chimeric UAS-Fz1/2 transgenes ( Brand and Perrimon 1993 ) with dpp-Gal4, en-Gal4, or omb-Gal4 ( Brand and Perrimon 1993 ; Yoffe et al. 1995 ; Lecuit et al. 1996 ; Morimura et al. 1996 ). tub -promoter-driven Fz chimeric constructs were generated by cloning the respective Fz1/2 constructs into the Casper4-tub vector (containing a 2.4-kb tub promoter fragment in Casper4— a kind gift from Stephen Cohen). fz P21 and fz R52 are null alleles of fz ( Jones et al. 1996 ). dsh V26 is a null allele of dsh ( Perrimon and Mahowald 1987 ). Immunohistochemistry Rat anti-DE-Cad was used at 1:200 ( Oda et al. 1994 ). Mouse anti-Myc (9E10) was used at 1:250–500 (Santa Cruz Biotechnology, Santa Cruz, California, United States). Rabbit anti-Dlg was used at 1:3500 ( Lee et al. 2003 ). Rabbit anti-GFP (Molecular Probes, Eugene, Oregon, United States) was used at 1:4000 to detect Dsh-GFP and Fz1-GFP. fmi E59 clones were induced in first instar larvae via the Flp/FRT system in the w, hs-flp; FRT42B fmi E59 /FRT42 arm-lacZ genotype . Larvae were dissected 4 d after clone induction during late third instar. fmi E59 is a null allele of fmi ( Usui et al. 1999 ). Adult wing and eye preparation Wings were soaked (with agitation) in 0.1% Triton X-100 PBS for about 30 min or longer, and then mounted in 80% Glycerol PBS. Eye embedding and sectioning was performed as described by Tomlinson (1987) . | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC449784.xml |
548303 | Pretreatment with ACE inhibitors improves acute outcome of electrical cardioversion in patients with persistent atrial fibrillation | Background Persistent atrial fibrillation (AF) is difficult to treat. In the absence of class I or III antiarrhythmic drugs sinus rhythm is maintained in only 30% of patients during the first year after electrical cardioversion (ECV). One of the remodeling processes induced by AF is fibrosis, which relates to inducibility and maintenance of AF. The renin-angiotensin system may play a important role in this. The aim of this study was to investigate the role of angiotensin-converting enzyme (ACE) inhibitor use on efficacy of ECV, and occurrence of subacute recurrences. Methods One hundred-seven consecutive patients with persistent AF underwent ECV. In twenty-eight (26%) patients ACE inhibitors had been started before initiation of the present episode of AF ('pre-treated' patients). Results ECV was successful in 96% of patients who were on ACE inhibitors before start of the present episode of AF compared to 80% of the patients not pre-treated (p = 0.04). After 1 month of follow-up 49% of the pre-treated patients and 50% of those not pre-treated with ACE inhibition were still in sinus rhythm (p=ns). Multivariate analysis showed that pre-treatment with ACE inhibitors and a smaller left atrial size were independent predictors of successful ECV (OR = 5.8, C.I. 1.3–26.1, and OR = 5.6, C.I. 1.2–25.3, respectively). Conclusions Pre-treatment with ACE inhibitors may improve acute success of ECV but does not prevend AF recurrences. | Background Persistent atrial fibrillation (AF) is difficult to treat. In the absence of class I or III antiarrhythmic drugs sinus rhythm is maintained in only 30–50% of patients during the first year after Direct Current electrical cardioversion (ECV)[ 1 , 2 ]. Furthermore, even following an aggressive approach with repeated ECVs and use of prophylactic drugs, arrhythmia-free outcome is still poor: only 39% of patients maintain sinus rhythm during two years of follow-up[ 1 , 2 ]. Notwithstanding the recent results of AFFIRM and RACE showing no beneficial effect of rhythm control over rate control a rhythm control strategy may be indicated in severely symptomatic patients and those with a tachycardiomyopathy[ 3 ]. In recent years research has focused on the atrial remodeling processes that are induced by AF itself and that trigger the arrhythmia to become sustained: "AF begets AF"[ 4 ]. One of the remodeling processes induced by AF is fibrosis. Fibrosis causes dispersion of conduction, which, in its turn is related to inducibility of AF[ 5 ]. The renin-angiotensin system seems to play an important role in the development of fibrosis in heart failure. It was shown that pre-treatment with enalapril may attenuate atrial fibrosis and conduction abnormalities in a canine model of heart failure, and the occurrence of AF in patients with left ventricular dysfunction [ 5 - 7 ]. A recent experimental study showed that angiotensin II blockers may prevent electrical remodeling when started before start of AF[ 8 ]. In the present study we report on the effects of ACE inhibition on the outcome of ECV and the prevention of early recurrences after ECV of persistent AF. Methods One hundred-seven consecutive patients with persistent AF, defined as the presence of AF for at least 24 hours were included in this study[ 9 ]. ECV was performed according to a previously described step up protocol[ 10 ]. Successful ECV was defined as the presence of sinus rhythm for at least 4 hours after ECV. No difference was made between unsuccessful ECV due to shock failure or due to an immediate recurrence of AF (within 2 minutes after successful ECV). ACE pre-treatment was defined as use of ACE inhibitors before onset of the current AF episode. Most patients on ACE inhibitors used these drugs for hypertension or congestive heart failure. To make sure that all patients were not completely remodeled at the very moment of start of the current episode of AF (and thereby verifying the fact that they were treated with ACE inhibitors before the process of electrical remodeling started), only patients with at least 1 month sinus rhythm before the current episode of AF were included in this study. Duration of AF was determined as precisely as possible by previous electrocardiograms, 24-hours Holter registrations, and by the patient's history. None of the patients were on class I or III antiarrhythmic drugs neither at the moment of ECV nor during follow-up. Statistical analysis Quantitative variables were compared between groups using a two-tailed t-test for normally distributed variables or a Wilcoxon two-sample test for skewed distributed variables. For qualitative variables (categorical or ordered), group differences were evaluated using a Fisher's exact test or a Chi-square test. Accordingly, baseline characteristics are given in mean ± SD, median and range (min-max) or percentages. To determine the predictive factors for successful ECV, an univariate logistic regression analysis was performed using the relevant baseline predictors. Variables with a p-value < 0.20 were selected for the multiple logistic regression analysis to derive a model with statistically significant predictors, by using a backward selection method. All p-values are two-sided and a p-value of < 0.05 was considered statistically significant. SAS version 6.12 (Cary, NC) was used for all statistical evaluations. Results Baseline characteristics are listed in table 1 . Twenty-eight patients (26%) were treated with ACE inhibitors before start of the current episode of AF (pre-treated patients). In the latter group, ECV was successful in 96% while in the group of patients not pre-treated with ACE inhibitors before start of the current AF episode only 80% had a successful ECV (p = 0.04). Table 1 Baseline characteristics All Successful ECV Unsuccessful ECV P N = 107 N = 90 N = 17 Duration in days, median (range) current episode 115(71–175) 113 (61–175) 126 (81–189) ns total AF duration 144 (95–232) 142 (86–213) 160 (102–340) Ns Number of previous ECVs, N(range) 1 (0–5) 1 (0–5) 1(0–3) Ns Underlying heart disease * (%) Left ventricular dysfunction 21% 25% 0% 0.02 Hypertension 30% 32% 17% ns Valve disease 19% 19% 22% ns Coronary artery disease 22% 21% 28% ns Lone AF 16% 16% 15% ns Echocardiography (mm ± SD) LA parasternal long axis 47 ± 8 47 ± 9 47 ± 8 ns LA 4 chamber view long axis 67 ± 9 67 ± 10 67 ± 6 ns LA 4 chamber view short axis 46 ± 8 46 ± 9 43 ± 5 ns RA 4 chamber view long axis 62 ± 8 62 ± 8 63 ± 7 ns LVEDD 50 ± 8 50 ± 8 50 ± 7 ns LVESD 35 ± 10 35 ± 10 33 ± 9 ns Medication Beta blocker 49% 50% 48% ns Calcium channel blocker 36% 36% 37% ns Digoxin 47% 46% 49% ns ACE pretreatment 26% 30% 6% 0.04 Diuretics 31% 31% 29% ns Angiotensin receptor blocker 1% 1% 0% ns * more then 1 entity per patient AF = atrial fibrillation, ECV = electrical cardioversion, LA = left atrium, RA = right atrium, LVEDD = left ventricular end diastolic diameter, LVESD = left ventricular end systolic diameter In general, patients treated with ACE inhibitors before start of the current episode of AF had more evidence for heart disease than those who were not pre-treated. Prevalence of hypertension (46% versus 24%, respectively, p = 0.03), and left ventricular dysfunction (40% versus 14%, respectively, p = 0.01) was significantly higher in the pre-treated group in comparison to those who were not pre-treated with ACE inhibitors. (Table 2 ) Furthermore, there was a trend towards a higher prevalence of coronary artery disease in the pre-treated group compared to the not pre-treated group (36% versus 18%, respectively p = 0.07). Table 2 Baseline characteristics divided in ACE pre-treatment and no ACE pre-treatment. No ACE pre-treatment ACE-pre-treatment P N = 79 N = 28 Duration in days, median (range) current episode 121 (74–178) 96 (59–135) ns total AF duration 161 (105–247) 117 (63–160) ns Number of previous ECVs, N(range) 1 (0–5) 1(0–3) ns Underlying heart disease * (%) Left ventricular dysfunction 14% 40% 0.01 Hypertension 24% 46% 0.03 Valve disease 19% 21% ns Coronary artery disease 18% 36% 0.07 Lone AF 17% 14% ns Echocardiography (mm ± SD) LA parasternal long axis 46 ± 7 50 ± 10 0.03 LA 4 chamber view long axis 66 ± 9 67 ± 12 ns LA 4 chamber view short axis 45 ± 8 49 ± 10 ns RA 4 chamber view long axis 61 ± 8 64 ± 7 ns LVEDD 49 ± 7 54 ± 8 0.01 LVESD 34 ± 9 38 ± 12 ns Medication Beta blocker 51% 47% ns Calcium channel blocker 35% 39% ns Digoxin 45% 51% ns Diuretics 29% 34% ns Angiotensin receptor blocker 1% 0% ns * more then 1 entity per patient AF = atrial fibrillation, ECV = electrical cardioversion, LA = left atrium, RA = right atrium, LVEDD = left ventricular end diastolic diameter, LVESD = left ventricular end systolic diameter An additional 21 patients received ACE inhibitors after start of the current episode of AF because of either insufficiently treated hypertension or left ventricular dysfunction newly documented with echocardiography. Success of ECV in these patients was comparable to patients who were not treated with ACE inhibitors at all (80% and 82%, respectively). Use of beta adrenergic receptor blockers, calcium channel blockers or digoxin, started before the present episode of AF or not, did not influence success of ECV. After 1 month of follow-up 49% of the pre-treated patients compared to 50% of those who were not pre-treated with ACE inhibition were still in sinus rhythm (Table 3 , Figure 1 ). Table 3 All No ACE-pretreatment ACE-pretreatment P Successful ECV (%) 84% 80% 96% 0.04 SR at 1 month follow-up (%) 50% 50% 49% ns ECV = electrical cardioversion, SR = sinus rhythm Figure 1 Effect of pretreatment with ACE-inhibitors before start of AF. Multivariate analysis showed that pre-treatment with ACE inhibitors and a smaller left atrial size were independent predictors of successful ECV (OR = 5.8, C.I. 1.3–26.1 and OR = 5.6, C.I. 1.2–25.3, respectively). Discussion Main results This post-hoc retrospective analysis shows that use of ACE inhibitors before the onset of AF enhances acute ECV outcome but it does not improve maintenance of sinus rhythm. Furthermore, when ACE inhibitiors are instituted later, i.e. after the start of AF, it does no longer improve success of ECV. Effect of ACE-inhibition on structural remodeling ACE inhibitors reduce the incidence of AF in patients with left ventricular dysfunction[ 7 , 6 ]. This may be related to the protective effects of ACE inhibitors, which help to maintain atrial integrity and attenuate fibrosis[ 5 ]. In the present study pre-treatment with an oweveHowACE inhibitor (i.e. use of ACE inhibitors before onset of the arrhythmia) improved acute outcome of ECV. In view of the above this suggests that ACE inhibitors may prevent or diminish AF induced structural remodeling. These clinical findings are compatible with experimental findings showing that ACE-inhibition could attenuate heart failure induced atrial functional remodeling and fibrosis in dogs[ 11 ]. In atrial tissue from AF patients an ACE-dependent increase of activated extracellular signal-regulated kinase (Erk) type 1 and 2 was found, which may contribute to the development of atrial fibrosis during AF[ 12 ]. Effect of ACE-inhibition on electrical remodeling In 1995 it was shown that AF induces several electrophysiological changes, called electrical remodeling[ 4 ]. Experimental data show that ACE inhibition prevents short-term (< 2 hours) tachycardia-induced atrial electrical remodeling[ 8 , 13 ]. However, enalapril could not attenuate or prevent the long-term (7 days) effects of tachycardia on remodeling[ 8 ]. Several studies have investigated the role of calcium channel blockers on electrical remodeling. In these studies it was also shown that although there was a short-term prevention of electrical remodeling, calcium channel blockers did not have a long-term protective effect on electrical remodeling [ 14 - 16 ]. Effect on AF recurrences Madrid et al. investigated the role of the angiotensin II type 1 receptor antagonist irbesartan in amiodarone treated patients[ 17 ]. Persistent AF patients were randomized to either treatment with amiodarone alone or treatment with amiodarone in combination with irbesartan. Drug treatment was started at least 3 weeks before cardioversion but after the start of AF. No differences were found in electrical cardioversion outcome between the two treatment groups which is in contrast to the present study. This may relate to the fact that all patients were in AF at the very moment of start of irbesartan. Two months after cardioversion it appeared that patients treated with irbesartan and amiodarone had a significantly lower AF recurrence rate compared to amiodarone alone treated patients, 15% versus 37%, respectively (p = 0.007). This difference was maintained during a median follow-up of 254 days. According to their figure 2, the benefit of irbesartan (in combination with amiodarone) was mainly achieved by reduction of recurrences after the first two weeks after cardioversion. An earlier study on the role of ACE inhibitors in patients with heart failure and AF showed a trend towards more patients maintaining sinus rhythm after ECV when instituted on lisinopril in comparison to patients not treated with lisinopril[ 18 ]. In the present study no difference in recurrence rate could be found between patients with and without ACE-pretreatment. Limitations of the study This was a non-randomized and post-hoc analysis. This implies that all findings can only be used to generate hypotheses. However, this is the first clinical study showing that ACE-inhibition initiated before start of AF enhances direct ECV outcome. Only a very small amount of patients was on angiotensin receptor blockers. Whether the effect of pretreatment with angiotensin receptor blockers on cardioversion outcome would be the same as the effect of pretreatment with ACE inhibitors could not be investigated in this study. However, in this context the results of the study of Madrid et al. are encouraging [ 17 ]. Conclusions Pretreatment with ACE-inhibitors significantly improves acute outcome of ECV when initiated before the onset of AF but it does not lead to better maintenance of sinus rhythm. When ACE inhibitiors are, however, instituted later, i.e. after the start of AF, it does no longer improve success of ECV. Competing interests The author(s) declare that they have no competing interests. Authors' contributions TVN carried out the study and drafted the manuscript, HC designed the study, and interpreted the data, MB participated in the draft of the manuscript and interpreted the data, DVV participated in the draft of the manuscript and interpreted the data, IVG participated in the design of the study and interpreted the data. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548303.xml |
539288 | Threshold intensity factors as lower boundaries for crack propagation in ceramics | Background Slow crack growth can be described in a v (crack velocity) versus K I (stress intensity factor) diagram. Slow crack growth in ceramics is attributed to corrosion assisted stress at the crack tip or at any pre-existing defect in the ceramic. The combined effect of high stresses at the crack tip and the presence of water or body fluid molecules (reducing surface energy at the crack tip) induces crack propagation, which eventually may result in fatigue. The presence of a threshold in the stress intensity factor, below which no crack propagation occurs, has been the subject of important research in the last years. The higher this threshold, the higher the reliability of the ceramic, and consequently the longer its lifetime. Methods We utilize the Irwin K-field displacement relation to deduce crack tip stress intensity factors from the near crack tip profile. Cracks are initiated by indentation impressions. The threshold stress intensity factor is determined as the time limit of the tip stress intensity when the residual stresses have (nearly) disappeared. Results We determined the threshold stress intensity factors for most of the all ceramic materials presently important for dental restorations in Europe. Of special significance is the finding that alumina ceramic has a threshold limit nearly identical with that of zirconia. Conclusion The intention of the present paper is to stress the point that the threshold stress intensity factor represents a more intrinsic property for a given ceramic material than the widely used toughness (bend strength or fracture toughness), which refers only to fast crack growth. Considering two ceramics with identical threshold limits, although with different critical stress intensity limits, means that both ceramics have identical starting points for slow crack growth. Fast catastrophic crack growth leading to spontaneous fatigue, however, is different. This growth starts later in those ceramic materials that have larger critical stress intensity factors. | Background Slow crack growth is most suitably described in a v (crack velocity) versus K I (stress intensity factor) diagram. Slow crack growth in ceramics is attributed to corrosion assisted stress at crack tips or at any defect pre-existing in the ceramic [ 1 ]. The combined presence of body fluid molecules (mainly water), which reduce the surface energy at the crack tip, and the presence of high stresses are the reasons for subcritical crack growth (SCCG) in ceramics. The presence of stress intensities above a critical value (K I > K Ic ) initiates fast catastrophic crack growth, followed by the deterioration of a dental or a body restoration machined from ceramics. The presence of stress intensities above a threshold value (K I > K I0 ) initiates SCCG in ceramics, followed by a slow, however continuous, erosion of the strength of a restoration which also may result in final fatigue. In an early stage of ceramic research it was believed that this lower limit for SCCG is very close to zero. In the mean time, however, one has learned that for most ceramic materials the lower limit for SCCG is significantly larger than zero. Indeed, it may even be just below K Ic . The threshold limit K I0 corresponds to a crack equilibrium at null crack velocity. Therefore, it allows a safety range of clinical use. The higher the value of K I0 , the higher the reliability, and hence the lifetime of a restoration. Bio-components should be designed to work in a region of the v-K I - diagram where the upper border line of that region corresponds to the threshold limit. In the present paper we preferentially focus on those ceramics that are important in dental research. Note, however, that alumina and zirconia have meaning in both fields of application (dentistry and medicine). We use soda lime glass as a well characterized standard and silicon nitride as important in the general field of ceramics. There are several methods available and in the literature extensively described how the threshold limit can be measured. The feasibility of these measurement procedures is mostly demonstrated with the help of soda lime glass as a brittle solid model. In principle, the proper test for existence of a threshold lies in the observation of reversibility of crack growth. The threshold can be regarded as a Griffith quiescent point, where forward and backward fluctuations just balance, i.e., the mean velocity of the crack tip becomes zero. The forward and backward fluctuations take place over discrete energy barriers definable as G = W = 2 γ , where G is the energy release rate, W is the Dupré work of adhesion, and γ is the surface energy. If G < W the crack should retract and heal; otherwise it should repropagate [ 2 ]. On the basis of this assumption, the authors in [ 2 ] (see also [ 3 ]) calculate equations prescribing the v G characteristics (crack velocity versus mechanical energy release rate; equivalent to v - K I crack velocity versus stress intensity factor) at specified chemical concentrations and temperatures, which can describe observed v-G dependencies. One common experimental method to determine the threshold limit is to measure slow crack growth rate down to velocities as low as 10 -14 m/s. Then one can extrapolate from the vertical branch of the function to the zero velocity limit on the stress intensity factor axis K I , with the intersection K I equal to K I0 [ 4 - 7 ]. Another method to determine the aforesaid threshold limit is the "interrupted static fatigue test" (ISF-test) [ 8 ]. For a bending experiment, the applied stress is chosen such that a significant fraction of samples fails in a "hold period". Samples that do not fail during this static phase are then fractured by the usual four point bending technique. The threshold is calculated either from the applied stress intensity factor at which 50% of samples fail during the stress hold, or by using the factor applied to the weakest specimen during the stress hold as calculated for various hold times. Once the value of the stress intensity factor becomes independent of hold time, it is equivalent to the threshold [ 9 ]. Another method uses a side grooved specimen with a crack propagating along its length, and under a bending condition similar to four point bending. The crack velocity can be obtained from the rate of load relaxation at constant displacement and the initial crack length. Having established the v - K diagram, the threshold is determined as described above. For further details refer to [ 10 ]. Other methods may be characterized by the phrase "decay of residual stress" [ 11 ]. Here, the threshold limit can be calculated from the residual stress factor attributed to this decay of residual stress. The current method of measurement used, however, is based on indentation cracking, analogous to other studies also utilizing flaw initiation for starting the test [ 11 - 13 ]. After this start, however, the subsequent procedure is different. A follow up of the decay of residual stress intensities near the crack tip is done over a period of about one year, determining K tip via the COD for different times after indentation [ 14 ]. Methods Using a micro-hardness testing machine, a Vickers indentation is made on the carefully polished surface of a sample of the ceramic to be investigated. Radial cracks emanate from each of the four indentated corner sources. To determine the stress intensity present at the crack tip due to the indentation, the near crack tip profile is determined using a scanning microscope (ESEM: "Environmental Scanning Electron Microscope"). A specific feature of this technique is that it is carried out at a moderate vacuum (p ≈ 10 -1 mbar). Hence, there is no longer need to sputter the samples with a gold or carbon layer. Our initial attempts to measure the crack opening displacement (COD) showed that sputtering resulted in blurring the crack banks or even partly hiding the crack. Thus, we abandoned those attempts and started again when the ESEM was available. Before the availability of the ESEM it was nearly impossible to precisely measure crack profiles at submicrometer resolution which, however, is mandatory. Images of the crack profiles (Fig. 1 ) were digitally stored and analyzed by imaging software (Paint Shop Pro, V. 6, Jasc Software, Eden Prairie, Maine, USA). Figure 1 Example of a Vickers indentation. Only one of four corners is shown (length of diagonal 115 μ m). With the crack tip as a starting point ( x = 0) the crack width 2*u(x) is measured at the distance x (COD after Irwin [15]; ceramic material for this example: Empress 1). The residual tensions cause crack growth over a long time interval until, at the end of the crack, K tip is equal to K I0 . Crack tip shielding by secondary effects (micro structural elements which toughen material as the crack extends) may slightly distort results (measured K I0 then lower than true K I0 ). Insert: idealized COD. The measured profiles can be attributed to the crack opening displacement (COD) near the crack tip [ 14 ]. The near crack tip profiles for stress-free crack surfaces are usually represented by the Irwin K-field displacement relation [ 15 ], with 2u being the total COD, x the distance from the crack tip, and the plane strain Young's modulus E' = E /(1- v 2 );( v = Poisson's constant) being. We assume that there is no crack shielding. Then, in equilibrium, the currently acting crack tip stress intensity factor K tip is balanced by the toughness of the material K Ic (mode I loading [ 15 ]): and by re-arrangement: If data are taken sufficiently close to the crack tip ( x ≤ 20 μ m), a linear relationship is experimentally observed between u(x) 2 and x . K tip can then be calculated from a regression analysis as the slope of a straight line, provided E' is known (see below). The residual stresses close to the crack tip initiated by the indentation impression gradually decay over time t , and one anticipates that they slowly fade away eventually approaching zero. Hence K tip = K tip (t) and it is plausible to assume K tip ( t →∞) ≈ K I0 . Therefore, in the present work, because of slow crack growth, we take the threshold value of the stress intensity factor as the time limit of the slowly decreasing K tip value. Provided that a suitable high resolution scanning microscope is at hand, there is no need of sputtering the samples, and the presently utilized method is very simple. A potential shortcoming, however, is that this method may need many months or even years until the residual stresses are relaxed and the threshold value is reached. The authors concede that they have chosen to consider a somewhat ideal situation since the assumption K tip ( t →∞) = K I0 assumes ideal behavior. In real ceramics, especially polycrystalline and composite materials, the crack tip may be shielded from residual load by micro structural elements, which toughen the material in the region just before the crack tip [ 2 ]. This behavior is reminiscent to R-curve behavior. We carried out ESEM analyses of crack profiles after 1 hour and then after up to 420 days, at 5 dates distributed over the whole time interval (Fig. 3 ). After indentation and between two measurements the samples were stored at normal lab environmental conditions (21°C, 65 % humidity). We determined the threshold stress intensity of the following ceramics (Soda lime glass and veneering ceramics as reference): Al 2 O 3 , coarse grained, load of indention 9,9 kg, Young's modulus 350 GPa (Frialit-Degussit, Mannheim/Ludwigshafen, Germany), Cerec Mark II, 4 kg, 69 GPa, HiCeram, 6,9 kg, 107 GPa, VMK 95, 4 kg, 91 GPa (all three Vita, Bad Säckingen, Germany), Cercon Base, 7,9 kg, 210 GPa, CergoGold, 4 kg, 70 GPa (both Degudent-Dentsply, Hanau, Germany), Dicor, 2 kg, 74 GPa (Corning Glass Works, Corning, USA), Empress 1, 5,9 kg, 67 GPa, Empress 2, 5,9 kg, 96 GPa (both Ivoclar, Schaan, Liechtenstein), Lava, 8 kg, 210 GPa (3M-Espe, Seefeld, Germany), Soda lime glass, 2 kg, 73 GPa (Saint Gobain, Aachen, Germany), Si 3 N 4 , 6 kg, 289 GPa and hipped 5%Y 2 O 3 -Zirkon, 8 kg, 210 GPa. The constitution of the soda lime glass was SiO 2 72.65, Al 2 O 3 0.28, MgO 3.98, CaO 8.84, Na 2 O 13.79, K 2 O 0.19, other 0.27. Results As examples, Fig. 1 shows a crack starting at the corner of a Vickers indentation (right hand) and Fig. 2 shows a plot representing data for "Cerec Mark II" two days after indentation, as a function of distance from crack tip x (2 μ m < x < 23 μ m), analyzed with the help of Eq. 1'. A linear relationship is observed, from which the value of K tip ( t = 48 h ) = 0,90 MPa√m was easily and precisely deduced. Fig. 3 shows all K tip values determined in an analogous manner for nine examples out of the thirteen investigated ceramics. The gradual decrease of K tip (t) due to decaying stress intensities at the crack tip becomes apparent. The manner in which K tip (t) decreases suggests an exponential relationship, as the decrease appears to be linear on a logarithmic scale. The truncation of the measurements after about 10 4 hours (for reasons of feasibility) appears somewhat arbitrarily, and it cannot be excluded that a further decay, although very small, may have been missed. Note that due to the apparent exponential relationship, the overestimation of the threshold value K I0 due to the truncation after 10 4 hours becomes smaller and smaller with time. We plan to do further measurements after another interval of 10 4 hours (417 days). Considering the mathematical aspect, 10 5 hours (11+ years) would make more sense; but such a long interval is obviously not practicable. As already mentioned, this time constraint is a decided disadvantage of our current method to determine the threshold value. Figure 2 Regression analysis representing data for "Cerec Mark II" two days after indentation, analyzed with the help of Eq. 1 ( u(x) 2 as a function of distance from crack tip x (2 μ m < x < 23 μ m)). A linear relationship is observed. Being aware of the these limitations, and having in mind the neglected possible crack tip shielding as discussed above, we identify K I0 = K tip ( t →∞). Fig. 4 displays all K I0 values in comparison with their K Ic counterparts. Figure 3 K tip ( t ) values of nine out of the thirteen ceramics investigated. The gradual decrease of K tip ( t ) with time due to decaying stress intensity at the crack tip becomes apparent. Figure 4 K I0 threshold values (hatched columns) in comparison with their counterpart critical stress intensities, K Ic (unfilled columns). Refer also to [22]. In the available literature, values for reference: Al 2 O 3 (K I0 = 2.5 ± 0.2 MPa√m); ZrO 2 (K I0 = 3.1 ± 0.2 MPa√m, both values after [4]); Soda lime glass (K I0 = 0,42 MPa√m, after [11]). Discussion K Ic is the lower limit for (fast) catastrophic crack growth. Stress intensities exceeding this limit cause fast crack growth at supersonic velocity, and eventually result in destruction of ceramic components. This kind of destruction, however, is not the most common or important, since it can be avoided by strictly limiting the stress intensities existing throughout a component by a suitable shape of construction. K I0 is the upper limit of stress intensities for absence of crack growth and the lower limit for (slow) subcritical crack growth (SCCG). Limiting stress intensities such that they stay always below K I0 means infinite life time for a component, since SCCG becomes irrelevant. Hence, the most favorable characteristic stress intensity values are obvious: K Ic as high as possible and K I0 as close as possible to K Ic . Such a selection minimizes the extension of the interval in which subcritical crack growth can take place, and it maximizes resistance to catastrophic crack growth due to overloading. Fig. 5 gives a ranking of all ceramics currently tested, based on threshold values related to the corresponding critical values K I0 /K Ic . Favorable ceramics within their class of toughness are situated at the right hand side of the chart. Note, however, that a perfect ceramic material dependent on the focused area of application has not only a favorable (threshold/critical) stress strength relationship but also a high K Ic value. Figure 5 Ranking of all ceramics as imposed by their ratio "threshold value to critical value" ( K I0 / K Ic ). Dicor: see [23]. At first glance zirconia may seem to be a ceramic material superior to alumina, since it has a critical stress intensity factor (Fig. 4 : 9.4 ± 1.5 MPa·√m) which is about three times larger than this of alumina. Values in the literature for zirconia are up to about 8 MPa·√m [ 16 ], compared with 5.4 MPa·√m and [ 5 ]: 5.0 ± 0.2 MPa·√m [ 17 ] for alumina. Naturally, this is a significant advantage when operations near the critical stress of a material are involved. However, in practical applications, stresses having an intermediate level are more common, thus initiating SCCG instead of catastrophic crack growth. Then, if the threshold stress intensities of two ceramics are equal, they are both subject to SCCG at the same rate. Apparently. zirconia vs alumina is an example for such a situation (Fig. 4 ): meaning that both ceramics have equal potential for SCCG. The different behavior of these ceramics is solely rendered to stress bearing capabilities near catastrophic crack growth. At such stresses near K Ic zirconia, of course, has properties superior to alumina. It becomes apparent that at moderate stresses alumina and zirconia may be equally suitable choices, and other criteria may become important for favoring the one or the other material. Such reasons may be the ease of shaping, questions of color, ease of veneering, esthetic considerations, availability, and other circumstances. There is one other aspect to be considered when comparing zirconia and alumina. The exponents n of SCCG of both ceramics are high (in principle meaning slow SCCG), and the answer to the question of which of the materials has the larger exponent depends on whether static or cyclic behavior is addressed: n static = 39 vs 104 and n cyclic = 28 vs 16 for Al 2 O 3 and Y-PSZ, respectively [ 16 ]. These parameters show that lifetimes are shortened and crack growth rates are significantly accelerated by cyclic loading compared to static loading. Zirconia is known to be sensitive to humidity, which is a particular important issue when prosthetic and orthopedic applications are considered. It is known that yttria stabilized zirconia ceramics can be destabilized during the process of steam sterilization. This is due to hydrothermal transformation, resulting in surface roughening of the zirconia ceramic femoral heads. These femoral heads may also undergo slow degradation during long term implantation in the human body. This low temperature degradation does not become significant before several years, but it does raise the question of the use of zirconia for load bearing systems [ 4 ]. In conclusion, it can be stated that SCCG of Y-TZP is activated by the influence of water [ 18 , 19 ], however, there is some controversy about this effect [ 20 ]. An analogous statement holds for MgO-partially stabilized zirconia (PSZ) [ 21 ]. Note that concerning the sensitivity to humidity, there is a notable difference between ceramics for dental or for orthopedic applications. Ceramics for dental applications are often veneered by a different ceramic, which means that there is a protective shield against humidity attacking from outside of the ceramic tooth (but not from inside or from the marginal region). Fig. 6 displays an example of a zirconia ceramic material developed for dental applications and which was formerly used. The sensitivity to humidity becomes apparent. Figure 6 Example (linear Weibull plot) for a zirconia based ceramic material developed for dental applications. Samples handled at 60 % relative humidity (lab environmental conditions; diamonds) vs samples stored in aqua dest for 10 days (triangles). The sensitivity to humidity is obvious. The bending strength due to water storage decreases from σ 63% = 1,346 MPa to 1,003 MPa (about 25 %). There are some other examples of ceramics for which a large difference in the critical stress intensities is observed whereas the threshold values are very similar. For these ceramics an analogous argument holds, as given above for alumina vs zirconia. From Fig. 4 , for Empress 1 or Empress 2 (e.g.) the following values are measured: K Ic = 1.17 ± 0.08 MPa·√m or K Ic = 2.48 ± 0.22 MPa·√m, respectively; and K I0 = 0.83 ± 0.16 MPa·√m or K I0 = 0.94 ± 0.12 MPa·√m, respectively. Again, the critical stress intensity values are largely different, the threshold values, however, are nearly identical. Compare also "Al 2 O 3 " with "Lava" and "Cercon". Authors' contributions RM conceived in the study, designed the study and drafted the manuscript. POW and FJ carried out the experimental work. All authors read and approved the final manuscript. All authors contributed equally to this work. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539288.xml |
534784 | Effects of heparin on the uptake of lipoprotein lipase in rat liver | Background Lipoprotein lipase (LPL) is anchored at the vascular endothelium through interaction with heparan sulfate. It is not known how this enzyme is turned over but it has been suggested that it is slowly released into blood and then taken up and degraded in the liver. Heparin releases the enzyme into the circulating blood. Several lines of evidence indicate that this leads to accelerated flux of LPL to the liver and a temporary depletion of the enzyme in peripheral tissues. Results Rat livers were found to contain substantial amounts of LPL, most of which was catalytically inactive. After injection of heparin, LPL mass in liver increased for at least an hour. LPL activity also increased, but not in proportion to mass, indicating that the lipase soon lost its activity after being bound/taken up in the liver. To further study the uptake, bovine LPL was labeled with 125 I and injected. Already two min after injection about 33 % of the injected lipase was in the liver where it initially located along sinusoids. With time the immunostaining shifted to the hepatocytes, became granular and then faded, indicating internalization and degradation. When heparin was injected before the lipase, the initial immunostaining along sinusoids was weaker, whereas staining over Kupffer cells was enhanced. When the lipase was converted to inactive before injection, the fraction taken up in the liver increased and the lipase located mainly to the Kupffer cells. Conclusions This study shows that there are heparin-insensitive binding sites for LPL on both hepatocytes and Kupffer cells. The latter may be the same sites as those that mediate uptake of inactive LPL. The results support the hypothesis that turnover of endothelial LPL occurs in part by transport to and degradation in the liver, and that this transport is accelerated after injection of heparin. | Background Lipoprotein lipase (LPL) hydrolyses triglycerides in chylomicrons and VLDL and thereby makes fatty acids available for cellular uptake and use in metabolic processes [ 1 , 2 ]. Relatively high levels of LPL mRNA are found in adipose tissue, heart, red skeletal muscle and lactating mammary gland [ 3 , 4 ]. Parenchymal cells, such as adipocytes and myocytes, synthesize and secrete the enzyme, which is then transferred to the endothelium and anchored to the oligosaccharide chains of heparan sulfate proteoglycans (HSPG) [ 1 , 2 ]. There is continuous recycling of the enzyme between the luminal and abluminal side of the endothelial cells, and perhaps to other extracellular sites in the tissue [ 1 , 5 , 6 ]. It is not known how the extracellular enzyme is turned over. One possibility is that it is transported with blood to the liver and degraded there [ 7 ]. LPL activity in the circulating blood is normally low and most of the LPL protein in blood is catalytically inactive [ 7 - 10 ]. Release of lipase from extrahepatic tissues into blood has been demonstrated [ 11 , 12 ]. Model studies with labeled LPL have demonstrated uptake and degradation of both active and inactive LPL in the liver [ 13 - 15 ]. Heparin releases LPL from its endothelial binding sites into the circulating blood. The uptake in the liver is retarded, but not abolished [ 13 , 14 ]. This has been taken as evidence that there are both heparin-sensitive and heparin-insensitive binding sites in the liver. An implication is that the high lipase activity in blood after heparin injection is due to release from peripheral tissues combined with retarded uptake in the liver. Studies in rats and in human subjects indicate that the net effect of heparin is an accelerated transport of LPL to the liver [ 16 , 17 ]. If this hypothesis is correct, LPL mass and activity should increase in the liver after injection of heparin, in contrast to the decrease that occurs in extrahepatic tissues [ 6 ]. To test these concepts we have followed LPL activity and mass in liver after injection of heparin, and we have used immunofluorescence to explore if heparin changes the pattern of where in the liver LPL binds. Results Amount and distribution of LPL in liver LPL activity in rat liver was 26 ± 1 mU/g (Table 1 ), similar to the activity reported by Peterson et al [ 15 ]. This is low compared to the activities in adipose tissue (around 1600 mU/g in fed rats [ 6 ]) and heart (around 1100 mU/g [ 18 ]). LPL mass was 120 ng/g. The relation between LPL activity and mass in plasma was similar to that in liver; activity was 8 mU/ml and mass was 29 ng/ml (Table 1 ). The specific activity of the enzyme in plasma increased to around 1.2 after injection of heparin. This indicates that most of the LPL in plasma or liver before heparin was inactive, in accord with studies on LPL in human plasma [ 7 - 10 ]. To study the distribution of endogenous LPL in rat liver we used affinity-purified chicken antibodies raised against bovine LPL. These antibodies have previously been used for ELISA of LPL in rat tissues [ 19 ]. There was faint immunofluorescence in a granular pattern (green) over hepatocytes, and stronger staining over scattered cells (Figure 1 ). Some of these reacted positively with the ED2 antibodies indicating that they were Kupffer cells. Sections treated with pre-immune IgG instead of anti-LPL (inset in Fig 1 ), as well as sections where the second antibody was omitted, showed no immunofluorescence. Effects of heparin LPL activity and mass in plasma increased many-fold after heparin injection (Table 1 ). The highest value was at 15 min, but even after 60 min the activity was still more then 50-fold higher than in normal plasma. In liver the level of LPL activity had increased already two min after heparin injection and it was about 10-fold increased both at 15 and 60 min compared to time 0. The amount of LPL protein in the liver, measured by the ELISA, had increased about two-fold at two min, 3.5-fold at 15 min and five-fold after 60 min. These data give direct evidence that after injection of heparin some of the LPL released into plasma was taken up by the liver. Hepatic lipase (HL) was also measured in the livers (Table 1 ). As expected, but in sharp contrast to what was found for LPL, the HL activity decreased after heparin. Already after two min the activity had decreased by two thirds. This reflects the release of HL into the circulating blood. The activity in liver remained constant at 15 min and had begun to increase again after 60 min. The pattern of immunofluorescence for LPL after heparin was similar to that before heparin, with faint staining over most cells and more intense staining over scattered cells, some of which were ED2 positive (not shown). To explore the origin of the LPL released into plasma and taken up by the liver we measured LPL activity in heart and adipose tissue before and 20 min after injection of heparin (Table 2 ). Data for plasma and liver in fed rats were similar to those in the experiment in Table 1 . In fasted rats, plasma post-heparin LPL activity was less than half of that in fed rats in accord with previous studies [ 18 , 20 ]. The LPL activity in liver increased after heparin, in concert with the results in Table 1 . There was no statistically significant difference of the LPL activity in liver between fed and fasted rats either before or after heparin. In adipose tissue the LPL activity was about 5-fold higher in fed compared to fasted rats. Heparin caused a washout of 45 % of the LPL activity from epididymal and 65% from perirenal adipose tissue (p < 0.01). Values for LPL activity in heart were somewhat higher in fasted than in fed rats, but this did not reach statistical significance (p = 0.16). Heparin caused no significant decrease of heart LPL in fed rats, but a highly significant (p < 0.01) washout in the fasted rats, about 40%. Injection of labelled bovine LPL The immunostaining of endogeneous LPL was faint and not suitable for detailed analysis or quantitation. The reason is that the only reagents available were chicken antibodies raised against bovine LPL. To further explore the hepatic binding and uptake of the enzyme we therefore injected bovine LPL. A trace amount of 125 I-labeled LPL was included so that we could quantitate the uptake/metabolism. Values are given in Table 3 for the times at which localization of the lipase was studied by immunofluorescence. These values agree with an earlier study when more complete time curves were obtained [ 13 ]. For active LPL earlier studies have shown that at short times after injection about half of the lipase locates in extrahepatic tissues and about half in the liver [ 13 , 21 , 22 ]. In the present study 33 % of the radioactivity was in liver after two min (Table 3 ). This increased to 52 % after 15 min and then decreased again to 20 % after 60 min. Heparin slows down the clearance of LPL from the blood [ 13 , 21 ]. Fifteen min after injection of the labeled lipase, about half is still in the circulating blood [ 13 ]. In our experiments 30 % was in the liver at this time (Table 3 ). Hence, 60 % or more of the removal from plasma had taken place in the liver. After 60 min radioactivity in the liver had decreased to 23 % of the injected dose (Table 3 ). Earlier studies have demonstrated that acid soluble breakdown products of the labeled lipase appear in blood [ 13 ] and in the perfusion fluid in experiments with isolated livers [ 14 ]. Hence, the decrease of label in liver at longer times probably occurred through degradation of the lipase. For inactive lipase earlier studies have shown that only a minor fraction locates in extrahepatic tissues, and more than 70% of the uptake occurs in the liver [ 13 ]. In the present study, 50 % of the radioactivity was in the liver 15 min after injection, while 9 % was in the blood. After 60 min the radioactivity in the liver had decreased to 19 % of the injected dose. To visualize the injected bovine LPL we used rabbit antibodies. These antibodies did not inhibit endogenous LPL in rat post-heparin plasma or in extracts of adipose tissue. No immuno-reaction was seen in sections from control rats not injected with LPL (inset in Figure 2E ). Likewise, there was no immunofluorescence in sections treated with non-immune rabbit IgG instead of anti-LPL, or when the second antibody was omitted. Two min after injection of the active lipase, intense immuno-staining was seen along sinusoids (Figure 2A ). This staining was strongest in the periportal areas. There was little staining outside the sinusoids. Occasionally a few fluorescent dots were seen in hepatocytes, possibly representing endocytic vesicles. At 15 min after injection there was still staining over the sinusoids (Figure 2C ), but most of the staining was now associated with hepatocytes and the number of granulae seen in hepatocytes had increased, indicating that the lipase had been internalized in vesicles (Figure 3A ). Some cells, localized predominantly around the portal area, had many fluorescent dots. This staining was mainly granular in contrast to the more continuous staining over sinusoids at this time. Double staining at 15 min after injection of LPL demonstrated that some of these cells were ED2-positive (Figure 3A ). At 60 min little or no staining remained at sinusoidal surfaces and the total staining had decreased (Figure 2E ), but there were still grains in cells close to the portal area. This was probably enzyme that had been taken up in intracellular vesicles and had not yet become degraded. To get more detailed information about the binding sites in liver we used electron microscopy. For this, bovine 125 I-labeled LPL was injected and 10 min later the livers were fixed by perfusion. The enzyme was visualized as silver grains by means of autoradiography (Figure 4A ). Counting of silver grains in the sections showed that about 55% of the lipase was within spaces of Disse, mostly associated with hepatocytes. Twenty-five to 30% was on the luminal side of endothelial cells. Only about 15% was inside hepatocytes and other cell types, probably Kupffer cells. The pattern of distribution was quite different for the inactive lipase. At the first time, two min, there was intense staining for LPL over scattered cells concentrated to the portal regions but very little staining over sinusoids or hepatocytes (not shown). Double staining with the ED2 antibody showed that the intensively LPL-positive cells were Kupffer cells. After 15 min the immunofluorescence had changed to a more punctuate pattern that still co-localized with Kupffer cells (Figure 5C ). This indicated that the lipase had been internalized in vesicles. There was some immunostaining over other cells, presumably hepatocytes, but this was much weaker than the staining over Kupffer cells (Figure 5A and 5C ). After 60 min the intensity of the staining had decreased, but the pattern with more intense staining over Kupffer cells and much weaker staining over other cells remained (not shown). Hence, there was no indication that the inactive lipase first bound to one type of cell and then transferred to another type for uptake. Electron microscopic autoradiography of sections from livers of rats ten min after injection the inactive LPL, confirmed that the labelled lipase was mostly associated to sinusoidal cells that morphologically seemed to resemble Kupffer cells. Heparin markedly changed the pattern of localization for the active lipase. The initial (two min after injection) staining along the sinusoids was much weaker than in sections from rats that had not received heparin. The staining was generally more intense at 15 min compared to at two min after injection of the lipase (Figure 2D ). This is in accord with the radioactivity data that showed that more LPL had been taken up (Table 2 ). Compared to the pattern without heparin, the staining was spread throughout the liver parenchyma rather than concentrated in the portal areas (compare Figure 2A,2C and 2E with 2B,2D and 2F ). There was more staining associated with scattered cells in the portal areas than in sections from rats that had not received heparin (Figure 2B ). These cells were ED2-positive (not shown). Already at 15 min the LPL-staining had taken on a granular character both in the ED2-positive cells and in hepatocytes (Figure 3 ). At 60 min the intensity of staining had decreased (Figure 2F ). The ED2 positive cells still dominated the picture but there was also granular staining over hepatocytes. More staining remained compared to the same time point without heparin (compare Figure 2F and 2E ). Electron microscopic autoradiography showed that when heparin was injected ten min after active LPL there was a strong reduction in the amount of LPL in the spaces of Disse and on endothelial cells, while the radioactivity found in hepatocytes and Kupffer cells remained (data not shown). Heparin had no marked effect on the distribution of the inactive LPL (Figure 5 ). Most of the staining co-localized with staining for ED2 positive Kupffer cells (Figure 5D ). Discussion This study shows that after injection of heparin, LPL activity and mass in liver increases several-fold, in concert with the hypothesis that heparin causes accelerated transport of LPL to the liver. In other parts of the body LPL is attached to HSPG [ 1 , 2 ]. Heparin competes efficiently with these binding sites. The rapid extraction of LPL by the liver in the presence of heparin implies that some other type of binding site must be present there. Members of the LDL receptor (LDL-R) family bind both the active and the inactive form of the lipase [ 23 ] and two recent studies indicate that LRP is involved in hepatic uptake of LPL [ 24 , 25 ]. The binding of active LPL to LRP is, however, strongly impeded by heparin [ 26 ]. Therefore, it is unlikely that the heparin-resistant binding of active LPL is mediated by LRP or some other receptor of the LDL-R family. Heparin markedly decreased the binding of LPL along the sinusoids. This presumably reflects that binding to HSPG was competed by heparin. Staining associated with Kupffer-like cells increased. This may be the same sites as those that bind inactive LPL. Another possibility is that the LPL-heparin complexes were recognized and taken up as such. There is evidence for binding and uptake of heparin by Kupffer cells [ 27 ]. Most of the binding was, however, to hepatocytes even in the presence of heparin. Our data are qualitative, based on the immunolocalization. For more accurate quantitation one should label the lipase with a non-degradable label like 125 I-tyramine cellobiose and isolate the different cell types. It has been suggested that LPL and HL bind, at least in part, to the same sites in the liver [ 24 , 28 ]. It is, however, unlikely that HL shares the heparin-insensitive sites. The response of the two enzymes to heparin was very different. HL activity decreased by 60% within two min after heparin injection, reflecting release of the lipase into blood. In contrast, LPL activity in the liver increased, reflecting binding to the heparin-insensitive sites. Earlier studies have shown that there are also heparin-sensitive sites that bind LPL in liver [ 13 ]. Wallinder et al perfused livers in situ with heparin 15 min after injection of 125 I-LPL to rats and found that about 10% of the lipase that had bound in the liver could be released [ 13 ]. Vilaró et al perfused isolated livers with 125 I-LPL in a recirculating system for 10 min. After wash the perfusion was then continued in single pass mode with a heparin-containing medium. About 50% of the LPL that had bound in the liver reappeared in the perfusion medium within four min [ 14 , 29 ]. At least some of these heparin-sensitive sites are likely HSPG, and they are probably the main sites that mediate the initial capture of LPL from blood. This binding may correspond to the decoration of the sinusoids seen by immunofluorescence at the earliest time after injection of the lipase. HSPG are present on virtually all cells in the body, including hepatocytes and endothelial cells in the liver [ 30 , 31 ]. Vilaró et al studied binding and uptake of LPL in cultured hepatocytes [ 32 ]. The enzyme was concentrated at the tips of the microvilli, a site where also HSPG are highly abundant [ 33 , 34 ]. Immunofluorescence now showed that at short times after injection the lipase located along the sinusoids, and electron microscopy showed that most of the lipase was in the spaces of Disse. The immunofluorescence was strongest in the portal areas, indicating that the lipase was extracted soon after it entered the liver. Most of the staining was over hepatocytes. With time the immunofluorescence shifted to a granular pattern, indicating that the enzyme had been internalized. The internalization may have occurred with lipase bound to HSPG, as demonstrated with cultured fibroblasts and with hepatocytes [ 32 , 35 , 36 ] and/or with lipase bound to LRP as suggested by several authors [ 24 , 25 , 35 , 37 ]. In fibroblasts, both these pathways contribute to LPL internalization [ 38 ]. The lipase may well recycle as demonstrated by Heeren et al in experiments with cultured hepatocytes [ 39 , 40 ]. The immunostaining gradually faded with time indicating that the lipase was degraded, in accord with previous studies, and with the decrease of 125 I-radioactivity in the liver observed in the present study. Inactive LPL, as prepared here, was taken up in Kupffer cells. Most of the LPL in plasma is inactive [ 8 , 10 ] and there is inactive LPL also in the tissues [ 19 ]. The nature and metabolic significance of this inactive LPL is not clear. Western blot analysis indicates that it is full-length LPL [ 7 ]. On heparin-agarose it elutes in the position expected for monomeric LPL [ 7 ]. Gel filtration of plasma indicates that it is associated with the lipoproteins [ 8 ]. The turnover of inactive LPL does not appear to be much influenced by heparin [ 6 , 8 , 13 ]. Earlier studies had shown that injected, inactive LPL is rapidly bound and degraded in the liver, both in vivo [ 13 ] and on perfusion through an isolated liver [ 14 ]. In these studies the inactive lipase was produced by complete denaturation in 6 M guanidinium chloride. In the present study the inactive lipase was gently prepared by incubation in rat plasma at 45°C. This probably results in dissociation to inactive but still folded monomers [ 41 , 42 ]. Our preliminary experiments had shown that under these conditions the enzyme slowly lost its catalytic activity. After 90 min, the time used here, less than 5 % of the activity remained. In terms of clearance rate and tissue distribution, the present preparation behaved as the fully denatured lipase used in previous studies. Whether any of these lipase preparations faithfully reproduces the metabolic behaviour of the inactive lipase present in plasma is not clear. A recent study has defined several conformational states of the LPL molecule, and the kinetics of conformational transitions [ 42 ]. Before heparin, the ratio of LPL activity to LPL mass was about 0.2 mU/ng in liver and 0.3 mU/ng in blood (Table 1 ). After heparin, the specific activity for LPL in plasma increased to about 1.2, indicating that heparin released mainly or almost exclusively the active form of the lipase as has been shown to be the case in humans [ 7 ]. In the liver, however, the ratio stayed well below one. At the 60 min time point LPL mass had increased by about 500 ng/g, but LPL activity had only increased by about 240 mU. This indicates that the lipase loses catalytic activity after it is taken up in the liver, but is degraded more slowly. These observations are in accord with earlier studies. Chajek-Shaul et al. perfused rat livers with LPL-containing media and found that the enzyme lost its catalytic activity soon after binding/uptake in liver [ 43 ]. Wallinder et al compared the uptake and degradation of 125 I-labeled LPL in liver to that for asialofetuin, which is taken up by the galactose receptor [ 13 ]. The half-life for asialofetuin was about 15 min, whereas that for the lipase was longer, about one hour. To explore the source of LPL released into plasma and taken up by the liver we measured LPL activity in adipose tissue and heart before and after injection of heparin. In fed rats there was a large decrease of LPL activity in adipose tissue, in accord with a previous study [ 6 ]. From these data and the tissue weights at least 7000 mU LPL activity was washed out from white adipose tissue during the first 20 min after heparin. To this should be added an unknown amount of LPL washed out from other tissues. During the same time the LPL activity increased by the 1600 mU in the liver and about 5100 mU in blood. These data indicate that the dominant source of LPL released to plasma in fed rats is the white adipose tissue. Post-heparin LPL activity was lower in fasted than in fed rats, less than half, in accord with previous studies [ 18 , 20 ]. LPL activity in adipose tissue is suppressed during fasting, and we did not find any significant loss of activity after heparin. Hence, the adipose tissue releases much less LPL activity into plasma in fasted than in fed rats. The main contributors are presumably heart and skeletal muscle. We observed a large washout of LPL activity from heart in the fasted rats, about 40%. In other studies we have noted a similar washout from the Soleus muscle. Kuwajima et al perfused some of the rat hindlimb muscles (gastrocnemius, soleus and plantaris) with heparin in situ and observed a large release in fasted but not in fed rats [ 20 ]. These data suggest that in fasted rats, the main source of LPL released into plasma by heparin are skeletal muscles and heart. LPL in plasma is bound to lipoproteins and it has been suggested that the lipase serves as a ligand for binding and uptake of lipoproteins in the liver. Chevreuil et al injected doubly labelled chylomicrons to rats shortly after heparin [ 44 ]. The results showed accelerated lipolysis of the triglyceride moiety of the chylomicrons, as expected. In addition, clearance of chylomicron remnants, as traced by retinyl esters, was greatly accelerated. Together with the present results this suggests that after heparin, the large increase of LPL in blood may accelerate the hepatic uptake of some lipoproteins. Conclusions • In the liver, the active form of LPL initially binds to sinusoidal surfaces but then transfers to and is taken up mainly in hepatocytes • An inactive form of LPL, presumably monomers, was mainly taken up in Kupffer cells • Heparin retards the uptake of active LPL in liver, but there are heparin insensitive binding sites for LPL both on hepatocytes and on Kupffer cells • Release of LPL into blood by heparin results in accelerated transport of the lipase to the liver • The observation that rat liver contained substantial amounts of LPL, most of which was inactive, is in accord with the hypothesis that one route for turnover for endothelial LPL is transport to and degradation in the liver • The observations that most of the LPL in blood is inactive, that injected inactive bovine LPL located to Kupffer cells, and that the immunostaining for endogenous LPL was more intense over Kupffer cells than over hepatocytes suggest that a substantial fraction of the transport from extrahepatic tissues occurs with LPL that has lost its activity. • The main source of LPL released into plasma and taken up by the liver in fed rats is the adipose tissue, whereas in fasted rats the main sources are heart and skeletal muscles. Methods Animals Male Sprague-Dawley rats (Moellegard Breeding centre, Denmark) weighing 180–220 g were used. They were kept on a standard pellet diet in a 12-hour light cycle. In order not to disturb blood circulation or the metabolic functions of the liver, we performed all experiments on unanaesthetized rats. They were killed through decapitation at the time of tissue removal. Injections were made in the tail vein. Mean liver weight for the rats was about 9 g. In some of the rats we dissected out all visible adipose tissue. The mean total weight was 14 g, including fibrous tissue removed with the subcutaneous adipose tissue. To correct values for LPL mass/activity in the liver we used an estimated figure of 3% for the amount of blood plasma remaining in the liver after exsanguination. This was based on earlier experiments with 125 I-albumin and Cr 51 -labeled red blood cells [ 45 , 46 ]. The local Animals Care Committee in Umeå approved all animal procedures. Materials Vectashield mounting medium was from Vector Laboratories, Burlingame, CA. Tissue-Tec OCT compound was purchased from Sakura Finetek Europe BV, Zoeterwoude, The Netherlands. Microscope slides and cover slips were from Menzel – Gläser, Germany. Plasma, used in the preparation of catalytically inactive LPL, was taken from fasted rats with EDTA as anticoagulant. Heparin was obtained from Leo Pharma AB, Malmö, Sweden. The dose given was 500 IU/kg body weight. Lipase and antibody preparations LPL was purified from bovine milk as previously described [ 47 ] and was labeled with 125 I using the lactoperoxidase/glucose oxidase method [ 13 ]. The labeled LPL was separated from damaged protein and free iodine by chromatography on heparin-Sepharose using a gradient of NaCl. The labeled preparations were stored at -70°C in the presence of 2 mg BSA per ml. The specific activity of the labeled LPL was approximately 10 000 cpm/ng. Inactive LPL for the electron microscopy study was prepared by dissociation in guanidinium hydrochloride as described [ 13 ]. To find suitable conditions to prepare inactive LPL for the immunofluorescence experiments, we diluted bovine LPL in rat plasma to the concentration we would later use in the in vivo experiments and incubated this at different temperatures. On incubation at 37°C the LPL activity remained essentially stable for one hour. At higher temperatures the lipase became unstable. Based on these results we decided to use 45°C for gentle inactivation of LPL aimed to prevent aggregation of the enzyme. Chromatography on heparin-Sepharose of active and inactivated LPL showed, as expected [ 19 ], that the active form of LPL eluted around 1 M NaCl, while the inactive form(s) eluted earlier in the salt gradient. After 30 min at 45°C most of the lipase eluted early in the gradient. Only about 20 % remained in the form with high heparin affinity. At 60 min this form had been reduced to 8 % and after 90 min it had virtually disappeared. From this we decided to use incubation at 45°C for 90 min to transform LPL to the inactive (presumably monomeric) form with low affinity for heparin. A trace amount of 125 I-labeled LPL was included in the preparation, to enable us to follow the distribution and metabolism of the injected material. Each rat received about 40 μg lipase protein, except in the electron microscopy studies where only a trace amount of the labeled lipase was injected. Antibodies against bovine LPL were raised in a chicken (chicken no 225) and IgG were isolated from egg yolks as previously described [ 48 ]. Antibodies against bovine LPL were also raised in a rabbit and IgG were isolated on a Protein A-Sepharose column. Both the chicken and the rabbit antibodies were affinity purified on LPL-Sepharose. They were eluted with 0.2 M glycine at pH 2.7, and 50 mM diethylamine at pH 12, respectively, and immediately dialysed against 10 mM Tris/HCl, pH 7.4. Monoclonal antibody 5D2 to LPL was a kind gift from Dr. J. Brunzell, Seattle. A mouse monoclonal antibody (ED2) against a surface antigen expressed on rat Kupffer cells was obtained from Becton Dickinson, San Diego, CA. Goat anti-rabbit IgG labeled with Alexa Fluor 488, goat anti-chicken IgG labeled with Alexa Fluor 488, and goat anti-mouse IgG labeled with Alexa Fluor 546 were from Molecular Probes, Leiden, The Netherlands. Goat IgG, used for control sections, was from Sigma, St.Louis, MO. Preparation of tissue for immunofluorescence studies and confocal microscopy Small pieces of liver were mounted in Tissue – Tec OCT and snap frozen in propane chilled with liquid nitrogen. The tissue pieces were then stored at -70°C until sectioning. Cryosections were fixed for 10 min in 4 % paraformaldehyde. After rinsing, the sections were blocked in 5 % goat serum for 10 min and then incubated overnight with the primary antibody. All these procedures were made at room temperature. Incubation with the secondary antibody was then for 30 min at 37°C. The sections were rinsed in 0.01 M phosphate 0.15 M NaCl at pH 7.4 and mounted in Vectashield medium (Vector laboratories, Burlingame, CA). The immunostained samples were analyzed by confocal laser scanning microscopy (Leica SP2 or Nikon Eclipse E 800). To avoid potential signal crossover the two fluorophores were sequentially scanned. Data were collected with sequential laser excitation to eliminate bleed through and with confocal parameters such as pinhole size set to minimize the thickness of the optical sections. The images were digitally optimized using the Adobe Photoshop software. Electron microscopy For autoradiographic studies, 125 I-labeled LPL was injected to rats and 10 min later, the livers were perfused with 2% paraformaldehyde, 2.5% glutaraldehyde in 0.1 M phosphate buffer at pH 7.4 for 10 min. After fixation the livers were washed in 0.1 M phosphate buffer, cut in small pieces, dehydrated through graded acetone solutions and embedded in Spurr resin. Postfixation with 1% osmium tetroxide was not performed due to its known effect of fading latent images in autoradiography. Ultrathin sections, 70 nm thick, were collected over Formvar-carbonated copper grids. These sections were coated with a monolayer of Ilford L4 nuclear emulsion, diluted 1:4 with distilled water, by means of a tungsten wire loop following the "loop interference" technique. After an exposure of 8 months, the silver grains were revealed with Phenidon development. Sections were stained with uranyl acetate and lead citrate and examined in a Hitachi MT 800 electron microscope at 75 kV. Quantitative analysis was performed by counting the number of silver grains per area in each experimental condition. The area analyzed was about 25 × 10 4 μm 2 . Lipase assays The activities of LPL and of hepatic lipase were determined as described [ 49 ]. In the assay for LPL the substrate was an emulsion of soybean triglycerides and a trace amount 3 H oleic acid-labeled triolein in egg yolk phospholipids. Hepatic lipase (HL) was inhibited by incubation of the samples on ice for two h with rabbit anti-HL IgG. In the HL assay, LPL is inactivated by 1 M NaCl. Both assays were run at 25°C for 30 min. All determinations were carried out in triplicate. The activities are expressed in mU/ml plasma. One mU corresponds to 1 nmol of fatty acid released / min. All determinations were carried out in triplicates. Plasma samples were stored frozen at -70°C before the analysis. Tissue samples were rinsed in cold 0.9 % NaCl, blotted dry, weighed and then immediately frozen in liquid nitrogen in 9 volumes of buffer at pH 8.2 containing per ml: 1 mg BSA, 10 mg Triton X-100, 1 mg SDS, 5 IE heparin, and protease inhibitor Complete Mini (Roche) 1 tablet / 50 ml buffer. They were stored at -70°C and later thawed and homogenized with a Polytron homogenizer (PT-MR 3000; Kinematica AG, Littau, Switzerland). The homogenates were centrifuged for 15 min at 3000 rpm in a Beckman Microfuge and the supernatants were used for the assays. For assay of LPL in liver and post-heparin plasma, the activity of hepatic lipase was suppressed by incubating the extract with an excess of anti-HL immunoglobulins before assay. Detergent containing extraction buffers are needed to solubilize and stabilize active and inactive forms of LPL efficiently [ 19 , 50 ], but the detergents may interfere with the assay. Bergö and Olivecrona [ 19 ] used the same assay conditions as in the present study and found that the assay system tolerated at least 10 μl of the detergent buffer without any decrease in LPL action. We have repeated these studies in the context of the present experiments. With extracts from adipose tissue, the assay system showed good linearity between the amount of extract added and the lipase activity displayed, but with extracts from heart, kidney or liver there was a definite nonlinearity. Our interpretation is that other tissue proteins, solubilized by the detergents, interfere. We have therefore used a small volume of tissue extract (usually 2 μL), to stay within, or close to, the linear range of the assay. To explore the recovery of LPL activity in liver extracts as prepared here, we added purified bovine LPL to the homogenate which was then treated and assayed as the other samples. The recovery of the added bovine LPL was complete within experimental error. LPL protein mass was measured by an ELISA, using chicken antibodies for capture and the monoclonal 5D2 antibody coupled to peroxidase for detection [ 19 ]. Bovine LPL was used as standard. List of abbreviations LRP – low density lipoprotein receptor-related protein, LPL – lipoprotein lipase, HSPG – heparan sulphate proteoglycan, HL – hepatic lipase, ELISA – enzyme-linked immunoassay, BSA – bovine serum albumin, SDS – sodium dodecyl sulphate, VLDL – very low density lipoprotein Authors' contributions LN carried out the immunolocalization studies, C L-I carried out the electron microscopic studies, SV participated in the design of the study and supervised the electron microscopy, JG carried out the studies on wash-out of LPL from tissues after heparin, TO conceived of the study, participated in its design and drafted the manuscript. GO participated in the design of the study and coordinated the work. All authors read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC534784.xml |
548667 | The role of glucocorticoid action in the pathophysiology of the Metabolic Syndrome | Glucocorticoids are stress hormones that modulate a large number of physiological actions involved in metabolic, inflammatory, cardiovascular and behavioral processes. The molecular mechanisms and the physiological effects of glucocorticoids have been extensively studied. However, the involvement of glucocorticoid action in the etiology of the Metabolic Syndrome has not been well appreciated. Recently, accumulating clinical evidence and animal genetics studies have attracted growing interest in the role of glucocorticoid action in obesity and insulin resistance. This review will discuss the metabolic effects in the context of glucocorticoid metabolism and establish the association of glucocorticoid action with the features of the Metabolic Syndrome, especially obesity and insulin resistance. Special discussions will be focused on corticosteroid-binding globulin and 11β-hydroxysteroid dehydrogenase type 1, two proteins that mediate glucocorticoid action and have been implicated in the Metabolic Syndrome. Due to the complexities of the glucocorticoid biology and the Metabolic Syndrome and limited space, this review is only intended to provide a general link between the two areas with broad rather than in-depth discussions of clinical, pharmacological and genetic findings. | Introduction Insulin resistance and hyperinsulinemia are often associated with a group of risk factors such as obesity, dyslipidemia, hypertension and impaired glucose tolerance. This cluster of metabolic abnormalities, first defined as Syndrome X by Reaven in 1988 [ 1 ] and supported by additional evidence [ 2 , 3 ], is now more often referred to as the Metabolic Syndrome and has been increasingly recognized as important risk factors for coronary artery disease (CAD). The point of view became institutionalized and although the National Cholesterol Education Program's Adult treatment Panel III (ATP III) and the World Health Organization (WHO) have slightly different definitions [ 4 - 6 ], the Metabolic Syndrome is consistently characterized by a collection of metabolic abnormalities such as insulin resistance, obesity, dyslipidemia, hyperglycemia, and hypertension [ 7 ]. Not all of the disorders in the Metabolic Syndrome may be observed in the same individual. Most people with the syndrome have insulin resistance that could lead to glucose intolerance and diabetic hyperglycemia. Although the mechanisms underlying the pathogenesis of the Metabolic Syndrome are not exactly clear, obesity, insulin resistance and other independent factors such as vascular and immunologic origins appear to be involved [ 7 ]. The prevalence of the Metabolic Syndrome is more than 20% among the US adults adjusted for age [ 8 ], which is far greater than observed in an earlier study with European participants at least partly due to differences in the criteria used to define the condition [ 9 ]. Increased cardiovascular and mortality risks are associated with the Metabolic Syndrome [ 10 ]. The condition is usually managed with pharmaceutical agents for correcting dyslipidemia, anti-hypertensives, and insulin sensitizing agents or a combination of the above. Most existing agents only treat individual metabolic abnormalities. To date, no single agent can ameliorate all the features of the Metabolic Syndrome. There is an increasing need for novel agents to treat multiple abnormalities of the syndrome. Glucocorticoid (GC) excess has been linked to clinical observations associated with the Metabolic Syndrome. In Cushing's syndrome [ 11 ], increased secretion of GCs largely due to pituitary adenoma leads to central obesity, hypertension, hyperlipidemia and glucose intolerance, a group of metabolic abnormalities reminiscent of the Metabolic Syndrome. Correction of hypercortisolism by transsphenoidal surgery at least normalizes blood pressure [ 12 , 13 ]. In addition, clinical administration of GCs to treat acute and chronic inflammatory diseases has been associated with metabolic adverse effects such as hypertension, obesity, hyperlipidemia and insulin resistance as seen in the Metabolic Syndrome [ 14 - 16 ]. These clinical findings suggest that GC action could play a role in the pathophysiology of the Metabolic Syndrome. GC metabolism and action Cortisol, the principal active GC in humans, is secreted by the adrenal gland and is converted to cortisone, the inert GC, primarily in kidney [ 17 - 19 ]. Two isozymes of 11β-hydroxysteroid dehydrogenase (11β-HSD) are responsible for the tissue-specific interconversion of cortisone and cortisol at the endoplasmic reticulum: type 1 and 2 (11β-HSD1 and 11β-HSD2) [ 20 ]. The two isozymes are products of two different genes and have distinct tissue distributions, with 11β-HSD1 expressed primarily in liver, adipose, kidney and brain and 11β-HSD2 mainly in kidney and salivary glands [ 20 ]. 11β-HSD1 converts inactive cortisone to cortisol in human or inactive 11-dehydrocorticorsterone (11-DHC) to corticosterone in rodents and 11β-HSD2 catalyzes the opposite reaction. Bidirectional activities (both reductase and dehydrogenase) have been observed with 11β-HSD1 in vitro but it is mainly a reductase in vivo [ 21 ]. Since GC action is largely mediated by the ligand-induced activation of the GC receptor (GR), the local concentration of cortisol (or corticosterone) dictates GR activation. In tissues such as liver and adipose where 11β-HSD1 is expressed, there are two sources for cortisol (or corticosterone) accumulation: the fraction produced by 11β-HSD1 within the tissue and that from the plasma by diffusion. Obviously, 11β-HSD2 activity is responsible for reducing the cortisol level in kidney [ 17 - 19 ]. In addition, cortisol metabolism in liver is part of the balance maintaining the tissue-specific cortisol concentration. The circulating cortisol level undergoes circadian variations peaking in the early morning at approximately 800 nM and reaching a nadir of about 200 nM at midnight [ 22 ]. The plasma cortisone level is much lower and shows no significant circadian rhythm [ 22 ]. The salivary cortisol level exhibits a similar trend of diurnal rhythm [ 23 ]. Rodents housed under 12-h light, 12-h dark illumination conditions exhibit an opposite pattern of circadian variation with lowest circulating corticosterone levels in the early morning and the peak concentration at the light/dark transition phase before declining to nadir [ 24 ]. The plasma GC level is regulated by the activity of the hypothalamic-pituitary-adrenal (HPA) axis, a neuroendocrine feedback circuit that can be activated by physiological stimuli such as stress [ 25 ]. Plasma cortisone is largely in the free unbound form but approximately 6% cortisol is bound to albumin and 90% is bound to corticosteroid-binding globulin (CBG), a protein synthesized in liver and secreted in blood [ 26 , 27 ]. Since only free cortisol is active, CBG binding may restrict the access of cortisol to target cells and regulate its bioavailability and metabolic clearance. On the other hand, CBG may act as a carrier protein for cortisol mediating its delivery to sites of inflammation [ 28 , 29 ]. CBG is also present in several tissues and may be involved in the regulation of tissue-specific GC action. For example, the significantly lower CBG level in the adipose tissue of obese Zucker rats may contribute to insulin resistance [ 30 ]. CBG levels are down regulated by physiological changes such as stress [ 31 - 33 ]. Both cortisol and cortisone are metabolized in liver first by the A-ring reductases followed by several steps of further structural transformation catalyzed by other enzymes [ 20 ]. The final metabolites, 5α – and 5β-tetrahydrocortisol (5α – and 5β-THF) and 5β-tetrahydrocortisone (THE), are eliminated through urinary excretion and are often used as biomarkers for GC metabolism [ 20 , 34 ]. While the total urinary tetrahydro metabolites (THF and THE) may serve as an indicator for GC metabolism or activity, using the ratio of the urinary THF to THE to predict the interconversion of cortisol and cortisone by 11β-HSDs is questionable for the following reasons: First, the ratio is a reflection of the total metabolism of cortisol and cortisone in the whole body instead of one particular tissue because the two isozymes have distinct tissue distribution patterns. Second, other enzymes, including the A-ring reductases and those involved in the subsequent metabolic steps forming THF and THE, also contribute to the balance between cortisol and cortisone. Therefore, the urinary ratio of THF to THE is determined by the combined activities of different enzymes in multiple tissues. Another convenient way to measure GC metabolism is to measure the salivary cortisol levels [ 20 ]. GC action is mediated by GR, a nuclear receptor that regulates physiological events through activation or repression of target genes involved in inflammation, gluconeogenesis and adipocyte differentiation [ 35 , 36 ]. Upon activation, a GR dimer binds to GC response elements (GREs), interacts with components of the transcription machinery and activates the transcription of downstream genes [ 35 , 36 ]. The ligand-bound GR could also bind to negative GREs (nGREs) that mediate the repression of gene transcription, or the starting point of transcription and thus interferes with the general transcription machinery [ 35 , 36 ]. Some transrepression effects of GC action are achieved through a DNA binding-independent process, in which GR interacts with transcription factors such as AP-1 and NFκB and represses their activity on gene expression [ 37 - 39 ]. Repression of NFκB mediated transcription by GC can also be achieved by induction of IκB synthesis [ 40 , 41 ]. Examples of genes regulated by GR and involved in the hepatic gluconeogenesis, adipocyte differentiation, hormonal control, and inflammation are summarized in Table 1 [ 39 , 42 - 66 ]. The gene stimulation or suppression effects mediated by activated GR sequentially regulate a myriad of physiological actions in response to GCs. Since the pool of active cortisol or corticosterone is the active ligand for GR, the availability of free cortisol or corticosterone mediated largely by CBG-dependent protein binding and tissue-specific activities of 11β-HSDs are critical for GC action. The role of GC action in obesity and insulin resistance is implicated by the biological or physiological consequences of deficiency or activation of CBG or 11β-HSDs (see below). The GC production and tissue-specific conversions are illustrated in Figure 1 . Table 1 Examples of genes regulated by GR Gene Names Function Regulation Reference Glutamine synthetase Amino acid metabolism Up 42 TAT Amino acid catabolism Up 43, 44 Tryptophan oxygenase Amino acid catabolism Up 45 PEPCK (liver) Gluconeogenesis Up 46 G6Pase Gluconeogenesis Up 47, 48 Angiotensinogen Precursor of angiotensin I; vasoconstriction, electrolyte balance, etc. Up 49 Leptin Energy metabolism Up 50 VLDLR Lipoprotein metabolism Up 51 PEPCK (adipose) Glyceroneogenesis Down 52 aP2 Intracellular lipid shuttling and metabolism Up 53 GLUT4 Glucose transport Up 53 HSL Lipolysis Up 53 LPL Lipid metabolism Up 53 TNF-α Inflammation and apoptosis Down 53 Osteocalcin Marker for mature osteoblasts Down 54, 55 CRH Stress mediated/feedback hormone release Down 56 POMC Precursor of pituitary hormones Down 57, 58 Prolactin Hormone critical for reproduction Down 59 Proliferin Angiogenesis Down 60, 61 Glycoprotein hormone α-subunit Common subunit of gonadotropin hormones Down 62, 63 IL-6 Proinflammatory cytokine Down 64 IL-8 Proinflammatory cytokine Down 65 Collagenase Matrix protease Down 66 ICAM-1 Inflammatory response Down 39 Abbreviations : TAT, tyrosine aminotransferase; PEPCK, phosphoenolpyruvate carboxykinase; G6Pase, glucose-6-phosphatase; VLDLR, very low density lipoprotein receptor; aP2, adipocyte fatty acid binding protein or A-FABP; GLUT-4, glucose transporter 4; HSL, hormone sensitive lipase; LPL, lipoprotein lipase; TNF-α, tumor necrosis factor α; CRH, corticotrophin-releasing hormone; POMC, proopiomelanocortin; IL-6, interleukin 6; IL-8, interleukin 8; ICAM-1, intercellular adhesion molecule 1. Figure 1 Glucocorticoid metabolism. The secretion of glucocorticoids by the adrenal gland is regulated by the HPA axis via secretion of ACTH. The main plasma cortisol (F) is protein bound with 4–5% free fraction. The plasma cortisone (E) is in the free unbound form. The equilibrium of cortisol and cortisone between the plasma and tissues are illustrated with the dotted bidirectional arrows. Tissue-specific GC metabolism are also depicted. GCs are metabolized primarily in liver and the metabolites are excreted in the urine. Only tissues relevant to the Metabolic Syndrome are shown. THE, tetrahydrocortisone; THF, tetrahydrocortisol. Clinical association of GC action and the Metabolic Syndrome Accumulating clinical evidence has demonstrated the association of abnormal GC metabolism and the Metabolic Syndrome. The plasma cortisol levels were increased in an elderly cohort with one or more features of the Metabolic Syndrome [ 67 ]. Further, a good correlation was observed between total urinary GC metabolites and the number of features of the Metabolic Syndrome in these patients [ 67 ]. Both the secretion rate and peripheral clearance of cortisol in these patients were positively correlated with systolic blood pressure, fasting glucose and insulin [ 67 ]. In agreement with this finding, stress-related cortisol secretion in a population of 51-yr-old men showed associations with diastolic blood pressure, fasting glucose and insulin [ 68 ]. Several additional reports also suggest correlation of increased GC activity with insulin resistance, hyperglycemia and hypertension [ 69 - 71 ]. Although one study indicated that plasma cortisol levels decreased in obese women due to increased metabolic clearance [ 72 ], stress-induced cortisol response is consistently correlated with obesity in independent studies suggesting increased HPA activity in obesity [ 73 - 77 ]. Higher adrenocortical activity was also observed in children with higher body fat mass [ 78 , 79 ]. Weight loss led to lower plasma cortisol and reduced insulin resistance [ 79 ]. A study in the general population indicates that even modestly increased cortisol levels contribute to obesity [ 80 ], and insulin resistance is positively associated with cortical activity [ 81 , 82 ]. These clinical findings demonstrate the strong correlation of increased GC activity with the features of the Metabolic Syndrome in humans. The metabolic effects of GCs The clinical correlation studies raised the possibility that GC action could play a role in the origin of the features of the Metabolic Syndrome. This notion was further established and supported by animal studies to address the metabolic effects of GCs. Adrenalectomy in young ob/ob or db/db mice slowed body weight gain [ 83 ]. Upon GC administration, these animals retained body weight gain with concomitant increase in food intake [ 83 ]. Likewise, obese Zucker rats lost body fat mass after adrenalectomy and remained so even after exogenous administration of low doses of GCs [ 84 ]. The adrenalectomy resulted in significantly reduced plasma insulin, glucose and triglyceride levels [ 84 ]. As the doses of administered GCs increased, the plasma insulin and triglyceride levels were elevated [ 84 ]. Similar results were observed in another study using adrenalectomized rats with diet-induced obesity demonstrating the effects of GC action on plasma and liver triglyceride levels, plasma insulin, and adipose tissue weight [ 85 ]. These effects appear to be minimized when there is restriction on high-energy diet [ 86 ], suggesting they may be exerted via mediating the central ingestive behavior. These findings highlight the central role of GCs in the development of obesity and other features of the Metabolic Syndrome. The metabolic effects of GCs are mediated by several mechanisms that are physiologically relevant to hepatic and peripheral insulin resistance, dyslipidemia, obesity and hyperglycemia. Events driven by these mechanisms take place across the tissues contributing to the abnormalities in the Metabolic Syndrome (Fig. 2 ). In liver, GCs increase the activities of enzymes involved in fatty acid synthesis and promote the secretion of lipoproteins [ 87 , 88 ]. The hepatic lipogenic effect of GCs is consistent with clinical findings that GC therapy causes triglyceride accumulation within the liver [ 89 - 91 ]. Since liver fat appears to be involved in the negative regulation of hepatic insulin sensitivity [ 92 ] and is associated with certain features of the Metabolic Syndrome independent of visceral fat mass [ 93 - 96 ], hepatic fat accumulation promoted by GCs is likely to contribute to the pathophysiology of the Metabolic Syndrome. GCs also induce the hepatic gluconeogenic pathway via the activation of GR, which stimulates the expression of phosphoenolpyruvate carboxykinase (PEPCK) and glucose-6-phosphatase (G6Pase), the rate-limiting enzymes in gluconeogenesis [ 97 , 98 ]. This results in increased hepatic glucose output and hyperglycemia. In adipose tissue, GCs promote the differentiation of pre-adipocytes to adipocytes, which could lead to increased body fat mass [ 99 , 100 ]. However, once differentiated, the adipocytes develop insulin resistance in the presence of GCs with decreased insulin-stimulated glucose uptake without changing their ability to bind insulin [ 101 ]. The reduced insulin sensitivity appears to be mediated by GC antagonizing the insulin-stimulated translocation of glucose transporters from intracellular compartments to the plasma membrane [ 102 - 104 ]. A similar mechanism is likely responsible for the GC-induced insulin resistance in skeletal muscle [ 105 ]. GCs also inhibit insulin-stimulated amino acid uptake by adipocytes [ 106 ]. Increased lipolysis or lipid oxidation could be also involved in the peripheral insulin resistance induced by GCs [ 107 , 108 ]. GCs inhibit insulin secretion by the pancreatic β cells in animals and perturb high-frequency insulin release in the fasting state in human [ 109 , 110 ]. GC action has been implicated in hypertension as well. GCs are agonists of mineralocorticoid receptor (MR), which upon activation leads to renal salt retention and elevated blood pressure. The expression of both 11β-HSD1 and 11β-HSD2 in kidney suggests the interconversion of inert and active GCs is maintained in a balance so that MR activation can be controlled tissue-specifically [ 111 ]. GC excess as a result of either increased 11β-HSD1 activity or reduced 11β-HSD2 activity leads to MR activation and hypertension. GCs also increase aortic vasoconstriction through unknown mechanisms. The expression of 11β-HSD1 in aortic endothelial cells is consistent with such a notion and suggests this could be a second pathway for GC induced hypertension [ 112 - 114 ]. Figure 2 The link between the metabolic effects of glucocorticoids and the features of the Metabolic Syndrome. The major effects in different tissues are summarized and the potential physiological links to the Metabolic Syndrome are shown. These data, both physiologically and mechanistically, suggest that the metabolic effects of GCs are exerted in multiple tissues and increased GC action contributes to the etiology of the Metabolic Syndrome. Through molecular and genetic studies, more information has become available to dissect the role of tissue-specific GC action in the features of the Metabolic Syndrome. Genetic studies with the main players in GC action have been most revealing. Since GR has been well reviewed in other publications, this review will only discuss CBG and 11β-HSD1. Modulation of GC action by CBG is associated with adiposity CBG is not only in the blood but also found in tissues [ 115 , 116 ]. Since CBG is the main GC binding protein, its tissue distribution and local levels play important roles in GC action. Intuitively, CBG level should be negatively correlated with the free cortisol or corticosterone concentration because of its role in restricting free GC fraction. This is especially true in a tissue-specific manner. For example, the reduced adipose CBG level in obese Zucker rat results in elevated free local corticosterone that may have contributed to the obesity and insulin resistance phenotype [ 30 ]. In general, in the human population, serum CBG levels are negatively correlated with a variety of parameters important in defining the Metabolic Syndrome: body mass index (BMI), waist to hip ratio (WHR), blood pressure and HOMA [ 117 ]. However, over-expression or secretion of CBG in the liver could lead to compensatory activation of the HPA axis and consequently elevated adrenal production of cortisol or corticosterone. This feedback response leads to a global effect of elevated total and free cortisol or corticosterone levels. This was observed in a pig genetic model with high fat deposits and low muscle content, where the hepatic CBG expression was significantly higher than in another population and the total and free cortisol levels were elevated [ 118 ]. On the other hand, drastic reduction of CBG concentration or its capacity to bind cortisol or corticosterone can also cause compensatory response by the HPA axis. A familial CBG deficiency led to decreased total and free plasma cortisol levels and hypotension [ 119 ]. Likewise, a human CBG polymorphism associated with reduced affinity for cortisol only led to a marginal increase in serum free cortisol, possibly due to the negative regulation of cortisol production by the HPA axis [ 120 ]. Together, these data demonstrate the importance of CBG level and its cortisol binding capacity in modulating GC action and origination of the Metabolic Syndrome. Further, these studies also suggest that the variation in CBG level or capacity may trigger compensatory response of the HPA axis to balance plasma free cortisol concentrations. Despite the compensatory response by the HPA axis to balance the plasma free cortisol or corticosterone concentrations under conditions of CBG reduction or deficiency, tissue-specific GC excess can still occur. This is especially important with respect to GC-stimulated differentiation of pre-adipocytes and insulin resistance of mature adipocytes, with the former effect increasing fat content and the latter reducing the tissue sensitivity to insulin. For instance, preadipocytes from an individual with CBG deficiency had increased proliferation and enhanced differentiation compared to normal cells [ 121 ], which may be responsible for the increased adiposity in CBG deficiency. This notion was observed in genetic models of obesity and insulin resistance. The CBG capacity the white adipose tissue of Zucker rat is lower than that in its lean counterpart [ 30 , 122 ], suggesting increased GC action in the obese adipose tissue that could contribute to the obese and insulin resistance phenotype. 11β-HSD1 and obesity and insulin resistance Both 11β-HSD1 and 11β-HSD2 are located at the endoplasmic reticulum (ER) but with distinct topologies. 11β-HSD1 has one short N-terminal transmembrane region with the catalytic domain protruding into the ER lumen; in contrast, the N-terminus of 11β-HSD2 is lumenal with the catalytic domain facing the cytoplasm [ 123 - 125 ]. The primary role of 11β-HSD2 is to prevent renal GC excess and consequent MR activation by inactivating cortisol or corticosterone, as mice deficient in 11β-HSD2 had hypokalemia and hypertension [ 126 ]. Given the growing interest in 11β-HSD1 and its role in the Metabolic Syndrome, this section will primarily focus on this isozyme. Dysregulation of tissue-specific 11β-HSD1 expression and activity has been observed in obese diabetic animal models and humans. Compared with their lean littermates, ob/ob mice have reduced hepatic 11β-HSD1 activity but higher corticosterone level in liver due to their elevated plasma corticosterone [ 127 ]. As a result, the liver PEPCK expression is elevated at least partly contributing to hyperglycemia. However, the hepatic 11β-HSD1 activity is marginally increased in db/db mice [ 128 ]. As in ob/ob mice, the 11β-HSD1 activity is decreased in liver but increased in omental fat in obese Zucker rats [ 129 , 130 ]. Although both impaired hepatic regeneration of cortisol by 11β-HSD1 and elevated adipose 11β-HSD1 activity were observed in obese humans [ 131 , 132 ], the association of adipose 11β-HSD1 activity with obesity, insulin resistance and other features of the Metabolic Syndrome has been consistently observed in different groups of obese subjects, including obese men and women [ 131 , 133 , 134 ]. However, no difference in 11β-HSD1 activity was detected between obese type 2 diabetics and their obese controls, suggesting the dysregulation of 11β-HSD1 is better associated with obesity than the diabetic phenotype [ 135 ]. In-situ hybridization revealed that 11β-HSD1 mRNA is increased in both subcutaneous and visceral fat in obese subjects [ 136 ]. The association of adipose 11β-HSD1 with BMI and other features of the Metabolic Syndrome was also found in populations of different ethnic backgrounds [ 137 ]. In a group of young adult monozygotic twins, the intrapair differences in BMI are positively correlated with those in adipose 11β-HSD1 expression [ 138 ]. This association is clearly established on the same genetic background, confirming the direct link of adipose 11β-HSD1 activity and adiposity. Most of these association studies were done with subcutaneous fat. It is important to note that 11β-HSD1 activity is higher in omental fat and subject to stimulation [ 139 ]. The activity of 11β-HSD1 in adipocytes is relevant for the correlation since the activity in cultured preadipocytes does not appear to be correlated with obesity [ 140 ]. These association studies suggest that the adipose 11β-HSD1 may be a contributing factor to obesity and insulin resistance. In agreement with this conclusion, treatment of obese Zucker rats with carbenoxolone slightly improved lipid profile but had no effect on obesity and insulin resistance, because only the hepatic 11β-HSD1 but not that in adipose tissue was inhibited [ 141 ]. It is important to note that carbenoxolone also inhibits 11β-HSD2 and further studies with selective 11β-HSD1 inhibitors are needed to confirm this observation. In contrast to increased adiposity in the Metabolic Syndrome, some human immunodeficiency virus (HIV)-infected patients treated with combined highly active antiretroviral therapy (HAART) develop a lipodystrophic syndrome. The condition is characterized with loss of subcutaneous fat, accumulation of abdominal fat, hypertriglyceridemia and insulin resistance [ 142 ]. The condition is also referred to as pseudo-Cushing's syndrome because the distribution of fat accumulation in these patients is similar to that in Cushing's syndrome but their circulating cortisol levels are not elevated [ 143 ]. Interestingly, patients with lipodystrophy were shown to have higher levels of subcutaneous adipose 11β-HSD1 expression and higher ratios of urinary cortisol:cortisone metabolites than non-lipodystrophic patients [ 144 ]. These findings suggest that 11β-HSD1 could play a role in mediating the metabolic abnormalities of the HAART-associated lipodystrophy with the almost complete loss of subcutaneous fat. This further suggests that the expression of 11β-HSD1 seems to be more important to the metabolic state than the amount of subcutaneous fat though further investigation is required. Genetic studies using animal models support the findings in the clinical studies. In mice deficient in 11β-HSD1 generated through targeted gene disruption, there was no conversion of the inert 11-dehydrocorticosterone to corticosterone and attenuation of the hepatic activities of PEPCK and G6Pase, two key gluconeogenic enzymes [ 145 ]. These mice consumed more calories but were resistant to high fat diet-induced obesity, insulin resistance and hyperglycemia with improved lipoprotein profile [ 145 - 147 ]. Concomitant with these phenotypes, the adipose expression of TNF-α decreased, and adiponectin, PPARγ, and UCP-2 increased indicating insulin sensitization [ 146 ]. There were no bone marrow adipocytes in these knockout animals but bone formation appeared to be normal, suggesting that intracellular GC action may not play a role in bone formation [ 148 ]. The HPA axis appears to be activated in the 11β-HSD1 knockout mice. There was compensatory adrenal hyperplasia, increased secretion of corticosterone and exaggerated ACTH and corticosterone response to stress [ 145 , 149 ]. The plasma CBG levels were slightly reduced [ 149 ]. These findings with 11β-HSD1 deficiency suggest inhibition of this enzyme could help ameliorate some of the features of the Metabolic Syndrome. However, compensatory response from the HPA axis and the induced adrenal activity can occur. Interestingly, 11β-HSD1 knockout ameliorated age-related learning impairments but the underlying mechanism is not clear [ 150 ]. The importance of 11β-HSD1 in the Metabolic Syndrome was also demonstrated with 11β-HSD1 transgenic animals. Mice with adipose-specific overexpression of the rat 11β-HSD1 had increased adipose levels of corticosterone and acquired features of the Metabolic Syndrome: diet-induced visceral obesity, insulin resistance, hyperlipidemia and hyperphagia [ 151 ]. The transgenic mice also developed hypertension, at least in part due to the increased adipose expression of angiotensinogen and the consequent activation of the rennin-angiotensin system (RAS) [ 152 ]. In contrast, selective overexpression of 11β-HSD1 in liver only caused mild insulin resistance with no effect on fat depot mass [ 153 ], although impaired hepatic lipid clearance and hypertension were observed in these animals. These transgenic studies demonstrate that both the hepatic and adipose 11β-HSD1 activities contribute in some way to insulin resistance and other features of the Metabolic Syndrome. However, the adipose activity appears to be correlated with a stronger phenotype of obesity and insulin resistance and therefore is likely the primary target for the treatment of insulin resistance. The hepatic 11β-HSD1 activity, although secondary, appears to be more important in improving lipid metabolism and controlling blood pressure. Several cases of human 11β-HSD1 deficiency have been reported. The ability of these subjects to convert cortisone to cortisol upon dexamethasone suppression was apparently compromised [ 154 - 158 ]. These patients appeared to be normal except for mild adrenal hyperplasia in some cases, and hirsutism, and elevated plasma cortisol levels [ 154 - 158 ]. Unfortunately, insufficient insulin sensitivity data have been reported with these patients. Although both obese and lean patients with 11β-HSD1 deficiency have been identified, it is not clear if the body weight is associated with 11β-HSD1 deficiency. However, polymorphisms in the 11β-HSD1 gene have been linked to adiposity in association studies with human subjects [ 159 , 160 ]. Inhibitors of GC action Given its important role in the Metabolic Syndrome, antagonizing GC action has been taken as an approach to treat some features of the Metabolic Syndrome. Targeting GR is a direct approach to antagonize the GC action. The global effect on GC action by this approach could lead to the activation of the HPA axis as well as blocking the anti-inflammatory function of GCs. Inhibition of 11β-HSD1 activity offers more tissue specificity due to the limited expression pattern of this enzyme. Inhibitors for both 11β-HSD1 and GR include naturally occurring and pharmaceutically developed compounds. The expected effects of 11β-HSD1 inhibition include reduced hepatic PEPCK and G6Pase expression to reduce hepatic glucose output; reduced adiposity and improved peripheral insulin sensitivity. Since 11β-HSD1 mediated GC action inhibits glucose-dependent insulin secretion [ 161 ] and the expression of 11β-HSD1 is significantly increased in diabetic islets [ 162 ], 11β-HSD1 inhibitors can potentially help reduce postprandial glucose excursion. Several inhibitors of 11β-HSD1 were described in the literature prior to the pharmaceutical targeting of this enzyme in recent years but none of them is selective and highly potent. Metyrapone, often used in the diagnosis of adrenal corticoid-related disease such as Cushing' syndrome, is a weak competitive inhibitor of 11β-HSD1 [ 163 ]. Other inhibitors include liquorice derivatives carbenoxolone (CBX) and glycyrrhetinic acid (GE) [ 164 ]. GE is more potent against the dehydrogenase activity and CBX is almost equally potent against activities of both directions (dehydrogenase and reductase). Although far more potent than other inhibitors, CBX and GE are not selective because they also inhibit 11β-HSD2. Chenodeoxycholic acid (CDCA) inhibits 11β-HSD1 with a potency of micromolar range but studies of its activity against 11β-HSD2 have generated conflicting results [ 165 - 167 ]. Although not selective, CBX has been used in human studies where it reduced glucose production during hyperglucagonemia largely due to its suppressive effect on glycogenolysis in lean male patients with type 2 diabetes [ 168 ]. Interestingly, CBX also improved verbal frequency and memory in healthy elderly men and patients with type 2 diabetes [ 169 ]. This is consistent with findings in 11β-HSD1 knockout mice [ 150 ]. Selective 11β-HSD1 inhibitors have been developed for pharmaceutical use in recent years. These inhibitors have been shown to be efficacious in diabetic animal models [ 170 - 173 ]. GR antagonists were developed on the rationale that activated GR stimulates PEPCK and G6Pase, the two key enzymes in hepatic gluconeogenesis that increases the hepatic glucose output [ 97 , 98 , 174 ]. Since hepatic gluconeogenesis in diabetics is increased [ 175 ], inhibition of hepatic GR action is expected for glucose lowering in diabetics. A well-known GR antagonist is RU-486, which was also found to have agonist activities [ 176 ]. Although efficacious [ 177 ], long-term systemic treatment with a GR antagonist may activate the HPA axis and increases cortisol secretion [ 178 ]. Other GR antagonists were also reported but without resolving the issue of HPA activation [ 179 ]. Selective inhibition of the hepatic GR activation in a non-systemic manner could provide advantages with no undesirable side effects. Liver selective targeting of the drug appears to be a good strategy [ 180 ]. Conclusions GCs are stress hormones with a wide spectrum of physiological effects and have been implicated in the pathophysiology of the Metabolic Syndrome. This notion has been supported by animal studies and clinical findings. The GC action appears to mediate certain aspects of the Metabolic Syndrome. In that regard, targeting key players in the GC action is expected to be a viable approach to treat some or all the features of the Metabolic Syndrome. However, cautions should be taken because the GC metabolism is regulated by the HPA axis and inhibition of GC pathways could lead to the activation of HPA axis and elevated adrenal cortisol secretion. To avoid the compensatory feedback response, efforts to separate the effect of GC modulators from HPA activity is needed. Although challenging, this could be achieved by tissue-specific modulation of GC action by targeting drugs to tissues of interest while sparing others, especially the CNS where HPA activation occurs. The availability of small molecule compounds will facilitate this type of studies in animal models to further dissect the regulatory function of the HPA axis and help assess whether tissue selective modulation of GC action without triggering the HPA axis is achievable. Declaration of competing interests The author is an employed researcher in a biopharmaceutical company. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548667.xml |
490026 | Genome-Wide Mapping of the Cohesin Complex in the Yeast Saccharomyces cerevisiae | In eukaryotic cells, cohesin holds sister chromatids together until they separate into daughter cells during mitosis. We have used chromatin immunoprecipitation coupled with microarray analysis (ChIP chip) to produce a genome-wide description of cohesin binding to meiotic and mitotic chromosomes of Saccharomyces cerevisiae . A computer program, PeakFinder, enables flexible, automated identification and annotation of cohesin binding peaks in ChIP chip data. Cohesin sites are highly conserved in meiosis and mitosis, suggesting that chromosomes share a common underlying structure during different developmental programs. These sites occur with a semiperiodic spacing of 11 kb that correlates with AT content. The number of sites correlates with chromosome size; however, binding to neighboring sites does not appear to be cooperative. We observed a very strong correlation between cohesin sites and regions between convergent transcription units. The apparent incompatibility between transcription and cohesin binding exists in both meiosis and mitosis. Further experiments reveal that transcript elongation into a cohesin-binding site removes cohesin. A negative correlation between cohesin sites and meiotic recombination sites suggests meiotic exchange is sensitive to the chromosome structure provided by cohesin. The genome-wide view of mitotic and meiotic cohesin binding provides an important framework for the exploration of cohesins and cohesion in other genomes. | Introduction Sister chromatid cohesion ensures that daughter cells inherit complete copies of their genome. Cohesion in eukaryotic cells is mediated by a multisubunit protein complex called cohesin. Cohesin consists of four proteins: Smc1, Smc3, Scc1/Mcd1, which is called kleisin and is the target of the protease separase, and Scc3. These proteins have recently been proposed to form a ring structure that encircles sister chromatids ( Gruber et al. 2003 ). Alternately, the ring may act as a snap ( Milutinovich and Koshland 2003 ). Cohesion is established during replication and maintained until metaphase in mitosis ( Uhlmann and Nasmyth 1998 ). All members of the cohesin complex are essential in Saccharomyces cerevisiae, since mutation results in the precocious dissociation of sister chromatids. Cohesion serves at least three roles in the cell with respect to accurate genome transmission. Firstly, cohesion proximal to the centromere facilitates biorientation of chromosomes with respect to the spindle ( Tanaka et al. 2000 ). Secondly, it prevents splitting of chromosomes once bipolar attachments are made ( Tanaka et al. 2000 ). Thirdly, cohesin bound along chromosome arms may be essential for proper chromosome condensation in yeast ( Guacci et al. 1997 ). In meiosis, cohesin at most arm sites disappears prior to the first nuclear division. The meiotic cohesin complex contains Rec8 instead of Scc1/Mcd1 ( Klein et al. 1999 ). Cohesion is maintained distal to crossovers between homologs, which links them and facilitates their biorientation on the meiotic I spindle. Cohesin is also maintained at pericentric regions, which is essential for biorientation of chromosomes at the spindle for the second nuclear division ( Buonomo et al. 2000 ). We are interested in understanding the cis determinants of cohesin binding. Cohesin-associated regions have been identified in yeast using chromatin immunoprecipitation. In these studies cohesin association with chromatin was followed at low resolution along the entire length of Chromosome III (3-kb intervals) or high resolution (300-bp intervals) at limited regions on Chromosome III, V, and XII ( Blat and Kleckner 1999 ; Megee et al. 1999 ; Tanaka et al. 1999 ; Laloraya et al. 2000 ). These studies showed associations of cohesin with specific regions of chromosomes; one of the regions of intense association is centromeres. In addition to the enrichment of cohesin in the pericentric region of Chromosome III, Blat and Kleckner (1999) found a spacing of cohesin-associated regions of 13 kb along the arms of Chromosome III. A similar spacing was observed in a limited region of Chromosome XII ( Laloraya et al. 2000 ). These studies also noted a correlation of cohesins with elevated AT content. The average size of the mapped sites was 0.8–1 kb ( Laloraya et al. 2000 ). Based on three sites mapped to high resolution, cohesin was proposed to associate with the boundaries of transcriptionally silent regions ( Laloraya et al. 2000 ). Despite these insights into cis determinants of cohesin binding, many aspects of the cohesin–DNA interaction remain obscure. The high resolution studies sampled a small portion of the genome, and the low-resolution analysis of Chromosome III does not address questions about the position of cohesin relative to smaller-scale genome features, such as individual transcription units. Furthermore, Chromosome III is the sex chromosome of budding yeast, and, similar to other organisms, it has unusual properties including large domains of repressed recombination, silent mating type loci, and different patterns of replication ( Reynolds et al. 1989 ; Wu et al. 1997 ). Some discrepancies between high- and low-resolution studies have emerged, including, for example, whether cohesin is found at telomeres ( Blat and Kleckner 1999 ; Laloraya et al. 2000 ). One approach to better understand cis determinants of cohesin binding is to analyze them across the whole genome. To obtain a genome-wide picture of cohesin binding to S. cerevisiae chromosomes at 1–2-kb resolution, we used a combination of chromatin immunoprecipitation (ChIP) and microarray methods, often referred to as ChIP chip technology. To aid identification of peaks of cohesin binding we developed a program, PeakFinder, for extraction of peaks from raw ChIP data. We further used this approach to map all the cohesin-binding sites on an “ectopic chromosome,” a yeast artifical chromosome containing a 334-kb insert from human Chromosome VII. Information from a large number of sites greatly facilitates the assessment of cohesin distribution and of the significance of correlations with many local properties of the genome, such as base composition and coding content. Furthermore, it allows us to evaluate the impact of several factors, such as strain background, transcription, and developmental programs like meiosis, on cohesin binding, and to test the predictions by engineering individual cohesin sites. Results Determining Sites at Which Cohesin Interacts with the Yeast Genome We used the genome-wide approach of ChIP chip to identify and evaluate cis determinants of cohesin sites (detailed protocol for ChIP available at http://www.uchsc.edu/sm/bbgn/images/ChIP%20protocol.htm ; see also Protocol S1 ). We began the study using Mcd1-18Myc as the protein target in the W303a strain background. Cells were arrested in metaphase by a temperature-sensitive mutation in CDC16, a subunit of the anaphase-promoting complex required to degrade Pds1p, a negative regulator of anaphase ( Yamamoto et al. 1996 ; Cohen-Fix and Koshland 1997 ). We were interested in determining potential correlations between cohesin binding and genome features such as base composition, transcriptional state, and known cis determinants of chromosome transmission. Earlier studies have used simple ratio “thresholds” to define binding sites in ChIP chip data ( Iyer et al. 2001 ; Lieb et al. 2001 ; Wyrick et al. 2001 ). A single genome-wide threshold would be of limited value in our experiments because (1) peaks representing the intensity of cohesin binding are much higher at pericentric regions than towards the end of chromosomes, therefore, a threshold constant would have the effect of skewing all the binding sites towards the centromere-proximal regions; (2) binding sites in ChIP chip data are frequently defined by several array elements, complicating the identification of cis determinants; and (3) much of the analysis has to be done manually. A better approach would be to use the local parameters of cohesin binding to identify the peaks. To aid such a task, we have written a Windows program, PeakFinder, which discerns and filters the peaks from a variable local background and maps the tip of the peak to a single array element. The program is freely available ( http://research.stowers-institute.org/jeg/2004/cohesin/peakfinder ). We validated our methods by comparing the results to previously collected data. Laloraya et al. (2000) discovered nine sites in the arms of Chromosomes III and XII using ChIP followed by semiquantitative PCR. All of those sites could be identified as peaks in the microarray data using PeakFinder (for Chromosome III see Figure 1 ). Blat and Kleckner (1999) mapped 23 cohesin sites to 3-kb resolution on Chromosome III. Although the number of identified peaks is increased in our data (33 versus 23), the increase can largely be accounted for by higher-resolution mapping. Qualitatively the results are comparable, including that (1) the peak height at CEN3 is the highest for the chromosome, (2) the height of the peaks declines towards the ends of the chromosome, and (3) peaks correlate with AT content ( Figure 2 ). Therefore, the results of previous studies are reproduced by our methods. Figure 1 Interactions between Cohesin and CHRIII in S. cerevisiae The centromere is indicated with a black circle; the smoothed data are indicated with a green line. 50-kb intervals are indicated by vertical grey lines. (A) Data generated from a cdc16 -arrest ChIP for Mcd1-18Myc in W303a. The midpoint of each feature is used to represent the log 2 of the median red:green ratio (left y-axis) with a black line, high firing replication origins are indicated with black triangles, and previously mapped CARC2, CARC1, CARC3, CARC4, CARC5, and CARC6 (Laloraya et al., 2000) correspond to peaks 9, 10, 29, 30, 31, and 32, respectively. Peaks are located and numbered by PeakFinder (with the exception of telomeres) using the parameters described in the Materials and Methods . (B) Smoothed data from cdc16 -arrest ChIP for Mcd1/Scc1-6HA in A364a. (C) Smoothed data from cdc16 -arrest ChIP for Smc3-6Myc in A364a. Figure 2 Visual Representation of the Interactions between Mcd1-18Myc and the S. cerevisiae Genome in W303a For each of the 16 chromosomes the centromere is indicated with a black circle, the smoothed data (based on the log 2 of the ratio) is indicated with a green line (left y-axis), and the percent GC is indicated by a red line (right y-axis). Vertical grey bars mark 50-kb intervals. Peaks are located and numbered by PeakFinder (with the exception of telomeres) using the parameters described in the Materials and Methods . For Chromosome XII, peaks 41 and 42 correspond to the previously described peaks CARL1 and CARL2 ( Laloraya et al. 2000 ). Three additional controls were performed to validate our methods. First, immunoprecipitation from a strain without any epitope tags on the cohesin complex was performed and did not yield any signal ( Megee et al. 1999 ). Second, for each ChIP performed with the anti-Myc antibody, a second ChIP was performed with the same chromatin solution in which the anti-Myc antibody was omitted. The immunoprecipitated DNA was subjected to semiquantitative PCR for centromere sequence (SGD coordinates 114318–114561) on Chromosome III. When the amplification was in the linear range, the difference in signal between the two templates was 11-fold ± SD 3.2, demonstrating that the enrichment for this particular sequence is specific to the interaction between the Myc epitope and the anti-Myc antibody. The ChIP samples were subjected to 20–25 cycles of random PCR amplification ( http://microarrays.org/protocols.html ; see also Protocol S2 ) prior to labeling and hybridization to microarrays. This amplification procedure was performed side by side on ChIP samples obtained with and without primary antibody. After 25 cycles of PCR, equal amounts of the samples were loaded on an agarose gel. The sample generated in the absence of primary antibody did not contain any detectable DNA, while the sample obtained with primary antibody generated a robust smear of DNA (unpublished data). While additional cycles of PCR did produce detectable DNA for the sample generated in the absence of a primary antibody, the lack of DNA under the amplification conditions used for the microarray experiment demonstrates that nonspecific immunoprecipitation of DNA was not a confounding factor for our microarray analysis. Third, the same chromatin solution was subjected to immunoprecipitation with an anti-Mif2 antibody. Mif2 is a centromere-binding protein. Centromeres in S. cerevisiae are approximately 125 bp. The peak of Mif2 binding spanned approximately 500 bp, as assessed by PCR amplification (see Weber et al. 2004 , Figure 4 ). This demonstrates that the shearing of fragments in the chromatin solution was sufficient to give resolution on the order of 500 bp in a case where this level of resolution is expected. Figure 4 Peaks and AT Content (A) The AT content for each array element was calculated and put into bins in 1% intervals (grey bars, left y-axis). The AT content for each array element that is a cohesin peak was also put into bins (red bars, right y-axis). (B) The AT content for each intergenic array element was put into bins in 1% intervals (grey bars, left y-axis). The AT content for each intergenic array element that is a cohesin peak was also put into bins (red bars, right y-axis). (C) The AT content for each convergent intergenic array element was put into bins (grey bars, left y-axis) and the AT content for each convergent intergenic array element that is a cohesin peak was also put into bins (red bars, right y-axis). To demonstrate the internal consistency and reproducibility of our data, we compared peaks of cohesin binding for Mcd1-18Myc in W303a (see Figure 1 A), Scc1/Mcd1-6HA in A364a ( Figure 1 B), and Smc3-6Myc in A364a ( Figure 1 C). There is good agreement between the location of cohesin peaks in different strain backgrounds (correlation coefficient = 0.76 for Mcd1/Scc1 ChIP in strains A364a and W303a). This is the relevant comparison for the data in Figure 1 here and the data in Figure 1 of Weber et al. (2004) , which shows the genome-wide results of ChIP for Scc1/Mcd1-6HA in the A364a background. The agreement is even stronger when different members of the cohesin complex are used as ChIP targets in the same genetic background (correlation coefficient = 0.96 for Mcd1/Scc1 and Smc3 ChIP in strain A364a). All data from individual arrays and datasets are available at http://research.stowers-institute.org/jeg/2004/cohesin/data/index.html and as supporting information ( Datasets S1–S58 ). Genomic Distribution of Cohesin The levels of cohesin on all the chromosomes are similar and follow a clear pattern: large regions of intense binding in the pericentric domain and less intense, smaller regions distributed in a semiperiodic manner throughout the arms. We evaluated whether cohesin was associated with cis determinants of chromosome transmission including centromeres, telomeres, and origins of replication. Cohesin shows a large (30–50 kb), dense region of binding in pericentric domains ( Figure 2 ). Although it has been proposed that telomeres do not associate with cohesin ( Blat and Kleckner 1999 ), we found that nine of the 32 telomeres in fact were associated with cohesin. However, the height of the peaks associated with telomeres and subtelomeric regions is lower than at internal regions, which may reflect lower affinity or occupancy of cohesin at these regions ( Figure 3 A). On Chromosome III, cohesin peaks appear to be associated with replication origins that have been functionally mapped ( Poloumienko et al. 2001 ) (see Figure 1 A, only the origins with the strongest signal are indicated). Cohesin enrichment at the centromeres clearly supports previous studies implicating a requirement for the coupled function of cohesion and the centromere in chromosome segregation ( Hill and Bloom 1987 ; Megee et al. 1999 ), while the significance of cohesin association with telomeres and origins is unclear. Figure 3 Features of Peaks (A) Using all cohesin-binding peaks within 40 kb of a telomere ordered based on distance from the telomere, we calculated a five-point moving average for distance in kilobases from the telomere (x-axis) and plotted this as a function of the five-point moving average of the log 2 value for the associated peaks (y-axis). (B) Chromosome length (x-axis) is plotted as a function of the number of cohesin peaks (y-axis). A line was fitted using the least squares method and R 2 = 0.96. (C) The distance between peaks was put into 1-kb bins; the average distance between peaks is 10.9 kb and the median is 9.3 kb. While some cohesin is associated with the known cis determinants of chromosome transmission, the vast majority of sites are not. The number of cohesin-binding peaks per chromosome shows an excellent correlation to chromosome length ( R 2 = 0.96; Figure 3 B). The mean distance between sites was 10.9 kb, with a standard deviation of 6.7 kb ( Figure 3 C). There are 50 regions of the genome with large gaps between neighboring peaks (24 kb or greater, i.e., more than 2 s.d. from the mean); these appear to be randomly scattered throughout the arms of the larger chromosomes, and are never located on any of the four smallest chromosomes. The spacing of peaks is conserved for the most part in pericentric regions, with an additional “baseline” level of binding. Cohesin distribution therefore appears to be nonrandom with a tendency for even distribution over the genome. Cohesin Tends to Bind AT-Rich Sequences Cohesin peaks were strongly associated with AT-rich regions ( Figure 4 ). We found that 810 of the 1,095 array elements defined as cohesin-binding sites have AT content above the yeast median of 62.6% ( p < 0.0001) ( Figure 4 A). Cohesin peaks are significantly associated with intergenic regions, with 765 out of 1,095, or 70%, of all peaks located in such regions ( p < 0.0001) even though intergenic regions make up only 27% of the genome length. Intergenic regions in S. cerevisiae are more AT-rich than open reading frames (ORFs). Therefore, we tested whether the AT bias could be explained by the bias towards binding intergenic sequences by comparing the AT content of all intergenic regions with the AT content of intergenic regions associated with cohesin (598 out of 765 are above the median, p < 0.0001; Figure 4 B). The peaks observed at ORFs are also higher in AT content than ORFs on average ( p = 0.0005). Thus, AT content appears to be a major determinant for cohesin association. We observed local oscillations of AT content in a 5-kb sliding window, which corresponded to cohesin-binding peaks in chromosome arms (see Figure 2 ), thus extending to the whole genome the observation for Chromosome III ( Blat and Kleckner 1999 ). Furthermore, all 16 pericentric regions have local peaks of AT content ( Figure 2 ). Interestingly, the sequence elements associated with cohesin in the pericentric domain contain nearly equal numbers of ORF and intergenic sequences, as might be expected if binding is mainly directed by the centromere and base composition and disregards other genomic features such as transcription units ( Weber et al. 2004 ). Distribution of Cohesin on a YAC The semiregular spacing of cohesin and the correlation with local oscillations of base composition suggested that AT content and/or a measuring mechanism might control cohesin distribution on the chromosome. In order to test these possibilities, we used a nonessential ectopic yeast artificial chromosome (YAC). We used ChIP followed by quantitative PCR to map cohesin-binding sites in a YAC containing 334 kb of human DNA from Chromosome VII. The pericentric region on the right end of the YAC shows a broad (approximately 35 kb), intense association with cohesin. This is similar to the cohesin association with endogenous pericentric regions. However, the spacing of cohesin does not have the same periodic nature as on endogenous chromosomes. For example, the leftmost 83 kb of the YAC contains only one peak of cohesin binding, resulting in two large gaps for cohesin of 38 and 45 kb ( Figure 5 A). These two gaps are larger than any of the gaps found on the smaller endogenous yeast chromosomes, which have a comparable size to the YAC. The human DNA fragment does not contain oscillations of AT content similar to those observed in the yeast genome, nor does the pattern of cohesin association appear to reflect base composition in this context. Thus, the difference in the distribution of cohesin-binding sites in the arms of the YAC supports the idea that sequence contributes to cohesin distribution in yeast and that evolution has selected for an even distribution on endogenous chromosomes. Figure 5 Cohesin Sites Mapped Using ChIP Followed by Semiquantitative PCR with Primers at 1-kb Intervals in a YAC Containing Human DNA (A) Cohesin binding for the entire YAC is shown. (B) Cohesin binding in the region spanning 135–180 kb is shown for the wild-type YAC (black diamonds) and for the YAC containing a replacement of the sequences at 156–162 kb with the gene encoding geneticin resistance (grey squares). To further test the contribution of sequence to cohesin binding, we took advantage of the fact that none of the human sequences were essential for yeast survival. When we replaced the region from 156 to 162 kb, which contains a cohesin-binding site, with the gene encoding for geneticin resistance, this region was no longer associated with cohesin ( Figure 5 B). Neighboring regions were unaffected, and de novo cohesin binding was not observed. This suggested that some property of the sequence, rather than its precise location or context, was responsible for cohesin binding. Transcription and Cohesin Inspection of the intergenic regions associated with cohesin revealed a strong preference towards the intergenic regions in which transcription is converging, and additionally, an extreme bias against association with intergenic regions in which transcription is diverging. Among the cohesin-associated intergenic regions that could be assigned to a category, 86% were in intergenic regions with converging transcription, 12% were in intergenic regions with surrounding unidirectional transcription, and only 2% were in intergenic regions that are between two divergently transcribed genes. In contrast, the genome as a whole has these regions in approximately a 1:2:1 ratio, respectively, making this result highly statistically significant ( p < 0.0001). In fact, 39% of the convergent intergenic regions in the genome have peaks of cohesin binding, and nearly half of all cohesin-binding sites are in convergent intergenic regions. These percentages approach the predictive power of consensus sequences for identifying binding sites of their cognate transcription factor ( Chu et al. 1998 ; Lieb et al. 2001 ). Convergent intergenic regions have high AT content compared to the genome at large; the bias of the intergenic sequences associated with cohesin can be partly explained by the AT bias of convergent intergenic regions (see Figure 4 C). Of the sites, 865 of 1,095 can be explained by one or more of the following three factors: (1) location within 25 kb of a centromere, (2) above average AT content, or (3) location in an intergenic region with converging transcription. This leaves 230 sites unexplained. Of these, 43 are intergenic. Most of these are simply difficult to assign to a transcriptional category. Interestingly, of the 230 “unexplained” sequences, 187 are in ORFs, which is more than half of all the ORFs associated with cohesin. These ORFs do not appear to have any unifying theme with regard to function, dubiousness, transcription level, or essentiality. The 187 peaks have similar height to other peaks. Thus, unlike the cis determinants in the intergenic regions, the factors within the ORFs that enable cohesin binding are not well understood. Attempts to identify a genome-wide linear consensus binding site for cohesin using BioProspector ( Liu et al. 2002 ), MobyDick ( Bussemaker et al. 2000 ), AlignACE ( Hughes et al. 2000 ), and MEME ( Bailey and Elkan 1994 ) did not return any sequence model with predictive value. However, when we took the group of “unexplained” ORF sequences and looked for a common motif using BioProspector, we identified two repetitive sequences that were strongly enriched relative to the genome: (CAR) 5 ( p = 10 −65 ) and (GAN) 10 ( p = 10 −76 ). The p values reflect the significance of these sequences as calculated based on a Monte Carlo simulation for the yeast genome. These sequences were rare in intergenic regions in yeast. These sequences were not present in the YAC. Further experiments will be required to determine whether these repeats are targeted by cohesin. Interestingly, human cohesin has been localized in Alu repeats ( Hakimi et al. 2002 ), suggesting that repetitive DNA may be prone to a particular structure or chromatin configuration preferred by cohesin, or may accumulate in regions that are bound by cohesin. Alu repeats do not bear any obvious similarity to the sequences we identified. Binding of cohesin may help modulate transcription of these repeated sequences. The Mechanism of the Negative Association between Transcription and Cohesin Binding The link between intergenic tail-to-tail regions and cohesin suggests a general incompatibility between transcription and cohesin association . The observations that transcription can disrupt centromeric cohesin and results in chromosome missegregation and cell death ( Tanaka et al. 1999 ), and that cohesin binds at the boundaries of silent chromatin in several loci ( Laloraya et al. 2000 ) are in agreement with this. We analyzed whether changing the transcriptional program could change the association of cohesin with a locus. We grew cultures with either 2% glucose or 2% galactose as the carbon source and arrested them in metaphase using nocodazole. The main difference between metaphase arrest in the presence of nocodazole and in cdc16 -ts is that cohesin binding at pericentric regions is increased in the former case (unpublished data). We carried out ChIP chip analysis in parallel with monitoring gene expression in the same cells using ORF arrays. Of the regions where gene expression changed 5-fold or more, only two regions were associated with cohesin. One peak of cohesin binding in glucose was associated with the promoter of the GAL2 gene ( Figure 6 A), which was induced 42-fold in galactose. This had a dramatic effect on the local profile of cohesin binding ( Figure 6 B). The promoter region of GAL2 became a trough of cohesin binding, and the single peak observed in glucose was split into two peaks, in effect adding a new peak to the region. The second region was an uncharacterized ORF, YDL218W, a membrane protein distantly related to secretory factor NCE102 /YPR149W and to metazoan synaptogyrin family (unpublished data). Expression of YDL218W was induced 11-fold in the presence of glucose compared to galactose. This ORF is associated with cohesin in the presence of galactose, but this association is reduced when glucose is present (unpublished data). The peaks surrounding both regions were unaffected. These results demonstrate that high levels of transcription are incompatible with cohesin binding. It also supports the results observed with the human DNA that neighboring cohesin sites behave independently. Figure 6 Transcription Affects the Cohesin Peak at the Promoter of GAL2 SGD coordinates 260–320 kb (x-axis) and a gene map are depicted for Chromosome XII. The strain 1827-22D (isogenic to the strain in Figure 1 except CDC16 ) was grown with either 2% glucose (A) or 2% galactose (B) as the carbon source. Cultures were arrested with nocodazole, and ChIP chip was performed. The smoothed data (as the log 2 of the ratio) is depicted in green, the peaks found by PeakFinder are indicated with black dots, and the region corresponding to the GAL2 promoter is indicated with a grey bar. Transcription of GAL2 is up-regulated 42-fold in (B). There are a number of mechanisms that could account for the incompatibility between transcription and cohesin binding. Transcription of a region during G1 may prevent new association of cohesin, or transcription may displace cohesin in G2. We explored the mechanistic link between transcription and cohesin association at the previously characterized cohesin sites cohesin-associated region on Chromosome III (C) (CARC1) and cohesin-associated region on Chromosome XII (L) (CARL2) ( Laloraya et al. 2000 ) by inserting 0.8 kb of CARC1 of 1.4 kb of CARL2 into a plasmid construct next to a galactose-inducible promoter (pGAL1-10). Strains containing one of these two plasmids were grown with 2% raffinose as the carbon source, arrested in G1 with alpha factor, and then released from G1 in the presence of nocodazole, producing a metaphase arrest. Galactose was added to half the culture. One hour after the addition of galactose, cultures were fixed with formaldehyde and processed for ChIP. Semiquantitative PCR was used to monitor the distribution of cohesin. With the appropriate use of primers, cohesin association with CARC1 and CARL2 on the plasmid and the endogenous locus could be distinguished. Galactose-induced transcription had no effect on association of cohesin with the endogenous loci (unpublished data) but disrupted cohesin associated with the 5′ end of both CARC1 and CARL2 on the plasmid ( Figure 7 A). This result demonstrates that transcription during G2 can displace cohesin. Figure 7 Effect of Transcript Elongation on Cohesin Associated with CARC1 and CARL2 Located on a Plasmid Next to a Galactose-Inducible Promoter (A) The fold reduction in cohesin binding in the presence (+) or absence (−) of galactose-induced transcription is depicted as a function of the 5′ or 3′ end of the locus. (B) The fold reduction in cohesin binding at CARL2 during galactose-induced transcription in the presence (+) or absence (−) of thiolutin, an inhibitor of transcript elongation. The displacement of cohesin may be due to a competition between RNA pol II/chromatin–remodeling factors and cohesin for DNA, or transcript elongation may remove cohesin. We tested if the incompatibility was dependent on transcript elongation. A culture was arrested with nocodazole, and galactose-responsive transcription was induced by the addition of galactose, as described above. Thiolutin was added to half of this culture immediately prior to the addition of galactose. Thiolutin inhibits transcript elongation but presumably does not inhibit the binding of RNA pol II ( Parker et al. 1991 ). The effect as monitored at CARL2 was dependent upon elongation since the addition of thiolutin prevented the displacement of cohesin ( Figure 7 B). This result demonstrates that the binding of RNA pol II per se does not affect cohesin binding, but transcript elongation can displace cohesin within the G2 portion of a single cell cycle. Meiotic Cohesin The transcriptional program of a cell changes under different conditions, such as the developmental program of sporulation. We used ChIP chip to analyze the location of the meiosis-specific cohesin complex ( Klein et al. 1999 ; Watanabe and Nurse 1999 ). The protein composition of the meiosis-specific complex has been described, but no information on the cis determinants of this complex has been reported. We expressed Rec8-3HA in SK1, a rapidly and synchronously sporulating strain, and analyzed the location of the cohesin–DNA complex in ChIP experiments ( Figure 8 ). The pattern of cohesin association in meiotic and mitotic cells appears to be similar (correlation coefficient = 0.77 across SK1 genome comparing Rec8 to Mcd1; see Figure 8 for coordinates 295–345 kb and 440–460 kb on Chromosome XII; additional data regarding the timecourse of sporulation is provided in Figure 9 ). In both cases, pericentric regions contain broad, intense regions of cohesin association and there are nonrandomly spaced cohesin sites in the arms. Figure 8 Meiotic Cohesin DSB data are shown in red, Rec8 data in black, and Mcd1 data in grey. (A) Ratios for meiotic cohesin are compared to mitotic cohesin in SK1 for kilobasepairs 440–461 on Chromosome XII. For the mitotic culture, cells were arrested with nocodazole. For meiotic cells, timepoints were collected every 2 h from hour 4 to hour 12 after transfer to SPM. The median ratio value was used to represent the data. Meiosis is slower in an SK1 strain with an HA-epitope-tagged Rec8 than in a wild-type strain (see Figure 9 ). The gene structure for this locus is shown below the graph, with genes encoded by the Watson strand labeled on top and genes encoded by the Crick strand labeled on the bottom. (B) Ratios for meiotic cohesin are compared to mitotic cohesin and DSBs for kilobasepairs 295–345 on Chromosome XII. Figure 9 Meiotic Timecourse for an SK1 Strain Containing Rec8-3HA Cells were collected at the indicated timepoints throughout meiosis using the same experimental regime used to collect the binding sites of Rec8-3HA presented in Figure 8 . The epitope tag appears to slow meiosis by 3–4 h as compared to an untagged strain. Three assays were developed to monitor culture synchrony during meiosis. (A) FACS profile of the REC8-3HA strain. Aliquots of cells were fixed with 70% EtOH, followed by FACS analysis. (B) Nuclear division of the REC8-3HA strain. Aliquots of cells were fixed with 1% formaldehyde for 1 h at room temperature. Nuclear DNA was stained by DAPI and visualized under a fluorescence microscope. At least 200 cells were scored at each timepoint. (C) Rec8-3HA protein level. Protein extracts were prepared and subjected to SDS-PAGE and Western blot. The Rec8-3HA protein level was detected by an anti-HA antibody (12CA5). The same blot was stripped and reprobed with anti-β-tubulin antibody to detect the level of β-tubulin, which served as a loading control. PCD1 has been shown to be a cohesin-binding site ( Laloraya et al. 2000 ). Binding to this site is diminished for meiotic cohesin (see Figure 8 A). The transcription of this gene is induced early in meiosis, which may explain why cohesin binding to this site is diminished ( Chu et al. 1998 ). Other genes show a similar pattern, namely that they are binding sites for cohesin in mitotic cells, but their transcription is induced early in meiosis, and they do not appear to be binding sites for the Rec8-containing cohesin complex (e.g., YPR006C, YDL238C, and YER179W ). This suggests that binding of the meiotic complex, like the mitotic complex, is not compatible with transcription. We compared the location of meiotic cohesin to the location of the double-strand breaks (DSBs) that initiate meiotic recombination ( Gerton et al. 2000 ). We found a negative correlation (correlation coefficient = −0.26); DSBs tend to occur in regions where meiotic cohesin is absent, and meiotic cohesin tends to be located in regions that contain low levels of DSBs (see Figure 8 B). Cohesin has been shown to be required for the formation of the axial elements that become the lateral elements of the proteinaceous structure known as the synaptonemal complex (SC) ( Klein et al. 1999 ). The SC organizes meiotic chromosomes and aids interhomolog recombination. In fact, meiotic cohesin has been shown to be required for meiotic recombination ( Klein et al. 1999 ). This result suggests that recombination proteins can recognize chromosome structure/organization provided by cohesin. The most notable difference between meiotic and mitotic cohesin is at the ribosomal DNA (rDNA) locus. (Nocodazole arrest does not affect cohesin binding in the rDNA in A364a or W303a, unpublished data) The rDNA is encoded in an approximately 1–2-Mb region on the right arm of Chromosome XII consisting of 100–200 tandem copies of a 9.1-kb repeat. There is a peak of cohesin binding that localizes to the left border of the rDNA repeat ( Laloraya et al. 2000 ) that is absent for meiotic cohesin (see Figure 8 A). Although information regarding the transcription of rDNA in meiosis is unavailable, genes involved in the processing of the 35S transcript, such as ROK1, RRS1 , and EBP2 ( Wade et al. 2001 ), are down-regulated 5- to 15-fold by 0.5 h in meiosis, and ribosomal protein genes are also down-regulated an average of 5-fold ( Chu et al. 1998 ), suggesting that transcription of this region is significantly reduced in meiosis. The rDNA is located in the nucleolus in meiotic cells and is associated with proteins that repress interhomolog recombination ( Petes and Botstein 1977 ; San-Segundo and Roeder 1999 , 2000 ). This region may have a chromatin structure in meiosis that suppresses transcription, recombination (so as to avoid chaotic exchange between repeated elements), and cohesin binding. Discussion We have used ChIP chip to map, to 1–2-kb resolution, the genome-wide pattern of cohesin association under several different growth conditions (metaphase arrest by cdc16 -ts or nocodazole, galactose versus glucose as a carbon source, and induction of meiosis) and in three different yeast-strain backgrounds (W303a, SK1, and A364a). Using PeakFinder, a program that assesses cohesin binding by comparison of signal to variable local background, we find that the majority of cohesin-binding sites are remarkably constant under these different circumstances. Distribution of cohesins throughout the genome appears to depend on a combination of base composition, sequence, and transcriptional activity. We find evidence for three types of cohesin sites in the genome: (1) the centromere and pericentric domain, (2) intergenic regions in chromosome arms, and (3) ORFs in chromosome arms. The association of cohesin with these three types of sites is subject to different genomic parameters. Cohesin at centromeres and pericentric regions is spread over a broad domain with an elevated “baseline” level and is not affected by the natural transcriptional and coding status. Much of the cohesin in chromosome arms is located in transcriptionally converging intergenic regions. ORFs in chromosome arms where cohesin is found are enriched for repetitive sequences. This suggests that there may be three mechanisms to load cohesin, consistent with what has been proposed for cohesin in meiotic chromosomes for S. pombe ( Kitajima et al. 2003 ). A unifying feature of all three types of sites is high AT content. Pericentric regions contain the most intense and broadest levels of cohesin in the genome (for a more complete analysis of pericentric cohesin see Weber et al. [2004] ). This finding is consistent with a model in which a centromere contains determinants of two opposing processes: (1) pulling the chromosomes apart, via the assembled kinetochore attached to a microtubule, and (2) keeping chromosomes together, via pericentric cohesion. The intensity and breadth of cohesin binding at pericentric regions is similar for all chromosomes, implying microtubules pull all chromosomes with comparable force, regardless of their length. On the other hand, the number of binding sites per chromosome is proportional to chromosome length. This result implies that arm cohesion is not a direct measure of the force exerted by spindle microtubules, and may serve a different function, for instance, to achieve similar levels of condensation. The model in budding yeast that cohesin can participate in genome maintenance in two ways, namely condensation via arm cohesin and biorientation via pericentric cohesin, is intriguing in light of the recent finding that cohesin complexes with different subunits are found on arms and pericentric regions on meiotic chromosomes in S. pombe and apparently serve different functions ( Kitajima et al. 2003 ). Cohesin cannot stay bound to DNA in the face of active transcript elongation based on three independent cohesin sites (promoter of GAL2, CARC1, and CARL2). If cohesin and transcript elongation were incompatible, then we would also expect to find sites biased towards intergenic regions, which we do. However, we find a strong bias towards intergenic regions with converging transcription, and a bias against intergenic regions with surrounding unidirectional transcription or diverging transcription, suggesting that intergenic regions with converging transcription may have especially low transcription. These regions may have evolved particularly strong transcriptional stops since they are quite short on average and the cell may need to avoid transcription from one side extending to the other to prevent the synthesis of antisense RNA. The protection of sequence elements important for the replication and segregation of eukaryotic chromosomes from transcription may be a general necessity for their proper function in vivo. For instance, transcription through an autonomous replicating sequence ( Snyder et al. 1988 ) or a centromere ( Hill and Bloom 1987 ) disrupts their function. The observed antagonistic relationship between transcription and cohesin binding in chromosome arms can be explained in two ways. Firstly, transcript elongation may be directly responsible for displacing cohesin. In this type of model, cohesin loading/binding is random, and transcription (and possibly other DNA metabolic processes) “pushes” cohesin into place or strips cohesin from inappropriate locations in each cell cycle. Secondly, transcript elongation may be indirectly responsible for localizing cohesins, for example by accumulation of “nonpermissive” chromatin in transcribed regions and “permissive” chromatin in nontranscribed regions. This type of genome-wide demarcation of transcription units has been shown to occur in S. cerevisiae ( Nagy et al. 2003 ) and may depend on nucleosomes ( Lee 2004 ) and histone variants. The chromatin remodeling complex RSC (Remodels the Structure of Chromatin) has recently been shown to be important for establishment of cohesin in chromosome arms ( Baetz et al. 2004 ; Huang et al. 2004 ). The preferential location of cohesin in heterochromatin in S. pombe also supports the idea of chromatin modification/structure as the basis for cohesin localization ( Bernard et al. 2001 ; Nonaka et al. 2002 ). The possibility also exists that cohesin itself may influence transcriptional status and act as a transcriptional boundary ( Hagstrom and Meyer 2003 ; Rollins et al. 1999 ). Despite the subunit difference between the meiotic and mitotic cohesin complex, we find that the association of cohesin with DNA in meiotic cells is similar to that in mitotic cells. In addition, we find that the constitutive peaks of meiotic cohesin binding are negatively correlated with DSB sites. This negative correlation is consistent with the model proposed by Blat et al. (2002) for the relationship between recombination and cohesin. This model suggests that cohesins are at meiotic chromosome cores and that recombination occurs in chromatin loops emanating from these cores where the part of the loop undergoing recombination is transiently localized to the axis. Thus, the recombination machinery can sense chromosome organization provided by cohesin. The differences in binding of meiotic and mitotic cohesin suggest that the location of the meiotic complex is also dependent on gene transcription. Hence, meiotic recombination in a given organism may be somewhat dependent on the spacing of cohesin as established in premeiotic S phase, which is in turn responsive to transcription. The genome-wide distribution of DSBs is positively correlated with regions of high GC content, divergent promoters, and transcription factor binding ( Gerton et al. 2000 ). Thus transcription, recombination, and cohesion all display intimate connections to chromosome and chromatin structure. Genome-wide studies of protein–DNA complexes afford a better understanding of the role of these complexes in the biology of an organism and its genome. In the process of analyzing the first genome-wide map of cohesin in any organism, we developed PeakFinder, a program able to sensitively identify binding sites of protein–DNA complexes in their local genomic environment, and potentially useful for analysis of any other genome-wide measurements. While budding yeast appears to have largely opted for placement of cohesin in AT-rich, transcriptionally inactive regions, other organisms with much longer and more complicated transcriptional units, different base composition properties, or different levels of condensation may employ different strategies for the placement of cohesin, which may in turn affect the stability of those genomes. The genome-wide analysis of cohesin in S. cerevisiae will serve as a useful framework upon which to explore attributes of cohesin localization in higher eukaryotes. Materials and Methods ChIP methods. ChIPs were performed as previously described ( Meluh and Koshland 1997 ; Laloraya et al. 2000 ). Semiquantitative PCR analysis was performed as previously described ( Laloraya et al. 2000 ). ChIP using the same experimental regime in a strain lacking the Myc epitope was performed and did not yield any appreciable signal ( Megee et al. 1999 ). Cell culture. For the meiotic timecourse, cultures were grown in YPA, then transferred to SPM. Timepoints were removed for ChIP at 4, 6, 8, 10, and 12 h after transfer to SPM. Nocodazole-mediated arrest was accomplished by adding nocodazole to a final concentration of 15 μg/ml to the media. All cultures were grown at 30 °C. Shifting cultures to 37 °C in prewarmed media induced metaphase arrest in cdc16 -ts cells. DNA amplification, labeling, and hybridization. Preparation of Cy5- and Cy3-labeled DNA, hybridization, and analysis were performed as previously described ( Bohlander et al. 1992 ; Gerton et al. 2000 ). The polyL-lysine-coated spotted glass microarrays used in this study contained each ORF and each intergenic region in the yeast genome as individual spots ( Iyer et al. 2001 ). For each experimental condition, a minimum of two independent immunoprecipitations was performed. DNA from the immunoprecipitation was labeled with Cy5 and competitively hybridized with total genomic DNA labeled with Cy3. Hybridizations with fluor reversal were also performed for DNA from at least one of the immunoprecipitations for each condition. At least three arrays were analyzed per experimental condition, and the median values were used to represent the dataset. Hybridizations were performed at 63 °C overnight under standard conditions, and slides were washed successively with 0.6X SSC/0.03% SDS and then 0.06X SSC prior to scanning (see also http://microarrays.org ). The meiotic experiments were done in the SK1 strain background and although two independent timecourses were performed, the results from a single representative timecourse were used for analysis. The resolution of these genome-wide maps is limited by (1) the shear size of the DNA fragments (range of 200–1000 bp) and (2) the size of the elements on our microarrays (mean of 0.9 kb). We do not expect these fragment-size distributions to introduce a significant bias in our mapping effort since the previously estimated size of a cohesin-binding region at an arm site is 0.8–1.0 kb ( Laloraya et al. 2000 ). Computational methods. The arrays were scanned using an Axon Instruments (Union City, California, United States) 4000B scanner and quantitated using GenePix 4.0. Results were stored in the AMAD database. Data were normalized and filtered by requiring intensity to be 200 or more, and spots to have a correlation coefficient of 0.5 or more. For analysis purposes, any feature with less than two measurements was excluded (with the exception of the meiotic timecourse). Data were analyzed using PeakFinder, a program developed specifically for finding peaks in ChIP data, but generally applicable for plotting any measurement against genomic coordinates, smoothing the curves, and annotating peaks on the basis of local properties of the curve. Extensive documentation for PeakFinder is available at http://research.stowers-institute.org/jeg/2004/cohesin/peakfinder/ . Briefly, PeakFinder takes the fluorescence ratios and samples them at the indicated interval of basepairs. The log 2 of the data are then smoothed. The first derivative of the smoothed line is used to identify peaks, and the absolute value of the corresponding peak is then extracted from the raw data (this is necessary because the nature of the smoothing algorithm dampens the peak height). PeakFinder allows filtering of peaks based on the parameters of the peak. For example, cohesin peaks analyzed in the cdc16 -ts dataset were identified using the following set of parameters: (1) sampling log 2 -transformed ratios at 100-bp intervals, (2) smoothing over eight rounds using a nine-point Gaussian-weighted moving average, and (3) filtering of peaks with a left and right rise of less than 0.1 and a height less than 0.4 (log 2 space). These conditions identified all peaks mapped to high resolution on Chromosomes III and XII ( Laloraya et al. 2000 ). A current limitation of PeakFinder is that it is unable to identify one-sided peaks; therefore telomeres were manually inspected for cohesin binding. PeakFinder is written in Delphi, runs on a Windows platform, and is distributed under the GNU General Public License. YAC. PCR primers were designed to amplify 150–300-bp sequences, at 1-kb intervals along the entire length of the 1572 YAC ( Green et al. 1995 ). Nucleotides 1–3,683 contain the vector sequences from pYAC4 including the telomere and URA3 . Nucleotides 326,702—332,707 contain vector sequences from pYAC4 including the centromere, TRP1, and ARS1 . The YAC was introduced into the strain 1377 A1 4B, two independent cultures were grown to exponential phase, and nocodazole was added. After 3 h of growth at 23 °C, more than 90% of cells were arrested in metaphase. Cultures were processed for ChIP as described previously. Thiolutin. 0.8 kb from CARC1 and 1.4 kb from CARL2 were cloned into pUNI and then recombined with pYCE to form pCM34 and pCM38. This places the pGAL1-10 promoter immediately adjacent to cohesin-associated regions. pCM34 and pCM38 were introduced into 1377 A1 4B by transformation. Strains with pCM34 and pCM38 were initially grown in complete medium lacking uracil with raffinose as a carbon source. This medium selects for retention of the plasmids and prevents transcription from the Gal-inducible promoter. These cultures were diluted approximately 100-fold in YEP raffinose and grown to 7 × 10 6 /ml. Cultures were arrested in G1 with alpha factor, released from G1 in the presence of nocodazole, and grown for 3 h to generate an M phase arrest. Cultures were split in two, and one half received galactose to a final concentration of 4%. One hour after addition of galactose, cultures were fixed and processed for ChIP. Experiments with thiolutin were performed as described above except thiolutin was added to a final concentration of 3 μg/ml just prior to galactose addition. Primers were generated that amplified 5′ and 3′ regions of CARC1 and CARL2 in the endogenous locus and on pCM34 and pCM38. Reduction in cohesin binding was expressed as the ratio of (1) the amount of cohesin bound to the 5′ or 3′ ends of the CAR on the plasmids to (2) the amount of cohesin bound to the 5′ or 3′ regions of the CAR in the genome. Supporting Information Datasets S10–S58 correspond to the individual GenePix results (GPR) files for each array performed. For each dataset we have listed the Cy3 channel sample and the Cy5 channel sample. Dataset S1 W303 Strain Arrested by cdc16 -ts with ChIP Performed for Mcd1-18Myc File cdc16_Mcd1-18Myc_W303. (483 KB TXT). Click here for additional data file. Dataset S2 A364a Strain Arrested by cdc16 -ts with ChIP Performed for Mcd1-6HA File cdc16_Mcd1-6HA_A364a. (957 KB TXT). Click here for additional data file. Dataset S3 A364a Strain Arrested by cdc16 -ts with ChIP Performed for Smc3-6Myc File cdc16_Smc3-6Myc_A364a. (701 KB TXT). Click here for additional data file. Dataset S4 W303 Strain Grown in Galactose and Arrested by Nocodazole with ChIP Performed for Mcd1-18Myc File Mcd1_18Myc_W303_NZgalCHIP. (406 KB TXT). Click here for additional data file. Dataset S5 W303 Strain with Mcd1-18Myc Grown in Galactose and Arrested by Nocodazole with RNA Harvested for Gene Expression File Mcd1-18Myc_W303_NZgal_exp. (196 KB TXT). Click here for additional data file. Dataset S6 W303 Strain with Mcd1-18Myc Grown in Glucose and Arrested by Nocodazole with RNA Harvested for Gene Expression File Mcd1-18Myc_W303_NZglu_exp. (221 KB TXT). Click here for additional data file. Dataset S7 W303 Strain Grown in Glucose and Arrested with Nocodazole with ChIP Performed for Mcd1-18Myc File Mcd1-18Myc_W303_NZgluChIP. (694 KB TXT). Click here for additional data file. Dataset S8 SK1 Strain Arrested with Nocodazole with ChIP Performed for Mcd1-3HA File Mcd1-3HA_SK1_NZ. (339 KB TXT). Click here for additional data file. Dataset S9 SK1 Strain in Meiosis with ChIP Performed for Rec8-3HA File Rec8-3HA_SK1. (710 KB TXT). Click here for additional data file. Dataset S10 SIMRUP2_147 Cy3 = ChIP cdc16 -ts Mcd1-6HA in A364a; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S11 SIMRUP2_170 Cy3 = ChIP cdc16 -ts Mcd1-6HA in A364a; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S12 SIMRUP2_171 Cy3 = ChIP cdc16 -ts Mcd1-6HA in A364a; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S13 SIMRUP2_178 Cy3 = genomic DNA; Cy5 = ChIP cdc16 -ts Mcd1-6HA in A364a. (4.6 MB XLS). Click here for additional data file. Dataset S14 SIMRUP2_180 Cy3 = genomic DNA; Cy5 = ChIP cdc16 -ts Mcd1-6HA in A364a. (4.6 MB XLS). Click here for additional data file. Dataset S15 SIMRUP2_187 Cy3 = ChIP cdc16 -ts Smc3-6Myc in A364a; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S16 SIMRUP2_190 Cy3 = ChIP cdc16 -ts Smc3-6Myc Mcd1-6HA ChIP for HA in A364a; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S17 SIMRUP2_191 Cy3 = ChIP cdc16 -ts Smc3-6Myc Mcd1-6HA ChIP for Myc in A364a; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S18 SIMRUP2_226 Cy3 = genomic DNA; Cy5 = ChIP cdc16 -ts Smc3-6Myc in A364a. (4.6 MB XLS). Click here for additional data file. Dataset S19 SIMRUP2_244 Cy3 = genomic DNA; Cy5 = ChIP cdc16 -ts Smc3-6Myc Mcd1-6HA ChIP for HA in A364a. (4.6 MB XLS). Click here for additional data file. Dataset S20 SIMRUP2_254 Cy3 = genomic DNA; Cy5 = ChIP cdc16 -ts Smc3-6Myc Mcd1-6HA ChIP for Myc in A364a. (4.6 MB XLS). Click here for additional data file. Dataset S21 UP2_13 Cy3 = genomic DNA; Cy5 = ChIP cdc16 -ts Mcd1-18Myc in W303. (2.5 MB XLS). Click here for additional data file. Dataset S22 UP2_19 Cy3 = genomic DNA; Cy5 = ChIP cdc16 -ts Mcd1-18Myc in W303. (2.5 MB XLS). Click here for additional data file. Dataset S23 UP3_186 Cy3 = ChIP 6h Rec8-3HA in SK1; Cy5 = genomic DNA. (2.6 MB XLS). Click here for additional data file. Dataset S24 UP3_187 Cy3 = ChIP 8h Rec8-3HA in SK1; Cy5 = genomic DNA. (2.5 MB XLS). Click here for additional data file. Dataset S25 UP3_188 Cy3 = ChIP 10h Rec8-3HA in SK1; Cy5 = genomic DNA. (2.6 MB XLS). Click here for additional data file. Dataset S26 UP3_190 Cy3 = ChIP 6h Rec8-3HA in SK1; Cy5 = genomic DNA. (2.6 MB XLS). Click here for additional data file. Dataset S27 UP3_191 Cy3 = ChIP 8h Rec8-3HA in SK1; Cy5 = genomic DNA. (2.6 MB XLS). Click here for additional data file. Dataset S28 UP3_29 Cy3 = genomic DNA; Cy5 = ChIP 12h Rec8-3HA in SK1. (2.6 MB XLS). Click here for additional data file. Dataset S29 UP3_30 Cy3 = genomic DNA; Cy5 = ChIP 4h Rec8-3HA in SK1. (2.6 MB XLS). Click here for additional data file. Dataset S30 UP3_48 Cy3 = genomic DNA; Cy5 = ChIP cdc16 -ts Mcd1-18Myc in W303. (2.6 MB XLS). Click here for additional data file. Dataset S31 UP3_51 Cy3 = genomic DNA; Cy5 = ChIP cdc16 -ts Mcd1-18Myc in W303. (2.6 MB XLS). Click here for additional data file. Dataset S32 UP3_84 Cy3 = ChIP cdc16 -ts Mcd1-18Myc in W303; Cy5 = genomic DNA. (2.6 MB XLS). Click here for additional data file. Dataset S33 UP3_85 Cy3 = ChIP cdc16 -ts Mcd1-18Myc in W303; Cy5 = genomic DNA. (2.6 MB XLS). Click here for additional data file. Dataset S34 UP3_86 Cy3 = genomic DNA; Cy5 = ChIP 4h Rec8-3HA in SK1. (2.6 MB XLS). Click here for additional data file. Dataset S35 UP3_87 Cy3 = genomic DNA; Cy5 = ChIP 10h Rec8-3HA in SK1. (2.6 MB XLS). Click here for additional data file. Dataset S36 UP3_89 Cy3 = genomic DNA; Cy5 = ChIP 12h Rec8-3HA in SK1. (5.6 MB XLS). Click here for additional data file. Dataset S37 UP4_224 Cy3 = genomic DNA; Cy5 = ChIP nocodazole arrest Mcd1-3HA in SK1. (2.6 MB XLS). Click here for additional data file. Dataset S38 UP4_225 Cy3 = genomic DNA; Cy5 = ChIP nocodazole arrest Mcd1-3HA in SK1. (2.6 MB XLS). Click here for additional data file. Dataset S39 UP5_164 Cy3 = genomic DNA; Cy5 = ChIP nocodazole arrest glucose Mcd1-18Myc in W303. (2.7 MB XLS). Click here for additional data file. Dataset S40 UP5_80 Cy3 = genomic DNA; Cy5 = ChIP nocodazole arrest glucose Mcd1-18Myc in W303. (2.7 MB XLS). Click here for additional data file. Dataset S41 UP6_124 Cy3 = ChIP nocodazole arrest glucose Mcd1-18Myc in W303; Cy5 = genomic DNA. (2.6 MB XLS). Click here for additional data file. Dataset S42 UP6_210 Cy3 = genomic DNA; Cy5 = ChIP nocodazole arrest glucose Mcd1-18Myc in W303. (4.4 MB XLS). Click here for additional data file. Dataset S43 UP6_213 Cy3 = genomic DNA; Cy5 = ChIP nocodazole arrest glucose Mcd1-18Myc in W303. (4.4 MB XLS). Click here for additional data file. Dataset S44 UP6_214 Cy3 = genomic DNA; Cy5 = ChIP nocodazole arrest galactose Mcd1-18Myc in W303. (4.4 MB XLS). Click here for additional data file. Dataset S45 UP6_217 Cy3 = ChIP nocodazole arrest glucose Mcd1-18Myc in W303; Cy5 = genomic DNA. (4.4 MB XLS). Click here for additional data file. Dataset S46 UP6_218 Cy3 = ChIP nocodazole arrest galactose Mcd1-18Myc in W303; Cy5 = genomic DNA. (4.4 MB XLS). Click here for additional data file. Dataset S47 UP6_221 Cy3 = ChIP nocodazole arrest galactose Mcd1-18Myc in W303; Cy5 = genomic DNA. (4.4 MB XLS). Click here for additional data file. Dataset S48 UP6_223 Cy3 = genomic DNA; Cy5 = ChIP nocodazole arrest glucose Mcd1-18Myc in W303. (4.4 MB XLS). Click here for additional data file. Dataset S49 UP6_225 Cy3 = genomic DNA; Cy5 = ChIP nocodazole arrest galactose Mcd1-18Myc in W303. (4.4 MB XLS). Click here for additional data file. Dataset S50 YA1S4P2_106 Cy3 = polyA+ reference RNA; Cy5 = polyA+ RNA nocodazole arrest galactose Mcd1-18Myc in W303. (2.2 MB XLS). Click here for additional data file. Dataset S51 YA1S4P2_108 Cy3 = polyA+ RNA nocodazole arrest galactose Mcd1-18Myc in W303; Cy5 = polyA+ reference RNA. (2.2 MB XLS). Click here for additional data file. Dataset S52 YA1S4P2_109 Cy3 = polyA+ reference RNA; Cy5 = polyA+ RNA nocodazole arrest glucose Mcd1-18Myc in W303. (2.2 MB XLS). Click here for additional data file. Dataset S53 YA1S4P2_110 Cy3 = polyA+ reference RNA; Cy5 = polyA+ RNA nocodazole arrest galactose Mcd1-18Myc in W303. (2.2 MB XLS). Click here for additional data file. Dataset S54 YA1S4P2_111 Cy3 = polyA+ RNA nocodazole arrest glucose Mcd1-18Myc in W303; Cy5 = polyA+ reference RNA. (2.2 MB XLS). Click here for additional data file. Dataset S55 YA1S4P2_112 Cy3 = polyA+ reference RNA; Cy5 = polyA+ RNA nocodazole arrest glucose Mcd1-18Myc in W303. (2.2 MB XLS). Click here for additional data file. Dataset S56 YA1S4P2_114 Cy3 = polyA+ RNA nocodazole arrest galactose Mcd1-18Myc in W303; Cy5 = polyA+ reference RNA. (2.2 MB XLS). Click here for additional data file. Dataset S57 YA1S4P2_115 Cy3 = polyA+ RNA nocodazole arrest glucose Mcd1-18Myc in W303; Cy5 = polyA+ reference RNA. (2.2 MB XLS). Click here for additional data file. Dataset S58 YA1S4P2_125 Cy3 = polyA+ reference RNA; Cy5 = polyA+ RNA nocodazole arrest glucose Mcd1-18Myc in W303. (2.2 MB XLS). Click here for additional data file. Protocol S1 ChIP for Yeast (61 KB DOC). Click here for additional data file. Protocol S2 Round A/B/C Random Amplification of DNA (37 KB DOC). Click here for additional data file. Accession Numbers The Saccharomyces Genome Database ( http://www.yeastgenome.org/ ) accession numbers for the genes and gene products discussed in this paper are CDC16 (SGDID S0001505), EBP2 (SGDID S0001655), GAL2 (SGDID S0004071), Mif2 (SGDID S0001572), NCE102 /YPR149W (SGDID S0006353), PCD1 (SGDID S0004141), Pds1p (SGDID S0002520), Rec8 (SGDID S0006211), ROK1 (SGDID S0003139), RRS1 (SGDID S0005820), Scc1/Mcd1 (SGDID S0002161), Scc3 (SGDID S0001288), Smc1 (SGDID S0001886), Smc3 (SGDID S0003610), YDL218W (SGDID S0002377), YDL238C (SGDID S0002397), YER179W (SGDID S0000981), and YPR006C (SGDID S0006210). | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC490026.xml |
522870 | A first-draft human protein-interaction map | Using data from model organisms, the authors have generated a large-scale human protein-protein interaction map. The map can be used to predict the function of human proteins. | Background Physical interactions between proteins underpin most biological processes. For this reason, large-scale protein-interaction mapping projects have been initiated in several model organisms [ 1 - 6 ]. Unfortunately, projects of a similar scale have not yet been described for mammalian systems, with the result that our global understanding of protein function remains less advanced in mammals than in lower eukaryotes. However, many physical interactions are conserved between species, so it should be possible to infer information about human protein interactions and protein function using data from model organism protein-interaction datasets [ 7 , 8 ]. To transfer information on gene function between two genomes requires the identification of orthologous genes in the two genomes (that is, genes that are descended from a common ancestor and share biological functions). However, the identification of gene orthologs is often not a trivial problem; gene duplications can result in a single gene having multiple potential orthologs in a second species. In addition, it is necessary to distinguish true gene orthologs from 'out-paralogs' (that is, genes that arose from a gene-duplication event before the divergence of two species, and so are unlikely to share functions) [ 9 ]. One method that addresses both these problems is the InParanoid algorithm, which first identifies potential orthologs by best pairwise similarity searches, and then clusters these orthologs into groups of likely co-orthologs, with each ortholog assigned a score representing the confidence that it is the main ortholog [ 9 ]. We have used the orthology relationships identified by the InParanoid algorithm to construct a putative human protein-interaction map based solely on high-throughput interaction datasets from model organisms. We show that this approach successfully identifies functionally related human proteins, and so can be used to assign putative functions to many novel human genes. The resulting network provides a framework for human biology and acts as a guide for a future experimental human protein-interaction mapping project. Results Generation of a human protein-interaction map Protein interactions are often evolutionarily conserved between orthologous proteins from different species [ 7 ]. Hence we reasoned that a human protein-interaction map could be constructed using data from model organism protein-interaction mapping projects. We obtained the data from seven experimental and four computationally predicted protein-interaction maps from Saccharomyces cerevisiae [ 1 - 4 , 10 , 11 ], Drosophila melanogaster [ 5 ] and Caenorhabditis elegans [ 6 ]. For each interacting protein, we identified potential human orthologs using the InParanoid algorithm [ 9 ]. A human protein interaction is predicted if both interaction partners from a model organism have one or more human orthologs. Using this strategy, we were able to generate a human interaction network comprising 71,496 interactions between 6,231 human proteins. The sources of these predicted interactions are summarized in Table 1 and Figure 1a , and all the interactions are available in Additional data file 1 available online with this article and can also be searched or downloaded from our website [ 12 ]. Assessment of the accuracy of the interaction datasets In the absence of a comprehensive set of verified human protein interactions, we required another method to assess the accuracy of the interaction network. Proteins that interact physiologically are expected to have related functions. Therefore high-quality interaction datasets should predict a greater proportion of interactions between functionally related proteins than low quality datasets. The functions of human proteins can be systematically described using the Gene Ontology (GO) annotations [ 13 ] available from Ensembl [ 14 - 17 ]. GO annotations provide a hierarchical description of gene functions with general functions described by GO annotations at the top levels of the hierarchy and very precise functions described by terms deeper in the hierarchy. Because physiologically interacting proteins are expected to have related, but non-identical functions, they are expected to share some, but not all GO annotations. Therefore, one method to evaluate an interaction dataset is to count the proportion of interactions that connect proteins that share common GO terms [ 5 ]. For the complete predicted human interaction network, 25% of interaction partners share at least one GO term, which is many more than observed with a randomly generated network of the same size (15% of interactions). To confirm that this result did not just apply to quite general GO annotations, we calculated the proportion of interaction partners that share GO annotations at depths 3 to 8 and greater than 8 in the GO hierarchy. We found that the predicted interaction network preferentially connects proteins that share GO annotations at any level of the GO hierarchy (see Figure 2 ). This suggests that the interaction network indeed preferentially connects functionally related human proteins. We then used the same strategy to compare the accuracy of human interactions predicted by data from the three different model organisms. If the interactions from a particular model organism dataset predict fewer interactions between functionally related human proteins than the other datasets, then this dataset should be considered less reliable as a source of candidate human protein interactions. As shown in Table 1 and Figure 2a , interactions predicted by the complete yeast and worm datasets are slightly better at connecting functionally related human proteins than those predicted by the fly dataset, suggesting that these interactions can be considered with higher confidence. This result is especially interesting given that the yeast interaction map is an order of magnitude larger than the fly or worm maps, confirming that the fly and worm interaction maps currently have a relatively low coverage. Next we asked how the confidence in the assignment of gene orthologs affects the accuracy of an interaction. For each predicted interaction, an orthology confidence score was calculated by summing the InParanoid orthology confidence scores for the two human and two model organism proteins (see Materials and methods). Of the predicted interactions, 24,897 have the maximum possible confidence score of 4. Of these interactions, 28%, 24% and 13% connect proteins that share GO terms at depths of 3, 5 or 7 in the GO hierarchy (excluding proteins without GO annotation). In contrast, for interactions with an orthology confidence score less than 4, these figures are 24%, 20% and 10%. Hence we conclude that the predicted human interactions with high-confidence orthology assignments can be considered more reliable than those interactions with less confidence in their orthology assignments. This confirms that the confidence scores assigned using InParanoid are indeed likely to be useful predictors of functional conservation. A core dataset of high-confidence protein interactions The worm and fly interaction mapping projects both defined a subset of high-confidence 'core' interactions that have the greatest experimental support (Figure 1b ). For the worm interaction map these were defined as interactions identified more than once, or that reconfirmed when retested in the two-hybrid assay [ 6 ]. In the fly interaction map each interaction has an associated confidence score, and interactions with a score greater than 0.5 are considered core interactions (the interaction score mainly depends upon the number of times each interaction was detected, the total number of interactions made by each protein and the local network clustering; see [ 5 ]). To generate a similar subset of yeast protein interactions, we defined core yeast protein interactions as those identified more than once by any single assay, consistent with previous analyses of the individual datasets [ 1 - 3 , 11 ]. As shown in Figure 2a and Table 1 , for all three species these core interactions predict a greater proportion of human interactions that share GO terms than the total datasets. Indeed all three core interaction maps are of similar accuracy, so we combine their predicted interactions into a core network of 11,487 higher-confidence human protein interactions (summarized in Table 2 and available as Additional data file 2). Of these core interactions, 38%, 35% and 24% connect proteins that share GO terms at depths of 3, 5 or 7 in the GO hierarchy (excluding proteins with no GO annotations). Combining interaction datasets to generate high-confidence networks It has been shown previously that protein interactions detected by more than one high-throughput interaction assay are more accurate [ 11 ]. We find that this is also true for human protein interactions predicted by yeast protein interactions detected by more than one method (see Figure 2b and Table 1 ). It has also been suggested that protein interactions are more likely to represent physiologically important interactions if they have been detected between orthologous protein pairs from two or more species [ 7 , 18 ]. To test this hypothesis we identified 288 human protein interactions predicted by interactions in two or more model organisms (Figure 1 , Table 1 ). Remarkably, 75%, 70% and 56% of these interactions share GO terms at depths of 3, 5 or 7 in the GO hierarchy, respectively (Figure 2b ). Indeed, for interactions derived from core interaction datasets, these figures rise to 88%, 80% and 67% of interactions. Hence, protein interactions predicted by data from multiple species can be considered with very high confidence. Using the interaction network to predict human gene function Because physiologically interacting proteins often have similar functions (Figure 2 ), it should be possible to predict the functions of a novel human protein if it interacts with proteins of known function. To address how well our interaction map could be used for this purpose, we asked whether the known GO terms of a protein could be predicted using only the GO terms of its interaction partners. As shown in Table 3 , GO terms associated with at least one of a gene's core interaction partners predict GO terms associated with that gene with an accuracy of around 8%. However, GO terms associated with at least two, three, four or five of a gene's interaction partners have 22%, 30%, 37%, 42% and 45% probabilities, respectively, of also being associated with that gene (Table 3 ). Although these values may vary for different GO terms, as shown in Additional data file 3, the accuracy and coverage of these GO term predictions are very similar for GO terms at different levels in the GO hierarchy, and so can be used as an approximate indication of the confidence in a prediction of gene function. Hence the network can be used to predict GO terms for a human gene of unknown function, with the approximate confidence in the GO prediction determined by the number of interaction partners that share the GO term. The ability to provide a reasonably accurate prediction of a gene's GO terms means that we can use the interaction network to provide probabilistic gene function predictions for novel human proteins and also to predict additional functions for proteins with some known functions. The core interaction map contains 864 proteins with no functional annotations. About 10% of these proteins interact with two or more proteins that share GO terms. The probabilistic predictions of the functions of these novel proteins are listed in Additional data file 4. Often these predicted functions are also supported by the known functions of the protein domains predicted to be encoded by these novel genes (see Additional data file 4). For example, ENSG00000028310 encodes a bromodomain and interacts with six proteins annotated as 'GO:0006355 regulation of transcription, DNA-dependent', ENSG00000080608 encodes an RNA-binding domain and interacts with five proteins annotated as 'GO:0006364 rRNA processing', and ENSG00000104863 encodes a PDZ domain and interacts with three proteins with the annotations 'GO:0005887 integral to plasma membrane, GO:0007242 intracellular signaling cascade' (Additional data file 4). The complete and core interaction maps also predict interactions for 448 and 292 human disease genes (listed in Additional data file 5), of which 55 interact with two or more proteins in the core interaction network that share a GO annotation. The functional predictions for these 55 genes are listed in Additional data file 6. Discussion A framework for human biology We report here the use of data from model organism protein-interaction mapping projects to predict a network of human protein interactions. This network consists of over 70,000 interactions that connect over one-third of all the predicted human proteins, including 1,482 proteins of unknown function and 448 proteins encoded by human disease genes. The physiological relevance of this network is supported by its ability to preferentially connect human proteins that share biological functions (Figure 2 ). Indeed the network can be successfully used to predict the functions of a gene using the known functions of its interaction partners (Table 3 ). As such, the network should provide a rich source of functional hypotheses for researchers interested in the functions of one or many human proteins. The accuracy and coverage of the interactions predicted in this network depend primarily on two parameters: the quality of the original model organism interaction datasets; and the ability to identify the human orthologs of a model organism protein. Our analysis suggests that the raw yeast and worm protein-interaction datasets are currently slightly more accurate than the raw fly interaction dataset, but that when filtered for high-confidence interactions the three interaction maps are of very similar accuracy (see Table 1 and Figure 2 ). The fly and worm interaction maps both have a much lower coverage than the yeast interaction network, most probably because they both only represent the results of a single interaction-mapping project. The continuation of these model organism protein-interaction mapping projects to generate higher coverage interaction maps will greatly enhance our ability to predict human protein interactions. For the identification of gene orthologs, we used the InParanoid algorithm. InParanoid offers several important benefits compared to simple 'reciprocal best hit' sequence-similarity searches [ 9 ]. First, many genes from lower eukaryotes have multiple co-orthologs in humans, which can be identified using InParanoid, but not by simple one-to-one sequence-similarity searches. Second, InParanoid can successfully distinguish these true co-orthologs from paralogs that arose before a speciation event (which are unlikely to retain similar functions). Finally, each potential ortholog in a group of co-orthologs identified by InParanoid has an associated score that represents the likelihood that it is the main ortholog of a gene. We have summed these confidence scores to provide an orthology confidence score for each predicted human protein interaction in our network. These high-confidence ortholog interactions connect a greater proportion of functionally related human proteins, suggesting that the InParanoid confidence score is indeed a useful tool for predicting the likely physiological relevance of a predicted protein interaction. The ability to successfully predict human protein functions using the results of model organism protein-interaction mapping projects highlights both the relevance of model organism protein-interaction mapping projects to understanding human biology and also the benefits that would result from an experimental human protein-interaction mapping project. Although the interaction network can currently accurately predict only a subset of the known functions of a gene, this should improve as more protein-interaction data becomes available. For this reason, we strongly encourage the continuation of model organism protein-interaction mapping projects. Methods of verifying protein-interaction datasets We also assessed the relative merits of three different methods to improve the accuracy of protein-interaction maps. The first strategy is to define a subset of interactions detected more than once with a single assay [ 1 - 3 , 6 ]. We found that this approach leads to an approximately 1.5- to 2.7-fold increase in the proportion of predicted human interactions that share GO terms (Figure 2b ). The second strategy is to define a subset of interactions that have been identified by more than one interaction assay. This results in around a 2.3- to 8-fold improvement in the prediction of associations between proteins that share GO terms (Figure 2b ). The final strategy is to define a subset of interactions that are predicted by interactions from more than one model organism, which results in around a 3- to 12-fold improvement in the proportion of interactions between proteins sharing GO terms (Figure 2b ). With all these filtering methods, the greatest improvements are seen when considering the proportion of interactions that share GO terms deep within the GO hierarchy; that is, the filtering steps dramatically improve the proportion of interactions between proteins with very closely related functions. We conclude that using interaction data derived from a second interaction assay or from a second species both represent excellent methods to improve the accuracy of protein-interaction maps. Because of the small number of protein-interaction assays that have been adapted to a high-throughput format, we suggest that constructing a second interaction map in a related organism using the same assay may be an efficient way to produce a high-confidence interaction map. This strategy is somewhat similar to using phylogenetic footprinting to identify functional noncoding DNA, so we suggest it should be named 'interaction footprinting'. Using the relatively low-coverage model organism interaction datasets currently available, only a small proportion of interactions can be verified by interaction footprinting. The continuation of these model organism interaction mapping projects will not only provide a much richer framework of predicted human protein interactions, but will also allow many more interactions to be verified using the interaction footprinting strategy. However, such an approach will be limited to providing information on those proteins and interactions that are conserved between vertebrates and invertebrates. Strategies for completing the human interaction map The interactions described here provide a first-draft human protein-interaction map that can be used to predict interactions and functions for genes of interest to a particular researcher. However, the map also provides a framework from which a complete human protein-interaction map could be generated. Firstly, the map could be used to identify subsets of high-confidence, evolutionarily conserved interactions from the results of large- or medium-scale human interaction-mapping projects. For example the map verifies 51 of 296 yeast two-hybrid interactions detected for human proteins involved in mRNA decay [ 19 ]. Alternatively, the interactions predicted here could be directly experimentally validated using an assay that allows rapid testing of binary interactions (such as the yeast or mammalian two-hybrid assays [ 20 ] or protein fragment complementation assays [ 21 ]). This would represent a cost-effective strategy to produce a high-confidence human protein-interaction map because it massively reduces the number of candidate interactions that need to be tested. Finally, the map identifies 17,300 (23,531 - 6,231) human genes for which no protein interactions are predicted from model organism interaction datasets. Many of these proteins are likely to be vertebrate- or mammalian-specific, and are the most logical choices for bait proteins for the discovery phase of an experimental human protein-interaction mapping project. Materials and methods Model organism protein-interaction datasets The interaction datasets used to generate the draft human protein-interaction network were two-hybrid-based interaction maps for D. melanogaster [ 5 ] and C. elegans [ 6 ] and a list of S. cerevisiae protein-interactions compiled by Von Mering et al. [ 11 ] from two two-hybrid [ 1 , 2 ], two complex purification [ 3 , 4 ], one genetic [ 10 ], and four in silico -predicted interaction datasets (which used correlated mRNA expressions, conserved gene neighbourhood, gene co-occurrence or gene fusion events to predict protein interactions [ 11 ]). Table 4 shows the number of unique interactions in each dataset, the methods used to generate each dataset, and the URLs from which the datasets were obtained. Identification of gene orthologs and construction of the interaction network The human orthologs of yeast, worm and fly genes were identified using the InParanoid algorithm, which is designed to distinguish true orthologs from out-paralogs that arose from gene duplications before the divergence of two species [ 9 ]. The InParanoid algorithm first identifies potential orthologs by best pairwise similarity searches, and then clusters these orthologs into groups of probable co-orthologs, with each ortholog assigned a score representing the confidence that it is the main ortholog. For each interaction data source, we obtained SWISS-PROT/TrEMBL accessions for each interacting protein using the Ensmart data-mining tool [ 16 , 17 ] (for worm and fly genes) or both SWISS-PROT [ 22 ] and a TrEMBL conversion file kindly provided by Paul Kersey, EBI, Hinxton, UK (for yeast genes). Potential human orthologs of these genes were then identified using the pre-computed InParanoid results (version 2.3, available from [ 23 ]), and the results converted to nonredundant Ensembl (v19.34a.1, genome assembly NCBI34) gene IDs using Ensmart (v19.1) 1 [ 16 , 17 ]. In total, InParanoid identifies 9,500 human genes with at least one ortholog in at least one of worm, fly or yeast. For each potential ortholog in a group of co-orthologs, the InParanoid algorithm calculates a score that represents the confidence that it is the main ortholog. In this scoring system, the main ortholog always receives a score of 1, with the other co-orthologs receiving scores ranging between 0 and 1, calculated according to their similarity to the main ortholog [ 9 ]. As an indication of the confidence we have in the orthology relationships between a pair of interacting proteins from a model organism and a predicted pair of interacting human proteins, we calculate a confidence score by summing the InParanoid confidence scores for each of the four proteins. Hence, each interaction has an associated score ranging from 0 to 4 that represents the confidence that both human proteins represent the main orthologs of the model organism proteins, and vice versa. Core interactions were defined as those predicted by worm interactions identified more than once or that reconfirmed when retested in the two-hybrid assay [ 6 ], by fly interactions with an interaction score greater than 0.5 [ 5 ], or by yeast interactions detected two or more times by a single assay [ 1 - 3 , 11 ]. Assessment of the interaction data Human GOs (at levels 3 or deeper in the GO hierarchy) were obtained from Ensembl (v19.34a.1) [ 14 , 15 ] using Ensmart (v19.1) [ 16 , 17 ]. The GO terms 'unknown molecular function/biological process/cellular compartment' were discarded in all subsequent analyses. To validate the accuracy of the interaction data, we calculated the percentage of interactions that shared at least one GO term. To confirm that the results did not just apply to very general GO annotations, we calculated the proportion of interacting proteins that shared a GO annotation at levels 3 to 8 and greater than 8 in the GO hierarchy. For all of these analyses we ignored proteins with no associated GO annotations. Moreover, self-interactions were excluded because they will always share GO terms and so bias the results. Prediction of gene functions To predict the GO terms of a protein, we identified all the GO terms associated with x or more of its interaction partners (where x varied from 1 to 6). To validate the accuracy and coverage of this approach we predicted GO terms for genes that already have associated GO terms. The accuracy was calculated as the total number of correct GO term predictions divided by the total number of GO term predictions. The coverage was calculated as the total number of correct GO term predictions divided by the total number of known GO terms. This analysis was repeated, but only considering individually GO terms at depths of 3 to 8 and greater than 8 in the GO hierarchy (see Additional data file 3). To avoid biasing the results we again ignored self-interactions. For the same reason, we also only counted once GO terms associated with more than one interaction partner predicted by the same source interaction from a model organism. The InterPro protein domains [ 24 ] encoded by each human gene were obtained from Ensembl using Ensmart. Genes of unknown function were defined as those having no associated GO terms, and disease genes were as defined by Ensembl using the Online Mendelian Inheritance in Man (OMIM) database as a reference [ 25 ]. Additional data files The following additional data files are available with the online version of this article: Additional data file 1 contains a complete list of predicted human protein interactions; this dataset contains every human protein interaction that is predicted by a protein interaction from any of seven experimental and four computationally-predicted protein interaction maps from Saccharomyces cerevisiae [ 1 - 4 , 10 , 11 ], Drosophila melanogaster [ 5 ] and Caenorhabditis elegans [ 6 ]. Additional data file 2 contains a list of all core human protein interactions. This represents a subset of high-confidence human protein interactions that is predicted by model organism protein interactions with greater experimental support. In the worm interaction map, these are defined as interactions that reconfirmed when retested in the Y2H assay [ 6 ]. In the fly interaction map, each interaction has an associated confidence score, and interactions with a score greater than 0.5 are considered core interactions (the interaction score mainly depends upon the number of times each interaction was detected, the total number of interactions made by each protein and the local network clustering [ 5 ]). To generate a similar subset of yeast protein interactions, we defined core yeast protein interactions as those identified more than once by any single assay. Each entry in the core and complete interaction networks contains the following tab delimited information: Gene 1 Id, Ensembl gene ID for human interaction partner 1; Gene 1 description, alternative names for human Gene 1 (from Ensembl); Gene 2 Id, Ensembl gene ID for human interaction partner 2; Gene 2 description, alternative names for human Gene 2 (from Ensembl); Source Organism, the model organism protein interaction dataset that predicts this human protein interaction; Ortholog 1, model organism interaction partner 1 from the model organism protein interaction that predicts the human protein interaction; Ortholog 2, model organism interaction partner 2 from the model organism protein interaction that predicts the human protein interaction; and Ortholog score, a confidence score for the human protein interaction based on the likelihood that the two human proteins are the functional orthologs of the two model organism proteins. The score ranges from 0 (no confidence) to 4 (high confidence). The score is calculated as the sum of the InParanoid confidence scores for each gene orthology assignment. A score of 4 means that both of the human genes and both of the model organism genes are all the main orthologs in their groups of co-orthologs according to InParanoid. These represent higher confidence human protein interactions. Description, this field contains the original annotation for the model organism protein interaction; for worm interactions this indicates whether the interaction is in the core dataset of interactions found more than once (CORE_1), or interactions that reconfirmed when retested (CORE_2), or non-core interactions that did not reconfirm (NON_CORE) [ 6 ]. For fly interactions this indicates the interaction score. This score mainly depends upon the number of times each interaction was detected, the total number of interactions made by each protein and the local network clustering, see [ 5 ] for details. A score >0.5 is considered high confidence. For yeast protein interactions, these are the annotations of von Mering et al . [ 11 ] and contain the following information: experimental/computation method (and the number of times the interaction was detected); Von Mering et al .'s confidence assignment; and whether the interaction was previously known in the literature. For more information, please see [ 11 ]. Additional data file 3 lists the accuracy and coverage of GO term predictions at different levels in the GO hierarchy; Additional data file 4 lists gene function predictions for 85 human genes of unknown function; Additional data file 5 lists human disease genes with predicted protein interactions; and Additional data file 6 lists gene function predictions for 55 human disease genes. Supplementary Material Additional data file 1 A complete list of predicted human protein interactions Click here for additional data file Additional data file 2 A list of all core human protein interactions Click here for additional data file Additional data file 3 The accuracy and coverage of GO term predictions at different levels in the GO hierarchy Click here for additional data file Additional data file 4 Gene function predictions for 85 human genes of unknown function Click here for additional data file Additional data file 5 Human disease genes with predicted protein interactions Click here for additional data file Additional data file 6 Gene function predictions for 55 human disease genes Click here for additional data file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC522870.xml |
549035 | Bayesian and maximum likelihood phylogenetic analyses of protein sequence data under relative branch-length differences and model violation | Background Bayesian phylogenetic inference holds promise as an alternative to maximum likelihood, particularly for large molecular-sequence data sets. We have investigated the performance of Bayesian inference with empirical and simulated protein-sequence data under conditions of relative branch-length differences and model violation. Results With empirical protein-sequence data, Bayesian posterior probabilities provide more-generous estimates of subtree reliability than does the nonparametric bootstrap combined with maximum likelihood inference, reaching 100% posterior probability at bootstrap proportions around 80%. With simulated 7-taxon protein-sequence datasets, Bayesian posterior probabilities are somewhat more generous than bootstrap proportions, but do not saturate. Compared with likelihood, Bayesian phylogenetic inference can be as or more robust to relative branch-length differences for datasets of this size, particularly when among-sites rate variation is modeled using a gamma distribution. When the (known) correct model was used to infer trees, Bayesian inference recovered the (known) correct tree in 100% of instances in which one or two branches were up to 20-fold longer than the others. At ratios more extreme than 20-fold, topological accuracy of reconstruction degraded only slowly when only one branch was of relatively greater length, but more rapidly when there were two such branches. Under an incorrect model of sequence change, inaccurate trees were sometimes observed at less extreme branch-length ratios, and (particularly for trees with single long branches) such trees tended to be more inaccurate. The effect of model violation on accuracy of reconstruction for trees with two long branches was more variable, but gamma-corrected Bayesian inference nonetheless yielded more-accurate trees than did either maximum likelihood or uncorrected Bayesian inference across the range of conditions we examined. Assuming an exponential Bayesian prior on branch lengths did not improve, and under certain extreme conditions significantly diminished, performance. The two topology-comparison metrics we employed, edit distance and Robinson-Foulds symmetric distance, yielded different but highly complementary measures of performance. Conclusions Our results demonstrate that Bayesian inference can be relatively robust against biologically reasonable levels of relative branch-length differences and model violation, and thus may provide a promising alternative to maximum likelihood for inference of phylogenetic trees from protein-sequence data. | Background The inference of phylogenies from molecular sequence data, like most other quantitative problems in science, is most powerful within a model-based statistical framework. Sophisticated models are available to describe how sequences change along branches of a tree, and how the rate of sequence change varies among sites. Statistical measures describe both the quality of inferred trees, and the confidence that can be assigned to the existence and position of subtrees. Likelihood-based approaches have proven especially powerful for inferring phylogenetic trees [ 1 , 2 ] but are computationally expensive owing both to the form of the likelihood function itself, and to the need to search the multidimensional space of possible outcomes (tree space) for optimal trees. This computation then must be repeated, typically 100–1000 times, if the nonparametric bootstrap [ 3 ] is used to estimate the support for specific subtrees. As a result, maximum-likelihood inference can be prohibitively slow for problems that involve large numbers of aligned sequences, comprehensive search of tree space, and/or many bootstrap replicates. The much faster RELL approximation [ 4 , 5 ] can in principle replace the bootstrap, although so far it has not been extensively investigated with large datasets [ 6 ]. At the same time, the ongoing success of genomic sequencing – new microbial genome sequences are now appearing at the rate of at least one per week – is yielding a wealth of ever-larger gene and protein datasets suitable for large-scale analysis of deep issues in comparative and evolutionary genomics, e.g. the relative contributions of vertical and lateral gene transfer to genomic diversity [ 7 , 8 ]. However, these datasets are too numerous, and many of them too large, for ready analysis by likelihood inference. For example, using an automated phylogenetics pipeline [ 9 ] we have generated more than 22400 protein datasets having up to 144 sequences each, for which we must infer trees. There is consequently much interest in approaches that offer improved search efficiencies while remaining within a model-based statistical framework. Among the most interesting of these is Bayesian inference, in which the posterior probability of a hypothesis ( i.e. a tree) is associated with its probability of being correct, given the prior probability, model and data [ 2 , 10 ]. Although posterior probabilities cannot be computed analytically for interestingly large datasets, Markov chain Monte Carlo (MCMC) methods can be used to find and examine equilibrium distributions of trees, on the basis of which we can make probability statements about the true tree [ 10 - 14 ]. Bayesian inference of phylogeny supports sophisticated evolutionary models, while MCMC, particularly with heated chains (Metropolis-coupled MCMC), recovers from the posterior probability distribution a sample of topologies within which the empirical relative frequency of a given topology converges to its corresponding marginal posterior probability [ 15 ]. The topology with highest relative frequency in this sample is typically reported, and posterior probabilities of subtrees can be estimated by consensus from the topologies visited [ 10 , 13 ]. Bayesian phylogenetic inference has been applied to simulated [ 16 - 18 ] as well as empirical nucleotide datasets (see below). The results establish the applicability and computational efficiency of the Bayesian MCMC approach to molecular phylogenetic inference. However, concerns have arisen about (1) finding optimal trees, (2) overly liberal confidence estimates on subtrees [ 19 - 24 ], and (3) the possibility that Bayesian inference can resolve topological features ( e.g. internal edges, hence subtrees) that do not actually exist [ 16 ]. Certain other issues have not been systematically addressed with nucleotide data, notably the robustness of Bayesian inference to relative branch-length differences and to model violation. Much less is known about the behaviour of Bayesian inference with protein-sequence data. While there is no a priori reason that protein-sequence data should be more or less problematic than nucleotide data for Bayesian phylogenetics, gene and protein sequences have distinct statistical properties, and are subject to different selective constraints; so it is not inconceivable that, in practice, the corresponding models of sequence change might tend to fail in different ways, or to different extents. Bayesian inference has been applied to inference of phylogenetic trees for cytochrome b [ 25 ], elongation factor 1α [ 26 ], hydroperoxidases [ 27 ], 3-hydroxy-3-methylglutaryl coenzyme A reductase (HMGR) [ 18 ], membrane-intrinsic protein [ 28 ], and concatenated mitochondrial protein [ 29 ] and larger [ 30 ] datasets. Douady et al. [ 18 ] report a linear, if noisy, correlation between bootstrap proportion and Bayesian posterior probability for a 15-taxon HMGR protein dataset. As in the case of nucleotide data, the robustness of Bayesian inference to branch-length differences and model violation with protein-sequence datasets remains unexplored. To better characterize the behavior of Bayesian phylogenetic inference with protein-sequence data, we have applied MrBayes [ 31 , 32 ] to both empirical and simulated data. Based on the analysis of 21 empirical protein datasets, we compare maximum likelihood bootstrap proportion and Bayesian posterior probability as estimates of subtree confidence. From analyses of simulated data known to contain phylogenetic signal, we address the fidelity with which the correct topology is recovered under progressively extreme ratios of branch-length differences, both under the correct model of sequence change (the model under which the data were evolved) and under a model that incorporates different amino acid substitution probabilities. Given our ongoing research on lateral gene transfer (above), we were particularly interested in the number of discrete events (edits: [ 33 ]) separating inferred from known trees. In this work we compare and contrast results obtained using two popular software programs, PROML [ 34 ] and MrBayes [ 31 ], as well-developed implementations of the ML and Bayesian approaches to phylogenetic inference respectively. Although the comparison is illustrative, it would be an oversimplification to view these two approaches as diametric opposites, or even as fundamentally mutually exclusive. Both likelihood and Bayesian are general statistical frameworks, with high-level decision criteria (the Akaike Information Criterion, or AIC [ 35 ] and Bayesian Information Criterion, or BIC [ 36 ], respectively: see also [ 37 ]) and associated apparatus for e.g. examining solution space, estimating support, and assessing stability to stochastic error. Only a subset of these broad bodies of theory and practice has so far been applied to phylogenetic inference, and even less implemented in platform-independent software. If we apply BIC to alternative trees and assume equal prior probabilities, it becomes possible to estimate Bayesian posteriors from their likelihood differences, linking the two approaches at this level [ 6 , 38 ]. Stochastic approaches related to MCMC, including simulated annealing [ 39 ] and the generalised Gibbs sampler [ 40 ], can be used to search tree space in ML. The nonparametric bootstrap, more typically applied in conjunction with parsimony and ML, has proven useful in assessing subtree support in Bayesian inference [ 6 , 18 ]. The application of likelihood in hybrid methods [ 41 - 43 ], the likelihood ratchet [ 44 ], and a metapopulation genetic algorithm [ 45 ] lie farther beyond the scope of this discussion, but illustrates the potential for further development of both of these phylogenetic approaches beyond the specific implementations used in this study. Results Empirical data Topology We inferred maximum likelihood (ML) and Bayesian (B) trees for the 21 empirical protein-sequence datasets. For 7 of these datasets, every combination of approach and model that we investigated (ML-JTT-HMM, ML-JTT-gamma, B-JTT, B-EQ: see Empirical data under Methods) yielded the same topology. Interestingly, for these, the bootstrap consensus ML trees were topologically identical to the ML and Bayesian trees, indicating that the sequences in these datasets show a high degree of internal consistency across positions ( i.e. bear few homoplasies). For another 10 datasets, one or more of these four approaches yielded a tree that differs slightly (edit distance ≤ 2) from the others. No pattern was obvious among these disagreements: the differences do not, for example, systematically separate ML from Bayesian trees. For these 10 datasets, the differences are simple edits, e.g. -(A(BC)) to -(B(AC)), or -((AB)(CD)) to -(A(B(CD))). For the remaining 4 datasets, one or more of the four approaches yielded a tree that differed more substantially (edit distance ≥ 3). Over these examples, the datasets that yield more-conflicted trees are slightly larger (mean, 12.25 sequences each) than those yielding slightly conflicted (mean, 11.50 sequences each) or identical trees (mean, 10.86 sequences each), although the numbers of datasets involved are too few for this observation to be generalized. Support for subtrees We compared PROML bootstrap proportions (BPs) with Bayesian posterior probabilities (PPs) separately for all subtrees among the three groups of trees inferred from these 21 empirical datasets: the 7 trees for which all four sets of approaches and models (ML-JTT-HMM, ML-JTT-gamma, B-JTT, B-EQ) yielded the same topology, the 10 for which one or more approach yielded a slightly different tree, and the 4 for which one or more tree differed more substantially. In Figure 1 , we show the relationship between BP (from PROML using the 8-category gamma distribution: see Methods) and PP for subtrees in these three groups of trees; results for PROML using the hidden Markov model (HMM) are very similar (results not shown). Where ML and Bayesian approaches yield the same topology, the relationship between BP and PP can most simply be fit by a straight line (P-values for linearity are between e-10 and e-13 depending on the subset of data examined). With very few exceptions, however, the PP values are greater, and almost all of the BP values above 80% correspond to 100% PP (Figure 1 , panel A). For the 14 datasets for which at least one of the four approaches yields a conflicting tree, the relationship between BP and PP appears much more complex (Figure 1 , panels B,C), although for the subset of non-conflicting subtrees among these 14 datasets (Figure 1 , panel D) the relationship between BP and PP is similar to that for topologically identical trees (Figure 1 , panel A). Panel E combines data for all non-conflicting subtrees (panels A and D). In all of these views on the data (Figure 1 , panels A-E), however, most points lie above and to the left of the diagonal (Table 1 ), indicating that for empirical protein-sequence datasets, as for DNA-sequence datasets (see Introduction), Bayesian PPs tend to be more generous than nonparametric BPs as estimates of confidence in subtrees. Figure 1 Empirical data: relationship between ML consensus bootstrap proportion and Bayesian posterior probability. Comparison of PROML bootstrap proportions (horizontal axes) with Bayesian posterior probabilities (vertical axes) for all internal nodes in trees inferred from 21 empirical protein-sequence datasets. Data are for trees inferred by gamma-corrected ML under JTT, versus those inferred by gamma-corrected Bayesian inference under JTT (open diamonds) or under EQ (closed squares), (A) for the 7 datasets for which the two ML and two Bayesian trees (see text) are topologically identical, (B) for the 10 datasets for which at least one ML or Bayesian tree (see text) differs slightly (edit distance ≤ 2) from the other three, (C) for the 4 datasets for which at least one tree differs more substantially (edit distance ≥ 3), (D) for the subset of internal nodes, within the latter 14 non-identical trees, that subtend identical subtrees, and (E) for data in panels (A) and (D) plotted together. Table 1 Linear fit equations for data in Figure 1. Slope, y-intercept, significance, and R 2 values for linear equations relating bootstrap proportion and Bayesian posterior values shown in panels A, D and E of Figure 1, i.e. for all nodes subtending identical subtrees among the 21 empirical protein-sequence datasets, regardless of whether the corresponding full ML and Bayesian trees are topologically identical or not. Data 1 Panel Slope SE 2 Signif y -Intcpt SE Signif Mult R 2 Adj R 2 JTT model, all data A 0.4993 0.0536 0.001 51.823 4.568 0.001 0.6207 0.6136 D 0.5150 0.0398 0.001 50.125 3.441 0.001 0.5279 0.5247 E 0.5101 0.0322 0.001 50.625 2.773 0.001 0.5508 0.5486 EQ model, all data A 0.4557 0.0572 0.001 55.661 4.871 0.001 0.5452 0.5367 D 0.4517 0.0352 0.001 56.269 3.050 0.001 0.5227 0.5195 E 0.4531 0.0298 0.001 56.077 2.569 0.001 0.5297 0.5274 JTT model, BP <85% A 0.7536 0.1771 0.001 38.180 10.819 0.01 0.4880 0.4610 D 0.7816 0.1258 0.001 34.931 8.115 0.001 0.4081 0.3976 E 0.7694 0.1020 0.001 36.112 6.487 0.001 0.4251 0.4176 EQ model, BP <85% A 0.6909 0.1809 0.01 43.124 11.054 0.001 0.4342 0.4044 D 0.6837 0.1099 0.001 43.086 7.089 0.001 0.4088 0.3982 E 0.6843 0.0922 0.001 43.174 5.866 0.001 0.4170 0.4095 1 All data: values based on all data shown in the respective panel in Figure 1; BP <85%: values based on only those data for which the value of the PROML bootstrap proportion is less than 85%. 2 SE, standard error; Signif, significance (probability level that estimate >|t|); y -Intcpt, y -intercept; Mult R 2 , multiple R 2 ; Adj R 2 , adjusted R 2 . From our data, it is not possible to reject the hypothesis that the relationship between BP and PP has the same slope whether the Bayesian inference is conducted using JTT, or EQ, as the model of sequence change. Analysis of covariation (ANCOVA) yields probabilities 0.579 (Panel A), 0.235 (Panel D) and 0.195 (Panel E) that the lines described in Table 1 differ in slope between the JTT and EQ models. When data having >85% BP are removed from analysis, the probabilities become 0.806, 0.559 and 0.537 respectively, but equivalence still cannot be rejected. Given the limitations of these data, we did not attempt a more-complete analysis, e.g. involving minority subtrees (those not in the extended 50% majority-rule consensus) or higher-order (sigmoidal) fit curves. Because for these trees the true molecular phylogeny is unknown, these results do not speak to the accuracy of the inferred topologies. For this, it is necessary to examine inferences based data simulated on trees of known topology. Simulated data Topology We first examine cases where tree inference was carried out under the same model (JTT) as that used to generate the data, and where a single branch was progressively extended in length (see Methods). When trees were inferred using gamma-corrected ML, the correct tree was recovered in 100% (50/50) of the cases in which the relative branch-length difference was 5-, 10- or 20-fold (Robinson-Foulds symmetric distance in Figure 2 , panel A, and edit distance in Figure 3 , panel A). The frequency of inaccurately reconstructed trees increased with further increase in relative branch-length difference, and the inaccurately reconstructed trees were increasingly inaccurate as judged by Robinson-Foulds symmetric distance (which measures the number of bipartitions involved in topological incongruence) although not by edit distance (which measures the number of break-and-reanneal differences without reference to bipartitions). Figure 2 Comparative performance with simulated data: correct model, single long branch, symmetric distance Performance at different branch-length ratios of ML and Bayesian inference with simulated protein-sequence data evolved on a tree having a single long branch, measured as Robinson-Foulds symmetric distance. The JTT model was used for both sequence evolution and tree inference. Number (out of 50) of accurately reconstructed topologies (vertical axes) versus branch-length ratio (horizontal axes), where inference was by (A) gamma-corrected PROML, (B) Bayesian uncorrected for ASRV, with uniform prior, (C) gamma-corrected Bayesian with uniform prior, and (D) gamma-corrected Bayesian with exponential prior. Shading codes for each different distance are shown in the small box at the right of each panel (A-D). Thus the right-hand bar in panel B shows that using Bayesian inference uncorrected for ASRV and assuming a uniform prior, with a dataset generated on a tree in which one branch was lengthened 70-fold, 33 of 50 independent trees recovered the correct topology (Robinson-Foulds symmetric distance zero); 6 differed topologically in ways that involved a single node (distance two); 2 differed in ways that involved two adjacent nodes (distance four); 4 were at distance six; and the remaining 5 were at the maximum symmetric distance, eight. See text for explanation of dual bars in Panel A. Figure 3 Comparative performance with simulated data: correct model, single long branch, edit distance. Performance at different branch-length ratios of ML and Bayesian inference with simulated protein-sequence data evolved on a tree having a single long branch, measured as edit distance. The JTT model was used for both sequence evolution and tree inference. Models, panels and axes are as in Figure 2. We investigated two ways of assessing the performance of ML inference. In panel A of Figures 2 and 3 , a pair of bars is shown at each value of branch-length difference. The left-hand bar shows performance assessed over 50 single ML reconstructions (one from each of the 50 datasets evolved at that relative length increment), while the right-hand bar shows performance assessed over 50 consensus trees (each of which summarizes 10 bootstrap replicates for each of the same 50 datasets). For datasets having a single long branch, the two representations yield very similar results, with the individual ML results usually showing slightly better performance. By contrast, the situation was reversed for datasets with two long branches. Although consensus is an appropriate way to summarize bootstrap results, nonparametric bootstrap proportions do not measure support for subtrees in a simple, direct and unbiased manner [ 46 - 48 ]. For this reason, one might question whether an approach based on bootstrap and consensus appropriately summarizes the performance of ML for comparison with Bayesian inference, as Bayesian posteriors do directly measure subtree probabilities (given the priors, model and data). The similarities we observe in both magnitude and trend for the two approaches demonstrate that the comparison we are making between ML and Bayesian inference does not, in these cases at least, depend on whether or not the performance of ML is assessed using an approach that involves the nonparametric bootstrap. With Bayesian inference, inaccurately reconstructed trees were also first seen at the 30-fold branch-length ratio (Figure 2 , panels B-D, and Figure 3 , panels B-D). Compared with the ML consensus result (panel A, right-hand bar), Bayesian inference almost always yielded a higher frequency of accurate reconstructions. However, unless correction was made for ASRV, the inaccurate trees, although fewer in number, could be more inaccurate as judged by symmetric distance. Gamma correction for ASRV greatly reduced the frequency of the most inaccurate reconstructions, yielding results (Figure 2 , panels C,D) noticeably better than with gamma-corrected ML. In our simulations, use of an exponential prior (Figure 2 , panel D) gave slightly fewer inaccurate trees at the most-extreme branch-length ratios, although the difference is not statistically significant (Wilcoxon matched-pairs signed-rank test). In Figures 4 and 5 we present the results of tree inference carried out under the same model (JTT) as that used to generate the data, but where two branches were progressively extended relative to the others. For each of the four sets of approaches and models considered, the first inaccurate tree reconstruction was observed at 20-fold relative difference. At higher branch-length ratios, relative performance among the four suites of approaches and models is much more striking than was seen when only a single long branch was present. With gamma-corrected ML, for example, by 50-fold ratio only 9/50 tree topologies are accurately inferred, and at 70-fold ratio only 1/50 (Figure 4 , panel A). ASRV-uncorrected Bayesian inference (Figure 4 , panel B) performs even worse, with no accurate inferences at branch-length ratios 50 or greater. However, gamma correction (Figure 4 , panels C,D) yielded a much higher frequency of accurate reconstructions, with the uniform prior performing better than the exponential prior at the more extreme ratios (Wilcoxon P ≤ 0.003906 at 70-fold) as judged by symmetric distance. About three-quarters (74.9%) of the inaccurate topologies inferred in the case of two differentially lengthened branches showed long-branch attraction, i.e. the long branches were topologically adjacent in the reconstructed tree. Figure 4 Comparative performance with simulated data: correct model, two long branches, symmetric distance. Performance at different branch-length ratios of ML and Bayesian inference with simulated protein-sequence data evolved on a tree having two long branches, measured as Robinson-Foulds symmetric distance. The JTT model was used for both sequence evolution and tree inference. Models, panels and axes are as in Figure 2. Figure 5 Comparative performance with simulated data: correct model, two long branches, edit distance. Performance at different branch-length ratios of ML and Bayesian inference with simulated protein-sequence data evolved on a tree having two long branches, measured as edit distance. The JTT model was used for both sequence evolution and tree inference. Models, panels and axes are as in Figure 2. Support for subtrees In Figure 6 we compare the quantitative support for subtrees, in trees inferred from these simulated datasets by ML and Bayesian approaches, as assessed by bootstrap proportion and posterior probability respectively. Panels A-C show the comparisons based on 1600 extended majority-rule consensus trees for datasets with one long branch (50 ML trees at each of eight branch-length ratios, compared with 50 Bayesian trees at each of the same ratios, over three combinations of ASRV correction and prior probability distribution), and panels D-F are based on 1600 consensus trees for datasets with two long branches. The values shown were derived by summation of BP, and of PP, values over all internal nodes only for the trees that were accurately inferred ( i.e. identical with the known topology). By structuring the comparison in this way, we avoid cases where the ML consensus might be topologically different from the best component tree, and avoid dealing with the plethora of cases and sub-cases that arise in comparing topologically non-congruent trees. Figure 6 Simulated data: relationship between ML consensus bootstrap proportion and Bayesian posterior probability. Relationship between bootstrap proportion for ML consensus trees, and posterior probability for Bayesian trees, for datasets with one (A-C) or two (D-F) branches of relatively greater length. Bayesian trees were inferred (A and D) without ASRV correction and with a uniform prior, (B and E) with gamma correction for ASRV and with a uniform prior, and (C and F) with gamma correction and with an exponential prior. Panel D does not show data at relative branch-length ratios ≥ 50 because none of the trees inferred at these branch-length ratios recovered the known topology. For all three combinations of ASRV correction and prior (corresponding to panels B-D of Figures 2 , 3 , 4 , 5 ), the relationship between BP and PP, structured in this way, is best fit by a smooth curve that reaches 100% PP only at BP 99.75% (Figure 6 , panels A-D) or BP 100% (panels E-F), i.e. shows little or no "saturation". For both the single- and two-long-branches cases, the PP is greatest, compared to BP, for Bayesian trees inferred without correction for ASRV, and least generous for gamma-corrected trees where the prior distribution was assumed to be uniform. Unsurprisingly, the lower values of subtree support, as measured both by BP and by PP, arise from the trees with the most extreme relative branch length differences. Performance under model violation We next compared the performance of ML and Bayesian inference under violation of the model of sequence change, by evolving datasets under a mammalian mitochondrial model (mtmam), but inferring trees under the JTT model (see Methods). Performance of each of the four sets of approaches and methods was assessed by comparing four measures: the branch-length ratio at which inaccurate trees were first observed; the total number of steps (summed over the eight ratios) by which the 400 trees differ from the true topology; the weighted sum ("burden") of these steps; and the mean number of steps by which each inaccurate tree differs from the known tree. The latter two measures were each calculated using both Robinson-Foulds symmetric distance, and edit distance, yielding six comparisons in all. A more-complete description is provided at footnote 2 of Table 2 . Performance in the case of one long branch is summarized in Figures 7 and 8 , and in the case of two long branches in Figures 9 and 10 . In Table 2 we summarize and compare the performance of ML and Bayesian inference with these datasets under the correct, and an incorrect, model. Table 2 Simulated data: comparative performance under correct and incorrect models. Performance of maximum-likelihood and Bayesian phylogenetic inference without, and with, violation of the model of protein sequence change, for trees with one, or two, relatively long branches. One long branch No model violation 1 Model violation 1 First 2 Wrong Burd SD Mean SD Burd ED Mean ED First Wrong Burd SD Mean SD Burd ED Mean ED ML 3 30 56 144 2.57 56 1.00 20 65 164 2.52 65 1.00 BUU 30 46 232 5.04 46 1.00 20 50 222 4.44 50 1.00 BGU 30 36 96 2.67 36 1.00 20 39 118 3.03 39 1.00 BGE 30 28 88 3.14 28 1.00 20 39 116 2.97 39 1.00 Two long branches No model violation Model violation First Wrong Burd SD Mean SD Burd ED Mean ED First Wrong Burd SD Mean SD Burd ED Mean ED ML 20 186 1124 6.04 273 1.47 20 174 1166 6.70 207 1.19 BUU 20 237 1854 7.82 326 1.38 10 244 1900 7.79 299 1.23 BGU 20 87 270 3.10 104 1.20 20 105 468 4.46 119 1.13 BGE 20 86 314 3.65 101 1.17 20 115 650 5.65 131 1.14 1 Protein-sequence data were evolved under the Jones et al. (JTT) or, alternatively, mammalian mitochondrial (mtmam) model of sequence change, and trees were inferred assuming the JTT model. 2 Performance was measured by six indices: First , the lowest investigated branch-length ratio at which at least one inaccurately reconstructed tree was found; Wrong , the number of inaccurately inferred trees out 400 (8 branch-length ratios × 50 replicates at each ratio); BurdSD , the bipartition burden, calculated as the Robinson-Foulds symmetric distance by which each tree differs from the known tree, summed over the 400 trees; MeanSD , the mean Robinson-Foulds symmetric distance per inaccurate tree; BurdED , the edit burden, calculated as the edit distance by which each tree differs from the known tree, summed over the 400 trees; and MeanED , the mean edit distance per inaccurate tree. 3 Inference methods: ML , protein maximum likelihood with gamma ASRV correction; BUU , Bayesian inference, uncorrected for ASRV, uniform prior; BGU , Bayesian inference, gamma ASRV correction, uniform prior; and BGE , Bayesian inference, gamma ASRV correction, exponential prior. See text for further details. Figure 7 Comparative performance with simulated data: incorrect model, one long branch, symmetric distance. Performance at different branch-length ratios of ML and Bayesian inference with simulated protein-sequence data evolved on a tree having a single long branch, measured as Robinson-Foulds symmetric distance. Data were evolved under the mtmam model, but trees were inferred under the JTT model. Panels and axes are as in Figure 2. Figure 8 Comparative performance with simulated data: incorrect model, one long branch, edit distance. Performance at different branch-length ratios of ML and Bayesian inference with simulated protein-sequence data evolved on a tree having a single long branch, measured as edit distance. Data were evolved under the mtmam model, but trees were inferred under the JTT model. Models, panels and axes are as in Figure 2. Figure 9 Comparative performance with simulated data: incorrect model, two long branches, symmetric distance. Performance at different branch-length ratios of ML and Bayesian inference with simulated protein-sequence data evolved on a tree having two long branches, measured as Robinson-Foulds symmetric distance. Data were evolved under the mtmam model, but trees were inferred under the JTT model. Models, panels and axes are as in Figure 2. Figure 10 Comparative performance with simulated data: incorrect model, two long branches, edit distance. Performance at different branch-length ratios of ML and Bayesian inference with simulated protein-sequence data evolved on a tree having two long branches, measured as edit distance. Data were evolved under the mtmam model, but trees were inferred under the JTT model. Models, panels and axes are as in Figure 2. For datasets in which a single branch was of relatively greater length, violating the model of sequence change degraded performance of the four approaches (Table 2 ). In each case, inaccurate trees were first observed at 20-fold branch-length ratio, earlier than the 30-fold ratio seen in the absence of model violation. Inaccurate trees were more numerous, in comparison with inference under the correct model. With ML, each inaccurate tree was about as inaccurate under the incorrect model as under the correct one, as measured by symmetric distance (Table 2 ). With Bayesian inference, inaccurate trees produced under the wrong model were, unexpectedly, sometimes less inaccurate than those inferred under the correct model (Table 2 ), and in one case (no correction for ASRV, uniform prior distribution) the total burden of changes was less, as measured by symmetric distance. Exclusion of results from the 70-fold data (results not shown) demonstrated that this effect is not due to a loss of dynamic range at extreme values. For datasets containing two long branches, model violation affected performance of ML and Bayesian inference differently. With ML, inference under the wrong model produced a somewhat lower frequency of topologically inaccurate trees, although each inaccurate tree was more inaccurate as judged by symmetric distance (Table 2 ). With Bayesian inference, use of the wrong model increased the frequency of inaccurate trees, and each inaccurate tree tended to be more inaccurate as measured by symmetric distance. With Bayesian inference uncorrected for ASRV and using a uniform prior, the first inaccurate tree appeared at a ratio of only 10, and no accurate trees were recovered at ratios 50 or higher; although by most indices the performance was not further degraded by violation of the model of sequence change, performance was already quite poor, and not much dynamic range remained available. Use of an exponential prior again made a significant difference only with two long branches and assessment using Robinson-Foulds symmetric distance (Wilcoxon P ≤ 0.00097 and P ≤ 0.00003 for degraded performance at 60- and 70-fold branch-length ratios respectively). Discussion Unlike the situation with established approaches based on pairwise distances, parsimony or maximum likelihood, relatively little experience has accumulated so far on the application of Bayesian approaches to phylogenetic inference, especially for protein-sequence datasets. In this work we (a) extend the comparison of Bayesian posterior probabilities with nonparametric bootstrap proportions as measures of confidence in subtrees, (b) systematically investigate the robustness of ML and Bayesian inference to branch-length differences, and (c) compare the behavior of these two approaches to one specific violation of the model of sequence change. We used two measures to compare topologies (Robinson-Foulds symmetric distance, and edit distance), and it is clear that they captured different facets of topological incongruence. Support for subtrees Using 21 empirical protein-sequence datasets, we compared Bayesian posterior probabilities with bootstrap proportions based on ML as measures of support for subtrees. To make this comparison as fair as possible, we restricted our analysis to a model of sequence change (JTT) and a correction for ASRV (discrete approximation to the gamma distribution) available in both PROML and MrBayes. We did not optimize models separately for each approach or for each dataset, as JTT+gamma represents the most-parameterized combination that these two programs support in common. It is therefore possible that some of the difference observed between the two measures results from differential sensitivity of ML and Bayesian inference, as implemented in these programs, to deviation of JTT and the discrete gamma distribution from an optimal description of the processes of sequence change that actually gave rise to these sequences (but see the final paragraph under Model violation , below). As it is unlikely that any existing model – certainly any that fails to account for lineage-specific processes and temporal variations along these lineages – fully represents the historical complexity of molecular evolution, the same criticism could be levelled, albeit perhaps in lesser degree, against all current applications of statistically based phylogenetic inference to empirical datasets. The data presented in Figure 1 demonstrate that, at least for these protein-sequence datasets, Bayesian PPs tend to offer a more-generous estimate of subtree reliability than does the nonparametric bootstrap combined with ML. This result supports and extends previous studies with DNA- [ 16 - 18 , 20 - 22 , 49 ] and protein-sequence data [ 18 , 19 ]. Bayesian PPs and nonparametric bootstrap BPs are not commensurate [ 17 , 48 ] and may be seen as "potential upper and lower bounds of node reliability" respectively (page 248 of [ 18 ]). (Being more-generous than a too-conservative measure does not, of course, imply that Bayesian PPs must be too-generous.) Our results strongly suggest that the interpretation of BPs and PPs being developed for nucleotide sequences will be applicable, as well, to protein sequences. For sets of consensus trees inferred from simulated protein-sequence data (Figure 6 ), Bayesian PPs tend to be more generous than nonparametric BPs in estimates of subtree support. However, whereas for empirical protein-sequence data (and nucleotide-sequence data: see references cited immediately above) PPs tend to "saturate", i.e. reach 100% at BP values less than 100% (here around 80%), with our simulated data the relationship between BP and PP resembles a smooth curve reaching 100% PP only at BP greater than 99%. Further studies will be required to disentangle why little or no saturation was observed; possibilities include the structure of our simulated trees ( e.g. their symmetry, or an usually regular spacing of internal nodes), the way that data were evolved on these trees ( e.g. assuming strict independence among sites, or rigorous adherence to the JTT model), and/or the way we summarize the support data for ML ( via extended majority-rule consensus trees). Relative branch-length differences Dissimilar sequences (represented in phylogenetic trees as long branches) create difficulties in phylogenetic analysis. The issue has been most extensively explored in parsimony analysis, where branch length can be an important consideration, e.g. in selection of outgroups and resolution of topologically problematic regions. Parsimony analysis is particularly susceptible to "long branch attraction" (LBA) artefacts, in which two or more branches are resolved adjacent in a tree solely because they are highly divergent from the others [ 2 ]. ML inference can be more robust against LBA, although to our knowledge this has been not been specifically examined for protein-sequence data. We are unaware of any systematic examination of the degree to which Bayesian phylogenetic inference is robust against branch length-based artefact. Our results (Figures 2 , 3 , 4 , 5 ) indicate that for protein-sequence datasets of this size, both gamma-corrected ML and Bayesian inference can be robust to artefact arising from the levels of dissimilarity likely to be encountered in empirical biological data. Both ML and Bayesian inference can be fully robust (within our limits of detection) to at least a 20-fold relative length ratio for a single branch, and both perform nearly as well when two branches are relatively lengthened. When a single branch is lengthened, performance (accurate retrieval of the known topology) degrades slowly as relative branch length increases thereafter; Bayesian inference with gamma correction for ASRV performs best among these alternatives. When two branches are relatively lengthened, the performance of ML, and of ASRV-uncorrected Bayesian inference, falls off much more rapidly, whereas in our simulations ASRV-corrected Bayesian inference was more robust than ML. These performance characteristics have been demonstrated only for protein-sequence datasets of the size, length, composition, divergence and tree shape we examined, and for these implementations of ML (PROML) and Bayesian inference (MrBayes). Applicability to larger, longer, and more divergent protein-sequence datasets, to more-diverse tree shapes, and to different implementations seems highly probable, although further nuance will doubtlessly emerge, and scope may remain for further optimization. Model violation Both the mammalian mitochondrial (mtmam) and JTT models embody empirical probabilistic models of amino acid substitution. Codon usage is highly skewed in mitochondrial genomes compared with the cognate nucleocytoplasmic components [ 50 ], and the amino acid transition probabilities in mtmam differ correspondingly from those in JTT. Nonetheless, for the datasets we examined, both ML and Bayesian inference perform well, at biologically reasonable ratios of branch-length difference, even when the JTT model is used to infer trees from protein datasets evolved under mtmam (Figures 7 , 8 , 9 , 10 ). With one exception, the first inaccurately reconstructed trees were observed at the 20-fold ratio (Table 2 ). Model violation increased the inaccuracy of reconstruction (as measured by the total number of inaccurate trees over the eight branch-length ratios) by 8 to 39% (mean 18%) in the case of one differentially extended branch, and by -6 to 34% (mean 13%) where two branches are lengthened (Table 2 ). In the former case, the total burden of these inaccuracies was 16% and 18% as assessed by symmetric and edit distances respectively. The effect of model violation on accuracy for trees with two differentially lengthened branches was more variable; little change (or even a reduction in burden) was observed for ML and Bayesian inference without ASRV correction, whereas violating the model greatly decreased the accuracy of reconstruction by gamma-corrected Bayesian inference. Nonetheless, even at this reduced accuracy, gamma-corrected Bayesian inference performed much more-accurately than either ML or uncorrected Bayesian inference at branch-length ratios of 20-fold and greater. Particularly in simulations where a single branch was differentially lengthened (Table 2 ), using the wrong model of sequence change sometimes improved some aspects of performance. Thus with ML inference, inference under the wrong model increased both the total number of inaccurate trees and the bipartition burden over the 8 branch-length ratios (400 trees), but each inaccurate tree was, on average, slightly less inaccurate (as assessed by symmetric distance) than those inferred under the correct model of sequence change. The same phenomenon was observed with Bayesian inference using gamma ASRV correction and an exponential distribution of prior probability over branch lengths. With Bayesian inference uncorrected for ASRV, both the total bipartition burden, and the mean inaccuracy of inaccurate trees as assessed by symmetric distance, were lessened under the wrong model. In simulations with two long branches as well (Table 2 ), we observed that with ML inference, model violation reduces the number of inaccurate trees and the burden of edits required to generate them, although the latter was not seen when using symmetric distance as the metric. Others have reported situations in which using the wrong model improves the performance of ML ([ 51 - 53 ] and pp. 272–274 of [ 2 ]). Some of these cases appear to result from the specific placement of long branches in the "anti-Felsenstein zone", where biased estimation can increase the efficiency of finding the correct topology [ 52 , 54 ]. However, this does not explain our results, as we separated the long branches from each other. The degree of insensitivity to model violation we observe for gamma-corrected ML and Bayesian inference goes some way toward mitigating possible concern (see above under Support for subtrees ) that the relative performance of these approaches as reported herein might, in part, reflect their differential sensitivity to sub-optimality in the models used. Measures of tree comparison Our results (Figures 2 , 3 , 4 , 5 , Figures 7 , 8 , 9 , 10 and Table 2 ) illustrate how Robinson-Foulds symmetric distance and edit distance provide non-identical, complementary views of topological incongruence. The former metric enumerates the number of internal nodes that must be collapsed to make two topologies identical, whereas the latter counts the number of break-and-reanneal operations needed to convert one topology into another. The scores are identical if all incongruent subtrees can be reconciled by collapse through, or transfer across, a single internal node, but diverge from each other to the extent that incongruent subtrees are positioned more distantly ( i.e. across more internal nodes) from each other. Our results also illustrate the difference in dynamic range offered by these metrics, while simulation studies [ 17 , 55 , 56 ] indicate their differential sensitivity to overall tree shape and/or local topology. Other tree-comparison metrics are available and may offer advantages, e.g. in distinguishing transformations that affect large numbers of termini from those that affect small numbers of termini, in robustness against displacement of particular termini, or in application to very large trees [ 57 - 59 ]. Conclusions Bayesian inference can be as robust as ML against relative branch-length differences of 20-fold or greater in inference of correct topologies from protein-sequence data, although details depend on the number of relatively long branches, the presence or absence of an effective correction for ASRV, and (presumably) other factors. One might doubt that sequences so dissimilar as to produce a 20-fold (or more) difference in branch lengths could be believably recognised as homologous, or reliably aligned. Bayesian inference can also be as robust as ML to violation of the model of amino acid transition probability. For empirical protein-sequence data that might reasonably be encountered in biological research, then, both gamma-corrected ML and gamma-corrected Bayesian inference perform well in recovering the correct topology. As Bayesian inference is typically very much faster than even a single ML run, not to mention than e.g. 100 or 1000 replicate runs required to estimate bootstrap proportions, ASRV-corrected Bayesian inference must be seen as an important alternative for statistically based phylogenetic analysis of protein-sequence data when computational resources are limiting. It appears that the interpretation of bootstrap proportions and posterior probabilities being developed for nucleotide sequences will apply as well to protein sequences. Our interest in lateral genetic transfer (LGT) [ 7 - 9 ] led us to investigate different measures with which to characterise topological difference among trees. Whether LGT tends to occur primarily among closely related lineages, or alternatively whether the frequency of transfer depends more critically on some other factor (oligonucleotide frequency, common environment) – or indeed is purely random – remains an open question. Attention has recently been focused on hypotheses that accord to close-range LGT the central role in metabolic and physiological innovation [ 60 ] and in shaping organismal phylogeny [ 61 ]. A statistic that captures both the number of transfer events (as does edit distance), and the topological breadth of transfer (as does symmetric distance), would thus be valuable in elucidating the pattern and significance of LGT. For such a statistic to be meaningful in a biological context, it must be sensitive to the annotation (specific phyletic value) of the subtrees involved. Implementation of this, and of a broader range of tree-comparison metrics, in platform-independent software should be a matter of some urgency. Methods Simulated data Simulated data were evolved using the "evolver" program within PAML version 3.13a [ 62 , 63 ]. First, we generated random trees, each with 7 species (sequences), using the settings birth rate 0.2, death rate 0.2, sampling fraction 0.5, and mutation rate 0.5. In one set of runs, 8 additional trees were then produced, in which 1 of these 7 branches (selected at random) was progressively lengthened to be longer than the others by the factors 5, 10, 20, 30, 40, 50, 60 and 70. In a second set of runs, 8 other trees were produced, in which 2 branches were progressively lengthened by these same factors (each long branch in a given tree was extended by the same factor). The branches to be lengthened were selected to be as distant from each other as possible in the tree; for a strictly bifurcating tree with seven termini (leaves), this means that it would have a Robinson-Foulds symmetric distance ([ 64 ]; see below) of 8 if its long branches were forced to become adjacent. For the trees with 1 or 2 branches differentially lengthened 70-fold, branch lengths were reduced proportionally (very slightly) to maintain all absolute values less than 10, so as not to exceed bounds set on Bayesian prior distributions (below). Protein data sets were then generated on each of the 16 trees with differentially lengthened branches, using the "evolver" program in PAML. On each tree we evolved 100 replicate protein datasets under the JTT model of sequence change [ 65 ], with among-sites rate variation (ASRV) modelled as an 8-category discrete approximation to a gamma distribution with alpha (shape) parameter 0.5. In a second set of runs, we similarly evolved 100 datasets under the mtmam [ 62 , 63 ] model, originally named REV [ 66 ], estimated from a set of mammalian mitochondrial proteins. Each protein dataset was of initial length 1000 amino acids. From each of the 32 sets of 100 replicate protein datasets, we then selected 50 replicate protein datasets at random for further analysis. Maximum likelihood inference All maximum-likelihood (ML) trees were inferred using PROML version 3.6a3 in Felsenstein's PHYLIP package [ 34 ] implemented on an 8-processor SGI Origin 2100 under IRIX, a 128-processor SUN Netra-1 cluster under Linux, and a 16-processor IBM p690 Regatta under AIX. In all ML inference we assumed the JTT model of sequence change, randomized (jumbled) the order of sequence addition, used global rearrangements, and selected the "not rough" analysis option in PROML. More information on these settings is available online [ 34 ]. We assumed an 8-category discrete approximation to a gamma distribution, with values for the gamma shape parameter estimated separately for each dataset using Tree-Puzzle [ 67 ], but frequencies for each category estimated by PROML; rates were assumed to be uncorrelated at adjacent sites. For both empirical and simulated data, models and parameter values were selected to facilitate, as much as possible, a fair comparison of ML and Bayesian approaches. Some details of ML inference differed for empirical vs simuated data. Here we present methods for the simulated data; methods specific to the empirical data are given below. For simulated data, we report results from both (1) single ML inference runs based on each of the 50 replicate protein datasets at each branch-length ratio increment, and (2) bootstrapping ( N = 10) each of the 50 replicate datasets, as described in the preceding paragraph. Bayesian inference Bayesian inference (B) was carried out using MRBAYES version 2.01 [ 31 ] implemented on a 16-processor IBM p690 Regatta under AIX, and on a 508-processor Compaq ES45 cluster under Linux. (Version 3.0 of MRBAYES was not used because, at the time these analyses were carried out, no documentation was available on how to force the shape parameter to remain fixed after initialization). Priors were defined over the branch-length interval (0.0,10.0), and the JTT model of sequence change was assumed (or known to be correct) for all analyses. For empirical data, trees were inferred using two models of sequence change: JTT, and a variant ("equalin", EQ) of the F81 model of Felsenstein [ 1 ]. ASRV was modeled as an 8-category gamma distribution, and the shape parameter was optimized by MRBAYES. The prior distribution on branch lengths was assumed to be uniform, and following initial trials (data not shown) the Markov chain temperature was set to 0.2000. For each dataset, 8 Markov chains were propagated for 30,000 generations each and sampled every 100 generations. As preliminary analyses showed convergence within a few thousand generations, burn-in was conservatively set at 10,000 generations. Posterior probabilities were obtained using allcompat ( i.e. , extended 50% majority-rule consensus) among these sampled trees. For simulated data, we examined three models of different complexities: (1) a uniform prior distribution over branch lengths, and a single rate category; (2) a uniform prior, and an 8-category gamma model of ASRV; and (3) an exponential prior, and an 8-category gamma. The gamma shape parameter was, as above, estimated using Tree-Puzzle, and was fixed ( i.e. did not merely serve to initialize estimation by MRBAYES). In runs where an exponential prior was used, the value of the exponent was estimated from the simulated data, and differed according to sequence-change model: under JTT, 0.10 for datasets with both one and two long branches, and under mtmam, 1.04 for one long branch, and 0.60 for two long branches. Comparing topologies and subtree reliabilities For trees inferred from the 16 sets of simulated data (1 or 2 long branches, 8 ratios of branch-length difference), topologies were compared against that of the (known) tree on which the data had been evolved. For this we employed two metrics: (1) the minimum number of break-and-reanneal edits required to convert one tree into the other. This metric goes under various names, including subtree prune and regraft distance [ 33 ]; we refer to it simply as edit distance ; and (2) the Robinson-Foulds symmetric distance [ 64 ] as implemented in TREEDIST in the PHYLIP package [ 34 ]. The values of these metrics were not normalized (cf. [ 68 ]) because all simulated trees have the same number of internal edges. Subtree support was assessed as bootstrap proportion (BP) for ML, and as posterior probability (PP) for Bayesian inference. Empirical data Methods and procedures followed those for simulated data (above), except as described subsequently here. Aligned protein sequence datasets (see Additional file 1 ) were obtained from Dr Nick Goldman (EBI). We selected 21 datasets (240 sequences in total, mean 11.4 sequences per dataset), requiring each to be of interestingly large size (minimum 8 sequences) but not too large for analysis by bootstrapped protein likelihood, given the computational resources available to us (maximum 16 sequences). These were reformatted for further analysis, assigning new designators to anonymize individual sequences and to avoid the use of characters that are not supported within the rule sets of the software programs we used ("illegal characters"). For empirical data, we inferred ML trees in two ways: (1) using a user-defined hidden Markov model (HMM) with 8 categories, each set to 12.5% of sites, and with rates in each category estimated using Tree-Puzzle version 5.0 [ 67 ]; and (2) assuming an 8-category discrete approximation to a gamma distribution, with values for the gamma shape parameter estimated using Tree-Puzzle (but frequencies for each category estimated by PROML), rates at adjacent sites assumed to be uncorrelated, and the among-sites rate variation (ASRV) gamma-shape parameter estimated for each dataset using Tree-Puzzle. In all our use of Tree-Puzzle, we assumed 8 rate categories (each covering one-eighth of the aligned positions) and JTT. For simulated data, ML trees were inferred using only the 8-category discrete approximation to a gamma distribution. ML analyses were bootstrapped ( N = 100 for the 21 empirical data sets, N = 10 for the 1600 simulated datasets) with preservation of rate-class information as described in the SEQBOOT documentation. Nonparametric bootstrap proportions (BPs) were computed under extended majority rule consensus using CONSENSE. Both SEQBOOT and CONSENSE are in PHYLIP [ 34 ]. For the 21 sets of empirical data, no "true" tree is available. Topologies resulting from the different inference approaches and models (ML-JTT-HMM, ML-JTT-gamma, B-JTT, B-EQ) were therefore compared amongst themselves, using edit distance (determined manually) as the comparison metric. Topologies of the bootstrap ML consensus trees were compared as well, although as consensus trees they do not necessarily reflect most-likely topologies. Availability of data The 21 empirical protein-sequence datasets from Dr Nick Goldman, and our simulated datasets with one or two long branches, are available for download at [ 69 ]. Authors' contributions JCM was responsible for data simulation and analysis. TJH was responsible for high-performance computing, and generated the figures. MAR initiated and supervised the project, and wrote the manuscript. Supplementary Material Additional File 1 Description of 21 empirical datasets This PDF file contains information on each of the 21 empirical datasets provided by Dr Nick Goldman, including: number of sequences, GenBank ID (gi number) of first sequence in dataset, key words from description line of first sequence, PAUP* parsimony score of dataset, number of internal nodes, and number of zero-length internal edges observed with PROML to have support in three non-overlapping intervals: P < 0.01, P < 0.05 but not P < 0.01, and worse than P < 0.05. Click here for file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549035.xml |
534791 | Occurrence of symptoms and depressive mood among working-aged coronary heart disease patients | Background The typical symptoms of coronary heart disease (CHD), chest pain and breathlessness, are well-known. They are considered quite dramatic, and can thus be fairly reliably mapped by a survey. However, people might have other clearly unpleasant symptoms impairing quality of life. The aim of this study is to evaluate the appearance of these complaints of working-aged people with self-reported CHD. Methods The study consists of a postal questionnaire of randomly selected Finns in age groups 30–34, 40–44 and 50–54, a response rate of 39% (N = 15,477). The subjects were asked whether or not a doctor had told them that they had angina pectoris or had had myocardial infarction. Four randomly selected age and sex matched controls were chosen for every patient. The occurrence of self-reported dyspnoea, chest pain during anger or other kind of emotion, palpitation and perspiration without physical exercise, irregular heartbeats, flushing, trembling of hands and voice, jerking of muscles, depression and day-time sleepiness were examined. Odds ratios (OR) with 95% confidence intervals (CI), between occurrence of symptoms and CHD with and without heart infarction, were computed by multivariate logistic regression analysis. Results The sample eventually comprised 319 CHD patients. Dyspnoea, chest pain during anger or other kind of emotion, palpitation, perspiration without physical exercise, irregular heartbeats daily or almost daily, trembling of hands and voice, and jerking of muscles occurred statistically significantly more frequently among CHD patients than among controls. The CHD patients also reported more depressive mood according to Beck's inventory scores and poorer sleep and more frequent day-time sleepiness than controls. In the multivariate logistic regression analysis chest pain during anger or other kind of emotion (ORs 4.12 and 3.61) and dyspnoea (ORs 2.33 and 3.81) were the symptoms most associated with CHD. Conclusions Working-aged people with self-reported coronary heart disease evince a number of symptoms limiting the quality of their every day life. This aspect should be paid attention to when evaluating functional capacity of these patients. | Background In Finland, as in most industrialised countries mortality from cardiovascular diseases has shown decreasing trends since around 1970 [ 1 , 2 ]. The Mini-Finland survey from the years 1979–80 revealed that the angina pectoris symptom (i.e. reported chest pain under physical strain) may already appear in both sexes at the age of 30, though it was not until the age of 65 that it becomes more common among men compared to women [ 3 ]. The typical symptoms of coronary heart disease (CHD), chest pain and breathlessness, are well-known. They are considered quite dramatic, and thus can be fairly reliably mapped by a survey [ 4 ]. However, coronary heart disease patients also have other complaints in respect of their health, for example fatigue and sleep problems [ 5 ]. It has also been estimated that 17% to 27% of patients with coronary artery disease have major depression and a significantly larger percentage has subsyndromal symptoms of depression [ 6 ]. The diagnosis of CHD is usually based on medical examinations or register data and not on what people by themselves experience. However, people act and use health services according to what they experience to suffer from and what they experience to limit their ability to work. Since CHD patients have many other symptoms than the traditional and well-known, it is important to know what the spectrum of symptoms and complaints among working-aged people with self-reported coronary heart disease is in relation to functional capacity. Methods Design The Health and Social Support study (HeSSup) is a prospective etiological follow-up study on the psychosocial health of the Finnish working-aged population. The HeSSup population consisted of a random sample of 39,563 individuals drawn from the Finnish Population Register in three age groups: 30–34, 40–44, and 50–54. The survey was carried out by postal questionnaire. Forms were returned by 15,477 individuals (approximately 5,000 in each age group), a response rate of 38.9% (37.6% in 30–34, 37.9% in 40–44, and 41.1% in 50–54). The sample was subjected to a thorough analysis of non-response [ 7 ]. The analysis was made using the official statistics of the Finnish population for the corresponding age groups in 1998 to assess whether the study population adequately represented the Finnish population. Diagnosed epilepsy and diagnosed hypertension were selected to represent chronic diseases. The major reasons for refuse were the length of the questionnaire and above all suspicion of the purpose behind the request for written consent. Less educated, divorced, widowed, unemployed and those on disability pension were least willing to participate. Differences in physical conditions between the study participants and the whole population were, however, small. It was also noted that people suffering from hypertension returned the questionnaire somewhat less readily than others. Material and methods The subjects were asked whether or not a doctor had told them that they had angina pectoris or had had myocardial infarction. The perceived state of health was determined according to Likert's five-step scale (good, quite good, fair, rather poor and poor). In order to avoid small frequencies in the analyses this was modified to a three-step scale (good, fair, poor). The appearance of dyspnoea was categorised into four degrees of difficulty according to a widely used cardiovascular survey method [ 4 ]. Persons who suffered from dyspnoea when walking uphill or upstairs comprised the group of mild symptoms. Those out of breath when walking on level ground at normal speed with other people of the same age comprised the intermediate group, those who had to stop walking on level ground due to breathlessness comprised the group of difficult dyspnoea and those who became breathless even while standing still or while washing or dressing themselves, comprised the group of extremely difficult symptoms. Participants were also asked whether or not they had experienced daily or weekly chest pain during anger or other kind of emotion, palpitation and perspiration without physical exercise, flushing, trembling of hands and voice, and jerking of muscles. Irregular heartbeats daily or weekly were asked, too. Depression was estimated by Beck's [ 8 ] depression scale ranging from 0 to 63. The normal score on this scale is below 10. In mild depression the scores are between 10 and 19 [ 9 ]. It was also asked how well and how many hours a day the participants had usually slept and how often they had felt day-time sleepiness, which when occurring daily or almost daily has been proved to be associated with depression, insomnia and breath interruptions during sleep [ 10 ]. Analyses In interpretation of results the coronary heart disease (CHD) patients were divided into two groups. The first comprised coronary patients not having had heart infarction (angina pectoris group) and the second patients having had heart infarction (infarction group). In order to have the best available comparison groups, four randomly selected age and sex matched controls for comparison were selected for every patient. Thus there were altogether 740 controls for the angina pectoris group and 536 controls for the infarction group. Stroke was ruled out in the control groups, but otherwise there were no differences between the CHD groups and their respective controls. The associations, odds ratio (OR) with 95% confidence intervals (CI), between symptoms and coronary heart disease with and without heart infarction, were computed by multivariate logistic regression analysis. The analyses were made using the SAS System for Windows, release 8.2/2000. Results The data comprised 319 patients: 185 coronary heart disease patients who had not experienced heart infarction (55.1% were men) and 134 patients who had (78.4% were men) (Table 1 ). Most of the CHD patients were in the oldest age group, and almost 90% of those who had had a heart infarction were in the age group 50–54. In all age groups the prevalence of self-reported CHD was higher among men than among women (Table 2 ). Table 1 The coronary heart disease patients studied according to age and gender Angina pectoris group Myocardial infarction group Women Men Total Women Men Total Age group N N N % N N N % 30–34 16 14 30 16 4 3 7 5 40–44 20 28 48 26 3 9 12 9 50–54 47 60 107 58 22 93 115 86 Total 83 102 185 100 29 105 134 100 Table 2 Prevalence of self-reported coronary heart disease according to age and gender Angina pectoris Myocardial infarction Women Men Women Men Age group % % % % 30–34 0.5 0.7 0.1 0.2 40–44 0.7 1.4 0.1 0.4 50–54 1.7 2.8 0.7 4.0 Perceived state of health State of health was perceived as good or quite good by 37.3% in the angina group and by 24.6% in the infarction group. The corresponding figures in the control groups were 76.0% and 67.4%, the differences being statistically significant (p < 0.001). State of health was perceived as poor or rather poor by 28.1% in the angina group and by 32.8% in the infarction group. Symptoms and complaints At least mild breathlessness occurred in two thirds of the angina group and three fourths of the infarction group (Table 3 ). Difficult or extremely difficult breathlessness was reported by 20.2% in the angina group and by 27.8% in the infarction group. The corresponding figures in the control groups were 1.4% and 4.0%, the differences also being statistically significant (p < 0.001). Chest pain during anger or any kind of emotion, palpitation and perspiration without physical exercise, irregular heart beats, and jerking of muscles were all both daily and weekly statistically significantly more common among CHD patients than among controls. Almost daily CHD patients reported more daytime sleepiness and trembling of hands and voice than controls. CHD patients also slept more poorly than controls, and sleeping hours ≤ 6 in a day was more common among then than among controls. CHD patients scored higher on the depression scale than the controls, the average score being 10.2 (95% CI 9.0–11.4) in the angina group and 5.8 (95% CI 5.4–6.2) in the control group. In the infarction group the average score was 9.7 (95% CI 8.4–11.0). The corresponding figure in the control group was 5.9 (95% CI 5.4–6.5). In both coronary heart disease groups at least mild depression was twice as common as among controls. Flushing was the only complaint, which was not statistically significantly more common among CHD patients than among controls. Table 3 Occurrence (%) of symptoms and complaints in coronary heart disease (CHD) patients and the control population Angina pectoris Controls Myocardial infarction Controls N = 177–185 N = 728–736 N = 129–134 N = 514–525 % % p % % p At least mild dyspnoea (Rose and Blackburn 1968) 66.5 33.2 <0.001 75.4 35.3 <0.001 Chest pain during anger or emotion Almost daily 12.2 1.0 <0.001 16.3 0.8 <0.001 Weekly 12.8 2.5 <0.001 16.3 3.3 <0.001 Palpitation without physical exercise Almost daily 14.1 3.4 <0.001 20.8 3.5 <0.001 Weekly 15.3 5.2 <0.001 16.9 5.4 <0.001 Perspiration without physical exercise Almost daily 22.7 9.7 <0.001 26.0 11.0 <0.001 Weekly 18.2 10.2 0.003 16.8 8.3 0.001 Irregular heart beats Almost daily 15.7 3.5 <0.001 24.0 3.7 <0.001 Weekly 12.4 6.1 0.004 14.0 4.4 <0.001 Depression (Beck ≥ 10) 41.6 20.4 <0.001 43.3 21.0 <0.001 Sleeping hours ≤ 6 in a day 16.8 10.3 0.015 19.4 10.1 0.003 Poor sleep usually 25.0 15.4 0.002 32.1 13.0 <0.001 Daytime sleepiness Almost daily 33.2 11.2 <0.001 32.1 14.9 <0.001 Flushing Almost daily 11.1 7.0 0.066 10.9 7.6 0.228 Weekly 12.8 7.1 0.014 9.3 8.0 0.624 Trembling of hands Almost daily 13.8 2.3 <0.001 9.2 3.3 0.004 Weekly 8.3 4.7 0.053 10.7 4.0 0.003 Trembling of voice Almost daily 3.3 1.1 0.030 5.3 1.2 0.002 Weekly 3.3 1.2 0.049 3.8 2.3 0.341 Jerking of muscles Almost daily 13.7 2.7 <0.001 12.3 4.4 0.001 Weekly 11.5 2.7 <0.001 9.2 3.8 0.011 ORs of reported symptoms In the multivariate logistic regression analysis chest pain during anger or other kind of emotion and dyspnoea were the symptoms most associated with CHD (Table 4 ). Irregular heart beats and perspiration without physical exercise were also strongly associated with heart infarction, but not with CHD without heart infarction. On the other hand, jerking of muscles was strongly associated with CHD without heart infarction, but not with heart infarction. Table 4 Age- and sex-matched ORs with 95% CI in the multivariate logistic regression analysis for reported symptoms of coronary heart disease (CHD) without and with heart infarction. All of these symptoms were in the same model. Statistically significant associations are bolded. Angina pectoris Myocardial infarction OR (95% CI) OR (95% CI) Dyspnoea 3.81 (2.16–6.72) 2.33 (1.21–4.50) Chest pain during anger or emotion* 3.61 (1.68–7.77) 4.12 (1.72–9.84) Palpitation without physical exercise* 1.19 (0.55–2.59) 1.39 (0.57–3.39) Irregular heart beats* 1.46 (0.69–3.08) 3.12 (1.28–7.60) Perspiration without physical exercise* 1.44 (0.87–2.38) 2.19 (1.15–4.14) Flushing* 1.11 (0.62–1.97) 2.02 (0.89–4.59) Trembling of hands* 1.36 (0.58–2.93) 1.01 (0.37–2.67) Trembling of voice* 2.58 (0.85–7.87) 1.17 (0.29–4.69) Jerking of muscles* 2.44 (1.20–4.96) 1.37 (0.55–3.41) Depression (Beck ≥ 10) 1.57 (0.99–2.48) 1.01 (0.52–1.98) Poor sleep 1.30 (0.74–2.27) 1.39 (0.70–2.77) Daytime sleepiness 1.24 (0.76–2.01) 1.41 (0.78–2.56) Sleeping hours ≤ 6 in a day 1.21 (0.64–2.26) 1.08 (0.50–2.33) * almost daily or weekly Discussion The principal finding in this study was that many working-aged coronary heart disease patients experience unpleasant symptoms such as dyspnoea, chest pain during anger or emotion, irregular heart beats, perspiration without physical exercise, and jerking of muscles. In addition, the frequency of most of the self-reported symptoms among the study population is higher also in respect of those symptoms, which would not be expected at least among those CHD patients whose disease is in good balance. Working-aged CHD patients may be regarded as a special group compared with the main part of CHD patients who are already by age entitled to a pension. It is likely that working-aged CHD patients have experienced the most widespread and intense exposure to risk factors, which thus has caused them this disease among the first ones within their age group. As working-aged they are wished, however, to return back to normal life and work as soon as possible. According to our study they still have a lot of symptoms concerning their every day life, which harm their recovery and rehabilitation. It is also noteworthy that many of the working aged CHD patients are still in working life. Although chest pain and dyspnoea do not prevent them to work at customer service, many of the symptoms such as trembling of hands interfere their normal jobs while appearing mostly in rest. Trembling of hands and voice are also very irritating symptoms, and they may be considered shaming. Thus they may interfere social life and reduce the quality of life. The study material may be considered representative of the Finnish working-aged population, although the response rate was only 39%. Careful non-response analysis indicated that respondents and non-respondents were comparable in respect of the most important demographic variables [ 7 ]. It is possible that CHD patients respond to the questionnaire more actively than other people. On the other hand, there are certainly those among CHD patients who neglect their disease and are not willing to respond. However, we do not know for sure whether there is an over or under estimation of the associations, but we can presume that these two factors compensate each other. Moreover, it is unlikely that the principal association studied, i.e. the association between CHD and appearance of symptoms would be a substantially different one in non-participants. The findings reflect the respondents' own conception of their symptoms. The own conception of symptoms is important, since according to findings from a 3 years' follow-up of 4,000 men, self-reported coronary heart disease predicts very strongly a new coronary event [ 11 ]. In addition, the presence of anginal symptoms may be an important independent correlate of prognosis in patients with CHD [ 12 ]. Our method to determine the existence of CHD is based on the patients' report on whether a doctor had told them that they suffered from this particular disease. Thus, we cannot know for sure the accuracy of the information reported. Nowadays, CHD is rare among young people [ 1 , 2 ]. In our data there still is some. It is possible that there is a combination of several risk factors in the background. There are validated instruments, such as the generic The Short Form 36 Health Survey [ 13 ] (SF-36), the Nottingham Health Profile [ 14 ], and the Seattle Angina Questionnaire [ 15 ], to investigate health-related quality of life. Our method to examine the subject was to ask about complaints and symptoms. Most of these questions have been successfully used in cardiovascular surveys [ 4 ] and in the Finnish Twin Cohort Study [ 16 ]. In addition, mood was estimated according the Beck's depression scale [ 8 ]. The occurrence of dyspnoea and chest pain even during anger or other kind of emotion may be considered a finding that was expected. In an American study on the care of coronary heart patients at the emergency department the most frequently reported symptom was chest pain (70% among men and 71% among women) and dyspnoea (30% men and 29% among women) [ 17 ]. The typical chest pain is also a symptom more predictive of an acute coronary attack in working-aged than in older patients [ 18 ]. It was suggested as far back as 1987 that palpitations are not an independent risk factor for increased cardiac morbidity or mortality [ 19 ]. However, according to a British study those experiencing palpitations at work and while asleep were more likely to have a cardiac cause for their palpitations [ 20 ]. Our finding was that working-aged CHD patients report palpitations more often than the control population. The high occurrence of trembling of hands and voice, and jerking of muscles may be considered an unexpected finding. Most CHD patients use beta-blocking medicines, which in addition to protecting the heart muscle also reduce the adrenergic stimulation and thus relief the symptom of trembling. Concerning depression our findings support those of previous studies. One out of four of symptomatic coronary heart disease patients have namely been found to have a probable depressive disorder, but none of them had previously been identified as suffering from depression or been treated for this reason [ 21 ]. In primary care it is of vital importance to notice symptoms of this illness, since continuing depression has been found to be associated with increased risk of mortality among CHD patients following hospital discharge [ 22 ]. It has also been verified that depression is common after coronary heart disease events such as bypass grafting, coronary angioplasty, myocardial infarction and myocardial ischaemia [ 23 ]. In our study about 40% of CHD patients had at least minor depression compared with 20% among controls. The high depression rates are probably due to our method to diagnose depression at ≥ 10 points in Beck's inventory scale. Thus we do not think there are any selection bias, since the controls were randomly selected. Furthermore, it is not probable that depressed people respond to our questionnaire more readily than people not suffering from low mood. It was also of noteworthy that daytime sleepiness was connected with coronary heart disease. A cross-sectional study of 5,419 Finnish adult men found a higher prevalence of diagnosed myocardial infarction among those who slept more than nine hours, whilst those sleeping less than six hours per night had more symptomatic coronary disease [ 24 ]. In a Swedish study concerning working-aged women poor sleep was associated with an increase in spasmodic chest pain and irregular heart beat [ 25 ], whereas in men an association between difficulties falling asleep and CHD mortality has been found [ 26 ]. From previous research we know that despite having survived a life-threatening clinical event, CHD patients appear to have continued adverse behaviours such as smoking, being obese and having frequent hangovers more than the control population [ 27 ]. The follow-up of our cohort will show in what extent the symptoms we found are indicators of increased risk of CHD among working-aged people and to what extent the symptoms are result of CHD and its care. In both cases particular attention should be paid to these aspects in primary care. Conclusions According to the present findings many working-aged people with self-reported coronary heart disease perceive their state of health as poor or rather poor. They suffer from a wide range of symptoms limiting their every day life. It is noteworthy that many of these symptoms are not only irritating, but constitute a threat to health. The health related quality of life is poor among working-aged coronary heart disease patients. Competing interests The author(s) declare that they have no competing interests. Authors' contributions MPTS drafted the manuscript; SBS participated in drafting of the manuscript; MJK participated in the design of the study and the statistical analyses; LHS participated in the statistical analyses; KJM conceived of the study, and participated in its design and co-ordination. All authors have read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC534791.xml |
539289 | Transthoracic coronary flow reserve and dobutamine derived myocardial function: a 6-month evaluation after successful coronary angioplasty | After percutaneous transluminal coronary angioplasty (PTCA), stress-echocardiography and gated single photon emission computerized tomography (g-SPECT) are usually performed but both tools have technical limitations. The present study evaluated results of PTCA of left anterior descending artery (LAD) six months after PTCA, by combining transthoracic Doppler coronary flow reserve (CFR) and color Tissue Doppler (C-TD) dobutamine stress. Six months after PTCA of LAD, 24 men, free of angiographic evidence of restenosis, underwent standard Doppler-echocardiography, transthoracic CFR of distal LAD (hyperemic to basal diastolic coronary flow ratio) and C-TD at rest and during dobutamine stress to quantify myocardial systolic (S m ) and diastolic (E m and A m , E m /A m ratio) peak velocities in middle posterior septum. Patients with myocardial infarction, coronary stenosis of non-LAD territory and heart failure were excluded. According to dipyridamole g-SPECT, 13 patients had normal perfusion and 11 with perfusion defects. The 2 groups were comparable for age, wall motion score index (WMSI) and C-TD at rest. However, patients with perfusion defects had lower CFR (2.11 ± 0.4 versus 2.87 ± 0.6, p < 0.002) and septal S m at high-dose dobutamine (p < 0.01), with higher WMSI (p < 0.05) and stress-echo positivity of LAD territory in 5/11 patients. In the overall population, CFR was related negatively to high-dobutamine WMSI (r = -0.50, p < 0.01) and positively to high-dobutamine S m of middle septum (r = 0.55, p < 0.005). In conclusion, even in absence of epicardial coronary restenosis, stress perfusion imaging reflects a physiologic impairment in coronary microcirculation function whose magnitude is associated with the degree of regional functional impairment detectable by C-TD. | Introduction Percutaneous transluminal coronary angioplasty (PTCA) has deeply modified the effective management of coronary artery disease [ 1 ]. Coronary artery restenosis is unfrequent when PTCA is associated to coronary stenting application which is able to enlarge the lumen area stenosis [ 2 , 3 ]. However, even in absence of coronary artery restenosis, the results of revascularization can be suboptimal because of coronary microvessel dysfunction subsequent to the procedure [ 4 , 5 ]. This issue may be intriguing for management of patients undergone PTCA. The non-invasive assessment after PTCA is usually performed by gated single photon emission computerized tomography (g-SPECT) [ 6 , 7 ] and by stress echocardiography [ 7 , 8 ]. However, both these tools present technical limitations, g-SPECT having a low specificity [ 9 ] and semi-quantitative echocardiographic wall motion analysis low sensitivity [ 10 ]. In the last years, great interest has been developed about new echocardiographic techniques as Doppler-derived coronary flow reserve (CFR) [ 11 , 12 ] and color Tissue Doppler (C-TD) [ 13 , 14 ]. The first tool provides reliable information about coronary microvascular function in absence of epicardial coronary stenosis [ 15 ] while C-TD is able to quantify left ventricular (LV) myocardial performance both at rest and during pharmacological stress [ 13 , 14 ]. On these grounds, aim of the present study was to assess C-TD derived myocardial performance, both at rest and during pharmacologic stress, in relation to the function of coronary microcirculation determined by non invasive CFR after successful PTCA of LAD. Methods Study population Among 30 patients who had undergone PTCA with stenting for significant LAD stenosis between September and October 2000, 24 patients (age = 50–64 years) free of coronary angiographic evidence of LAD restenosis 6 months after the procedure, entered the study and performed non-invasive test screening in the same period of the invasive assessment (±7 days). The informed consent of all patients and approval of Institutional Committee were obtained. Patients were excluded for acute and previous myocardial infarction (according to ECG at rest), concomitant coronary stenosis of right coronary artery and/or circumflex artery, congestive heart failure, valvular heart disease, primary cardiomyopathy, atrial fibrillation, inadequate quality echocardiograms. On the basis of g-SPECT dipyridamole induced perfusion defects, the study population was divided into 2 groups: without and with perfusion defects. Procedures Patients underwent dipyridamole gated myocardial perfusion g-SPECT acquisition, transthoracic echocardiography, C-TD (both at baseline and during dobutamine stress) and non-invasive CFR determination by dipyridamole test. All echocardiographic measurements were analyzed without knowledge of the clinical data. According to the rules of the Institutional Committees, all patients withdrew cardiac drugs at least 2 days before the performance of the non invasive assessment. Dipyridamole g-SPECT Single day rest/dipyridamole g-SPECT was performed according to the standard methods by injecting patients with technetium 99m ( 99m Tc) tetrophosmin 8 mCi (296 MBq) at rest and 24 mCi (888 MBq) after dipyridamole infusion by volumetric pump (dose of 0.14 mg/kg/min in 4 minutes) [ 16 , 17 ]. A single stress SPECT corresponds to a dose exposure of about 500 chest x-ray. Qualitative assessment of reconstructed gated images was obtained on mid-short axis slices, vertical and horizontal long axis slices. Transthoracic Echocardiography Standard echocardiographic examinations were performed using a System FiVe, Vingmed Sound AB machine (GE, Horten, Norway), by a 2.5 MHz transducer equipped with second harmonic capability. M-mode echocardiographic analysis was performed according to the criteria of the American Society of Echocardiography [ 18 ] and LV mass indexed for height powered to 2.5 [ 19 ]. LV end-diastolic and end-systolic volumes were estimated according to the Simpson method [ 20 ] and LV ejection fraction derived. Stress protocols Dobutamine stress protocol was performed according to the standard method [ 21 ] using low and high-dose (up to 40 μg/Kg/min) by using the System FiVe Vingmed machine. C-TD of posterior septum was recorded at rest and during each dobutamine stage. CFR assessment was performed by HDI 5000 ultrasound machine (ATL Ultrasound, Bothell, Washington, USA), using a high-frequency (7 MHz) transducer. The visualization of the distal portion of the left anterior descending artery and the recording of PW-Doppler derived coronary blood flow velocities performed at baseline and after dipyridamole infusion (0.56 mg/kg over 4 minutes) according to the standards of our laboratory [ 14 ]. Blood pressure and a 12-lead ECG were recorded at rest and at the end of each stage of both dobutamine and dipyridamole tests. C-TD dobutamine stress echocardiography analysis Echocardiographic images were recorded on S-VHS videotapes and digitally stored on magneto-optical disk for subsequent analysis. Images were evaluated by 2 experienced observers. Baseline and stress wall motion analysis was performed by 2 experienced readers blinded to the other data. Regional wall motion was assessed with a 16-segment model of the left ventricle and semiquantitatively graded from 1 to 4 as follows: 1 = normal; 2 = hypokinesia; 3 = akinesia; and 4 = dyskinesia. A wall motion score index (WMSI), obtained dividing the sum of each segment scores by the number of the segments, was assessed both at baseline and at high-dose dobutamine. C-TD acquisition of posterior septal wall was performed in real time, superimposed on 2-D images, at baseline and at the end of each dobutamine infusion stage. C-TD imaging was stored in digital format and analyzed off line on cine-loop as previously described [ 14 ]. The region of interest was the middle segment of the posterior septum where myocardial velocity profile was obtained. The middle posterior septum for measurements of C-TD was chosen since the perfusion of this myocardial segment is provided by a branch of LAD, where also CFR was determined. The reproducibility of C-TD of our laboratory has been reported, the intra- and inter-observer variability being <3% and <6% for all the measurements both at rest and at high-dose dobutamine [ 22 ]. CFR Analysis Methods and reproducibility (intra-observer and inter-observer variability 1.9% and 4.2% respectively) of our laboratory in measuring coronary blood flow reserve has been described [ 22 ]. By placing sample volume on the color signal, spectral Doppler of LAD flow showed the characteristic biphasic flow pattern with a larger diastolic and a smaller systolic component. Diastolic peak velocities were measured at baseline and after dipyridamole, by averaging the highest 3 spectral Doppler signals for each measurement. CFR was defined as the ratio of hyperemic to basal diastolic peak velocities. All images were recorded on a magneto-optical disk and analyzed off-line by 2 independent observers, blinded to the other data. Statistical Analysis The analyses were performed by SPSS for Windows release 8.0 (Chicago, Illinois, USA). Data are presented as mean value ± SD. Analysis of variance was used to assess intergroup differences. Linear regression analyses and partial correlation test was done using Pearson's method. Differences were considered significant at p < 0.05. Results Characteristics of the study population The characteristics of the study population and both heart rate and blood pressure at baseline and at high-dose dobutamine are listed in Table 1 . The 2 groups were comparable for heart rate and blood pressure values both at rest and at stress dobutamine peak. Of note, the prevalence of arterial hypertension, diabetes mellitus, hypercholesterolemia and smoke was not different between groups and no patients of both groups presented g-SPECT derived myocardial perfusion defects at rest (data not reported in Table). Table 1 Characteristics of the study population Variable Normal Perfusion n = 13 Perfusion Defect n = 11 P Age (years) 55.9 ± 4.1 58.4 ± 3.1 NS Body mass index (Kg/m 2 ) 26.1 ± 1.1 26.3 ± 0.8 NS Baseline Systolic BP (mm Hg) 147.0 ± 7.5 149.2 ± 11.6 NS Baseline Diastolic BP (mm Hg) 85.1 ± 7.5 85.5 ± 9.1 NS Baseline Heart rate (bpm) 74.8 ± 5.9 73.7 ± 6.9 NS DOB Systolic BP (mm Hg) 151.1 ± 7.5 151.9 ± 10.2 NS DOB Diastolic BP (mm Hg) 83.2 ± 6.3 83.4 ± 7.3 NS DOB Heart rate (bpm) 139.6 ± 5.4 141.0 ± 5.5 NS BP = Blood Pressure, DOB = Dobutamine Echocardiographic analysis The comparisons of echocardiographic measurements and CFR between the 2 groups are reported in Table 2 . Because of higher septal and posterior wall thickness, patients with myocardial perfusion defects after PTCA had greater LV mass index (p < 0.05). LV ejection fraction was comparable between the 2 groups. Table 2 Standard Doppler echocardiographic and CFR analysis Variable Normal Perfusion Perfusion Defect P Septal wall thickness (mm) 10.1 ± 1.4 11.2 ± 0.4 <0.02 Posterior wall thickness (mm) 10.2 ± 1.4 10.6 ± 0.5 NS LV internal diastolic diameter (mm) 54.7 ± 2.6 56.7 ± 3.5 NS LV internal systolic diameter (mm) 39.2 ± 2.8 39.6 ± 2.9 NS 2-D Ejection Fraction (%) 54.8 ± 6.0 54.9 ± 3.4 NS LV mass index (g/m 2.7 ) 49.6 ± 10.7 57.9 ± 8.0 <0.05 LV = left ventricular Dobutamine test and Color TD analysis WMSI was comparable between the 2 groups at baseline (1.07 ± 0.10 versus 1.15 ± 0.11) whereas it was higher at low-dose dobutamine (1.07 ± 0.11 versus 1.17 ± 0.12) and at high-dose dobutamine (1.07 ± 0.12 versus 1.20 ± 0.14) (both p < 0.05) in patients with SPECT-derived perfusion defects than in controls. Positive dobutamine stress-echo involving LAD territory (and in particular mid-septal region) was observed in 5/11 patients (45.4%) with SPECT perfusion defects. C-TD diastolic measurements of mid-septum (E m , A m , E m /A m ratio) were similar between the two groups at rest (E m /A m ratio = 1.04 ± 0.1 and 1.03 ± 0.3 in patients with and without perfusion defects respectively, NS) and at low dose dobutamine (E m /A m ratio = 1.00 ± 0.1 and 1.13 ± 0.4 respectively, NS) while E m /A m ratio was mildly different at high-dose dobutamine (0.83 ± 0.2 and 0.70 ± 0.2 respectively, p < 0.05). S m peak velocities were lower in patients with perfusion defects at low- (p < 0.05) and at high-dose dobutamine (p < 0.01) and were significantly lower also in patients with perfusion defects showing stress induced wall motion abnormalities in comparison with patients with perfusion defects but no change of wall motion during dobutamine infusion (Figure 1 ). Figure 1 In the left panel comparison of S m peak velocity of middle posterior septum of patients without and with SPECT perfusion defects at rest, at low and at high-dose dobutamine. In the right panel comparison of S m peak velocity of middle posterior septum during dobutamine stress echocardiography in patients with perfusion defects having or not dobutamine-induced wall motion abnormalities. CFR analysis Coronary diastolic peak velocities were similar at rest between the two groups (21.5 ± 5.2 cm/s in patients without perfusion defects and 22.0 ± 19 cm/s in patients with perfusion defects, NS) but significantly different after dipyridamole (60.0 ± 16.1 cm/s versus 49.1 ± 10.8 cm/s respectively, p < 0.05). Thus, CFR was 2.87 ± 0.6 in patients without defects and 2.11 ± 0.4 in patients with perfusion defects (p < 0.002). Of note, analyzing the group with SPECT-derived perfusion defects, the patients with stress inducible wall motion abnormalities had lower CFR (1.91 ± 0.1) than those without change of WMSI during dobutamine infusion (2.28 ± 0.4) (p = 0.06). Relationship between CFR and Dobutamine stress measurements In the overall population, CFR was negatively related to WMSI at low-dose (r = -0.46, p < 0.02) and high-dose dobutamine (r = -0.50, p < 0.01). Among C-TD Doppler measurements, CFR was positively related to S m peak velocity of middle septum at low dose (r = 0.39, p < 0.05) and high-dose dobutamine (r = 0.55, p < 0.0005) (Figure 2 ) while the relation of S m at baseline (r = 0.12) did not achieve the statistical significance. No relation of CFR was found with C-TD diastolic measurements of middle septum at any stage of dobutamine stress. Figure 3 shows a patient with SPECT derived normal perfusion: CFR is >2 and middle septal S m peak velocity has a significant increment from baseline to high-dose dobutamine (Δ = +9). Figure 4 displays a patient with a perfusion defect: CFR is reduced and middle septal S m increase from baseline to high-dose dobutamine is lower (Δ = +5). Figure 2 Positive association between CFR and C-TD derived S m peak velocity of middle septum at high-dose dobutamine. Full circles indicate patients with SPECT-derived myocardial perfusion defects; empty circles indicates patients without perfusion defects. Figure 3 CFR and S m peak velocity of middle septum at high-dose dobutamine in a patient with SPECT derived normal perfusion. The upper panels show coronary artery flow velocity in the LAD at baseline and with a normal increase with dipyridamole (DIP). In the lower panels, myocardial systolic velocity (S m ) shows a normal increase at high-dose dobutamine. Figure 4 CFR and septal S m peak velocity at high-dose dobutamine in a patient with SPECT derived perfusion defect. The upper panels display a reduced CFR. In the lower panels, the increase of septal S m from baseline to high-dose dobutamine is low. Discussion The present study used new ultrasound tools, as off-line quantitative C-TD [ 13 , 14 ] and Doppler-derived CFR [ 15 ], to evaluate long-term effects of PTCA on the relation between myocardial performance and coronary microvascular function, in the absence of angiographic coronary artery restenosis. According to dipyridamole g-SPECT, the population was divided into 2 groups, the first without and the second one with perfusion defect in the LAD territory, comparable for resting LV ejection fraction. Our findings show that, six months after successful PTCA of LAD, patients with myocardial perfusion defects present both lower CFR and reduced peak dobutamine myocardial systolic function of wall involved by LAD perfusion (i.e., middle posterior septum) in comparison with the control group and that CFR is positively related to stress peak S m velocity measured at middle septum in the overall population. CFR and perfusion defect after PTCA According to the study design, we intentionally selected patients without angiographic evidence of post-PTCA LAD restenosis. Worthy of note, the incidence of coronary restenosis was very low 6 months after the procedure, in accord with previous experiences about the combined use of stenting and PTCA [ 23 ]. Nevertheless, 11 our patients without restenosis showed dipyridamole g-SPECT LAD perfusion defects. Myocardial hypoperfusion may occur in patients without overt epicardial coronary artery stenosis having coronary microvessel damage [ 24 , 25 ] and coronary microvascular dysfunction corresponds to a reduced CFR in the absence of epicardial coronary stenosis [ 26 ]. Accordingly, the patients of the present study with long-standing SPECT perfusion defects showed lower CFR than the control group. An abnormal CFR had been already described immediately after balloon angioplasty [ 5 , 27 , 28 ], probably because of a slow recovery of autoregulation in the microvascular bed [ 29 ]. This reduction is primarily due to an increased flow velocity at rest [ 5 , 27 , 28 ], in relation to the failure of microvessel bed to vasoconstrict appropriately and/or to epicardial vasoconstriction mediated by a myogenic response and/or neural mechanism [ 30 ]. In contrast to previous studies showing normalization of CFR after three [ 31 ], five [ 32 ] or six months [ 5 ], CFR was persistently reduced in our patients with SPECT-derived perfusion defects. A suboptimal Doppler flow wire derived CFR had been observed six months after PTCA without restenosis by DEBATE investigators [ 33 ]. In the suboptimal CFR group the reduction of CFR was mainly due to a long-standing elevation in resting peak velocities while in our patients with perfusion defects it was due to a blunted maximal vasodilator response to dypiridamole. It is conceivable that this alteration could depend on endothelial damage of coronary microcirculation [ 34 ] preceding the procedure and persisting long time after PTCA. Coronary microcirculatory vasoconstriction induced by endothelial dysfunction has been described as effect of spontaneous myocardial ischemia [ 35 ] as well as in conditions other than epicardial coronary artery stenosis, as diabetes mellitus [ 36 , 37 ] arterial hypertension [ 26 , 38 ] and LV hypertrophy [ 39 , 40 ], which can alter microvascular function. However, an alternative interpretation of our findings include the possibility that a residual coronary stenosis might be anatomically insignificant but hemodynamically important, thus explaining a discrepancy between the percentage of lumen reduction and the amount of regional flow reserve. Myocardial systolic function and perfusion defect after PTCA The reduction of myocardial systolic function expressed by the decrease of low and high-dose dobutamine S m in middle septum, i.e. in the territory supplied by LAD, is not surprising in patients with perfusion defects after PTCA. Of note, myocardial systolic performance of middle septum was not significantly different between the 2 groups at rest. These findings are consistent with an altered myocardial systolic velocity response to exercise already described by C-DT in patients with coronary artery disease [ 13 , 41 ]. Also WMSI, not different at rest, became significantly higher at low and high-dose dobutamine in patients with perfusion defects. This increase (involving LV segments of LAD territory) during stress occurred only in 5/11 patients who had lower S m peak velocities at peak dobutamine stress and lower CFR than patients without inducible wall motion abnormalities. Myocardial reperfusion injury may include LV regional systolic dysfunction as irreversible manifestation, it depending by a reduction of myocardial blood flow reserve [ 5 ]. Inducible wall motion abnormalities in the presence of a successful coronary revascularization might indicate a very severe microvascular damage. Association between CFR and myocardial systolic function It is recognized that the extent of stress dobutamine-induced dissinergy is associated to the degree of CFR reduction in patients with significant coronary artery stenosis, an invasive myocardial fractional flow reserve ≤0.75 having a sensitivity of 76% and a specificity of 97% [ 42 ]. In agreement with these findings, we found a positive association between the functional degree of vasodilator microvascular coronary circulation and the magnitude of regional myocardial systolic function at peak dobutamine stress, i.e. S m peak velocity of the wall (middle septum) supplied by LAD. Accordingly, we also found a lower but significant negative relation between CFR and stress peak WMSI. Since patients undergoing successful reperfusion procedures generally present a good stress-echo LV functional response [ 6 ], our data suggest the ability of C-DT to detect even minor forms of LV regional myocardial dysfunction occurring under these circumstances. Study Limitations The main limitation of the present study include the fact that the negativity of coronary angiography can not exclude definitely the presence of coronary restenosis while an invasive measurement of CFR by Doppler flow wire or an intra-coronary ultrasound could have been crucial to clarify this issue. Unfortunately, these evaluations were not included into our study protocol. In addition, it has also to be underscored that patients with perfusion defects of the present study had greater LV mass, a factor which can itself induce a reduction of CFR [ 39 , 40 ]. Clinical implications The results of the present study suggest that CFR impairment may be detectable after PTCA even in the absence of coronary restenosis, it depending by an altered coronary microvascular function. In this clinical scenario, SPECT stress perfusion defects have to be interpreted as false positive results for PCTA restenosis while they reflect a true physiologic impairment in regional CFR with some associated degree of systolic impairment detectable by C-TD. Quantitative parameters of CFR and C-TD can provide an additive value over conventional stress echocardiographic assessment. Competing interests Financial competing interests • In the past five years we didn't receive reimbursements, fees, funding, or salary from an organization that may in any way gain or lose financially from the publication of this manuscript, either now or in the future. • No organization financied this manuscript. • We didn't hold any stocks or shares in an organization that may in any way gain or lose financially from the publication of this manuscript, either now or in the future. • We didn't hold or are currently applying for any patents relating to the content of the manuscript We didn't receive reimbursements, fees, funding, or salary from an organization that holds or has applied for patents relating to the content of the manuscript. • We don't have any other financial competing interests. Non-financial competing interests There are not any non-financial competing interests (political, personal, religious, academic, intellectual, commercial or any other) to declare in relation to this manuscript. We have not a competing interest, please discuss it with the editorial office. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539289.xml |
526380 | Two Drosophila suppressors of cytokine signaling (SOCS) differentially regulate JAK and EGFR pathway activities | Background The Janus kinase (JAK) cascade is an essential and well-conserved pathway required to transduce signals for a variety of ligands in both vertebrates and invertebrates. While activation of the pathway is essential to many processes, mutations from mammals and Drosophila demonstrate that regulation is also critical. The SOCS ( S uppressor O f C ytokine S ignaling) proteins in mammals are regulators of the JAK pathway that participate in a negative feedback loop, as they are transcriptionally activated by JAK signaling. Examination of one Drosophila SOCS homologue, Socs36E, demonstrated that its expression is responsive to JAK pathway activity and it is capable of downregulating JAK signaling, similar to the well characterized mammalian SOCS. Results Based on sequence analysis of the Drosophila genome, there are three identifiable SOCS homologues in flies. All three are most similar to mammalian SOCS that have not been extensively characterized: Socs36E is most similar to mammalian SOCS5, while Socs44A and Socs16D are most similar to mammalian SOCS6 and 7. Although Socs44A is capable of repressing JAK activity in some tissues, its expression is not regulated by the pathway. Furthermore, Socs44A can enhance the activity of the EGFR/MAPK signaling cascade, in contrast to Socs36E. Conclusions Two Drosophila SOCS proteins have some overlapping and some distinct capabilities. While Socs36E behaves similarly to the canonical vertebrate SOCS, Socs44A is not part of a JAK pathway negative feedback loop. Nonetheless, both SOCS regulate JAK and EGFR signaling pathways, albeit differently. The non-canonical properties of Socs44A may be representative of the class of less characterized vertebrate SOCS with which it shares greatest similarity. | Background The vertebrate JAK signaling pathway is an essential component of cellular response to a wide array of cytokines and growth factors. The JAK cascade is reutilized for signaling events in numerous tissues and at multiple stages of mammalian development [reviewed by [ 1 - 3 ]]. Many interleukins, interferons, and growth factors are among the ligands that stimulate signaling through the JAK pathway. The pathway can also be stimulated through activation of some receptor tyrosine kinases, including epidermal growth factor receptor (EGFR). As a result of its broad utilization, JAK signaling is essential for many developmental events. Though the JAK pathway is vital to many developmental processes, strict control of JAK signaling is equally important. As with other signaling pathways, mechanisms must be in place to balance the activation of JAK pathway activity. Regulation serves to "reset" the pathway so that it will be responsive to subsequent signals and it restricts the level or duration of the signal so that it is properly interpreted by the cell. Inappropriate JAK activation is the direct cause of a specific form of acute lymphocytic leukemia (ALL) [ 4 - 6 ]. In addition, JAK/STAT activation has been strongly correlated with a variety of cancers, including many blood cell and immune cell transformations [reviewed by [ 7 - 9 ]]. Furthermore, in cell culture, constitutive activation of c-Eyk, v-src, or v-abl results in the constitutive activation of specific STATs or JAKs [ 10 - 13 ]. These examples highlight the necessity of regulating JAK/STAT activation. Because of the need to limit JAK activity, it is not surprising that there are several conserved protein families that regulate JAK activation [reviewed by [ 3 , 14 , 15 ]]. These include phosphatases, Protein Inhibitors of Activated STATs (PIAS), and, the best characterized, the suppressors of cytokine signaling (SOCS) family. In mammals, eight different SOCS genes have been found [ 16 ]. These SOCS proteins have a distinctive modular architecture: a central SH2 domain followed by a carboxyl terminal SOCS domain, while the amino termini are quite divergent. Biochemical investigations have revealed that SOCS proteins use multiple mechanisms to regulate activity of the JAK pathway [see reviews, [ 3 , 9 ]]. First, the SOCS SH2 domain can bind to the phosphorylated receptor, thereby prohibiting access to positive effectors of the pathway. Second, at least some SOCS can specifically inhibit the catalytic activity of JAKs. Lastly, SOCS binding to activated JAK pathway components may target those proteins for degradation. The SOCS motif interacts with the elongins B and C, which bind to cullins and are E3 ubiquitin ligases [ 17 , 18 ]. Addition of ubiquitin to the bound proteins would target them for proteasomal degradation. Therefore, the negative influence of SOCS on its substrates may be due to multiple distinct mechanisms. Use of the JAK signaling pathway for developmental processes is not restricted to mammals. Indeed, the JAK cascade is evolutionarily conserved, and can be found as an intact signaling pathway even in insects [ 3 , 19 - 21 ]. In Drosophila , the JAK pathway is involved in embryonic patterning, sex determination, blood cell development, patterning of adult structures, planar polarity of photoreceptor clusters, maintenance of stem cells in spermatogenesis, and follicle cell patterning and function [see reviews [ 19 , 21 ]]. Furthermore, the fly JAK pathway must also be properly regulated to avoid deleterious effects. As in vertebrates, hyperactive JAK signaling has also been shown to directly cause neoplastic cell growth in Drosophila . Two dominant gain-of-function alleles of hopscotch result in hypertrophy of the larval lymph glands, the hematopoietic organ, and melanotic masses [ 22 - 24 ]. Excess activity in the blood system causes overproliferation and differentiation of the macrophage-like blood cells, creating leukemia-like effects. Inappropriate activity in the developing tissues of the adult fly can also cause alteration of the development of the adult thorax, wing veins, head, eyes, and ovaries [ 22 , 25 - 27 ]. Of the eight mammalian SOCS, four have been studied extensively (CIS, SOCS1-3). These genes have been shown to respond to JAK pathway activation and subsequently are able to downregulate its activity as described above, completing a classical negative feedback loop. In comparison, very little is known of the remaining four. Here we present the identification and characterization of Drosophila Socs44A. It contains the same modular domain architecture as mammalian SOCS and shows greatest sequence similarity to the relatively uncharacterized SOCS6 and SOCS7. We show that, unlike the previously studied Drosophila Socs36E [ 28 , 29 ], Socs44A expression in embryogenesis is independent of JAK pathway activity. However, Socs44A is able to regulate the JAK cascade in embryogenesis, but not in oogenesis. Finally, Socs44A genetically interacts with and upregulates the EGFR/MAPK pathway. The characteristics of Socs44A that distinguish it from the canonical Socs36E may be representative of features that are shared with the class of less-defined mammalian SOCS genes. Results The Drosophila genome encodes three putative SOCS genes Based on the consensus protein sequence for a SOCS box derived by Hilton and colleagues [ 30 ], a tBLASTn search of the Berkeley Drosophila Genome Project (BDGP) database [ 31 ] was conducted to examine all possible reading frames. Three putative loci containing both a SOCS box and an SH2 domain were identified using this strategy. All three match the arrangement of mammalian SOCS genes in that the SOCS box is at the carboxyl terminus with the SH2 domain directly preceding it. Each of these putative homologues also overlaps with a predicted gene from the BDGP. We named these three genes Socs16D (overlapping with CG8146), Socs36E (overlapping with CG15154), and Socs44A (CG2160), based upon their cytological location. Comparison of these three fly SOCS genes with vertebrate SOCS reveals that Socs36E is most similar to the mouse SOCS5, while Socs16D and Socs44A are less similar to specific mouse SOCS (see Fig. 1 ). While the amino termini are quite different, SOCS5 and Socs36E are 62% identical (71% similar) at the carboxy terminus from the region just before the SH2 domain to the end of the SOCS domain (region shown in Fig. 1A ). Within that same C-terminal region, Socs44A is most similar to SOCS6 and SOCS7 (46% and 39% similar, respectively). Socs16D also has highest similarity to SOCS6 and SOCS7 (47% to each) over the same carboxyl region. These similarities suggest that the ancestral versions of Socs36E and a common predecessor of Socs16D and Socs44A existed as two separate SOCS genes at the time of divergence of mammals and dipterans (Fig. 1B ). Figure 1 Protein sequence comparison of Drosophila and mouse SOCS. (A) The predicted carboxyl terminal protein sequences of Drosophila (d), mouse (m), and C. elegans (ce) SOCS genes, including the SH2 and SOCS box domains, are aligned and shaded to indicate similarities and identities. (B) Based on the protein alignments, the neighbor-joining method was used to construct a phylogenetic tree of these SOCS. Socs44A expression is not regulated by JAK pathway activity In mammals, regulation of JAK signaling through SOCS proteins is based on a simple negative feedback mechanism. Specifically, the activity of the JAK pathway stimulates the expression of SOCS genes, because activated STATs bind to enhancers for the SOCS genes and induce transcription. Socs36E is similarly regulated during embryogenesis by Drosophila JAK signaling [ 29 ]. Socs36E is expressed dynamically, in a striped pattern that later becomes restricted predominantly to the tracheal pits [[ 28 , 29 ], and Fig. 3 ], very similar to upd , the gene encoding the embryonic ligand for the JAK pathway [ 32 ]. Indeed, activation of JAK signaling is both necessary and sufficient for Socs36E expression in embryogenesis [ 29 ]. Furthermore, the expression of Socs36E during oogenesis matches the known activation of JAK signaling. The expression of upd in the ovaries is restricted to the two polar follicle cells at either end of the egg chambers of the vitellarium [[ 26 ] and Fig. 2C ]. Socs36E is expressed in a larger number of follicle cells centered at the two poles of the egg chamber (Fig. 2D ). Given that secreted Upd protein is produced in the polar follicle cells and activates JAK signaling in neighboring cells [ 33 ], this suggests that Socs36E expression is controlled by JAK activity in oogenesis, as well as embryogenesis. Figure 3 Loss of JAK activity does not affect Socs44A expression. As compared with wild-type at various embryonic stages (A and B), germline clone derived embryos from hop c111 mothers (C-H) display dramatically reduced or eliminated expression of Socs36E (C and D). Only a stripe of mesodermal staining in germ band extended embryos (D) remains at nearly normal intensity in the mutant embryos. In contrast, expression of Socs44A in trachea persists in hop c111 germline clone-derived embryos that are unrescued (E) or paternally rescued (F). However, the trachea are morphologically altered and drastically reduced in unrescued (G) and paternally rescued (H) animals, as compared with wild-type (I), as evidenced by a trachealess enhancer trap (G-I). Figure 2 Socs36E and Socs44A are expressed in different spatio-temporal patterns. The embryonic expression patterns of upd and Socs36E are dynamic from early blastoderm throughout embryogenesis [see 28, 29 and Fig. 3]. Socs44A expression is not detected until very late stages in the trachea (A). Although such staining can be artifactual, sense strand probe never showed any staining (B). In the ovary, upd is expressed specifically in the polar follicle cells at each end of the chamber (C). Socs36E expression encompasses the anterior and posterior follicular epithelium, with highest expression at the poles (D). This is consistent with activation of Socs36E transcription due to reception of the Upd ligand which is secreted from the polar follicle cells and diffuses toward surrounding cells. Socs44A expression is restricted to the germline and only during later stages of oogenesis (E) The Socs44A gene that was predicted based on protein homology is identical to hypothetical gene CG2160. A single cDNA corresponding to the locus (LP02169) was isolated by the BDGP, has been completely sequenced and encodes the expected SH2 and SOCS domains at the carboxyl terminus (gb AF435923). To determine whether Socs44A is similarly regulated by JAK pathway activity, in situ hybridization to embryos and ovaries was performed. No specific expression of Socs44A was detected until very late in embryogenesis. The only striking staining pattern observed was in the trachea of late embryos (Fig. 2A ). Non-specific tracheal staining is sometimes seen with probes to late embryos, however this pattern was never observed when sense probe was used (Fig. 2B ). Unfortunately, embryos homozygous for any available deletions that remove Socs44A die prior to formation of trachea, therefore we cannot conclusively determine whether the late tracheal staining reflects RNA expression. Nonetheless, because the JAK pathway is activated in a segmentally repeated pattern during embryogenesis, the lack of Socs44A expression suggests that it is not responsive to JAK signaling. Consistent with this conclusion, expression of Socs44A in the ovary is restricted to only germline expression late in oogenesis, with no detectable RNA in the follicular epithelium (Fig. 2E ). To directly test whether Socs44A expression is regulated by JAK pathway activity, in situ hybridization to Socs44A RNA was performed in embryos that lack JAK pathway activity. The product of the hop gene is required in early embryogenesis and must be provided maternally for proper segmentation of the embryo. The dominant female sterile (DFS) technique was used to generate females that fail to produce hop in the germline [ 34 ]. In situ hybridization of hop germline clone embryos using Socs36E as probe demonstrates a strong reduction in Socs36E expression in the mutant embryos as compared with wild-type (Fig. 3 ). Similar results have been reported in embryos lacking upd activity [ 29 ]. These data demonstrate that hop is required to stimulate the normal segmentally-repeated Socs36E expression in the embryo. However, expression of Socs44A does not appear to be affected by maternal loss of hop . Although the trachea are malformed and dramatically reduced in embryos lacking JAK pathway activity [[ 35 , 36 ] and Fig. 3G,3H ], the remaining segments of trachea continue to express Socs44A at apparently normal levels (Fig. 3E,3F ). Thus the failure of endogenous Socs44A to be expressed in the normal pattern of JAK pathway activation and of Socs44A expression to be eliminated by loss of JAK activity indicate that Socs44A expression is not stimulated by the pathway. Activity of the JAK pathway is both necessary and sufficient for the expression of Socs36E . The ectopic activation of the JAK pathway by misexpression of upd results in expression of Socs36E in the same pattern [[ 29 ] and data not shown]. In contrast, similar misexpression of UAS-upd with the paired-GAL4 driver failed to stimulate any detectable expression of Socs44A in the embryo (not shown). We conclude that Socs44A expression is not responsive to JAK pathway activity, therefore cannot function via a traditional auto-regulatory feedback loop. Ectopic SOCS activity suppresses JAK signaling in the wing The lack of transcriptional regulation by JAK signaling does not preclude a role for Socs44A in the control of JAK activity. To test whether it can attenuate JAK signaling, Socs44A was misexpressed using the GAL4/UAS system. Similar experiments performed with Socs36E have demonstrated that expression in the developing wing reproducibly results in the production of ectopic wing vein near the posterior crossvein [Fig. 4C and [ 28 ]]. This phenotype is quite similar to that noted for viable mutants of hop or Stat92E [Fig. 4B and [ 37 ]], suggesting that Socs36E misexpression may cause a reduction in JAK signaling in the wing. But unlike observed JAK mutations, the anterior crossvein was also completely missing from Socs36E misexpression wings, perhaps suggesting an additional role for Socs36E that is independent of the JAK pathway. Callus and Mathey-Prevot [ 28 ] demonstrated that the additional influence on wing venation may be due to the suppression of the EGFR pathway. Figure 4 Socs44A misexpression reduces JAK signaling in the wing. Wild-type venation (A) is compared with a viable hop mutant, hop msv / hop M38 (B). hop reduction causes ectopic vein (arrow) near the posterior crossvein. (C) Expression of UAS-Socs36E using the engrailed-GAL4 driver (e16E-GAL) produces a similar ectopic vein phenotype, plus the loss of the anterior crossvein (arrowhead). (D) Similar misexpression of Socs44A causes ectopic wing vein production near the posterior crossvein (arrow) and arching of vein L3 (arrowhead). (E) Reduction of the dosage of hop enhances the Socs44A misexpression phenotype. (F) Misexpression of hop in the posterior compartment causes dramatic vein loss, but that loss is restored by the simultaneous expression of Socs44A (G). Using the engrailed-GAL driver, GAL-e16E, expression of Socs44A in the posterior compartment of the wing caused mild venation defects similar, but not identical, to Socs36E (Fig. 4D ). Expression of Socs44A caused production of ectopic wing vein near the posterior crossvein, but unlike Socs36E , the ectopic vein was seen predominantly posterior to L5, not between L4 and L5. Furthermore, the anterior crossvein was not reduced or eliminated by Socs44A expression, but a substantial arching of L3 was noticed. Both the ectopic vein and arching of L3 were enhanced in animals heterozygous for a null allele of hop (Fig. 4E ), indicating that the phenotype is sensitive to a reduction in JAK pathway activity. Misexpression of hop activates JAK signaling and causes reduction of wing venation in the posterior of the wing, somewhat the opposite of Socs44A misexpression (Fig. 4F ). The simultaneous misexpression of hop and Socs44A results in a phenotype similar to expression of Socs44A alone (Fig. 4G ). Therefore, the activity of Socs44A is capable of negating the influence of ectopic JAK activity in the wing. Loss of JAK function in embryos is lethal, but various combinations of weak alleles of hop show some viability (Table 1 ). If Socs44A were negatively regulating the JAK pathway, misexpression of Socs44A in a hop mutant background would be expected to further reduce viability. The ability of Socs44A misexpression to enhance the lethality of weak heteroallelic combinations of hop was tested. For all alleles examined, expression of Socs44A in the engrailed pattern caused complete lethality. For the weakest hop allelic combination, hop msv / hop M75 , misexpression of Socs44A caused viability to drop from 62% to 0% (Table 1 ). These data are consistent with the hypothesis that ectopic Socs44A acts to further reduce pathway activity in these JAK activity depleted animals, causing lethality. Table 1 Misexpression of Socs44A exacerbates the reduced viability of hop heteroallic mutants. Genotype hop M38 (n = 213) hop GA32 (n = 332) hop M75 (n = 172) A - hop x /FM7; en-GAL; TM3 33 52 21 B - hop x /FM7; en-GAL; UAS-socs44A 25 33 28 C - hop x /hop msv ; en-GAL; TM3 11 20 13 D - hop x /hop msv ; en-GAL; UAS-socs44A 0 (E = 8.33) 0 (E = 12.69) 0 (E = 17.33) Misexpression of Socs44A in a range of hop heteroallelic mutants resulted in lethality. For each mutant combination, x is the allele of hop designated in the column heading, n represents the total number of progeny scored in the cross. E represents the expected number of progeny of that genotype if Socs44A misexpression were to have no effect on viability. The expected value is calculated using the formula A/B=C/D , which takes into account the change in viability imparted by homozygosity for hop relative to heterozygosity and the change in viability for misexpression of Socs44A relative to the GAL4 alone. The progeny scored here are derived from the cross: hop X / Y Dp(1;Y)v + y + hop + ; en-GAL4/CyO males mated to hop msv / FM7; UAS-socs44A/TM3 females. While the above data indicate that ectopic Socs44A is capable of downregulating JAK activity, they do not address whether Socs44A has an endogenous role in JAK pathway regulation. To determine if endogenous Socs44A downregulates JAK activity, we assayed the effect of a Socs44A deficiency on hop mutant phenotypes. The hop M38 / msv heteroallelic mutant exhibits wing vein material at the posterior crossvein (Fig 4B ) that is 98% penetrant. Removal of a single copy of Socs44A using either of two deficiencies in the region reduced the penetrance of the hop phenotype by as much as 52% (Table 2 ). An overlapping deficiency that did not remove the Socs44A locus had little effect on penetrance of the phenotype. These results suggest that regulation of JAK activity in the wing is a normal endogenous function of Socs44A. Table 2 Endogenous Socs44A regulates JAK pathway activity. +/+ CA53/+ (n = 237) NCX10/+ (n = 292) Drl/+ (n = 242) hop M38 /hop msv 98% (of 89) 46% (of 13) 58% (of 12) 87% (of 15) hop msv / hop M38 heteroallelic females have a wing spur phenotype (Fig. 4B) that is 98% penetrant (n = 89). The penetrance of the spur phenotype is dramatically reduced by removal of one copy of Socs44A , as seen for heterozygotes of Df(2)CA53 (CA53)and Df(2)NCX10 (NCX10). Rescue of the phenotype was not seen with Df(2)Drl rv18 (Drl), an overlapping deficiency that does not include Socs44A . Total number of animals is indicated by n , and number of animals of the indicated genotype is in parentheses. Socs44A upregulates EGFR pathway activity In mammals, there are multiple points of cross-talk between the JAK and EGFR/MAPK signaling pathways [ 3 , 38 - 40 ]. EGFR signaling plays a prominent role in many developmental processes in Drosophila , including wing venation [ 41 , 42 ]. As mentioned above, expression of Socs36E has been reported to suppress EGFR signaling in the wings [ 28 ]. To determine the relationship of Socs44A to EGFR/MAPK signaling, wing phenotypes due to misexpression of Socs44A were examined in the background of heterozygous mutations for components of the EGFR signaling pathway. Engrailed-GAL4 driven misexpression phenotypes of Socs44A were suppressed in the background of heterozygous mutations for Ras85D, Son of sevenless (Sos), and Egfr (Fig. 5A,5B,5C,5D ). Consistent with these observations, reduction in the dosage of the EGFR negative regulator argos enhanced the Socs44A misexpression phenotype (Fig. 5E ). In contrast, concurrent misexpression of Socs44A and argos had antagonistic effects. Misexpression of two copies of an argos transgene under the engrailed-GAL4 driver resulted in wings lacking the 4 th lateral vein (L4) as well as both cross-veins (Fig. 5H ). Concurrent misexpression of a single copy of the Socs44A transgene in this background was able to rescue this phenotype, restoring the posterior crossvein and both the most proximal and distal portions of L4 (Fig. 5I ). The resulting wing phenotype mimicked that seen when only a single copy of argos was used in the misexpression assay (Fig. 5J ) or what is seen in heteroallelic Egfr mutants (Fig. 5G ). Finally, concurrent misexpression of a single copy of the argos and Socs44A transgenes produced a nearly wildtype wing (Fig. 5K ). These data indicate that Socs44A expression is able to suppress argos misexpression phenotypes in a dose-dependent manner. It should be noted that concurrent misexpression of UAS-GFP did not affect the UAS-argos phenotype (not shown), indicating that the suppression by UAS-Socs44A was not merely a consequence of titrating GAL4. Figure 5 Socs44A increases activity of EGFR signaling. The ectopic wing vein phenotype of Socs44A misexpression (A) is rescued by reduction of Egfr (B), Sos (C) or Ras85D (D), positive effectors of EGFR signaling. In contrast, reduction of argos , a negative regulator of EGFR signaling, enhances the Socs44A misexpression phenotype (E). The argos allele combined with en-GAL have no effect on venation without the UAS-Socs44A transgene (F). Certain heteroallelic Egfr mutants possess a distinct wing vein phenotype, whereby the anterior crossvein and the central portion of L4 is missing (G, arrows). Engrailed -driven misexpression of argos has a similar phenotype (H and J). Concurrent misexpression of Socs44A antagonizes argos misexpression to restore near normal wing venation (I and K). The designation "2xUAS-argos" refers to presence of 2 total copies of the transgene in the genome. Although these misexpression data indicate that Socs44A can enhance EGFR signaling, they do not necessarily demonstrate that this is a normal function of Socs44A. To address whether this is an endogenous function of Socs44A, we assayed the influence of a deficiency that removes Socs44A in the argos misexpression background. Engrailed-GAL4 misexpression of argos produces a range of phenotypic classes in which parts or all of L4 and/or the posterior cross-vein are missing (Fig. 6A ). Addition of a single copy of a deficiency that removes Socs44A shifted the distribution of phenotypes to the more severe classes (Fig. 6B ). In contrast, addition of an overlapping deficiency that does not include the Socs44A locus did not show such a shift. While it cannot be unambiguously stated that this effect is due to loss of Socs44A specifically, these results are consistent with the misexpression analyses and suggest that Socs44A normally plays a role in enhancing EGFR signaling in the Drosophila wing. Figure 6 Socs44A deficiencies enhance argos misexpression phenotypes. (A) The engrailed-GAL4 driven misexpression of argos produces a range of phenotypes which were classified based on severity. The combination of en-GAL and Df(2)CA53 had no effect on venation. (B) In flies that were also heterozygous for Df(2)CA53 , which removes the Socs44A locus, the distribution of phenotypes was significantly shifted to more severe classes as compared to animals heterozygous for Df(2)Drl rv18 , an overlapping deficiency that does not remove Socs44A or for Sco , a chromosome wild-type for the 44A region. Socs36E and Socs44A have different effects on oogenesis Evidence presented here and elsewhere indicates that Socs36E and Socs44A can downregulate JAK signaling in the wing [ 28 ]. However, the ability of specific mammalian SOCS to regulate JAK activity has been observed to differ, depending upon the tissue examined [ 43 ]. To determine whether there is a similar context specificity for the Drosophila SOCS, regulation was examined in another tissue in which JAK and EGFR functions have been well characterized. Both pathways are required for proper patterning of the follicular epithelium surrounding developing egg chambers during oogenesis [ 26 , 33 , 44 - 47 ]. One of the distinct cell populations requiring these pathways is the posterior terminal follicle cells [ 33 ]. These cells are molecularly identified by the expression of the ETS domain transcription factor, pointed [ 47 - 49 ]. In clones of cells that lack hop activity (Fig. 7C,7D,7E ) or egfr activity (not shown), there is a loss of pnt-lacZ expression, indicating failure to specify the posterior terminal follicle cells. Figure 7 Socs36E and Socs44A have different activities during oogenesis. In wild-type ovaries (A, B), pnt-lacZ (red) is expressed in a gradient in the posterior terminal cells. Cells that lack hop activity (marked by a lack of green, see outline), also fail to express pnt-lacZ (C-E). Similarly, UAS-Socs36E misexpressed in clones (marked by presence of green, see outline), lack pnt-LacZ expression (F-H, see insets). In contrast, UAS-Socs44A misexpressed in clones (marked by presence of green, see outline), had no effect on pnt-LacZ expression (I-K). DAPI nuclear staining is shown in blue. To test whether Socs36E and Socs44A can downregulate JAK or EGFR activity during oogenesis, clones of cells misexpressing these genes in developing egg chambers were examined. In clones misexpressing Socs36E at high levels in posterior cells of the developing egg chamber, there was a dramatic loss of the pnt-LacZ marker (Fig. 7F,7G,7H ). This loss was restricted to only those cells that misexpressed Socs36E and did not influence neighboring cells. These results indicate that JAK and/or EGFR signaling was attenuated by Socs36E activity. In contrast, for cells in which Socs44A was misexpressed in a similar fashion, there was no reduction of pnt-LacZ expression (Fig. 7I,7J,7K ). We conclude that Socs44A is unable to attenuate JAK activity in the follicle cells. This ability of Socs44A to regulate JAK signaling in the wing, but not in the ovary, indicates that SOCS activity in invertebrates can also be context specific. Furthermore, the differential ability of the fly SOCS to attenuate JAK and EGFR signaling in the ovary demonstrates distinct functions for these two proteins. Discussion The Drosophila genome encodes three homologues of the vertebrate SOCS. Each homologue contains the hallmark modular architecture, with a central SH2 domain followed by a carboxy-terminal SOCS domain. The genes are dispersed in the genome and are referred to by their cytological locations as Socs16D , Socs36E , and Socs44A . These fly SOCS genes are most similar to the vertebrate SOCS5, 6, and 7, none of which has been functionally characterized to date. Socs36E is the most similar in protein sequence to a vertebrate SOCS, SOCS5, but shares many characteristics with the extensively studied mammalian SOCS genes, SOCS1-3 and CIS. Each of these has been shown to be transcriptionally responsive to JAK pathway stimulation and act to downregulate JAK activity in a classical negative feedback loop [reviewed by [ 9 ]]. On the other hand, Socs44A is most similar to the less studied vertebrate genes, SOCS6 and 7. In this study, we demonstrated that Socs44A has properties that distinguish it from Socs36E and the canonical mammalian SOCS (compared in Table 3 ). First, the expression of Socs44A was not dependent on JAK pathway activity. Nevertheless, Socs44A was able to downregulate the JAK cascade in some, but not all tissues. In addition to regulating JAK pathway activity, Socs44A genetically interacts with the EGFR/MAPK pathway, acting to enhance its activity. Table 3 Comparison of Drosophila SOCS. Socs36E Socs44A Expression-Embryogenesis Matches known pattern of JAK activation, including pair-rule stripes, gut, and tracheal pits Distinct from JAK activation, with possible exception of trachea very late Expression- Oogenesis Matches known pattern of JAK activation, with graded expression highest at anterior and posterior poles Distinct from JAK activation, with expression only in nurse cells Requirement for expression Requires JAK signaling for embryonic expression Does not require JAK signaling for embryonic expression Inducibility Inducible by JAK activity in embryos Not inducible by JAK activity in embryos Regulation of JAK activity Can repress JAK signaling in wing and possibly in follicle cells of ovary Can repress JAK signaling in wing, but cannot in follicle cells Regulation of EGFR activity Can repress EGFR signaling in wing and possibly in follicle cells of ovary Can enhance EGFR signaling in wing The contrasting properties of Socs36E and Socs44A are summarized. The Drosophila genome encodes three SOCS genes Phylogenetically, SOCS fall into three general clades. The first includes the best studied vertebrate SOCS, CIS and SOCS1-3. Interestingly, there are no representatives of this group found in the fly genome. Vertebrate SOCS of the remaining two clades have yet to be fully characterized with regard to their physiological roles, as well as mechanistic roles in JAK/STAT signaling. Socs36E is most similar to the vertebrate SOCS of the second clade, containing SOCS4 and SOCS5. It shares similarity not only in the SH2 and SOCS domain, but also in the region upstream of the SH2 domain. Mutational analysis has shown that SOCS5 inhibits IL-6 [ 50 ], whereas nothing is known about the activity of SOCS4. Socs44A falls into the third clade occupied by vertebrate SOCS6 and SOCS7, as well as the only C. elegans homologue. SOCS6 has been shown to downregulate the insulin receptor [ 51 , 52 ]. Very little is known about SOCS7, other than its ability to interact with Nck, Ash, and PLCγ [ 53 ]. Because of the relative lack of information about these latter two clades, study of the Drosophila SOCS may identify general properties of these homologues that span each clade. Although mammalian genomes encode large families of specific JAK pathway components, Drosophila has only one characterized receptor, domeless , one Janus kinase, hop , and a single STAT, stat92E . Despite the simplicity of the transduction machinery for the JAK pathway, there are three SOCS genes in flies. Furthermore, there is only one Drosophila homologue of the PIAS negative regulatory family, zimp , and it is also capable of inhibiting JAK pathway activity [ 54 , 55 ]. In an organism with few functionally redundant genes, why are there three Drosophila SOCS? Two possible explanations for the apparent abundance of SOCS are that the different Drosophila SOCS may be expressed differently or they may differently regulate signaling through pathways other than JAK. Indeed, we presented evidence for both of these distinctions for Socs36E and Socs44A. Socs44A does not participate in an auto-regulatory negative feedback loop It has been demonstrated that, like the classical vertebrate SOCS genes, Socs36E is transcriptionally responsive to JAK pathway activity [[ 29 ] and this work]. In both embryos and ovaries, the expression of Socs36E mirrors the known pattern of JAK activation and, indeed, altered JAK activation in the embryo elicits a transcriptional alteration in Socs36E . Unlike Socs36E , the expression of Socs44A did not match that of JAK induction. In the embryo, detectable Socs44A expression was absent until late stages of embryogenesis, when it was restricted to the developing trachea. JAK activation does occur in the tracheal pits and has been implicated in tracheal morphogenesis [ 35 , 36 ], but Socs44A expression was lacking in the other tissues of the early embryo where JAK activation has been described. More telling was the finding that neither reduction nor expansion of JAK activation in the embryo had any effect on Socs44A expression. This disparity between Socs44A and Socs36E support the hypothesis that these genes are not redundant. Despite the difference in expression of the two SOCS genes, both are able to downregulate JAK activity in some tissues. Misexpression of Socs36E is able to suppress JAK activity in the developing adult (imaginal) wing and thorax [ 28 ]. Similarly, misexpression of Socs44A reduced JAK activity in the imaginal wing, as illustrated by the enhancement of that phenotype by reduction of endogenous hop . Furthermore, misexpression of Socs44A rescued wing vein loss resulting from misexpression of hop . Perhaps most importantly, introduction of deficiencies that remove Socs44A rescued a hop wing vein phenotype. Taken together, these data strongly suggest that Socs44A downregulates JAK pathway activity during normal wing development. However, misexpression of Socs44A had no effect on expression of a marker for JAK pathway activity during oogenesis. This indicates that there is context specificity to SOCS action in Drosophila , a phenomenon that has been observed in the study of mammalian SOCS [ 43 ]. In contrast, misexpression of Socs36E was able to downregulate expression of the pnt-lacZ marker in follicle cells, although it cannot be distinguished whether this is due to reduction of signaling through JAK or EGFR. However, because Socs36E is expressed in the pattern of JAK activation in follicle cells, it is likely that it has a function in regulating JAK signaling in the ovary. Socs44A upregulates EGFR/MAPK signaling Another distinction we noted between the Drosophila SOCS was in their abilities to regulate signal transduction cascades in addition to JAK/STAT. Precedence for such additional roles for vertebrate SOCS include regulation of Tec, Vav, TCR, c-kit, and FAK mediated signaling [ 56 - 60 ]. It has been previously shown that Socs36E can suppress signaling not only through the JAK pathway, but also through the EGFR/MAPK pathway [ 28 ]. Socs44A was also able to regulate EGFR/MAPK signaling, but acted in the opposite manner. Socs44A was able to rescue misexpression of the EGFR negative regulator argos in a dose-dependent manner. Furthermore, mutations in EGFR pathway components rescued Socs44A misexpression phenotypes. Importantly, a reduction of endogenous Socs44A activity enhanced the argos phenotype. Taken together, these data suggest that a normal function for Socs44A is to enhance the EGFR pathway. A potential mechanism for this genetic interaction can be found in a recent report describing physical interaction between SOCS3 and the p120 RasGAP [ 61 ]. p120 RasGAP, a GTPase-Activating Protein, is an antagonist of MAPK signaling that is responsible for inactivating Ras. It does so by stimulating Ras GTP hydrolytic activity, leaving Ras in a GDP-bound, inactive configuration. Upon interaction with SOCS3, p120 RasGAP is unable to inactivate Ras, resulting in an upregulation of the EGFR/MAPK pathway. Perhaps Socs44A is acting in an analogous manner. Indeed, there are three candidate RasGAP genes in the fly genome. Biochemical analyses will be required to address this hypothesis. Conclusions There are three Drosophila SOCS, all of which have greatest homology to the two classes of vertebrate SOCS that are least well characterized. One of these, Socs36E, is a member of the vertebrate SOCS4/5 class and has been previously characterized [ 28 , 29 ]. It is similar to classical SOCS in that its expression is regulated by activity of the JAK pathway and that it functions to suppress JAK activity. Here we provided the initial characterization of Socs44A, a member of the vertebrate SOCS6/7 class. In contrast to Socs36E , activation of the JAK pathway was neither necessary nor sufficient for the expression of Socs44A . We conclude that Socs44A is unlike classical SOCS because it does not participate in a JAK pathway negative feedback loop. Still, Socs44A was capable of repressing JAK signaling, but that activity was limited to certain tissues. This context specificity is a feature that is shared with classical SOCS. Finally, Socs44A and Socs36E had opposite effects on EGFR/MAPK signaling. The enhancement of MAPK signaling that was seen for Socs44A is reminiscent of the influence of SOCS3 on this pathway, which is exerted through physical interaction of SOCS3 with p120 RasGAP. Perhaps a similar mechanism explains the enhancement of MAPK activity due to Socs44A. The differences observed here between Socs36E and Socs44A strongly suggest that they have distinct functions in the fly. Furthermore, the differences between Socs44A and the well studied class of canonical vertebrate SOCS may be representative of undiscovered distinctions amongst the three classes of vertebrate SOCS. Methods Comparison of SOCS sequences Putative Drosophila SOCS genes were identified using a simple tBLASTn 2.0 query with a consensus sequence for the vertebrate SOCS domains [ 30 ] used to probe the complete genome contig sequences available from the BDGP. Identified homologies were compared with the predicted gene structures reported as "CG" sequences in the annotations of the genomic contigs. The translated sequences of the three putative SOCS gene genomic regions were scanned manually for possible alternative structures. The sequences surrounding the SOCS and SH2 domains were used to generate primers for the amplification of DNA corresponding to each putative gene. Amplification products were cloned and used to generate probes for the identification of cDNAs as described below. Phylogenetic comparison of SOCS proteins was performed using AlignX (VectorNTI 9.0), based on the ClustalW algorithm, to generate protein alignments and a neighbor-joining algorithm to create a phylogenetic tree. Identification of cDNAs A cDNA library constructed from RNA of 12–24 hr old embryos [ 62 ] was screened using 800bp of genomic DNA derived from the 3' end of the Socs36E coding region, including the SOCS box and SH2 domain. Two independent clones (Genbank accessions AF435838 and AF435839) were recovered, with the former being structurally similar to an EST from the BDGP (clot #7147). The BDGP also recovered two cDNA clones representing socs44A which have been designated as clot #8463. We have determined the complete sequence of the longer clone, LP02169 (Genbank AF435923). In situ hybridizations In situ hybridizations to embryos were performed as previously described [ 32 ]. Digoxigenin labeled probes for Socs36E and Socs44A were generated from the 5' ends of the respective cDNAs and did not include the coding region for the conserved SH2 and SOCS domains. Germline clone mutants for the hop c111 null allele were generated using the ovo D1 dominant female sterile technique [ 34 ]. Embryos derived from mutant mothers were collected overnight and prepared for hybridizations as previously indicated. Embryos misexpressing upd in a specific pattern were generated by crossing females carrying a UAS-upd transgene with males heterozygous for paired-GAL4, which expresses GAL4 in the seven stripe pair-rule pattern of the paired gene. Progeny were collected and hybridized as above. Trachea in germline clone-derived hop c111 embryos were visualized with the trh 10512 enhancer trap [ 63 ] using anti-β-gal antibody (Cortex Biochemical, at 1:1000) as previously described [ 26 ]. Misexpression studies To express Socs36E and Socs44A under control of GAL4, the full-length cDNAs described above were cloned into the pUAST vector [ 64 ]. Germline transformations were performed [ 65 ] and transgenic lines established. For wing phenotypes, engrailed -GAL4 (e16E-GAL) was used to drive expression of the transgenes in the posterior compartment. Wings were dissected and mounted in Hoyer's medium [ 66 ] for photography. Ovarian clones of the null allele, hop c111 , were generated by hsFLP mediated mitotic recombination as previously described [ 26 , 33 ]. Misexpression clones of Socs36E and Socs44A were generated using a GAL4 flip-out cassette [ 67 ]. Genotypes of those animals were w [hsFLP]1 ; [Act5C>y>GAL4] [UAS-GFP.S65T]/ [UAS-socs36E]11.2 and w [hsFLP]1 ; [Act5C>y>GAL4] [UAS-GFP.S65T]/ +; [UAS-socs44A]/ pnt-LacZ , respectively. For each, ovaries were fixed and stained with anti-β gal and anti-GFP as previously described [ 26 , 33 ]. Microscopy All in situ hybridization and wing images were acquired using a Spot Camera (Diagnostic Instruments) on a Nikon E800 microscope using differential interference contrast (DIC). A Leica TCS-SP laser scanning confocal microscope was used to capture all fluorescence micrographs. All images were then exported to Adobe Photoshop for manipulation and annotation. Authors' contributions JR performed experiments with Socs44A and participated in drafting the manuscript. GR performed most experiments with Socs36E. SH performed some experiments with Socs36E. RX performed immunofluorescence experiments in ovaries, except for Socs44A. DH conceived of the study, participated in its design and coordination, and participated in drafting the manuscript. All authors read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526380.xml |
549034 | Comparing the clinical effectiveness of different new-born hearing screening strategies. A decision analysis | Background Children with congenital hearing impairment benefit from early detection and treatment. At present, no model exists which explicitly quantifies the effectiveness of universal newborn hearing screening (UNHS) versus other programme alternatives in terms of early diagnosis. It has yet to be considered whether early diagnosis (within the first few months) of hearing impairment is of importance with regard to the further development of the child compared with effects resulting from a later diagnosis. The objective was to systematically compare two screening strategies for the early detection of new-born hearing disorders, UNHS and risk factor screening, with no systematic screening regarding their influence on early diagnosis. Methods Design: Clinical effectiveness analysis using a Markov Model. Data Sources: Systematic literature review, empirical data survey, and expert opinion. Target Population: All newborn babies. Time scale: 6, 12 and 120 months. Perspective: Health care system. Compared Strategies: UNHS, Risk factor screening (RS), no systematic screening (NS). Outcome Measures: Quality weighted detected child months (QCM). Results UNHS detected 644 QCM up until the age of 6 months (72,2%). RS detected 393 child months (44,1%) and no systematic screening 152 child months (17,0%). UNHS detected 74,3% and 86,7% weighted child months at 12 and 120 months, RS 48,4% and 73,3%, NS 23,7% and 60,6%. At the age of 6 months UNHS identified approximately 75% of all children born with hearing impairment, RS 50% and NS 25%. At the time of screening UNHS marked 10% of screened healthy children for further testing (false positives), RS 2%. UNHS demonstrated higher effectiveness even under a wide range of relevant parameters. The model was insensitive to test parameters within the assumed range but results varied along the prevalence of hearing impairment. Conclusion We have shown that UNHS is able to detect hearing impairment at an earlier age and more accurately than selective RS. Further research should be carried out to establish the effects of hearing loss on the quality of life of an individual, its influence on school performance and career achievement and the differences made by early fitting of a hearing aid on these factors. | Background Approximately one to three per 1000 children are born with at least moderate, bilateral hearing disorders [ 1 - 4 ]. Children with congenital hearing impairment benefit from early detection and treatment of their hearing loss [ 5 ]. The neurological development of hearing abilities requires acoustic stimulation in the first 18 months of life. Deficits due to lack of acoustic stimulation within the first two years are not or not easily recovered by later rehabilitation. The consequences include delayed development of speech and other cognitive and social functions. This delay is already measurable in the first 3 years of life [ 6 ]. If disorders are detected and treated in time, either by the use of a hearing device or cochlea implant, most of the children develop normally and do not need additional speech therapy [ 7 , 8 ]. Early diagnosis and treatment within the sensitive time frames are therefore essential. The German consensus conference on neonatal hearing screening proposed diagnosis in the first 3 months and the start of treatment in the first 6 months of life. Various tests and test combinations with acceptable sensitivity and specificity are available. Transient evoked oto-acoustic emissions (TEOAE) or brainstem evoked responses (BERA) can be measured. One common strategy consists of a two-step TEOAE with the first measurement within the first days of life and a second test a few days later [ 9 ] (chapter 2.2.7). While modern screening techniques for early detection are available there is still a gap between the consensus on detection as early as possible and the current situation. Only 50% to 60% of children with permanent hearing impairment are diagnosed before their second birthday with traditional health care services [ 2 ]. While congenital hearing loss is a serious health problem, there is little evidence to support the use of routine universal screening because of the following factors: • as the prevalence is very low, the positive predictive value of the tests is low • screening technologies are still in development • possible costs and consequences are not sufficiently known • benefits of early intervention are frequently expressed in qualitative terms without presenting unbiased measures of outcome. In 1995 the US Preventive Services Task Force found insufficient evidence in favour of universal neonatal hearing screening (UNHS) [ 10 ]. The Task Force proposed selective screening of new-borns with risk factors to improve the predictive value of the test. In the UK a national neonatal hearing screening programme aimed at detecting bilateral moderate to severe hearing impairments has been recommended [ 11 ] and partially implemented within a pilot project [ 12 ]. Several studies have modelled the outcomes of UNHS versus risk factor screening [ 13 - 15 ]. Keren et al [ 16 ] presented a cost-effectiveness analysis based on a decision tree model which reported short-term effectiveness of UNHS compared to risk factor screening. However, there is presently no model explicitly quantifying the effectiveness of UNHS versus other programme alternatives in terms of early diagnosis, nor has it been taken into account that diagnosis of hearing impairment within the first few months of life is more "valuable" for the child's further development than diagnosis later in life. Children identified within the developmentally sensitive time frame should therefore be given more weight compared to children with delayed diagnosis. The objective of this clinical decision analysis was to systematically compare two screening strategies for the early detection of new-born hearing disorders: UNHS and risk factor screening; with the option of no systematic screening regarding their influence on early diagnosis. Our specific objectives were to show differences between strategies expressed as the number of quality weighed detected child months (QCM), the number of true positive cases at the age of 6 and 12 months, and the number of false positive cases. We also wanted to investigate which parameters had the most influence on the reported differences and how likely these differences were. Methods We developed a clinical decision model to compare two different screening strategies with the option of no systematic screening. Parameters were extracted from the literature, empirically derived from a representative patient survey, and estimated by experts. Univariate and multivariate sensitivity analyses were performed on all relevant parameters. The decision model was used to predict absolute and incremental effectiveness of two new-born hearing strategies compared with the option of no screening in new-born infants. For the modelling of effectiveness the recommendations of the Panel on Cost-Effectiveness in Health and Medicine were followed [ 17 ]. Three possible strategies for neonatal hearing screening (NHS) were evaluated: - Universal neonatal hearing screening (UNHS): Every hospital-born baby is screened during the first days of life. - Risk factor screening (RS): The prevalence of hearing disorders is estimated to be higher at risk groups such as children with a positive family history, with congenital infections, cranofacial abnormalities, low APGAR-Score or low birth weight. This strategy screens all children with one or more risk factors for hearing disorders. - No systematic screening (NS): Children undergo the usual distraction test when presented to the paediatric service during the routine visit at the age of 12 weeks. Some neonates are screened in the hospital, but not in a systematic way. This reflects the present situation in Germany. UNHS is performed in some maternity hospitals and outpatient paediatric services with pilot screening projects in several regions. The target population of this analysis was all newborn infants. Health effects are expressed as quality weighed number of detected child months (QCM), and as true positive and false positive cases at certain developmentally important ages (6 and 12 months); for example, if a hearing impairment was diagnosed briefly after birth, the infant contributed six QCM at the age of six months. If the infant's hearing loss was diagnosed at the age of five months, the infant added only one detected child month at the age of six months. The term 'quality' is to reflect the idea that the early detection of impairment is a better and desired outcome, although there is no data on quality of life gained by this early detection. QCM, true positives and false positives are reported at the age of 6 and 12 months and with a time horizon of 120 months. Child months which were added up until the age of 6 months were multiplied with a weight of 1, child months added after the age of 6 months were multiplied with decreasing weighting. The derivation of this weight index is described in the section 'data and assumptions' below. We assumed that all children with hearing impairment would be detected before the age of 72 months, the age of school entry, regardless of the kind of screening strategy. In order to give outcomes in the present more weighting, compared to outcomes in the future, future effects have been discounted at an annual rate of 3%. Discounting reflects the higher value of money spent now as opposed to in the future. Similarly, discounting also weights outcomes experienced now (e.g., being diagnosed as true positive) more heavily than those experienced in the future. Table 1 gives the model parameters and their references. Table 1 Model parameters Parameter Base case (in %) Range for sensitivity analysis Source Prevalence of new-born hearing impairment 0.15 0.09–0.3 [3], [28], [29], [4], [30], [31], [25] Prevalence of one or more risk factors for hearing impairment 20 - [2, 25, 32] Prevalence of hearing impairment - In children with risk factors 0.38 - Author's calculation, [33] In children without risk factors 0.09 - Author's calculation Prevalence of risk factors in children with hearing impairment 50 48–56 [1, 28, 34] Sensitivity of screening 96 96–100 [11, 32, 35] Specificity of screening 89 77–96 [11, 32, 35] Sensitivity of diagnostic testing 98 - Specificity of diagnostic testing 98 - Coverage of screening 90 85–95 Author's estimate Follow-up after screening 80 75–85 Author's estimate Healthy children under suspicion of hearing impairment 0.1 - Author's estimate Discounting factor 3 per year 0–5 Weighs for quality adjustment Time to diagnosis ≤ 6 months of age 1 - Experts'estimate Time to diagnosis > 12 months of age 0.875 Time to diagnosis > 6 months and ≤ 12 months linear extrapolation Probability of "natural" discovery without systematic screening Weibull Distribution Median age at diagnosis 18 months - Empirical data *) *) author's calculations, derived from a representative survey, covering all diagnosed cases and the age of diagnosis in Upper Bavaria in 1998 and 1999 [20] The decision model We developed a state-transition (Markov) model [ 18 ] with monthly cycles to reflect the course of disease and diagnosis under the three screening strategies (figure 1 ). Probabilistic modelling has been performed by Monte Carlo simulations. The following health states were possible: Figure 1 Health states framework of the Markov model. Arrows indicate the possible transitions. "Unknown status" is the initial state, "True Positive" and "True Negative" are final (absorbing) states. - Unknown status - Healthy (hearing) confirmed by diagnostic test and/or screening – true negative - Healthy (hearing) not confirmed by diagnostic test - Hearing impaired confirmed by diagnostic test and/or screening – true positive - Thought to be healthy (hearing) but hearing impaired – false negative - Thought to be hearing impaired but healthy (hearing) – false positive - Not compliant/not followed up The baseline cohort consisted of infants with a certain prevalence of hearing impairment but unknown status regarding this disorder. Screened and impaired children are detected with the sensitivity of the screening test, whereas screened and healthy children are classified as healthy with the specificity of the screening test. All children with a positive screening test (i.e., true positives and false positives) undergo a second confirmatory diagnostic test, unless they did not adhere to the screening or did not present for the following tests (lost to follow-up). Impaired children who have not been screened in the first cycle can be diagnosed in the subsequent cycles according to the "natural history" of diagnosis, that is, because they don't develop speech adequately or become apparent during the routine visits to the paediatrician. In each cycle children can move to other health states according to the transition probabilities. QCM are only attributed to impaired children in whom hearing impairment is detected. Ultimately, all impaired children of the model cohort are diagnosed as impaired and all healthy children are either classified as healthy or remain unclassified. Data Professional (TreeAge Inc., Williamstown, MA) was used to construct and run the Markov model and Excel for Windows (Microsoft Corp.) was used to validate the model and to perform the Monte Carlo simulations. Data and assumptions A pre-defined and externally reviewed literature search was performed on new-born hearing screening using all relevant electronic databases. Search strategy and methods have previously been reported in detail [ 9 ]. In brief, we searched 13 medical databases including MEDLINE, EMBASE, Current Contents for published papers and HTA databases for published HTA reviews. Relevant articles were identified by a combined text word and thesaurus search. The references of the retrieved articles were then checked for further relevant articles. We restricted our search to publication dates from 1990 to September 2001. Reviewed publications were scored for study quality according to a standardised questionnaire developed by the German Scientific Working Group Technology Assessment for Health Care [ 19 ] and then either included or excluded. All model parameters are shown in Table 1 . A two-step screening strategy of Transient Evoked Oto-Acoustic Emissions (TEOAE) was chosen as the model for screening strategies, as this is one of the most widespread and commonly used technologies [ 9 ]chapter 2.2.7. The prevalence of congenital hearing disorders in children with risk factors was calculated using the prevalence of children born with one or more risk factors (20%) and the prevalence of risk factors in children with congenital hearing disorders (50%) using Bayes theorem. The probability of presentation with a falsely suspected hearing disorder in hearing children was estimated by a panel of four clinical experts. The probability of detection at a certain age without screening was calculated from a representative survey, covering all diagnosed cases and the age of diagnosis in Upper Bavaria in 1998 and 1999 [ 20 ]. A Weibull function for the probability to diagnosis, was fitted to the empirical data. The slope of the weight function has been previously estimated by experts making the following assumptions: each month detected before the age of 6 months is weighted with 1, assuming that children detected (and treated) within the first 6 months of life can develop normal speech and language abilities. If not detected within the first 12 months, profoundly and severely impaired children will conclude with a weight of 0.85, and moderately impaired children with a weight of 0.90. Presuming that 50% of the children with permanent congenital hearing disorders are moderately impaired, gives a weight of 0.875 for every month which is detected after the first birthday. The weights between 6 and 12 months were extrapolated in a linear fashion. Model assumptions Screening procedures and diagnostic procedures are based on different biological and clinical testing principles. We therefore assumed conditional independence of screening procedures and subsequent diagnostic procedures. Sensitivity analysis In order to investigate the influence of the parameter estimates on the outcome measures, one-way sensitivity analyses were performed on all relevant parameters. Ranges used for sensitivity analyses were derived from literature searches and are shown in table 1 . Multi-way probabilistic sensitivity analysis was performed on prevalence, sensitivity, specificity, coverage and follow-up using the Monte Carlo technique with 1,000 trials. Point estimates and 95% confidence limits were obtained by counting the number of trials in which a certain strategy has previously been found to be superior to the other strategies [ 21 ]. As we assumed that UNHS will always yield more QCM than RS or NS, we also calculated these confidence limits as a function of the difference between strategy effectiveness. This function results in a curve showing the cumulative relative frequency of trials (vertical axis) yielding a certain difference in QCM (horizontal axis) between two alternative strategies. The relative frequency gives an estimate of the probability of a certain difference in clinical effectiveness. The ranges for probability estimates, derived from the literature, assumed beta distribution. Model validation The decision model was validated on three levels: (i) Technical validation: The model was tested independently with two different software packages (TreeAge Data Professional and MS Excel). Routine tests (e.g., replacing the Weibull function by a constant, setting screening probability equal for all strategies) yielded the expected results. (ii) Internal validation: All data used to derive model parameter values, were reproduced exactly by the model (e.g., number of detected children at model end point). (iii) External validation: The derived values are consistent with external projections and estimates of recently published studies which were not used in our model [ 16 , 22 ]. for this section. Results Base-case analysis Table 2 presents the results of the base-case analysis. With the base-case prevalence of 150 per 100,000 in a hypothetical cohort of 100,000 children a maximum of 900 QCM would have been accumulated at the age of 6 months, 1800 at the age of 12 months, 18,000 at the age of 120 months (discounted: 892, 1771, 15503), if all children born with hearing impairment had been discovered at birth. UNHS discovered 644 weighted child months before the age of 6 months (72,2% of the expected value). RS yielded 393 child months (44,1%), no systematic screening 152 child months (17,0%). UNHS yielded 74,3% and 86,7% weighted child months at 12 and 120 months respectively, RS 48,4% and 73,3%, NS 23,7% and 60,6%. At the age of 6 months UNHS identified approximately 75% of all children born with hearing impairment, RS 50% and NS 25%. At the time of screening UNHS marked 10% of screened healthy children for further testing (false positives), RS 2%. Table 2 Results of modelling, base case assumption, for a hypothetical cohort of 100,000 children (QCM discounted at an annual 3%) Alternative strategies Expected value Outcome UNHS RS NS % % % QCM at 6 months 644 72.2 393 44.1 152 17.0 892 QCM at 12 months 1315 74.3 858 48.4 420 23.7 1771 QCM at 120 months 13436 86.7 11367 73.3 9394 60.6 15503 TP at 6 months 112 74.7 74 49,3 38 25.3 150 Incremental TP at 6 months 38 36 - TP at 120 months 150 150 150 150 FP after screening 9885 1973 - - UNHS = universal neonatal hearing screening RS = risk factor screening NS = no systematic screening QCM = quality weighed detected child months TP = true positives FP = false positives. Sensitivity analyses Results of sensitivity analyses are presented in table 3 . All sensitivity analyses are reported for 120 months follow up. Resulting QCM strongly depended on the prevalence of hearing disorders. Very low prevalence decreased the incremental benefit of UNHS versus RS. Comparing UNHS vs. RS, a prevalence of 9 per 1000 children yielded a gain of 1241 QCM, a prevalence of 15 per 1000 yielded a gain of 2027 QCM. The results were insensitive to varying assumptions about test parameters and the proportion of children lost to follow up. A decrease in slope of the linear weighting function resulted in decreasing incremental QCM. If detected child months were not weighted according to time of diagnosis, UNHS would still be superior to RS and RS to NS in terms of detected child months (data not shown). Figure 2 shows the frequency distributions of QCM per strategy as a result of multi-way probabilistic sensitivity analysis. QCM was higher in UNHS compared to RS or NS in 100% of the 1000 performed trials. Figure 3 shows the cumulative probability of obtaining a given fixed incremental value of QCM. With a probability of 95%, UNHS resulted in a gain of at least 1200 QCM compared with RS and a gain of at least 2500 QCM compared with NS. A gain of 1200 QCM means, for example, that 200 impaired children would be positively diagnosed at birth instead of at the age of 6 months, or that 1200 impaired children would be positively diagnosed before the age of 5 months instead of before the age of 6 months etc. Table 3 One-way sensitivity analyses for QCM at 120 months. The model was evaluated with a range of different values for one parameter while the other parameters were held constant. The ranges of the parameter values are given in table 1 Parameter Strategy Lower estimate Upper estimate Prevalence UNHS 8061 26872 RS 6820 22733 NS 5636 18787 Sensitivity of screening UNHS 13436 13608 RS 11367 11453 NS 9394 9394 Specificity of screening UNHS 13436 13436 RS 11367 11367 NS 9394 9394 Prevalence of risk factors in impaired children UNHS 13436 13436 RS 11284 11615 NS 9394 9394 Coverage of screening UNHS 13158 13714 RS 11204 11530 NS 9394 9394 Follow up after screening UNHS 13177 13694 RS 11237 11496 NS 9394 9394 Discounting factor UNHS 15725 12124 RS 13447 10180 NS 11276 8327 UNHS = universal neonatal hearing screening RS = risk factor screening NS = no systematic screening QCM = quality weighed detected child months Figure 2 Distributions of quality weighted detected child months (QCM, 100,000 screened children, 120 months) for newborn hearing screening strategies evaluated by Monte Carlo simulation (UNHS = Universal Newborn Hearing Screening). For example, if a hearing impairment was diagnosed briefly after birth, the infant contributed six QCM at the age of six months. If the infant's hearing loss was diagnosed at the age of five months, the infant added only one detected child month at the age of six months. Figure 3 Probability of a fixed level of incremental quality weighed detected child months (QCM) between strategies. These graphs are the results of a multi-way sensitivity analysis where all model parameters were varied simultaneously within the ranges described in table 1. They give the probability that the incremental gain of QCM between two screening strategies exceeds a certain value (given on the horizontal axis). This reads as follows: With a probability of 95% the difference between UNHS and RS will be 1200 QCM or more. Results of 1000 trials of a Monte Carlo simulation with a time horizon of 120 months. (UNHS = Universal Newborn Hearing Screening, RS = Risk Screening, NS = No Screening). Discussion We developed a decision-analytic Markov model for the evaluation of the effectiveness of different new-born hearing screening strategies. In our model, UNHS identified 72% of all detectable QCM at the age of 6 months, RS identified 44% QCM, the control group with no systematic screening identified 17% QCM. UNHS shows higher effectiveness even under a wide range of additional relevant parameters. QCM was introduced as a dynamic time-to-event measure which takes into account that the age of confirmation of congenital hearing impairments is important for further language development. The model results were not sensitive to test the accuracy of parameters within the assumed range but varied with the prevalence of hearing impairment. We have shown that UNHS leads to an earlier age of confirmed diagnosis compared to selective RS. The cost effectiveness of UNHS has already been modelled along secondary data [ 16 , 23 , 24 ]. This model is the first to establish a time-dependent and quality-weighted outcome, to introduce empirical data of the natural history of discovery and to present the results within a probabilistic framework. The strength of the presented model is that detected child months are multiplied by a weighted function which adds more benefit per month to children that were diagnosed before 6 months compared to those with late detection. In the existing literature the most interesting outcome, the proportion of children detected early enough, has not been modelled [ 16 ]. Our findings are consistent with study results and other projections of effectiveness. The Wessex Universal Neonatal Hearing Screening Trial Group found that 71 more babies with moderate or severe hearing loss per 100 000 target population were diagnosed before the age of 6 months during periods with neonatal screening than during periods without [ 25 ]. We found that UNHS would yield 112 true positive cases per 100 000 as compared to 38 without systematic screening, which is a difference of 74 babies. In our model UNHS identified 28% more children in time compared to RS. Thompson et al. estimated this difference to be between 19% and 42% [ 22 ] and stated that 77% of hearing impaired children would be identified before 10 months. Keren et al. assumed that UNHS detects 77% and RS 52% of hearing impaired children at the age of 6 months [ 16 ]. Any differences may be due to the fact that our function of discovery without systematic screening is more pessimistic. Our study has several limitations. Firstly, it does not differentiate between bilateral and unilateral hearing loss or moderate and profound impairment. We believe, however, that even with a more refined model the differences in effectiveness would not be substantial. Variations in the degree of impairment would result in variations in sensitivity of the test, and the model was rather insensitive to changes of test parameters. Secondly, even though QCM has been weighted, it is a surrogate parameter for the actual burden of disease for the child. Preference-based utilities have not been measured and the weighting for the impact of early or late identification of hearing loss used in our analysis were estimated by experts. Sensitivity analyses, however, revealed that without weighing, the difference in effectiveness would still be substantial. Thirdly, effectiveness was measured as a function of time to diagnosis. In routine health care, adequate treatment does not necessarily start immediately after the diagnosis is made. However, as knowledge on the consequences of delayed intervention is limited, including time to intervention as another variable in the model would have resulted in decreased precision [ 22 ]. Fourthly, our empirical data do not differentiate between congenital and acquired hearing impairment. The rather pessimistic estimation of the detection rate without screening, which might be due to the lack of differentiation between congenital and acquired hearing impairment, may bias the results in favour of UNHS. Similar data, however, have been published ([ 2 , 26 ], presenting a median age at diagnosis 18 months). Our estimate of discovery rate in a setting without screening might be biased by regional differences but can be easily replaced by a constant or a different rate function adapted to other local settings. Fifthly, the impact of late diagnosis on delayed language development, is not yet sufficiently known [ 22 ]. The economic effects on society in terms of lost productivity [ 24 ] and the quality-of-life effects on individuals have been discussed [ 27 ]. There is a lack of evidence concerning health care utilization due to hearing loss and the proportion of children following a regular school and professional career after timely fitting of a hearing aid. Further research should be undertaken to investigate the effect of the age of diagnosis and intervention, on the development of hearing impaired children and on their quality of life. Conclusions The value of this modelling exercise on effectiveness, lies in the facilitation and provision of information to decision makers- by quantitatively projecting available data, making explicit and transparent statements about assumptions and the degree of uncertainty involved in this area. The probability of timely intervention increases with UNHS. UNHS can reduce age of confirmation to a much greater extent than RS. In further studies, our model can be used to predict costs of real life situations to evaluate whether programme implementation costs would surpass cost-effectiveness thresholds. Policy makers can also base their decisions on the incremental effectiveness of UHNS, by introducing a screening programme. The model shows how likely an outcome is under the assumption of parameter uncertainty. In our model UNHS showed higher clinical effectiveness compared to RS and NS. The strength of our model lies in the naturalistic and generic structure, which makes it useful as a model for further evaluation. The model gives explicit, transparent and quantitative information about the effectiveness of different screening strategies for policy makers who have to decide on the potential impact of neonatal screening. The model presented here is easily adaptable to different settings. It will and should be verified and tested with longitudinal data of ongoing trials and model projects. This can be one of the first steps towards the needed transparency concerning universal new-born hearing screening. Competing interests The author(s) declare that they have no competing interests. Authors' contributions EG developed the decision model, carried out the statistical analyses and drafted the manuscript. FH designed and coordinated the study and participated in the development of the decision model and in the writing process. US participated in the statistical analyses, in the development of the decision model and in the writing process. PS-I, SK and AN performed the systematic review and data extraction. AN participated in the design of the study. JW conceived of the study and participated in its design and coordination. All authors revised the manuscript and read and approved the final version. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549034.xml |
545197 | Deadly Alliances: Death, Disease, and the Global Politics of Public Health | Most people threatened by AIDS, tuberculosis, and unsafe drinking water are poor and have little or no influence over the global politics of public health | The rancour surrounding the Bangkok AIDS summit of July 2004 has exposed a series of fundamental disagreements surrounding the global politics of public health. These tensions range from access to cheaper lifesaving drugs to disputes over the role of poverty and gender equality in the promotion of sexual health. The world is now experiencing the most profound public health challenge of the last forty years: we have witnessed the appearance of new diseases such as Ebola, SARS, and in particular, AIDS, combined with the alarming resurgence of diseases previously thought to have been under control, such as malaria and tuberculosis [ 1 , 2 ]. The AIDS pandemic in particular threatens to devastate entire regions and has already fundamentally altered the life expectancy and demographic profile of many countries in sub-Saharan Africa. Billions of people lack access to adequate sanitation and safe drinking water ( Figure 1 ), and the UN predicts that slums will become the dominant urban form within the next fifteen years ( Figure 2 ) [ 3 , 4 , 5 ]. Figure 1 A Recently Completed Water Supply Project in Amukoko, Lagos, Nigeria Unsafe drinking water is the major cause of disease in Lagos. Under 5% of households receive a piped water supply and under 1% are connected to a closed sewer system. (Photo: Matthew Gandy) Figure 2 Open Sewer with Drinking Water Pipes Passing Through This photograph was taken in Amukoko, one of the largest slum areas in Lagos, Nigeria, a rapidly growing megacity with an estimated population of 15 million. Lagos is predicted to be one of the world's largest cities within the next 10 years. (Photo: Matthew Gandy) Yet the very idea of “public health” sits uncomfortably alongside the current emphasis of bio-medical science on the molecular realm of DNA coding and the development of lucrative Western markets for new pharmaceutical products such as sildenafil (Viagra). The needs of the majority—the global poor—scarcely feature within this tactical alliance between the biomedical sciences and corporate power. Since most people threatened by AIDS, tuberculosis, unsafe drinking water, and other health threats are poor, they have little or no influence over the global politics of public health. Making Sense of the Crisis If we are to make sense of the current public health crisis, we need to explore interconnections between political, economic, and social developments that are ignored by the fragmentary emphasis of the biomedical sciences. The current impetus toward economic globalization is causing widespread social and economic disruption, ranging from wild currency fluctuations to the systematic collapse of viable agricultural systems [6] . The imposition of austerity packages—widely referred to as structural adjustment programmes—in combination with the forcible extension of global markets for Western products is plunging millions of people into poverty and economic dependence. Existing primary health care services in much of the developing world and the former states of the Soviet Union have been drastically cut back, and services that were once freely available are now increasingly beyond the reach of the poor. And these harmful trends also extend even to the wealthiest global cities, such as London and New York, where a combination of poverty, homelessness, and cutbacks in primary health care since the 1980s has contributed toward the spread of tuberculosis and other diseases [ 7 , 8 ]. An examination of the social impact of the global public health crisis shows that it is women and children who have been most badly affected. One of the least addressed dimensions to the externally imposed structural adjustment programmes on developing countries is the harmful impact on the sexual health of women caused by their increased economic dependence on men. Sexual health programmes based on abstinence, for example, ignore the difficulties women face in negotiating safer or non-penetrative forms of sex. It is married women in southern Africa and south Asia who make up the largest and most vulnerable group of women, since they are at risk of being infected by their husbands. A new generation of HIV-positive women activists, such as the Nigerian journalist and AIDS campaigner Rolake Odetoyinbo Nwagwu, are now engaged in a vital struggle to challenge the social attitudes and economic inequalities that have driven the devastating impact of AIDS on women and children in developing countries. The spread of AIDS in more traditional societies is closely linked with patriarchal power structures.. The sustenance of these structures is assured by the rise of poverty-fuelled ethnic and religious chauvinism, which undermines the prospects for developing more progressive approaches to social policy. And in regions where war or civil strife prevail, the vulnerability of women and children to sexual violence, economic exploitation, and disease is even greater so that we cannot consider public health questions separately from issues surrounding political stability and social justice. Spaces of Exclusion In order to understand better the political dynamics behind public health we need to recognize that the development of modern forms of governance has emerged in tandem with new approaches toward the administration of human populations. The rise of the 19th-century industrial city, for example, necessitated the development of much more sophisticated forms of urban governance in order to tackle the threat of epidemic disease and enable these new cities to function effectively as centres of economic activity. But these new spaces of public health control, which scholars such as Michel Foucault have so vividly described, had a shadowy “other”. This “other” was represented by the increasingly squalid conditions endured by the mass of the population in European colonies, exemplified by the devastating modern outbreaks of bubonic plague in cities such as Baghdad, Lagos, and Bombay [9] . The Italian political philosopher Giorgio Agamben has elaborated on this distinction between “inside” and “outside” under modern systems of governance to reveal the persistence of what he terms conditions of “bare life” within even the most sophisticated legal and political systems [10] . The contemporary exclusion of the world's poor from adequate medical care is thus a form of state-sponsored violence, in which millions are denied even the most basic human rights. These “wasted lives”, to use the sociologist Zygmunt Bauman's phrase, represent a literal as well as metaphorical process of permanent and deadly exclusion for the poor, the marginalized, and others who have no value within the global economy [11] . Excluding the Poor from Medical Care A striking manifestation of this systematic exclusion of the poor from medical care is provided by the Bush administration's efforts to stymie access to affordable antiretroviral drugs. A dramatic standoff in 2001 between the global pharmaceutical industry and public health activists in South Africa, following the import of cheap generic antiretroviral drugs manufactured in Brazil, led to the historic Doha Declaration on intellectual property rights and public health [12] . Generic drug production in countries such as India, Brazil, and Thailand has succeeded in bringing down treatment costs per patient from $10,000 to $300 per year, yet the World Health Organization has revealed that less than one in 20 people who need antiretroviral treatment in the developing world are currently receiving it [13] . In order to widen access to affordable drugs, the current production of generic medicines will have to be massively expanded, but the Bush administration and its corporate allies in the pharmaceutical industry has worked assiduously to undermine the potential impact of the Doha agreement [14] . The United States government, for example, has negotiated bilateral trade deals with countries such as Chile and Thailand in an effort to dissuade them from the production of cheaper drugs [15] . Tensions exploded into the open at the Bangkok AIDS summit of 2004, where lobbyists on behalf of the US government put forward a series of specious assertions about the distribution of much-needed AIDS drugs in sub-Saharan Africa. The lobbyists asserted that the costs of drugs have been lowered—yet discounted brands remain at least twice as expensive as generic drugs. And the lobbyists for the Bush administration claimed that generic drugs undermine profitability and hence the incentive for new research (which is, in any case, overwhelmingly focused on the bloated market for prescription drugs in the US). Of the $15 billion pledged by the Bush administration in the fight against AIDS most of this money will be focused on 15 selected countries. The countries were chosen by their willingness to abide by bilateral trade deals to prevent the production of cheaper generic drugs and their willingness to stress the centrality of abstinence as a strategy for AIDS prevention (a strategy that the religious Right has pushed for) [ 16 , 17 ]. It should also be noted that some of the most vocal critics of US policy, among them France, currently make a derisory financial contribution to global efforts to tackle HIV/AIDS so that the inadequacies of US policy must be viewed in a wider context of Western negligence toward the public health needs of the world's poorest countries [18] . The US government has sought to highlight the inadequacies of health care infrastructure in developing countries as a further justification for the diversion of attention from the cost of drugs. But primary health care services have themselves been undermined by the structural adjustment and trade policies promoted by Western financial institutions and their corporate backers. The global politics of AIDS is therefore caught in a neo-liberal vicious spiral from which it is impossible to disentangle the needs for social and institutional reform in the worst affected regions from the challenge of widening access to available treatments. The Contemporary Politics of Public Health Public health pioneers of the past, such as the German bacteriologist Robert Koch, were not only scientists but also political advocates for social change. Improvements in health care were perceived as part of a nexus of reforms that ranged from better housing and nutrition to an extension of voting rights to ensure that the poor had adequate political representation. The contemporary politics of public health needs to be considered, however, in relation to wider discourses on security and human welfare that are quite different from those of the 19th century. The current Western preoccupation with the threat of international terrorism, for example, has distracted attention from the much more real threats to human well-being created by poverty and ill health. The vast resources—both financial and logistical—now being poured into “security” contrast starkly with the inadequate funding provided for the Global Fund for HIV/AIDS, Tuberculosis, and Malaria. The increasingly unilateral and self-interested stance of the US illustrates the tensions between attempts to build international forms of governance for health care and the continuing centrality of national interests or global institutions that explicitly represent the economic power of a relatively small group of nations [19] . The paradox is that widening global disparities in wealth and poverty play a role in fostering the social and political conditions in which religious fanaticism and hatred for the West can flourish. So any “war on terror” that fails to address the causes of poverty, despair, and insecurity in the lives of the world's poor will ultimately create a more dangerous world for everyone. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545197.xml |
534785 | The measurement of health-related quality of life (QOL) in paediatric clinical trials: a systematic review | Background The goal of much care in chronic childhood illness is to improve quality of life (QOL). However, surveys suggest QOL measures are not routinely included. In addition, there is little consensus about the quality of many QOL measures. Objectives To determine the extent to which quality of life (QOL) measures are used in paediatric clinical trials and evaluate the quality of measures used. Design Systematic literature review. Review Methods Included paediatric trials published in English between 1994 and 2003 involving children and adolescents up to the age of 20 years, and use of a standardised QOL measure. Data Sources included MEDLINE, CINAHL, EMB Reviews, AMED, BNI, PSYCHINFO, the Cochrane library, Internet, and reference lists from review articles. Results We identified 18 trials including assessment of QOL (4 Asthma, 4 Rhinitis, 2 Dermatitis, and single studies of Eczema, Cystic fibrosis, Otis media, Amblyopia, Diabetes, Obesity associated with a brain tumour, Idiopathic short stature, and Congenital agranulocytosis). In three trials, parents rated their own QOL but not their child's. Fourteen different QOL measures were used but only two fulfilled our minimal defined criteria for quality. Conclusions This review confirms previous reports of limited use of QOL measures in paediatric clinical trials. Our review provides information about availability and quality of measures which will be of especial value to trial developers. | Review Introduction Chronic disease affects approximately 18% of children [ 1 ]. Although cure is not possible, survival rates have improved substantially for many conditions (e.g. cancer [ 2 ] and cystic fibrosis [ 3 ]). Many diseases require daily self-management and restrict children's physical and social activities. Consequently questions are increasingly raised about the quality of life (QOL) of children with chronic disease. Efforts to measure child QOL have proved complex but a number of generic and disease-specific measures have been reported [ 4 ]. Generic measures are designed to assess and compare health status in patients with different diseases and may provide valuable information for comparing outcomes between sick and healthy populations. They are generally well validated and reliable but are often not recommended for work involving evaluation of randomised controlled trials (RCTs), as they lack sensitivity to detect small but clinically significant changes in QOL over time or due to treatment for specific diseases [ 5 ]. Disease specific measures are more suitable for evaluation of clinical trials designed to assess a particular treatment. These measures include items that are likely to be affected by the specific disease or treatment and are therefore more responsive to clinically significant changes. The quality of measures must be evaluated according to performance characteristics. Guidelines suggest good measures of QOL are reliable and valid for the group of patients for whom they are used, include a form for self-report wherever possible, are brief and developmentally appropriate, and allow completion by proxy [ 4 ]. There is little evidence that QOL measures are routinely used in clinic practice [ 6 ] or clinical trials [ 7 ], despite the fact that the aim in many trials is to improve QOL. In both child [ 8 ] and adult work [ 9 ], few trials include measures of QOL, and amongst those, non-standardised measures continue to be used. QOL is also frequently insufficiently analyzed, reported or discussed in the study report or subsequent publications [ 5 ], despite the increasing emphasis in clinical practice and research to use patient centered outcomes and child perspectives [ 10 ]. We report a systematic review drawing on established methodologies [ 11 ] to determine first, the extent to which QOL measures are used in paediatric clinical trials and RCTs, and second, the quality of QOL measures currently used. Method Search Strategy The following databases were searched: MEDLINE 1966 to Nov Week 2 2003, CINAHL 1982 to December Week 1 2003, EMB Reviews: Cochrane Central Register of Controlled Trials, Your Journals at OVID, EMB reviews: ACP Journal Club 1991 to July / August 2003, EMB reviews: Database of Abstracts and Reviews of Effects 3 rd quarter 1993, AMED (Allied & Contemporary Medicine) 1985 to December 2003, British Nursing Index (BNI) 1985 – October 2003, EMBASE, PSYCHINFO 1872–2003. Text word and thesaurus searches were used to minimise the chance of missing relevant articles. The following keywords were searched: • child, childhood, children, adolescent, infant, pediatric, paediatric, • quality of life, QOL, • clinical trial, randomised controlled trial Searches were restricted to English language papers. Search engines were used to search the Internet with keywords and Boolean logic. Additional references from articles identified through these searches were also pursued. Inclusion Criteria These included: 1) Children and adolescents up to the age of 20 years, 2) RCT, formal cross-over trial, or studies evaluating one or more active drug treatment with or without placebo, 3) Standardised QOL measure (For these purposes we drew on a previous review [ 4 ] and defined minimal psychometric criteria to include some preliminary reliability and validity data), 4) Articles published in English between January 1994 and December 2003. Exclusion Criteria 1) Samples including both adults and children. 2) Comparison of surgical treatment, pain control, palliative medication, or psychological/homeopathic intervention. 3) Outcomes evaluated in terms of medical data only, non-standardised measures of QOL or standardised psychological measures including symptom checklists, measures of self-esteem, or coping. Procedure Abstracts were reviewed for relevance and full articles obtained where appropriate. A summary sheet was developed and both authors independently reviewed papers to ensure reliability. Data extracted by reviewers was second coded and compared and any discrepancies were resolved through discussion. Results Of the 917 records retrieved from the databases, initial inspection suggested that 27 abstracts met the inclusion criteria. On reading the full articles, nine failed to meet inclusion criteria. The resulting 18 articles were included in the review [ 12 - 29 ]. Study characteristics (summarised in Table 1 [see Additional file 1 ]) • Disease: QOL was most frequently included in trials in atopic diseases (Asthma = 4, Rhinitis = 4, Dermatitis = 2 and Eczema = 1). Single studies were identified in Cystic fibrosis, Otis media, Amblyopia, Diabetes, Obesity associated with a brain tumour, Idiopathic short stature, and Congenital agranulocytosis. • Location: Seven studies were conducted in the U.S.A, 4 in the U.K, 3 in the Netherlands, one in Taiwan, and one in Israel. Two studies were multi-national. • Child's age: Three studies recruited children across a broad age range (1 to 18 years), 2 focused on pre-school children (1–5 years), 4 on pre-school and middle childhood (2–10 years), 2 on middle childhood (6–12 years), 6 on middle childhood and adolescence (5–18 years), and one on adolescents alone (12–17 years). • Sample size: Sample size ranged from 19 [ 29 ] to 689 [ 15 ]. Power calculations were reported in six studies. • Design and trial aim: We identified 11 RCTs, 2 cross-over studies, and 5 studies comparing two or more active treatments without placebo or control group. Of the 11 RCTs, 1 was multi-national, 7 multi centre, and 3 single centre studies. Of the 7 non RCTs, 1 was multi-national, 2 were multi centre, and 4 single centre. Nine articles involved comparisons of two or more treatment and the remainder involved comparison of treatments with placebo. • Blinding: Seven RCTs reported blinding procedures. • Parent and caregiver QOL: Fifteen studies measured the impact of the disease on the child's QOL. Three included assessment of the caregivers QOL. • Respondent for child QOL: Of the 15 studies focusing on child QOL, 10 were based on child, and 3 on parent reports. In two studies both children and parents reported the child's QOL and in one of these clinicians also rated child QOL [ 28 ]. Quality of QOL measures (Table 2 [see Additional file 2 ]) • Generic or disease specific: In total, 12 disease specific and two generic measures were used [ 30 - 38 ]. The four asthma trials involved three different measures of asthma specific QOL. The four perennial rhinitis trials used two different measures of rhinitis specific quality of life, and the two atopic dermatitis trials and one atopic eczema trial used two different dermatology specific measures of QOL. In two studies authors had developed their own disease specific measure [ 25 , 29 ]. • Quality of measure: We assessed quality of measures based on minimal accepted criteria [ 4 ] whereby measures should be brief, allow proxy and self report and include reliability and validity data and age appropriate versions. Although all measures included some preliminary psychometric data, only two measures fulfilled all of these criteria [ 36 , 38 ]. Three measures fulfilled four criteria but lacked age appropriate versions. The remaining measures fulfilled three or less criteria. Discussion Despite extensive searches we identified only 18 published reports of paediatric trials including standardised QOL measures. This undoubtedly represents a very small percentage of paediatric trials and supports previous findings that QOL data is seldom reported in paediatric clinical trials [ 8 ]. Asthma and rhinitis were most frequently studied, perhaps because there is higher incidence for these conditions in children compared to other conditions such as cancer and cystic fibrosis [ 39 ]. Further explanations include the non-life threatening nature of these conditions as well as the availability of disease specific measures compared to rarer illnesses. In considering why there are relatively few trials including QOL measures, it is important to take into account the aims and purpose of the trial [ 40 , 41 ]. The aim of most trials is to assess the impact of treatment on clinical variables, with QOL viewed to be of secondary importance if at all. It is not necessarily appropriate that QOL measures are included in all trials. Where QOL assessment is appropriate however, inclusion of a QOL measure must be hypothesis driven and an integral part of the clinical development programme rather than an added afterthought [ 5 ]. Quality of measures Where QOL was measured, disease specific measures were most often used (N = 12) as is normally recommended for use in clinical trials. Only two trials included measures that satisfactorily fulfilled accepted criteria [ 4 ]. Typically, information about measures included some reliability data although a third of studies failed to provide information about the validity of the scale. Most measures were brief and contained less than 30 items but many lacked age appropriate versions or parallel versions for child and proxy raters. Selection of a measure of QOL is dependent on the psychometric properties of the instrument, as well as clinical and demographic variables characteristic of the sample. However, psychometric properties depend upon samples for which the scale has been validated. Hence it is important to ensure measures are used with clinical populations where psychometric data are available. There are some grounds for assuming that QOL changes during childhood, and therefore satisfactory measures target specific age groups [ 42 ]. There are difficulties identifying single measures that are appropriate across a wide age range and only half of measures identified in this review included age appropriate versions. It is also generally recommended that ratings of QOL should be made by children themselves whenever possible [ 43 ]. In cases of younger children proxy reports are necessary but there are questions about the relationship between child and parent report [ 4 ]. It is therefore positive that most (73.3%) studies obtained ratings from children with only four relying on parents alone to provide proxy ratings. CONSORT [ 44 ] guidelines recommend methods of reporting RCTs, but do not adequately deal with the issues concerning QOL assessment and psychometric validity. It is essential that trial developers select appropriate measures and are aware of the problems associated with QOL assessment. Barriers to inclusion of QOL measures Objections to inclusion of QOL measures in trials involve anticipated increased costs, extra time needed to gain patient and parent consent, and lack of sophistication of currently available measures [ 8 ]. A major restriction to inclusion of QOL assessment in clinical trials remains limitations in currently available measures, especially for less prevalent chronic conditions. However, it is only through including measures that we will learn more and be able to develop a second generation of measures that do show more sophisticated properties. A second problem is that disease specific measures may simply not be available for rare conditions. Attempts to develop such measures are promising and in this review instruments for ambylopia [ 25 ] and agranulocytosis [ 29 ] had been developed. In order to facilitate collection of QOL data from children with chronic illness, reliable and valid measures are increasingly required [ 46 ]. Other methodological limitations in current work include the lack of power calculations. Where the aim of the trial includes QOL assessment, power calculations must be performed and are an essential element of clinical trial design. In cases where measurement of QOL is a secondary endpoint, sample size calculations are rare and difficult to establish. However attempts should be made to hypothesise expected changes in QOL scores in relation to the agreed sample size prior to the trial [ 5 ]. Conclusion This review supports previous findings of limited use of QOL measures in paediatric cancer trials [ 9 ] and extends this to include a number of conditions other than cancer. QOL assessment is most common in trials where the aim is to compare the impact of treatment on clinical variables and is largely limited to common non-life threatening conditions. The measurement of QOL provides valuable information about the psychological and social impact of treatment on children especially where no differences in survival rates are anticipated. For this reason, the inclusion of QOL measurement in paediatric trials is becoming increasingly valued and mandatory [ 47 , 48 ]. There are still questions concerning selection of QOL measures and how best to report findings [ 49 ], but our review provides useful information for trial developers regarding the availability and quality of QOL measures. Author's contributions Both authors were responsible for planning, conducting and reporting this work and approved the final manuscript. Supplementary Material Additional File 1 Table 1: Study characteristics Click here for file Additional File 2 Table 2: Quality of life measures Click here for file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC534785.xml |
548302 | Genetical and functional investigation of fliC genes encoding flagellar serotype H4 in wildtype strains of Escherichia coli and in a laboratory E. coli K-12 strain expressing flagellar antigen type H48 | Background Serotyping of O-(lipopolysaccharide) and H-(flagellar) antigens is a wideley used method for identification of pathogenic strains and clones of Escherichia coli . At present, 176 O- and 53 H-antigens are described for E. coli which occur in different combinations in the strains. The flagellar antigen H4 is widely present in E. coli strains of different O-serotypes and pathotypes and we have investigated the genetic relationship between H4 encoding fliC genes by PCR, nucleotide sequencing and expression studies. Results The complete nucleotide sequence of fliC genes present in E. coli reference strains U9-41 (O2:K1:H4) and P12b (O15:H17) was determined and both were found 99.3% (1043 of 1050 nucleotides) identical in their coding sequence. A PCR/RFLP protocol was developed for typing of fliC -H4 strains and 88 E. coli strains reacting with H4 antiserum were investigated. Nucleotide sequencing of complete fliC genes of six E. coli strains which were selected based on serum agglutination titers, fliC -PCR genotyping and reference data revealed 96.6 to 100% identity on the amino acid level. The functional expression of flagellin encoded by fliC -H4 from strain U9-41 and from our strain P12b which is an H4 expressing variant type was investigated in the E. coli K-12 strain JM109 which encodes flagellar type H48. The fliC recombinant plasmid carrying JM109 strains reacted with both H4 and H48 specific antisera whereas JM109 reacted only with the H48 antiserum. By immunoelectron microscopy, we could show that the flagella made by the fliC -H4 recombinant plasmid carrying strain are constituted of H48 and H4 flagellins which are co-assembled into functional flagella. Conclusion The flagellar serotype H4 is encoded by closely related fliC genes present in serologically different types of E. coli strainswhich were isolated at different time periods and geographical locations. Our expression studies show for the first time, that flagellins of different molecular weigth are functionally expressed and coassembled in the same flagellar filament in E. coli . | Background Bacterial strains belonging to the Enterobacteriaceae species Escherichia coli are common as commensals in the intestinal flora of humans and warm-blooded animals [ 1 ]. Typing systems for identification of related E. coli strains were developed in the early 1940ies when it became evident that certain E. coli strains were agents of infantile gastroenteritis [ 2 ]. In 1944, Kauffmann established the method of serological typing for E. coli O- (lipopolysaccharide) and H-(flagellar) antigens which allowed the grouping of E. coli strains according to their O:H-types (serotypes) [ 3 ]. Serotyping proved to be widely useful for identification of enteropathogenic E. coli (EPEC) strains from stools of diarrhoeic infants [ 4 , 5 ] and is successfully employed for characterization of pathogenic E. coli strains from both humans and animals [ 2 , 3 ]. The genetic analysis of E. coli populations by multilocus enzyme electrophoresis (MLEE) and multilocus sequence typing (MLST) allowed the detection of clonal types of strains which carry specific virulence markers and are associated with disease in humans [ 6 , 7 ]. It was shown that the O:H serotype is a good marker for identification of strains belonging to clonal types of pathogenic E. coli [ 6 , 7 ]. At present, 176 O- and 53 H-antigens are described for E. coli which can occur in different combinations in wildtype isolates of strains [ 2 , 3 , 5 , 8 ]. However, the large number of O- and H-antisera which are needed for E. coli serotyping and the laborious typing procedure restricts its usage to a few reference laboratories. Therefore, alternative typing methods were developed including molecular characterization of genes coding for the O- and H-antigens in E. coli [ 9 - 12 ]. Typing of fliC genes by PCR was successfully employed for characterization of human pathogenic O157:H7 and O26:H11 strains [ 13 , 14 ]. Analysis of the nucleotide sequence of fliC genes coding for flagellar antigens H7 and H6 revealed large sequence similarities between strains sharing the same H-type but different O-types [ 15 , 16 ]. Moreover, molecular typing of the fliC gene allows H-typing of non-flagellated (non-motile) E. coli isolates which cannot be analyzed for their flagella with H-specific antisera [ 14 , 15 , 17 ]. The flagellar type H4 is frequently occurring in E. coli belonging to many different O-groups including strains of shiga toxin-producing E. coli (STEC) and extraintestinal pathogenic serotypes [ 18 - 20 ], (K.A. Bettelheim, The VTEC table, May 2003, ). Moreover, a cryptic fliC -H4 gene was described to be present and to be spontaneously expressed in E. coli strain P12b (O15:H17) which carries another type of flagella called H17 which is not encoded by the fliC gene [ 21 , 22 ]. Therefore, we became interested in the genetic variability of flagellar H4 genes in E. coli strains belonging to different O-serogroups and pathotypes. We have used the published nucleotide sequence of the fliC gene present in the E. coli H4 reference strain U9-41 (accession AB028472) to develop a PCR which allows discrimination between the fliC -H4 gene variants. Nucleotide sequence analysis of the fliC gene was performed on other E. coli H4 strains that either showed deviations in the PCR analysis or were reported to harbour allelic types of the H4- fliC gene [ 9 ] or revealed differences in the agglutination reaction compared to the reference strains. To study the expression of different flagellar H4 types we have cloned their corresponding fliC genes and have introduced them into the laboratory E. coli K-12 strain JM109 [ 23 ]. Expression of recombinant flagella was demonstrated by serotyping and by immuno electron microscopy. Results Serological detection of the flagellar H4 antigen in different E. coli wildtype host strains E. coli reference strains U9-41 (O2:K1:H4) and P12b (O15:H17) [ 3 ] and the E. coli K-12 strain JM109 (O-rough:H48) (this study), which was used for expression studies were investigated for inhibition of motility in swarm-agar containing 1:600 dilutions of either H4, or H48-antiserum (see Methods). U9-41 and P12b were fully inhibited for their motility in the presence of H4 antiserum derived from strains U9-41 or from P12b but not in the presence of H48 specific antiserum. The E. coli K-12 strain JM109 was not inhibited for motility by H4 antisera but by H48 antiserum. These findings indicate that H4 antisera specifically inhibited the motility of E. coli H4 strains and that the antigenically different flagellar type H17 was not expressed or lost in our P12b isolate, similar as previously described with spontaneously arising variants of P12b [ 24 ]. We became interested if other E. coli H17 strains would also carry the genes for expression of H4 type flagella as it was described for P12b [ 21 , 22 , 24 ]. For this, we have investigated five additional E. coli H17 strains (872-69, 107-74, 305-78, 870-69 and 327-01) for their serological reaction with different H4 specific antisera and all strains showed specific positive reactions (Table 1 ). H-serotyping performed with strains from the collection of the Robert Koch-Institute revealed 88 E. coli strains which agglutinated with H4 antisera and the results obtained with 10 representative strains are shown in Table 1 . All strains agglutinated with both, H4 U9-41 and H4 P12b antiserum, but were not agglutinating with antisera made against other H-antigens (data not shown). Differences in agglutinating titers between H4 U9-41 and H4 P12b antisera were not more than twofold with either strain indicating that both sera were similar for their specificity (Table 1 ). The strains P7d, E1541-68, 107-74 and 305-78 showed significantly lower agglutination titers with H4 antisera than did the reference strains U9-41 and P12b, which had been used for production of H4 typing sera, respectively (Table 1 ). These findings prompted us to compare all the H4 strains with all other E. coli H-types for polymorphisms in the fliC gene by HhaI digestion of fliC -PCR products as described in the Method section. HhaI-RFLP typing of fliC genes in E. coli PCR products obtained with primers fliC-1 and fliC-2 were digested with HhaI to obtain H-serotype specific RFLP patterns from reference strains for the 53 different E. coli H-serotypes [ 3 , 9 ]. HhaI-RFLP typing of E. coli fliC genes coding for flagellar types H1 to H56 revealed individual patterns corresponding to the different H-serotypes (Fig. 1 ). Flagellar antigens H3, H17, H35, H36, H44, H47, H53, H54 and H55 were reported to be not encoded by fliC but by other genes ( flkA , fllA, flmA and others) in the corresponding E. coli strains and the fliC HhaI patterns obtained from these strains do therefore not correspond to their H-serotypes [ 21 , 25 - 27 ]. HhaI-RFLP patterns were found conserved among strains sharing the same H-serotype independent of their O-antigen as previously described [ 9 ] (data not shown). An exception was found for H-serotypes 2, 8, 18, 19, 21, 33 and 47 in which single strains possessed different HhaI-RFLP patterns when compared with the corresponding H-type reference strain (data not shown). The HhaI-RFLP patterns of different H-types were distinguishable from each other (Fig. 1 ). The results from HhaI-RFLP typing corresponded with the earlier reports showing that the fliC gene in strain P12b codes for flagella of serotype H4. The relationship between fliC genes present in the different E. coli strains was further investigated by nucleotide sequencing. Nucleotide sequence analysis of the fliC genes present in representative E. coli H4 strains To obtain the entire coding region of the fliC gene in E. coli strains oligonucleotide primers fliC-5 and fliC-6 were deduced from the fliC -H4 chromosomal region and applied to amplify the corresponding chromosomal regions from E. coli strains U9-41 (O2:K1:H4), P12b (O15:H17), U1-41 (O5:K4:H4), P7d (O68:H4), C107-74 (O15:H17) and E1541-68 (O154:H4) as listed in Table 1 . These strains were taken as flagellar type H4 representatives according to previously published results [ 3 , 9 ] and according to the H-serotyping performed in this study. The analysis of the coding sequences of the fliC genes present in these strains revealed in all cases a length of 1050 bp. As a control, the fliC gene of the strain U9-41 was sequenced and found to be identical to the previously deposited fliC sequence (accession AB028472). The identity of the fliC -H4 sequences ranges from 97,6 % to 100 %. The greatest sequence difference to the fliC gene coding sequence of strain U9-41 was found in the fliC gene of strain P7d (exchange of 20 nucleotides), whereas the fliC sequences of strains U1-41 and U9-41 were identical to each other. All deduced flagellins have a length of 349 amino acids. The greatest deviation in the primary structure to the FliC protein of the reference strain U9-41 was again observed for the FliC protein of strain P7d (exchange of 9 aa in a stretch of 349 aa). Fig. 2 shows the alignment of the deduced FliC proteins of all investigated strains. The FliC protein of strain U1-41 is 100% identical to that of strain U9-41 and therefore not shown. Similarities of the deduced amino acid sequences of the FliC proteins (flagellins) are summarized in Table 2 . PCR based detection of fliC -H4 specific DNA sequences Amplification of the fliC genes present in E. coli strains U9-41 and P12b with primers fliC-1 and fliC-2 resulted in the generation of a 953 bp internal PCR product with both strains (Fig. 3 , lanes 4+5). The nucleotide sequences of this stretch of DNA of strains U9-41 and P12b were compared for restriction enzymes which cut at different sites in the fliC -H4 U9-41 and fliC -H4 P12b sequence. HhaI was found to cut at identical sites confirming the results from HhaI RFLP typing (Table 3 ). In contrast, enzymes HpaII and MboI were found to generate each different restriction fragments from PCR products of the fliC -H4 U9-41 and fliC -H4 P12b gene, respectively. Both enzymes were taken for RFLP typing of amplified fliC genes (primers fliC-1 and fliC-2) from 88 E. coli strains which showed agglutination reactions with H4 antisera (Table 3 ). Eighty-six of the 88 strains showed HpaII and Mbo I restriction profiles which corresponded to the patterns obtained with strain U9-41 (O2:K1:H4) (Table 3 and Fig. 3 ). Exceptions were made by strains P7d (O68:H4) and P12b (O15:H17) which showed individual restriction patterns which differed from all other H4 strains investigated in this study. (Table 3 ). Cloning and expression of fliC -H4 genes in the E. coli K-12 strain JM109 In order to study the functional expression of the fliC -H4 genes in a different genetic background we cloned the corresponding PCR products of strains U9-41 and P12b into the vector pLITMUS38 as described in the Methods section. The fliC coding regions were inserted downstream of the lacZ promoter of pLITMUS38 and the fliC recombinant plasmids were transformed into the laboratory E. coli K-12 strain JM109 [ 23 ]. JM109 was serotyped as O-rough:H48, and it showed the same HhaI-RFLP fliC -pattern as the E. coli reference strain P4 (O16:H48) [ 3 ] (Fig. 1 ). The functional expression of the cloned fliC -H4 genes in the JM109 derivative strains TPE1976 ( fliC -H4 U9-41 clone) and TPE1978 ( fliC -H4 P12b clone) was analyzed by tube agglutination with H4 and H48 antisera, respectively. The strains TPE1976 and TPE1978 showed agglutination with H48 and H4 sera whereas the parental JM109 strain reacted only with H48 serum (Table 4 ). To find out if the flagellins of the fliC recombinant plasmid carrying JM109 strains were co-assembled in all flagella or assembled separately we studied the parental strain JM109 and its fliC- H4 derivative TPE1978 by IEM (Fig. 4B–G ). Both strains were found to express 2-4 flagella per cell which appeared morphologically typical as long helical filaments (diameter 18 nm, length up to 20 μm) (Fig. 4A ), [ 29 ]. The reaction of H48 antiserum with bacteria followed by detection of adsorbed antibodies with anti-rabbit-IgG coupled with 10 nm immuno-gold particles showed a specific and homogeneous labelling of flagella present on the surfaces of JM109 (Fig. 4C ) and of TPE1978 (Fig. 4B ). When H4 serum was used, only flagella of strain TPE1978 became immuno-labelled (5 nm gold particles) (Fig. 4D ) and no labeling was observed with JM109 (Fig. 4E ), confirming H-serotyping results. Sequential double labeling experiments with H48 and H4 antibodies resulted in staining of all flagella on the surfaces of JM109 (Fig. 4G ) and TPE1978 (Fig. 4F ). However, both 5 nm (H4 label) and 10 nm (H48 label) gold particles were only bound to flagella of TPE1978 (Fig. 4F ) whereas the flagella of JM109 were exclusively labelled with 10 nm gold particles which indicates the H48 antigen (Fig. 4G ). These results demonstrate that both H48 and H4 flagellins are co-assembled in the flagella made by the fliC -H48/H4 genes in strain TPE1978. Discussion The fliC genes of representative E. coli strains for the 53 different H-types were recently investigated and compared for their nucleotide sequences [ 21 ]. Among the H-type reference strains which were not analyzed for their complete fliC gene sequences, were the reference strains for the E. coli H4 (U9-41) and H17 (P12b) flagellar antigen [ 21 , 22 ]. In this work, we performed a complete characterization of the fliC- H4 genes from different E. coli strains with primers which were generated on the basis of the previously published fliC sequence (accession AB028472) of the E. coli H4 reference strain U9-41. The comparison of the fliC gene sequence encoding H4 flagella in strain P12b with the fliC sequences of five other E. coli H4 or H17 strains revealed a high similarity on the DNA and on the amino acid level (Fig. 2 and Table 2 ). We established a PCR/RFLP typing assay for genotypic investigation of clinical E. coli isolates which reacted with H4 antisera. Genotyping of 88 E. coli strains comprising 20 different O-serogroups (Table 3 ) revealed that 86 of the strains gave RFLP patterns with HhaI, MboI and HpaII which were indistiguishable from the prototype fli C-H4 gene and only two strains showed alternative patterns. The detection of some genetic variants in the fliC -H4 gene of E. coli strains studied here points to a sequence diversity similar as described previously for the fliC genes of E. coli H6 and H7 strains [ 15 , 16 ]. These results correspond to previous findings indicating that the H4 antigens in different E. coli typing strains are not fully identical [ 20 , 24 , 28 ]. It was suggested earlier that the strain P12b encodes two flagellins, H4 and H17, which are subject to phase variation for their expression [ 22 ] and mutants of P12b could be isolated which expressed only the H4 antigen [ 24 ]. Recent studies have demonstrated that the fliC gene of strain P12b encodes flagellar type H4 and it was suggested that the gene for the H17 flagellin is encoded by a locus outside fliC , however the gene responsible for the flagellar type H17 was not identified in the study [ 21 ]. In our study, cultures of strain P12b were fully inhibited for motility and swarming in the presence of H4 antiserum but not in the presence of H48 antiserum which was used a non-specific control. This result indicates that H17 type flagella were not expressed or lost from our P12b isolate, similar as previously described with mutant strains of P12b [ 24 ]. The functional expression of the cloned fliC -H4 U9-41 and fliC -H4 P12b genes in the genetic background of E. coli K-12 strain JM109 which shows flagellar serotype H48 confirmed their coding capacity. Since we found coexpression of cloned flagellins with the parental H48 flagellin, we became interested in the composition of flagella made by the fliC recombinant plasmid carrying JM109 strains. Electron microscopy of flagella from JM109 and from the fliC -H4 P12b recombinant plasmid carrying strain revealed no differences between JM109 and TPE1978 showing both a typical helical organization of normal sized flagella on their surface (Fig. 4 ). Immuno-gold staining of bacteria which were prior incubated with H48 and H4 antisera revealed high specificity of the rabbit antisera for the flagellar structure (Fig. 4B–G ). Single (Fig. 4 B+D ) and double labelling (Fig. 4 F ) experiments with different sized gold markers demonstrated that all flagella present on the surface of TPE1978 were labelled after incubation with H48 and H4 antiserum. Our results indicate that both H48 and H4 flagellins are coassembled in the flagella made by the fliC -H4 P12b recombinant plasmid carrying strain. This finding is surprising in view of the molecular weight size differences found between H48 and H4 flagellin. To our knowledge, the assembling of two different flagellins in the same filament has not yet been demonstrated before. The H48 flagellin of E. coli K-12 (accession AE000285) is a 51.3 kDa protein consisting of 498 amino acids and thus much larger than the 36.3 kDa H4 flagellin which is composed of 349 amino acids. However, both flagellins share conserved N- and C-terminal sequences which are known to be involved in the structural assembly of flagella [ 29 ]. H48 and H4 flagellins differ largely for their central regions which are not involved in flagellar assembly and function but which contain flagellar antigenic epitopes [ 29 ]. We were able to show that the introduction of an isolated fliC gene in E. coli can change the antigenic properties of the flagella made by this strain. Horizontal transfer of fliC genes may contribute to the diversity of flagellar serotypes by recombination within E. coli recipient strains. The mammalian host immune system is the driving force for continuous selection of new flagellar antigens in E. coli . Published data indicate that both mutation and recombination events in the fliC gene have taken place in the evolution of E. coli flagellar antigens [ 15 , 16 ]. Conclusions Our fliC sequence data have shown that the flagellar type H4 which is present in E. coli strains of clinical importance covers several genetic variants which are closely related to each other. We have shown for the first time that flagellins of different molecular size are expressed and coassembled into functional flagella in a laboratory E. coli K-12 strain. Methods Bacterial strains The reference strains used for production of antisera for O- and H-typing of E. coli were obtained from the International Escherichia and Klebsiella Centre, Statens Seruminstitut, Copenhagen, Denmark and are described elsewhere [ 3 ]. Origin and serotype data on five additional strains with flagellar type H17 are listed in Table 1 . A laboratory collection of 88 E. coli isolates originating from humans and animals was investigated by PCR/RFLP typing for fliC -H4 genes. These strains were previously investigated for the O-types and for production of Shiga-toxins (Stx) and were isolated in different countries between 1941 to 2002 [ 3 , 19 , 30 ] (Table 3 ). The E. coli K-12 strain JM109 is described elsewhere [ 23 ]. Production of E. coli O and H-specific antisera Rabbit antisera against the different O- and H-antigens of E. coli were prepared according to ∅rskov and ∅rskov [ 3 ]. Antisera for typing of flagellar antigens H4 were produced with reference strains U9-41 (O2:K1:H4) and P12b (O15:H17) [ 3 ]. Our strain P12b was found to express its fliC encoded H4 antigen and produced flagellar type H4 (this work). Motility inhibition test Expression of flagella and swarming of E. coli strains was tested by inoculating bacteria in tubes containing 10 ml portions of swarm-agar (L-broth + 0.3% agar) as described [ 3 ]. Inhibition of motility of E. coli strains in the presence of flagellar-specific antiserum (H4 and H48) was tested in swarm-agar containing a 1:600 dilution of the respective antiserum. Cultures which were inhibited for motility were observed over two weeks for possible switch to motility by phase variation. Serological typing of H-antigens of E. coli H-serotyping was performed as described [ 3 ]. In brief, bacteria were grown in tubes containing 10 ml 0.3% semi-solid LB-agar [ 23 ] for two to three passages. Highly motile bacteria were transferred in LB-medium, incubated 6h at 37°C and inactivated by addition of 0.5% formaldehyde in solution. Agglutination reactions were performed in two fold dilutions of 0.5 ml portions of serum in phosphate-buffered saline pH 7.4 (PBS) [ 22 ] with 0.5 ml formalized bacteria in glass tubes which were incubated for 2h at 50°C. Agglutination tests were read by eye immediately after incubation as described [ 3 ]. PCR-typing of fliC genes The oligonucleotide primers fliC-1 (5' CAA GTC ATT AAT AC(A/C) AAC AGC C 3') and fliC-2 (5' GAC AT(A/G) TT(A/G) GA(G/A/C) ACT TC(G/C) GT 3') were used for amplification of internal parts of fliC genes present in the E. coli reference strains as described [ 9 ]. The PCR was performed for 25 cycles at 94°C for 60 sec, 55°C for 60 sec and 72°C for 120 sec [ 9 ]. PCR products of sizes varying between 950 to 2500 bp were obtained with E. coli reference strains for 53 different H-types [ 3 ] (Figure 1 ). Amplified DNA was digested with HhaI and the resulting restriction fragments were compared on 2% agarose gels. Restriction enzymes HpaII and MboI were used for characterization of fliC-H4 specific PCR products. Gel images were stored digitally and analyzed with BioNumerics software, version 2.5 (Applied Maths, Kortrijk, Belgium) for similarity (Dice, complete linkage) (Fig. 1 ). Nucleotide sequence analysis of fliC genes Two primers deduced from the published fliC -H4 sequence (AB028472) were used for the amplification of the entire fliC -H4 coding regions. The PCR was performed for 30 cycles: 30 sec at 94°C, 60 sec at 58.1°C and 90 sec at 72°C with primers fliC-5 (5'-TGA GTG ACC AGA CGA TAA CAG GG-3') and fliC-6 (5'-GGA CGA TTA GTG GGT GAA ATG AGG-3') and yielded a 1243 bp product. PCR products were purified with the QIAquick™ PCR Purification Kit (Qiagen, Hilden, Germany) and used for sequencing. Sequencing reactions were carried out using the dye terminator chemistry (PE Applied Biosystems, Darmstadt, Germany) and separated on an automated DNA sequencer (ABI PRISM ® 3100 Genetic Analyzer). The sequences were analysed using the Lasergene software (DNASTAR, Madison,WI, USA) and the Mac Vector software (Oxford Molecular Group, Campell, CA, USA) to assemblings and alignings. Nucleotide sequence accession numbers The nucleotide sequence of the genomic region of E. coli strain P12b (O15:H17) with a size of 1234 bp containing the fliC gene for flagellin has been submitted to EMBL data library under accession number AJ515904. The coding sequences of the different fliC genes from the following strains have been assigned the following accession numbers: AJ605764 for strain U1-41 (O5:K4:H4), AJ605765 for strain P7d (O68:H4), AJ605766 for strain C107-74 (O15:H17) and AJ536600 for strain E1541-68 (O154:H4). The origin of the strains is listed in Table 1 . Molecular cloning of fliC gene PCR products PCR products encompassing the complete coding sequences of the fliC -H4 genes were obtained from genomic DNA prepared as described [ 9 ] of E. coli strains U9-41 and P12b using primers fliC-5 and fliC-6. The amplification products were inserted into the vector pLITMUS38 (New England Biolabs, Beverly, MA, USA) digested with EcoRV. The orientation of the the insert PCR products was determined by using commercially available sequencing primers LITMUS forward 28/38 and LITMUS Reverse 28/38 (New England Biolabs). Immuno electron microscopy (IEM) of E. coli flagellar antigens Motile cultures of E. coli strains were produced by repeated passage on semi-solid agar followed by growth in L-Broth as described above. Cultures carrying recombinant pLITMUS38 plasmids were grown in the presence of 100 μg/ml ampicillin. IEM was performed using fresh, non-formalized cultures of motile bacteria. For IEM, aliquots of respective bacterial cultures were diluted 1:2 in PBS pH 7.2 and adsorbed onto glow-discharge treated 400 mesh grids coated with Pioloform and carbon (Wacker Chemie, Munich, Germany) [ 31 ]. Grids with adsorbed bacteria were preincubated for 30 min at room-temperature with blocking buffer (0.1 % bovine serum albumine (Sigma, Deisenhofen, Germany) in PBS). Rabbit anti-H48- and anti-H4 hyperimmune sera were diluted 1:1000 in blocking buffer. After conditioning, specimens were incubated for 30 min at room temperature on droplets of the specific, unlabelled antibodies. Non-bound antibody was removed by washing the grids twice for 10 min on blocking buffer. Immuno-specifically bound primary antibodies were detected using anti-rabbit-IgG-gold 5 or -gold 10 nm conjugates (British Bio Cell International Ltd, Cardiff, UK). The conjugates were diluted 1:20 in blocking buffer and reacted for 30 min at room temperature. Unbound conjugate was removed by a sequence of washing steps (two times with blocking buffer for 5 min each; once with PBS for 3 min and a final wash with double destilled water for 3 min) at room temperature. Before negative staining with 1 % uranyl acetate (pH 4.0–4.5), the grids were washed rapidly on 4 droplets of double destilled water. The preparations were analyzed with an EM 10 electron microscope (Zeiss-LEO, Oberkochen, Germany) at an accelerating voltage of 80 kV. To look for different antigenic determinants expressed on the flagella of strains JM109 or TPE1978, double immuno-labelling was performed using H48 and H4 antisera sequentially and two anti-rabbit-IgG-gold conjugates with different sized markers for the detection of the bound primary unlabelled rabbit antibody. Both E. coli strains were incubated with rabbit anti-H4 (1:1000) followed by anti-rabbit-IgG-5 nm gold and two washing steps on droplets of blocking buffer for 10 min. The samples were subsequently incubated with anti H48 antibody (1:1000) followed by incubation with anti-rabbit-IgG- 10 nm gold. Removal of surplus conjugate and negative staining were performed as detailed above. Author's contribution LB conceived of the study and carried out PCR genotyping and coexpression studies. ES carried out sequence determination and alignments and construction of recombinant plasmids. SZ and SK performed serological assays and PCR genotyping. CS performed analysis of fliC recombinant plasmid carrying strains. AM and HG developed the IEM methodology and HG contributed to the data analysis. All authors participated in review and preparation of the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548302.xml |
522871 | Comprehensive analysis of pseudogenes in prokaryotes: widespread gene decay and failure of putative horizontally transferred genes | A comprehensive analysis of the occurrence of pseudogenes in a diverse selection of 64 prokaryote genomes identified around 7,000 candidate pseudogenes. A large fraction of prokaryote pseudogenes seems to have arisen from failed horizontal-transfer events. | Background Genes that have recently fallen out of use for an organism are often detectable in the genome as pseudogenes - disabled copies of genes characterizable by disruptions of their reading frames due to frameshifts and premature stop codons [ 1 - 3 ]. Surveys of the pseudogene populations of eukaryotes (budding yeast, nematode worm, fruit fly and human) have recently been completed [ 4 - 10 ]. These pseudogene analyses have yielded insights into eukaryotic proteome evolution, showing that duplicated pseudogene formation tends to occur in younger, more lineage-specific, protein families, and is in many cases linked to the generation of functional diversity [ 3 ]. However, pseudogene formation in most prokaryotes has not been analyzed as a matter of course, and has, historically, been assumed to be minimal [ 11 ]. Some recent substantial populations of pseudogenes have been discovered in pathogenic bacteria, most notably in the leprosy bacillus Mycobacterium leprae , where around 1,100 pseudogenes (compared to around 1,600 genes) were found, with pseudogene formation providing a 'fossil record' of recent wholesale loss of pathways involved in lipid metabolism and anaerobic respiration [ 12 ]. Here we want to address the question of whether these large populations are exceptional, or whether there are substantial populations of pseudogenes in other prokaryotic genomes. If so, from a holistic 'polygenomic' perspective, what sorts of proteins tend to form prokaryotic pseudogenes? And are there any themes in common with the occurrence of pseudogenes in eukaryotes? To address these broad questions, we have adapted a pipeline developed for eukaryotic pseudogene identification to 64 prokaryotic genomes [ 4 ]. The species analyzed include archaea, pathogenic bacteria and non-pathogenic bacteria, and many of the pathogenic bacteria are also important organisms in current biodefense research. We have found nearly 7,000 pseudogenes, with notable numbers of pseudogenes for specific families linked to DNA transposition and also that have some role in environmental responses. Our results, which we have derived consistently across all the genomes, are available from our prokaryote pseudogene information website [ 13 ]. Results and discussion Pseudogenes are pervasive in prokaryotes To identify pseudogenes in prokaryotic genomes, we performed a conservative and comprehensive search, as outlined in Figure 1 and Materials and methods. We used a proteome set consisting of sequences from the 64 genomes and Swiss-Prot [ 14 ] with relatively high confidence in annotation (that is, excluding those annotated as hypothetical proteins). Intergenic regions in prokaryotic genomes were searched against the proteome set using FastX [ 15 ] for homology matches with disablements as pseudogene candidates. We then applied several checks to reduce false positives (see Materials and methods). Overall, we found 6,895 candidate pseudogenes. Previously, the pseudogene fraction was defined as the ratio of the number of pseudogenes to the number of all gene-like sequences (genes plus pseudogenes) [ 16 ]. By this measure, we find that pseudogenes are pervasive in prokaryotes (Figure 2 ). Pseudogenes are detectable at a low 'background' level in most prokaryotes, ranging from 1 to 5% of the genome (Figure 2 ). Application of a more restrictive cutoff (E-value less than 0.001, instead of E-value less than 0.01) in FastX alignment results in slightly smaller percentage of pseudogenes (0.1% less on average) in all the genomes, and generates essentially the same results (data not shown). Our census is in general agreement with previous assessments of pseudogene content in the genomes of M. leprae , Escherichia coli and Rickettsia prowazekii [ 12 , 16 - 19 ]. In these previous studies, however, different criteria were used for pseudogene identification in different genomes, leading to inconsistencies in comparing results. This is avoided in our study by using a method applied uniformly across all genomes. All these assessments suggest that most prokaryotes have similar net genomic DNA deletion rates, resulting in similar low-level 'background' pseudogene fractions in their genomes. To check for a correlation with microbial 'lifestyle', we classified the 64 species into three categories: archaea, pathogenic bacteria and non-pathogenic bacteria. The pseudogene fractions for these groupings were assessed. M. leprae has a very large pseudogene fraction (36.5%) and is clearly a unique outlier. When this genome is set aside, the three groups have similar pseudogene fractions (3.6%, 3.9% and 3.3%). Note that three other pathogenic species/strains have relatively large pseudogene fractions, including Neisseria meningitidis MC58 (12.4%), N. meningitidis Z2491 (11.6%) and Rickettsia conorii (9.7%). The higher pseudogene fractions of some pathogenic species have previously been suggested to be a result of a rapidly changing environmental niche, with loss of metabolic and respiratory pathways [ 12 ]. We found that about 2,300 of our 6,895 candidate pseudogenes overlap with more than 2,600 annotated hypothetical open reading frames (ORFs), whose fractions were indicated in Figure 2 . The overlap could arise from erroneous gene annotations or sequencing errors [ 16 ]. In either case, the pseudogene annotation in prokaryotic genomes is evidently an important part of decontaminating gene annotation. Pseudogene families We used the Pfam classification [ 20 ] to analyze the families and functions of candidate pseudogenes. The 20 top-ranking domain families in terms of pseudogenes are shown in Figure 3a . Many large divergent gene families are among the top pseudogene families, including 9 of the top 10 gene families such as: the ABC transporter (PF00005), short-chain dehydrogenases/reductases (PF00106), sugar transporter (major facilitator superfamily) (PF00083), and histidine kinase-like ATPase (PF02518). As the largest family of proteins in prokaryotes, the ABC transporter functions to translocate a variety of compounds across biological membranes [ 21 - 23 ]. It consists of two ATP-binding domains (PF00005) [ 24 , 25 ] and two transmembrane domains (PF00664). These domains are present in large copy numbers across genomes (2,172 and 245 gene copies as well as 67 and 13 pseudogene copies respectively). There are notable protein families that rank high in pseudogene number, but low in terms of gene number. They include the PPE family (PF00823) which is thought to be linked to antigenic variation in mycobacteria and is highly polymorphic [ 26 ]; the cytochromes P450 (PF00067), which are involved in processing diverse substrates; the GGDEF domain (PF00990), which is of unknown function and is associated with a wide diversity of other protein domains [ 27 ]; alpha/beta-hydrolase enzymes (PF00561), which have diverse catalytic functions; and pseudo-U-synthase-2 enzymes (PF00849), which help synthesize pseudouridine from uracil. Note that the first two families in this list have sequence diversity that has some link to environmental response. Figure 3b shows the relationship between the number of pseudogenes and genes for Pfam families. One might expect this relationship to be linear, with bigger families having more pseudogenes, but Figure 3b shows this is not the case. Two large families that have a relatively high ratio of pseudogenes to genes are the transposase DDE domain (PF01609) and integrase core domain (PF00665). Transposase facilitates DNA transposition and horizontal gene transfer and its DDE domain may be responsible for DNA cleavage at a specific site followed by a strand-transfer reaction [ 28 ]. Many transposons contain transposases for their transposition [ 29 , 30 ]. We found that two strains of N. meningitidis (MC58 and Z2491) carry 26 and 22 copies of transposase pseudogenes, respectively, and have only 11 and 5 copies of transposase genes. In the MC58 strain, transposase pseudogenes have been found in most of the 29 remnant insertion sequences [ 31 ]. This suggests that N. meningitidis strains probably undergo high selection pressure for transposases. The integrase core domain family (PF00665) is the catalytic domain of integrase, which mediates integration of a DNA copy of a viral/bacteriophage genome into the host genome [ 32 ]. It catalyzes the DNA strand-transfer reaction by ligating the 3' ends of the viral DNA to the 5' ends of the integration site [ 32 ]. The large number of transposase and integrase pseudogenes might result from harmful foreign genes being disabled in transposable elements. Several species contain many integrase pseudogenes, including Streptococcus pneumoniae, M. leprae, M. tuberculosis , and E. coli strain O157:H7. The large number of pseudogenes relative to genes for these two gene families may reflect an overall high selective pressure for them - that is, a gene family that is rapidly duplicating and evolving may generate many pseudogenes. Origins of pseudogenes Retrotransposition and genomic DNA duplication generate pseudogenes in mammals and other eukaryotes [ 2 , 3 ]. In contrast, in prokaryotes, based on the experience annotating E. coli and M. leprae [ 12 , 16 ], pseudogenes are suggested to arise from three process: the disablement of detectable native duplications; the decay of native single-copy host genes; and failed horizontal transfers. However, the complete extent of the processes forming prokaryotic pseudogenes is not yet well understood. We realize that there are many methods of defining horizontal transfer [ 33 - 36 ] and an active debate on the best way of doing this [ 37 , 38 ], so we applied two independent methods to predict horizontal gene transfer events. The first method (GC-content) is based on the GC content bias at particular codon positions of recently acquired genes [ 33 , 39 ]. The second method (GeneTrace) is based on the analysis of phylogenetic distribution of protein families on species tree [ 40 ]. In the GC-content method, the number of pseudogenes resulting from horizontal transfer in each genome was estimated by applying the same criteria to them as had been previously used to identify horizontally transferred genes. Overall, we found that the ratio (19.9%) of pseudogenes from potential horizontal transfer to those derived from the host is significantly higher than the ratio of genes in the host (8.6%). We dubbed the ratio of these two quantities the 'failed horizontal transfer index', and observed that it implies that pseudogenes are 2.3 times more likely to arise from horizontal transfer than host genes are (Table 1 ). To confirm our findings based on a method relying on GC content bias we applied the GeneTrace method (see Materials and methods). We analyzed a subset of pseudogenes and found that 18% result from failed horizontal transfer events, consistent with the previous method. Note that GeneTrace and the GC-content method are very different in the criteria they use to assess horizontal transfer and thus make for good independent verification of each other. In summary, we report here for the first time an estimate of how often horizontal transfer in prokaryotes introduces genes that are redundant, useless or even detrimental. Firstly, ORFs from dangerous genetic elements are under strong selection pressure to be deleted from the host's genome [ 11 ]. Secondly, horizontally transferred genes have a higher chance than non-transferred genes of becoming pseudogenes in most prokaryotes, which may be a result of deactivation/disablement of non-beneficial transferred genes. By examining closely related strains of the same species, we found that most close strains have a similar value for the failed horizontal transfer index. In particular, M. tuberculosis (strains H37Rv and CDC1551), N. meningitidis (strains Z1491 and MC8), and Helicobacter pylori (strains 26695 and J99) share similar index values within species. However, E. coli has different index values in the three strains studied. The free-living E. coli K12 strain has an index value of 4.6, comparable to values calculated from previous results [ 16 ], while the two pathogenic E. coli strains O157:H7 and O157:H7 EDL933 have much lower values (1.8 and 0.8). This can be readily explained in two ways: the intracellular pathogenic E. coli strains could have moved into a different environment that results in lower exposure to incoming DNA and thus to a lower rate of horizontal gene transfer [ 41 ]; or these strains could have an increased rate of gene loss or pseudogene formation of their host genes. A polygenomic power-law-like trend in pseudogene disablement To characterize the overall rate of decay of pseudogene populations, we plotted the fraction of disablements versus the average number of matching residues (to their closest homologs) per pseudogene for each species. This measure shows how the overall level of decay of a pseudogene population relates to age (which corresponds to the degree of overall match to the closest homologs). There is a general power-law-like behavior governing this measure, with recent pseudogenes having few disablements and divergent pseudogenes having many (Figure 4 ). Archaea and most non-pathogenic bacteria cluster together at higher rates of disablement (between 10 and 28 per 1,000 residues) and less significant matches, indicating comparatively greater retention of ancient gene remnants in those species and fewer young pseudogenes. On the other hand, obligate pathogenic bacteria tend to have younger pools of pseudogenes, even though they exhibit high disablement rates. Interestingly, four species of obligate bacterial pathogens clearly stand out from the general tendency: these are M. leprae and three closely related mycoplasma species: Mycoplasma pneumoniae , Mycoplasma pulmonis and Ureaplasma urealyticum . Pseudogenes in these four pathogenic bacteria carry several times more disablements, suggesting that these bacteria have an accelerated disabling mutation rate. It is known that M. leprae has lost the dnaQ -mediated proofreading activities of DNA polymerase III [ 12 , 42 ], which could contribute to a higher mutation rate. The higher mutation rates in these species might suggest that these pathogens are under adaptation to their new environment, or have specific genome regions that are hypermutable. It is important to note here that the current sequence databases are derived from an uneven sampling of genomes. Therefore, genomes of organisms with more sequenced relatives may appear to have, on average, a seemingly younger population of pseudogenes, while others may appear to have older and fewer identifiable pseudogenes. Using data from 64 genomes, our results indicate an overall trend for pseudogenes observed in most of the genomes studied. However, these results have to be viewed as preliminary until more genome data is available. Conclusions We have shown that pseudogenes in prokaryotes are not uncommon, occupying 1-5% of all gene-like sequences. We find that specific gene families with clear links to DNA transposition and environmental responses have higher pseudogene/gene ratios. The pseudogene data has many implications for the study of genome reduction and expansion [ 43 , 44 ]. A significant proportion of the pseudogenes arose from putative failed horizontal transfer - at more than two times the rate for genes. Obligate pathogenic bacteria have high rates of disablement in younger pseudogene populations, consistent with recent accelerated genome reduction [ 44 ], while, in contrast, archaea and non-pathogenic bacteria have relatively older pseudogene populations, but similar rates of disablement. In terms of methodological implications, it is evidently necessary to include prokaryote pseudogenes as part of systematic annotation pipelines in the future. In addition, it was also shown to be helpful to identify potential short ORFs [ 45 ]. Furthermore, our survey shows that trends can be observed 'polygenomically' for prokaryotes, where they are not obvious or significant in individual genomes. Materials and methods Database releases used We used the following datasets in our prokaryotic pseudogene analysis: Swiss-Prot (release 40.19 and updated to 27 May, 2002) [ 14 ] containing 43,094 prokaryotic protein sequences; nucleotide sequences from 64 prokaryotic genomes from EMBL database release 70 on March-2002 [ 46 ], including 11 genomes from archaea and 53 from bacteria as listed in Figure 1 ; Pfam release 7.3 of May 2002, containing 3,849 families and 498,152 protein domains in the alignments [ 20 ]. Pseudogene identification pipeline Figure 1a shows the basic procedure for identifying prokaryotic pseudogenes. The general schema was adapted from pipelines for pseudogene analysis in eukaryotes [ 4 ]. We generated a prokaryotic proteome set by collecting all the prokaryotic protein sequences in the Swiss-Prot database and those annotated in the 64 prokaryotic genomes. To be conservative, we did not include hypothetical or putative proteins, a large proportion of which might be overannotated [ 47 , 48 ]. All the protein sequences were masked by SEG using the default low-complexity filter parameters (122.22.5) [ 49 ]. To maximize the efficiency of the pseudogene search, we only considered the intergenic DNA regions in the 64 prokaryote genomes (including the regions encoding hypothetical proteins) as query sequences, and searched their forward and reverse complement sequences against the proteome set using FastX [ 15 ]. Significant homology matches (E-value less than 0.01) that contained more than one disablement (either a frameshift caused by insertion or deletion of nucleotides or a premature stop codon) were considered as potential pseudogenes. If an intergenic region had multiple matches, these matches were sorted by E-value (increasing) and then by the number of matching residues (decreasing), if they have the same E-value. The match with the most significant E-value and the maximum matching residues was selected and redundant matches were removed. To ensure that spurious disablements were not introduced at ends of sequences as an alignment artifact, we excluded homology matches whose disablements occurred only within a 'cutoff region' at either end. We used 16 residues for the cutoff region for short sequences (160 amino acids or fewer) - a parameter that has been applied previously [ 6 ]. For longer sequences (more than 160 amino acids), 10% of the sequence length was applied as the cutoff region as FastX tends to include more residues at the ends of alignments. We also assessed the potential pseudogenes by examining the distribution of the disablements within pseudogene sequences. Given that mutations within pseudogenes are unconstrained, we would expect disablements on pseudogenes to be evenly distributed. Figure 1b shows the position of disablements within pseudogene fragments whose length is normalized to 100 residues. By removing those potential pseudogenes that only had disablements at their flanking regions at both ends, the distribution is almost evenly distributed. We used it as a 'control filter' to minimize false-positive pseudogenes. In the final pseudogene set, the length of pseudogenes ranges from 33 to 4,969 amino acids, with a median length of 130 amino acids, as compared with the proteome set, where the length ranges from 7 to 10,920 amino acids with a median length of 291 amino acids. We considered non-standard codon usage in some bacteria, such as when TGA encodes tryptophan rather than a stop codon in mycoplasma species, including Mycoplasma pneumoniae , M. pulmonis and U. urealyticum . By manual examination of E. coli genes with translational frameshifts in the RECODE database [ 50 ], we found that those genes were included in coding sequences (CDS) and therefore were excluded from our pseudogene search. Sequencing errors could also be a potential problem in the detection of pseudogenes. However, this effect is expected to be small, as comparison of independently sequenced isolates of the same E. coli strains indicated that only about 7% of candidate pseudogenes could be due to sequencing error [ 16 ]. To further consider the possibility of sequencing error, we examined the stop codons in the pseudogenes detected in the S. pneumoniae genome (frameshift positions are not considered as they are difficult to locate.). This genome and eight others found in the trace archive of the National Center for Biotechnology Information (NCBI) [ 51 ] and Ensembl [ 52 ] were all sequenced by TIGR. We selected S. pneumoniae as a case study as it is a relatively big genome available in the archive. By adapting a previous method [ 53 ], we examined the overall quality values (Q) for each nucleic acid of stop codons in the pseudogenes. Pseudogene sequences were aligned to the archived sequences (≥ 95% identity), and the quality values for nucleotides in stop codons were summed up. We chose 10 -2 as a cutoff of the error rate (err = 10 SUM(-0.1Q) ) for all nucleic acids. The stop codons with all three nucleic acids above the cutoff were validated. Out of 116 pseudogenes in this genome, 73 were found to contain 150 stop codons in total. Using the available data in the trace archive, we identified 54 pseudogenes with stop codons being aligned with the original sequences, and validated 47 of these (87%). In addition, a similar fraction of stop codons (101 out of 116) was confirmed. Family classification of genes and pseudogenes All genes in the 64 genomes were assigned to Pfam families by cross-referencing of their Swiss-Prot ID. Pseudogenes were assigned to Pfam families through ID of their closest homologs. Only the homologs that cover more than 70% of the Pfam domain were selected. A pseudogene could be assigned to multiple Pfam families if it contains multiple domains. Estimation of horizontally transferred genes and pseudogenes Here we used a method (GC-content) to estimate horizontal transferred genes on the basis of their base compositions [ 33 , 39 ]. We analyzed each of the 64 genomes individually, and atypical genes and pseudogenes were identified if the GC content at first and third codon positions was two or more standard deviations higher or lower than the mean values at those positions in genes. To ensure that we had the codon positions accurately assigned for the GC-content method, we only analyzed codons for pseudogenes that aligned well with annotated protein sequences, specifically excluding the regions of the alignment around frameshifts. While it is true that the local alignment in some regions of a pseudogene may be ambiguous, causing some difference in the GC-content calculation in that region, the impact on the overall GC-content estimation is minimal, given how many positions we average over to calculate the failed transfer index score. The results for the 64 genomes are shown in Table 1 . The failed transferred index in the last column represents the ratio of the fraction of putative horizontally transferred pseudogenes to the fraction of horizontally transferred genes , similar to the measure previously used in E. coli [ 16 ]. This essentially gives a likelihood ratio for horizontal transfer for pseudogenes relative to that of genes. Note that to minimize the effect of more divergent sequence alignments, for the horizontal-transfer calculations we only analyzed 1,748 'recent' pseudogenes, which have more than 50% sequence identity to their closest matches over an aligned subsequence of more than 100 residues. We have investigated the statistical robustness of the failed transfer index using resampling approaches [ 54 ]. For each of the 64 genomes, we randomly picked 90% of its genes and calculated their GC content. Using the new GC content, we then identified the putative horizontally transferred genes and pseudogenes and calculated the failed transfer index. We applied the process 1,000 times, generating a distribution of 1,000 indexes, which has a mean value of 2.32 with standard deviation of 0.01. We also applied an alternative method (GeneTrace) to estimate horizontally transferred pseudogenes [ 40 ]. In this method, potential horizontal transfer events are inferred within a protein family when it is present only in distantly related species and is absent from members of the same phylogenetic clade. We analyzed a subset of pseudogenes - 225 pseudogenes across 62 genomes - whose closest Swiss-Prot homologs share more than 70% sequence identity across at least 100 amino acids, and identified 41 of them (18%) as from failed horizontal transfer events. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC522871.xml |
509410 | Regulation of Muscle Fiber Type and Running Endurance by PPARδ | Endurance exercise training can promote an adaptive muscle fiber transformation and an increase of mitochondrial biogenesis by triggering scripted changes in gene expression. However, no transcription factor has yet been identified that can direct this process. We describe the engineering of a mouse capable of continuous running of up to twice the distance of a wild-type littermate. This was achieved by targeted expression of an activated form of peroxisome proliferator-activated receptor δ (PPARδ) in skeletal muscle, which induces a switch to form increased numbers of type I muscle fibers. Treatment of wild-type mice with PPARδ agonist elicits a similar type I fiber gene expression profile in muscle. Moreover, these genetically generated fibers confer resistance to obesity with improved metabolic profiles, even in the absence of exercise. These results demonstrate that complex physiologic properties such as fatigue, endurance, and running capacity can be molecularly analyzed and manipulated. | Introduction Skeletal muscle fibers are generally classified as type I (oxidative/slow) or type II (glycolytic/fast) fibers. They display marked differences in respect to contraction, metabolism, and susceptibility to fatigue. Type I fibers are mitochondria-rich and mainly use oxidative metabolism for energy production, which provides a stable and long-lasting supply of ATP, and thus are fatigue-resistant. Type II fibers comprise three subtypes, IIa, IIx, and IIb. Type IIb fibers have the lowest levels of mitochondrial content and oxidative enzymes, rely on glycolytic metabolism as a major energy source, and are susceptible to fatigue, while the oxidative and contraction functions of type IIa and IIx lie between type I and IIb ( Booth and Thomason 1991 ; Berchtold et al. 2000 ; Olson and Williams 2000 ). Adult skeletal muscle shows plasticity and can undergo conversion between different fiber types in response to exercise training or modulation of motoneuron activity ( Booth and Thomason 1991 , Jarvis et al. 1996 ; Pette 1998 ; Olson and Williams 2000 ; Hood 2001 ). This conversion of muscle fiber from type IIb to type IIa and type I is likely to be mediated by a calcium signaling pathway that involves calcineurin, calmodulin-dependent kinase, and the transcriptional cofactor Peroxisome proliferator-activated receptor-gamma coactivator 1α (PGC-1α) ( Naya et al. 2000 ; Olson and Williams 2000 ; Lin et al. 2002 ; Wu et al. 2002 ). However, the targeted transcriptional factors directly responsible for reprogramming the fiber-specific contractile and metabolic genes remain to be identified. Muscle fiber specification appears to be associated with obesity and diabetes. For instance, rodents that gain the most weight on high-fat diets possess fewer type I fibers ( Abou et al. 1992 ). In obese patients, skeletal muscle has been observed to have reduced oxidative capacity, increased glycolytic capacity, and a decreased percentage of type I fibers ( Hickey et al. 1995 ; Tanner et al. 2002 ). Similar observations have been made in type 2 diabetic patients ( Lillioja et al. 1987 ; Hickey et al. 1995 ). Recently, it has been shown that increasing oxidative fibers can lead to improved insulin action and reduced adipocyte size ( Luquet et al. 2003 ; Ryder et al. 2003 ). We have previously established that peroxisome proliferator-activated receptor (PPAR) δ is a major transcriptional regulator of fat burning in adipose tissue through activation of enzymes associated with long-chain fatty-acid β-oxidation ( Wang et al. 2003 ). Although PPARδ is the predominant PPAR isoform present in skeletal muscle, its in vivo function has not been determined. Our current study uncovers PPARδ as the first transcription factor able to drive the formation of functional type I muscle fibers, whose activation entrains complex pathways both enhancing physical performance and creating a state of obesity resistance. Results Activation of PPARδ Leads to Muscle Fiber Transformation A role of PPARδ in muscle fiber was suggested by its enhanced expression—at levels 10-fold and 50-fold greater than PPARα and γ isoforms, respectively (unpublished data). An examination of PPARδ in different muscle fibers reveals a significantly higher level in type I muscle (soleus) relative to type II–rich muscle (extensor digitorum longus) or type I and type II mixed muscle (gastrocnemius) ( Figure 1 A); this expression pattern closely resembles that of PGC-1α ( Lin et al. 2002 ). A similar pattern but with more pronounced differences was found at the protein level ( Figure 1 B). Figure 1 Expression of Endogenous PPARδ and VP16-PPARδ Transgene in Muscle (A) Pooled RNA isolated from various muscles of five wild-type male C57B6 mice was hybridized with indicated probes. EDL, extensor digitorum longus; Gastro, gastrocnemius. (B) Pooled nuclear proteins (15 μg/lane) isolated from muscles of five wild-type male C57B6 were probed with anti-PPARδ antibody. RNA polymerase II (Pol II) is shown as a loading control. (C) Expression of the VP16-PPARδ transgene in various tissues. 10 μg of total RNA from each tissue was hybridized with a VP16 cDNA probe. Gastrocnemius muscle was used here. (D) Nuclear proteins (15 μg/lane) isolated from gastrocnemius muscle of the transgenic mice (TG) and the wild-type littermates (WT) were probed with indicated antibodies. The upper, nonspecific band that cross-reacted with the anti-PPARδ antibody serves a loading control. To directly assess the role of activation of PPARδ in control of muscle fiber plasticity and mitochondrial biogenesis, we generated mice expressing a transgene in which the 78-amino-acid VP16 activation domain was fused to the N-terminus of full-length PPARδ, under control of the 2.2-kb human α-skeletal actin promoter. In agreement with the previous characterization of this promoter ( Brennan and Hardeman 1993 ; Clapham et al. 2000 ), the VP16-PPARδ transgene was selectively expressed in skeletal muscle, with 10-fold less in the heart ( Figure 1 C). Among different types of muscle fibers, the levels of VP16-PPARδ expression appeared to be similar (unpublished data). As shown in Figure 1 D for gastrocnemius muscle, VP16-PPARδ fusion protein was produced at a level similar to that of endogenous PPARδ in wild-type littermates. Interestingly, the level of endogenous muscle PPARδ protein in the transgenic mice was much higher than in the control littermates. The substantial increase of endogenous PPARδ may have been caused by a switch to type I fiber (see below), which intrinsically expresses higher levels of PPARδ ( Figure 1 A and 1 B). Type I muscle can be readily distinguished from type II or mixed muscle by its red color, because of its high concentration of myoglobin, a protein typically expressed in oxidative muscle fibers. We found that muscles in the transgenic mice appeared redder ( Figure 2 A), which is particularly evident in the mixed type I/II fibers of the hindlimb ( Figure 2 B). Indeed, metachromatic staining revealed a substantial muscle fiber transformation ( Figure 2 C). In gastrocnemius muscle, we estimated that there was a 2-fold increase of type I fibers. A diagnostic component of oxidative fibers is their high myoglobin and mitochondrial content, which is supported by the mRNA analysis shown in Figure 3 A. In addition to myoglobin, mitochondrial components for electron transfer (cytochrome c and cytochrome c oxidase [COX] II and IV) and fatty-acid β-oxidation enzymes were elevated ( Figure 3 A; unpublished data). These effects appear to be direct consequences of PPARδ activation, as levels of PGC-1α, a coactivator involved in muscle fiber switch and mitochondrial biogenesis ( Wu et al. 1999 ; Lehman et al. 2000 ; Lin et al. 2002 ), remained unchanged. Southern blot analysis detected a substantially higher copy number of the mitochondrial genome–encoded COXII DNA in the transgenic mice ( Figure 3 B). Mitochondrial DNA was increased 2.3-fold in gastrocnemius muscle of the transgenic mice ( Figure 3 C). These results reveal a marked stimulation of mitochondrial biogenesis and further support the idea that there is a muscle fiber switch. This conclusion was also confirmed by Western blot analysis. As shown in Figure 3 D, the characteristic type I fiber proteins, such as myoglobin and cytochrome c and b, were significantly increased. More importantly, the specialized contraction protein troponin I (slow) of type I fiber was robustly induced; this was accompanied by a marked reduction of the specialized contraction protein troponin I (fast) of type II fiber, indicating a high degree of fiber transformation. We next examined whether acute activation of endogenous PPARδ would induce similar target genes. In agreement with the chronic effects in the transgenic mice, we found that, after treatment of wild-type C57B6J mice with the PPARδ-specific agonist GW501516 for only 10 d, genes for slow fiber contractile proteins, mitochondrial biogenesis, and β-oxidation were all upregulated ( Figure 3 E). This indicates that rapid, systematic, and coordinated changes of muscle fiber properties toward type I can be achieved by activation of the endogenous PPARδ pathway. Figure 2 Increased Oxidative Type I Fibers in the Transgenic Mice (A and B) Muscles in transgenic mice (TG) are redder than those in wild-type mice (WT). (C) Metachromatic staining of the type II plantaris muscle. Type I fibers are stained dark blue. Figure 3 Activation of PPARδ Induces Genes Typical for Type I Fibers and Promotes Mitochondrial Biogenesis (A) Total RNA (10 μg/lane) prepared from gastrocnemius muscle of transgenic (TG) and wild-type (WT) littermates was probed with indicated probes. The fold increase of induction of each gene is shown. (B) Total genomic DNA (10 μg/lane) prepared from gastrocnemius muscle was digested with Nco1 and subjected to Southern analysis with COXII (mitochondrial genome–encoded) and MCIP1 (nuclear genome–encoded) DNA probes. (C) Equal amounts of gastrocnemius muscle were collected from both transgenic mice and control littermates. Total mitochondrial DNA was isolated and separated on 1% agarose gel. The relative abundance of mitochondrial DNA in transgenic and wild-type mice is presented. (D) Western blot analysis of muscle fiber markers and mitochondrial components. Each lane was loaded with 80 μg of total gastrocnemius muscle extracts. (E) Wild-type C57B6 mice were treated with vehicle or PPARδ agonist GW501516 for 10 d. Total RNA (10 μg/lane) prepared from the gastrocnemius muscle was probed with indicated probes. Muscle Fiber Switch by PPARδ Protects Against Obesity A number of previous studies have shown that obese individuals have fewer oxidative fibers, implying that the presence of oxidative fibers alone may play a part in obesity resistance. To test this possibility, we fed the transgenic mice and their wild-type littermates with a high-fat diet for 97 d. Although the initial body weights of the two groups were very similar, the transgenic mice had gained less than 50% at day 47, and only one-third at day 97, of the weight gained by the wild-type animals ( Figure 4 A). The transgenic mice displayed significantly higher oxygen consumption on the high-fat diet than the control littermates (unpublished data). By the end of this experiment, the control littermates became obese, whereas the transgenic mice still maintained a normal body weight and fat mass composition ( Figure 4 A). A histological analysis of inguinal fat pad revealed a much smaller cell size in the transgenic mice ( Figure 4 B), due to the increased muscle oxidative capacity. While there was no significant difference in intramuscular glycogen content, the triglyceride content was much less in the transgenic mice ( Figure 4 C and 4 D), which may explain their improved glucose tolerance ( Figure 4 E). We also placed wild-type C57BJ6 mice on the high-fat diet and treated them with either vehicle or the PPARδ agonist GW501516 for 2 mo. GW501516 produced a sustained induction of genes for type I muscle fibers; this, at least in part, resulted in an only 30% gain in body weight, a dramatically reduced fat mass accumulation, and improved glucose tolerance, compared to the vehicle-treated group ( Figure 5 ). Thus, muscle fiber conversion by stimulation with the PPARδ agonist or the activated transgene has a protective role against obesity. Figure 4 Resistance to High-Fat-Induced Obesity in the Transgenic Mice (A) Seven-week-old transgenic (TG) and wild-type (WT) littermates ( n = 5–6 for each group) were fed with a high-fat diet for 97 d. Left panel shows net body weight gain, which was calculated for individual mice and then averaged. Right panel shows the body weights before (Day 0) and after (Day 97) high-fat feeding. (B) Histology of inguinal fat pad in the transgenic and wild-type littermates under a high-fat diet for 2 mo. (C and D) Intramuscular glycogen content (C) and triglyceride content (D) of mice in (A) after high-fat feeding ( n = 6). (E) Glucose tolerance test. Mice in (A) after high-fat feeding were fasted for 6 h and then injected with glucose at a concentration of 1g/kg body weight. Then blood glucose levels were measured periodically over 2 h ( n = 6). Figure 5 PPARδ Agonists Counteract Obesity Induced by High-Fat Diet (A) Eleven-week-old wild-type C57B6 mice were fed a high-fat diet in combination with vehicle or GW501516 for 57 d. Total RNA (10 μg/lane) prepared from the gastrocnemius muscle was probed with indicated probes. (B) Net body weight gain for mice in (A) after treatment was calculated for individual mice and averaged. Initial body weights were 28.54 ± 1.04 g for vehicle group ( n = 5) and 28.86 ± 0.80 g for GW501516 group ( n = 5). (C) Various tissue weights for mice in (A) after treatment. ifat, inguinal fat; rdfat, reproductive fat; retrofat, retroperitoneal fat. (D) Glucose tolerance test. Mice in (A) after treatment were fasted for 6 h and then injected with glucose at a concentration of 1g/kg body weight. Blood glucose levels were then measured periodically over 2 h. Activation of PPARδ Enhances Physical Performance Muscle oxidative capacity is a crucial factor for determining endurance and fatigue. Indeed, type I fibers adaptively generated through exercise training are considered to be fatigue resistant. However, whether the type I fibers generated molecularly via PPARδ expression can contribute to enhanced performance in the absence of previous training is unclear. In fact, the consequence of genetically induced fiber switch on running capacity has to our knowledge never been evaluated. We thus compared exercise performance between untrained, body-weight-matched transgenic and wild-type littermates. Mice were run on oxygen-infused, enclosed treadmills until exhaustion. Strikingly, the running time and distance the transgenic mice were able to sustain were increased by 67% and 92%, respectively ( Figure 6 A; also see Videos S1 and S2 ). The transgenic mice ran about 1 h longer than the controls, which translates to nearly a kilometer further. No significant differences in muscle mass (unpublished data) and daily activity (total counts of activity per hour: 1618 ± 209 for transgenic versus 1987 ± 301 for wild-type, p > 0.35, n = 4) were observed between the transgenic and control mice. Thus, the remarkable increase in endurance is the physiologic manifestation of muscle fiber transformation. This suggests that genetically directed muscle fiber switch is physiologically and functionally relevant. In addition, we looked at what effect the absence of PPARδ function has on exercise endurance. In the treadmill test, the PPARδ-null mice could sustain only 38% of the running time and 34% of the distance of their age- and weight-matched wild-type counterparts ( Figure 6 B). These results further support a role for PPARδ in enhancement of physical performance. Figure 6 PPARδ Regulates Exercise Endurance (A) Enhanced exercise performance in the transgenic mice. Fourteen-week-old male transgenic and wild-type littermates with similar body weights ( n = 4 for each group) were subjected to a forced treadmill exercise test. (B) Compromised exercise performance in PPARδ-null mice. Two-month-old male PPARδ-null mice and wild-type controls with similar body weights ( n = 6 for each group) were subjected to a forced treadmill exercise test. (C) Functions of PPARδ in skeletal muscle. Discussion Our data reveal that a PPARδ-mediated transcriptional pathway can regulate muscle fiber specification, enabling the generation of a strain of mice with a “long-distance running” phenotype. We show that targeted expression of an activated form of PPARδ produces profound and coordinated increases in oxidation enzymes, mitochondrial biogenesis, and production of specialized type I fiber contractile proteins—the three hallmarks for muscle fiber type switching ( Figure 6 C). While induction of muscle oxidation enzymes by PPARδ has been seen both in vivo and in vitro ( Muoio et al. 2002 ; Dressel et al. 2003 ; Luquet et al. 2003 ; Tanaka et al. 2003 ; Wang et al. 2003 ), its effects shown here on muscle fiber switching are unexpected. These progressive changes in oxidative capacity in conjunction with eventual changes in type I muscle fiber lead to a dramatically improved exercise profile and protection against obesity. This does not solely depend on achieving a directed muscle fiber type switch but also requires all the associated changes in neural innervation, motor neuron function, and peripheral metabolic adaptation to enable a new integrated physiological response. Accordingly, activation of muscle PPARδ essentially recapitulates the effects of exercise training even in the absence of training itself. To our knowledge, this has not been directly described for any other transcriptional factor. The muscle phenotypes described here are remarkably similar to those of transgenic mice expressing either calcineurin, calmodulin-dependent kinase, or PGC-1α ( Naya et al. 2000 ; Lin et al. 2002 ; Wu et al. 2002 ), indicating that PPARδ could be one of the hypothetical downstream transcription factors of these pathways. It is important to note that, from our ligand and gain-of-function transgenic studies, PPARδ needs to be activated in order to direct the muscle fiber switch. Indeed, in a recent report by Luquet et al. (2003) , simple overexpression of wild-type PPARδ in muscle was found not to be sufficient to promote a fiber switch or obesity resistance, although certain oxidation enzymes were increased. This supports the model in Figure 6 C that the activating signal or ligand, but not the receptor, is limiting. Thus, PPARδ activation, rather than merely an increase of PPARδ levels, is an essential element for fiber switching and its associated functional manifestations. How might endogenous PPARδ become activated naturally by exercise training? First, it is possible that exercise generates or increases endogenous ligands for PPARδ as tissues are undergoing substantial increases in fatty-acid internalization and burning. Fatty acids and their metabolites can activate PPARδ. A second model is that exercise may induce expression of PGC-1α ( Goto et al. 2000 ) and thereby activate PPARδ. This is consistent with previous work in which we have shown that PGC-1α physically associates with PPARδ in muscle tissue and can powerfully activate it even in the absence of ligands ( Wang et al. 2003 ). Alternatively, PPARδ may be activated by a distal upstream signaling component such as a kinase cascade. Further dissecting the interactions between PPARδ and its regulatory components will be necessary to fully understand the molecular basis of muscle fiber determination pertinent to exercise training. Skeletal muscle is a major site to regulate whole-body fatty-acid and glucose metabolism. We show that mice with increased oxidative fibers are resistant to high-fat-induced obesity and glucose intolerance. Moreover, ligand studies provide compelling evidence that activation of endogenous PPARδ can produce similar effects. Might PPARδ have any beneficial effects on glucose metabolism in the lean condition? This has not been explored; however, insulin resistance in the elderly is confined mostly to skeletal muscle and may be due to reduction of mitochondrial number and/or function ( Petersen et al. 2003 ). The ability of PPARδ to stimulate mitochondrial biogenesis and oxidative function suggests that PPARδ could be important for control of insulin resistance during normal aging. Together, these data indicate that PPARδ and its ligands comprise a key molecular switch to regulate muscle fiber specification, obesity resistance, insulin sensitivity, and, most surprisingly, physical endurance. This work demonstrates that complex physiologic properties such as fatigue, endurance, and running capacity can be genetically manipulated. Materials and Methods Animals. The transactivation domain (78 amino acid residues, corresponding to residues 413–490) of VP16 was fused in frame with the N-terminus of mouse PPARδ. The VP16-PPARδ fusion cDNA was placed downstream of the human α-skeletal actin promoter ( Brennan and Hardeman 1993 ), and upstream of the SV40 intron/poly(A) sequence. The transgene was purified and injected into C57BL/6J × CBA F1 zygotes. Transgenic mice were backcrossed with C57BL/6J for two generations. Wild-type littermates were used as controls throughout the study. On normal chow diet, the transgenic mice and control littermates used here had similar body weights. PPARδ-null mice were previously generated ( Barak et al. 2002 ). Mice were fed either a standard chow with 4% (w/w) fat content (Harlan Teklad, Harlan, Indianapolis, Indiana, United States) or a high-fat diet containing 35% (w/w) fat content (product F3282, Bioserv, Frenchtown, New Jersey, United States) as indicated. For ligand experiments, we synthesized the GW501516 compound and mice were orally gavaged daily (10 mg/kg or vehicle alone). Gene expression analysis and physiological studies Mouse EST clones were obtained from ATCC (Manassas, Virginia, United States), verified by sequencing, and used as Northern probes. Antibodies were obtained from Santa Cruz Biotechnology (Santa Cruz, California, United States). Total muscle protein extracts ( Lin et al. 2002 ) and nuclear proteins ( Wang et al. 2003 ) were prepared as described. Prior to the exercise performance test, the mice were accustomed to the treadmill (Columbus Instruments, Columbus, Ohio, United States) with a 5-min run at 7 m/min once per day for 2 d. The exercise test regimen was 10 m/min for the first 60 min, followed by 1 m/min increment increases at 15-min intervals. Exhaustion was defined when mice were unable to avoid repetitive electrical shocks. Muscle fiber typing and mitochondrial DNA isolation Muscle fiber typing was essentially performed using metachromatic dye–ATPase methods as described ( Ogilvie and Feeback 1990 ). Muscle mitochondria were isolated ( Scholte et al. 1997 ). Mitochondrial DNA was prepared and analyzed on 1% agarose gel. Statistical analysis Number of mice for each group used in experiments is indicated in figure legends. Values are presented as mean ± SEM. A two-tailed Student's t test was used to calculate p- values. Supporting Information Video S1 Beginning of Running Test This video shows the exercise performance of a representative of the transgenic mice (right chamber) and a representative of wild-type control littermates (left chamber) on the treadmill 15 min into the exercise challenge. (52.4 MB MOV). Click here for additional data file. Video S2 Running Test 90 Min Later This video shows the exercise performance of a representative of the transgenic mice (right chamber) and a representative of wild-type control littermates (left chamber) on the treadmill 90 min into the exercise challenge. (41.7 MB MOV). Click here for additional data file. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC509410.xml |
526381 | Extending the mutual information measure to rank inferred literature relationships | Background Within the peer-reviewed literature, associations between two things are not always recognized until commonalities between them become apparent. These commonalities can provide justification for the inference of a new relationship where none was previously known, and are the basis of most observation-based hypothesis formation. It has been shown that the crux of the problem is not finding inferable associations, which are extraordinarily abundant given the scale-free networks that arise from literature-based associations, but determining which ones are informative. The Mutual Information Measure (MIM) is a well-established method to measure how informative an association is, but is limited to direct (i.e. observable) associations. Results Herein, we attempt to extend the calculation of mutual information to indirect (i.e. inferable) associations by using the MIM of shared associations. Objects of general research interest (e.g. genes, diseases, phenotypes, drugs, ontology categories) found within MEDLINE are used to create a network of associations for evaluation. Conclusions Mutual information calculations can be effectively extended into implied relationships and a significance cutoff estimated from analysis of random word networks. Of the models tested, the shared minimum MIM (MMIM) model is found to correlate best with the observed strength and frequency of known associations. Using three test cases, the MMIM method tends to rank more specific relationships higher than counting the number of shared relationships within a network. | Background Most scientific fields are data-intensive, but perhaps even more so for biology and medicine. Sequencing efforts have generated billions of base pairs of genetic information across hundreds of thousands of species, and ushered in the relatively recent completion of the Human Genome Project[ 1 ]. Microarrays enable thousands of transcriptional measurements per experiment [ 2 ], and high-throughput chemistry enables the simultaneous screening of thousands of molecules at a time for activity[ 3 ]. New discoveries among research areas (e.g. genetics, medicine, chemistry) lead to a necessarily increasing amount of specialization as more objects (e.g. genes, diseases, phenotypes, chemical compounds, etc.) are discovered to be of research interest. This is reflected by the growth in the number of scholarly journals published every year as well as the number of total records indexed in biomedical literature reference databases such as MEDLINE[ 4 ]. In any field, the gain in our cumulative scientific knowledge has the unfortunate effect of narrowing our perspectives as individuals – providing us with far too much information to assimilate, and far too many variables to analyze. Yet the most valuable type of information is often what is not known or apparent to others – information implied by a set of data, facts or associations. History is replete with examples of insights into scientific problems coming from a series of observations from apparently unrelated fields, discoveries or events. But how could one retrieve or compile such information in cases where one is not certain what to look for and the search space is vast? This is the primary reason that methods of data-mining and knowledge discovery are becoming increasingly important in handling this explosion of information. Previous research Most scientific knowledge comes from peer-reviewed articles and is written in free-form text, which is difficult to analyze algorithmically. However, the idea that novel relationships within text could be computationally identified based upon existing relationships has its roots in an approach developed by a researcher named Don Swanson, who used software to identify words shared between article titles [ 5 ]. Using their software, called Arrowsmith, Swanson and Smalheiser were able to identify common intermediates between Raynaud's Disease (a circulatory disorder restricting blood-flow to the extremities) and the dietary effects of fish oil, leading to the hypothesis and subsequent proof [ 6 ] that compounds within dietary fish oil could alleviate the symptoms of Raynaud's Disease [ 5 , 7 ]. To explain why such a sensible hypothesis had gone unnoticed by researchers in either field alone, the term "non-interactive literatures" was coined. This term, in essence, implies that increasing specialization among all fields results in a relative lack of awareness of the findings in other, less related fields. These entities that do not have known or documented associations, yet share intermediate relationships, have been referred to as " transitive ", " implicit ", " indirect " or " inferable " relationships. Deciding that no relationship exists when no co-mentions exist is somewhat of an over-simplification, but a necessary one. Realistically, several co-mentions between terms could be observed without a definitive relationship present. However, if one uses a greater-than-zero cutoff to define when a relationship exists, false-negatives become a problem: Some co-mentions below the cutoff will constitute a real relationship. Using zero co-mentions as a cutoff is a convenience to avoid this problem even if the end result is that some relationships are declared "known" when they really are not. While pioneering, a keyword-based method such as Swanson and Smalheiser's is both limiting and highly burdensome, especially where a large body of literature is concerned, because the number of unique keywords grows quickly per record analyzed. Neither is the method amenable to open-ended querying – that is, telling a user what is implicitly related to a query term. Rather, one must essentially begin by postulating a relationship between a query term, A, and another term, C, where a set of intermediate terms, B, can be found that connect the two. Even improvements in visualizing or exploring records that share commonalities and/or define entities of interest [ 8 , 9 ] are limited because they require manual user navigation and analysis of results. Other approaches have attempted to utilize Medical Subheadings (MeSH)[ 10 ] or the Unified Medical Language System (UMLS) [ 11 ] to engage in open-ended discovery by pairing concepts, counting the number of relationships shared by two terms as a means of judging its implicit significance. However, these approaches do not take into account the fact that the more general the nature of the relationship is, the more connections are likely to be shared by two terms. It was previously demonstrated that, because the number of associations between terms follows a scale-free, or inverse power-law, distribution, the number of inferable associations with any given term rapidly approaches the maximum number of possible associations as the number of direct associations grows[ 12 ]. That is, even if one starts with a term that is only associated with several others, at least one of these is likely to be associated with a very large number of terms. Thus, the starting term will be implicitly associated with most of the network (the "small world" phenomenon). Therefore, the issue is not identifying implicit associations, but somehow judging which of the many implicit associations are worth further examination. Previous work demonstrated that it was feasible to identify pertinent implicit relationships by ranking inferred relationships and preferentially examining those at the top of the list[ 12 ]. One of its shortcomings, though, was that associations between terms are assigned based upon co-occurrence of terms within an abstract. This is a fairly well accepted means of assigning tentative relationships between terms, but when considering the scale-free distribution of objects within the literature, it is apparent that some frequently mentioned objects could be co-mentioned many times with other terms without any actual biological association being implied. Figure 1 uses an analysis of terms related to the term "capsaicin" to illustrate this point. Although the MIM may have drawbacks in identifying broad relationships (for example, see Table 1 – some very pertinent relationships receive modest MIM scores if the terms are common), it is a very straightforward and well-established means of measuring information content between two terms. Such a measure would enable us to pursue more specific relationships – those with high information content. MIM, however, can only be calculated using direct (A-B) or (B-C) relationships rather than implicit (A-C). Thus, the goal here is to test methods of extending the MIM calculation to include implied relationships such that a statement can be made about the implied mutual information content of two unrelated terms. Identifying literature-based associations The general approach to associating objects by searching for their co-occurrence within text has been used in many fields as a simple, yet comprehensive way to identify potential associations. In biology and medicine, co-occurrence has been used to identify potential relationships between genes [ 13 , 14 ], proteins [ 15 ] and drugs [ 16 ]. The disadvantage of this approach is that associations are very general – that is, no specifics on how two objects are related or associated are obtained by this method. False-positives can also be a problem, as terms far apart within the abstract with no apparent association may be included as "relationships". The advantages are that it is easy to implement and comprehensive. To begin a search for novel, inferable associations within the literature, relevant "objects" of interest in scientific research were first defined by assimilating database entries from relevant databases into one central database. By doing this, both words and phrases can be identified within text, and it permits synonymous terms to be mapped to primary terms. All electronically available literature was then analyzed for associations between objects of interest by searching for their co-occurrence within MEDLINE records (titles & abstracts), summing the total number found. The significance of this collective set of co-occurrences is evaluated using the mutual information measure (MIM), which was originally based upon Shannon's Entropy theory [ 17 ], but has also been successful in identifying lexical dependencies [ 18 ]. By processing a body of literature that comprehensively covers a topic, field or area, it can be asserted that the current state of knowledge has been approximated, at least on the level of broad object-object associations. All available literature was processed, creating a network of associations for each object. This network can in turn be analyzed for associations shared by two unassociated objects. That is, we can use the network to identify objects that share associations but are not themselves associated. Such objects are said to be implicitly associated with each other, and new associations can be potentially inferred by evaluation of their shared associations. Since there are many implicitly associated objects, the relevance of each one is also evaluated using the MIM. However, a MIM can be calculated to evaluate the relevance of an association between A and B and between B and C, but it is not clear how each of these individual scores extends to the inference of an association between A and C. Therefore, we explore and evaluate different methods. Methods and algorithms Code was written in Visual Basic 6.0 (SP5) using ODBC extensions to interface with an SQL-based database, with database queries written in SQL. Programs were executed on a Pentium 4 3.06 GHz machine with 1 GB of RAM and two ultra-fast SCSI hard drives. The National Library of Medicine graciously provided an electronic archive of MEDLINE records in XML format. To obtain a set of common words for analysis, the Merriam-Webster dictionary was parsed into individual words and each word summed by the number of times it was observed within the dictionary. 10,000 words were chosen with dictionary frequencies ranging from 322 to 28. This range was selected so that no extremely common or rare words would be within the list. To create a database of random word associations, only 100,000 titles/abstracts were used. This was done to avoid network saturation (i.e. having a significant number of objects related to every other object) and to ensure that the distribution in the number of associations between words resembled the same power-law distribution observed for biomedical objects. The occurrence of such objects within scientific text is identified by comparing phrases within MEDLINE records to entries in the object recognition database (ORD). This ORD is built by inputting terms found in several different biomedical databases, all freely available for download. Objects classified as diseases, disorders, syndromes or phenotypes were obtained from Online Mendelian Inheritance in Man (OMIM) [ 19 ]; chemical compounds and small molecules were obtained from the Medical Subject Headings (MeSH) database [ 20 ]; approved drug names from the Food and Drug Administration; genes were obtained from Locuslink [ 21 ], and ontological classifications for genes were obtained from the Gene Ontology consortium [ 22 ]. Assimilation of terms is done automatically, but a table within the ORD contains additional biomedical terms to be added or deleted as deemed necessary (e.g., some databases contain vague or uninformative terms such as "survey" or "extended", useless information such as "deleted entry" or errors such as "#NAME?"). Compared to the overall size of the ORD, this table is small (1,007 entries versus over 223,000 terms assimilated) and designed primarily to reduce clutter. Acronyms for entries, if not explicitly stated within the assimilated database, were obtained from an acronym database[ 23 ]. Similarly, spelling variants were also obtained from this database where possible. This database can be accessed online[ 24 ]. As an example of spelling variants detected, the user can go to this URL, enter the acronym "ICAM-1" and note the many subtle variations. The acronym resolving heuristic used to construct this database was also used to resolve acronyms within text when they occurred. The Mutual Information Measure A scoring scheme based upon the Mutual Information Measure (MIM) [ 17 ] is used to estimate strength of association between co-occurring terms within the literature. It should be noted that other statistical methods of association such as chi-square tests, log-likelihood ratios, z-scores or t-scores could be used as well – these are all means of judging the statistical significance of a relationship. In this paper, however, we will focus on the MIM only as a proof of principle that mutual information calculations can be extended into implicit relationships as well. The MIM has been widely used to quantify dependencies between variables, including co-occurring terms in text [ 25 ], and is shown in equation (1): P AB is the measured probability that A and B will be observed together in the same abstract, while P A and P B are the probabilities of observing A or B, respectively, in a given abstract. Furthermore, because scientific research and discovery is a time-dependant process, prior information can be incorporated to refine the probabilities in Equation (1). The describing of a disease or discovery of a gene, for example, will occur at a given point in time (illustrated in Figure 2 ) within the history of publications. Regardless of an object's overall frequency in the database, the probability it will appear in the literature prior to its discovery is zero. Thus P A and P B are calculated from their time of first appearance. P AB is then calculated using the later of these two dates. P AB, P A and P B are thus calculated as: Where T A and T B are the total number of records A and B are independently mentioned in, respectively, and T AB is the total number of records co-mentioning A and B. A f and B f represent how many records were read in before the first occurrences of A and B were observed, respectively. Max(A f , B f ) represents the larger of the two values between A f and B f . And A t is the total number of records processed. As an example of how the MIM score is used, assume that the probability A will appear in any given record within a database of records is 10% and the probability of B appearing is the same. If the appearance of A is completely independent of the appearance of B then no information about one can be gained by observing the appearance of the other. The probability both A and B will be observed in the same record is thus 0.1*0.1 = 0.01. The value of MIM in Equation 1 then evaluates to 1 and the log value to zero – the information gained on one object by observing the other. If the probability of observing A increases when B is mentioned, then MIM > 0. If A and B are rarely mentioned together, then MIM < 0. When considering scientific writing style with reference to biomedical objects such as genes, diseases and chemical compounds, there is a probability that two of them might be mentioned together in the same record without having an established association. For example, one of the objects may be very commonly used in many studies (e.g. the gene LacZ is used for staining assays, luciferase is used for luminescence, etc), or one of the objects may be of great scientific/medical interest and authors may make an extra effort to speculate how their results might relate to such objects (e.g. cancer, diabetes, heart disease, apoptosis). The MIM provides a way of quantifying literature-based object dependencies. However, taking the log value can provide a negative weighting to an association when two frequent terms are mentioned together. Optimally, irrelevant or uninformative associations (i.e. those with little mutual information) would be ignored entirely rather than penalized. Therefore, the log function is removed and the equation becomes: The possibility remains that rare associations might receive a very high MIM score [ 26 ], but it is hoped that the fact that many MIM scores are being summed and compared will ameliorate this effect when it occurs. Inferring new associations based upon commonalities Figure 3 shows the general conceptual approach undertaken here. Call the primary research object node "A" in a network constructed of MIM scores between objects. For each A there is a set of other objects, or nodes, associated with it by virtue of co-occurrence in the literature. We'll call this set "B" and assuming a total of t objects in this set, each individual object can be given the symbol "B n ", where 0 < n < t . For each B n , there is another set of objects related to it by literature co-occurrence, called "C". Each object in the set C may or may not be connected to the primary object, A. That is, an association may consist of A↔B↔C where an object in the set C also belongs to the set B. The symbol "↔" is used here to represent the existence of a non-directional association between two objects. Objects in the set C having no literature-based association with the primary object, A, represent associations that have not been previously made, or at least documented, by others. These represent new associations that can potentially be inferred by virtue of their shared associations. Because the number of implicit associations rapidly increases with each established association, the goal here is to provide a quantitative measure of the strength of an implicit association based solely upon the associations shared by two objects. After all, if no known relationship is documented, then these shared associations will be the only way to understand the nature of an relationship between A and C. Since directly associated objects also share associations with other objects, it is reasoned that the strength of known associations can be used to benchmark how well the scores from implicit associations correlate with the relative importance of an association. However, it is not clear how A-C relationships are best evaluated given a set of component A-B n and B n -C associations. Two models are thus proposed and evaluated, the numeric score obtained by any one of them will only be relevant in terms of how well it assigns a relative importance to each A-C connection within a list. Scoring inferred associations The first model to be tested assumes that the total information content of an implied A-C association can be approximated by the mutual information measure of each component connection. Thus, the MIM scores for each A-B n and B n -C MIM association is averaged over a total of t shared connections and then finally divided by t to normalize the total score by the total number of connections. The function for the normalized averaged MIM (AMIM) model is: As model by which A-B and B-C values were summed was also considered, but it would be functionally indistinguishable from the AMIM model in terms of ranking implicit relationships, so it was not included. The second model views the process of inferring an A-C connection as function of each of its component processes, limited in its potential by the mutual information in each step of the inference process. That is, inferring an A-C connection depends upon how much information is in the A-B n association as well as the B n -C connection, and the information potential an A-C connection will be no greater than the least mutual information given by A-B n or B n -C. This is equivalent to assuming that a chain can be no stronger than its weakest link. The equation for the normalized minimum MIM (MMIM) model is: Results A total of 12,899,016 MEDLINE records recorded from 1967 to May 2003 were processed in chronological order to create a network of 10,873,926 associations between a total of 112,805 unique objects assimilated from the databases mentioned. When including synonyms, the total number of recognizable phrases for these unique objects was 223,540 (e.g. "IL-6" is a synonym for "Interleukin-6", and the two are treated equivalently). The distribution of objects found in MEDLINE ranges from more general categories (e.g. "blood", "tumor", "stress", "lesions") that are found in a higher percentage of records ("blood" was the most abundant, being found in 17.5% of all records analyzed) to the more specific. The frequency of objects when plotted follows a power-law distribution and resembles that of a scale-free network, which is reasonable given that new objects are typically studied in terms of their relationship to known objects (law of preferential attachment). Records were chosen for analysis due to their electronic availability and are also because they are a good source of pertinent information due to their brief, focused nature that presumably contains a summary of the most important findings in each report. Several objects were examined to see if associated objects with high MIM scores correlated with the relative importance of the association. This was done by obtaining summary descriptions of an object from various authoritative sources such as review articles, glossaries or biomedical databases. Table 1 shows an example of associations to an object that were found by scanning all MEDLINE records. Note here that objects with higher MIM scores tend to be objects found in fewer MEDLINE records. Initially this was thought to be problematic because objects highly germane to the biological activity of another object could be down-weighted solely because of their relative abundance. However it was found that when analyzing sets of shared associations in both AMIM and NMIM models, these abundant objects that initially receive low MIM scores subsequently receive much higher scores because they share many high-information content associations with the primary object of analysis, and their cumulative score rises with each one. Table 1 can be said to reflect the current state of knowledge, as obtainable from scientific abstracts and with reference to biomedically relevant associations to capsaicin. From what is known, a list of what can be inferred is constructed. Each of these secondary associations is used to identify and score implicit relationships as illustrated in Figure 3 . As mentioned earlier, a subset of the objects in (C) identified by their associations to the secondary objects (B), will be other secondary objects themselves. That is, they will also be in the set B. Model evaluation When ranking inferred associations, the goal is for the score assigned to an inference to correlate well with the amount of mutual information gained from any given association. The only reference basis for this is the mutual information contained within established associations. Since they too will frequently contain shared associations, they can be evaluated independently using only their shared associations (Figure 4 ). Several different methods of ranking inferred associations were evaluated together using the object capsaicin for comparison. Because time must be spent analyzing shared associations to determine the nature of an inferred association, one inference ranking method could be considered superior to another if it yielded a high ratio of relevant to irrelevant associations during the analysis phase. For analysis purposes, the "relevance" of an association will be equivalent to its MIM score – the higher the MIM score, the more relevant the association. Thus, the higher that known , relevant associations are ranked within the set of all inferable associations, the better the ranking method is. To evaluate this, a graph is drawn to reflect the rate that established relationships are discovered within the set of all objects analyzed. The total of all MIM scores for known relationships is added together, in order from highest MIM score to lowest, to reflect the fastest rate by which they could be discovered. When plotted, this curve is what would be observed were mutual information preserved exactly (the "exact" curve). Because it's neither expected that all possible relationships are known, nor that mutual information is static as the scientific discovery progresses, it is not anticipated that this curve would or even should be followed exactly (if it were, then that would imply future discoveries could not be more informative than what is already known). However, it is reasonable to expect established relationships with high mutual information content to retain a relatively high mutual information content when evaluated on the basis of its shared relationships. Thus, it is expected that the implicit MIM curve follow the "exact" MIM curve. Figure 5 illustrates what percent of all established associations are identified by each scoring method. The mutual information of associations shared by two objects is ranked by several methods, including the Minimum MIM (MMIM – equation 6) and Averaged MIM (AMIM – equation 7). Objects are ranked here by their shared associations and included in this set are associations that have already been established within MEDLINE as well as those that are implicitly associated. When an established association is encountered within this ranked list, its MIM is added as a percentage of the sum of all MIM scores. When all established associations have been ranked by each method, the total will add to 100%. The faster an inference ranking method approaches 100%, the better it scores objects with high mutual information. Shown for comparison is what the curve would look like if each established association were ranked in the exact order of its highest to lowest MIM scores ("Exact"). Also shown is how quickly established associations would be found by guessing at random ("Random"), and how quickly established associations would be found when counting the number of intermediates ("Count of B"). To gain a better quantitative estimate of performance, 50 objects were chosen at random from both the MEDLINE and random word databases. Each object was analyzed to identify and rank other objects that shared relationships with it as described and the area under the curve (AUC) was taken for each of the ranking methods shown in Figure 5 . For the MEDLINE network, the average AUC for the MMIM was 43% ± 9%, for the AMIM it was 42% ± 8%, and using the count of shared relationships was 9% ± 7%. The difference between the MMIM and AMIM was not large (p < 0.29 using a 2-tailed paired t-test) but was slightly biased by a relatively few examples where AMIM performed very well. Out of the 50 trials, MMIM performed better 35 times, equally 11 times and worse 4 times. What is most pertinent is that both MIM methods ranked objects with high mutual information content significantly higher (p < .000001) than counting the number of shared relationships. A peculiar effect was noted with the average MIM-based scoring model: Some implicitly associated objects received higher MIM scores than the primary object itself, which is also analyzed as a control. There tend to be relatively few, sometimes none, such instances per analysis, but it occurs when a relatively rare object shares several or more associations with the primary object. This effect was not present in the minimum MIM model. Using a random word network to estimate significance intervals The scores assigned by inference methods so far have no meaning by themselves, but only as a means of ranking the potential relevance of an inference. Because the majority of database objects will be present in the list of implicit connections, the question naturally arises as to where a significance cutoff can be drawn. A range of significance for a given MIM score can be estimated by analysis of a random word network in which we would expect that meaningful relationships are only encountered by chance. Since the MMIM model performed slightly better than the AMIM model, we evaluated it using the random words database. Words in the random network were effectively chosen at random from the Merriam-Webster dictionary (see Methods and algorithms), and so relevant associations between these words co-occurring within MEDLINE records should occur predominantly by chance. An uninformative association (i.e. chance alone could explain the number of term co-occurrences) would have an average MIM score of 1 (e.g. see Equation 5). Thus, summing a set of t random associations and dividing by t would also be expected to have an average (normalized) MIM score of 1. This is true for any set of A-B associations as well as a corresponding set of B-C associations, thus an average value of 1 should still be obtained when calculating the average minimum MIM score. Figure 6 shows a plot of the average minimum MIM score (with standard deviation) by the number of shared associations of words in a random network. The average minimum MIM score trends towards a value below 1 (average value from 500 to 1000 shared connections = 0.7), which is not surprising given the nature of writing: It is not random, so two words would not necessarily co-occur together with a probability that is proportional to their relative frequencies. This also suggests that a log value of zero for a MIM score may not be the most effective dividing line between informative and non-informative associations. The values obtained from this analysis provide us with a way of estimating a significance cutoff for implied mutual information analysis. As Figure 6 also shows, the fewer shared associations between two objects, the higher the average normalized MMIM score is as well as its standard deviation. Evaluating capsaicin Using the example of capsaicin brought up earlier, we analyzed it using the methods described to identify and rank objects sharing relationships with it (Table 2 ). When ranked by counting the number of shared relationships, the more general relationships (as mentioned in Table 1 ) tend to rank towards the top, such as calcium and neurons. This seems a good means of identifying general relationships, but each of the objects on this list is hardly specific to capsaicin. Ranking by MMIM, however, changes the nature of the types of objects that are ranked highly to those that share molecular/physiological relationships with capsaicin by their effects upon nerves and transmission of impulses. For example, the ileum is frequently used to test capsaicin effects because of its contractile response. Tachykinin s Substance P and Neurokinin A as well as the neurotransmitter acetylcholine [ 27 ] are responsible for afferent nerve transmission in response to capsaicin, the response to which can be blocked by antagonists such as tetrodotoxin [ 28 ] or atropine [ 29 ]. These implicit objects share relationships with B objects of all different types mentioned in the methods & algorithms section, but the ones that tend to score highest are the ones that share several highly informative relationships with the A object. In general, these informative relationships tend to be objects that are mentioned much more frequently with the A object than any other object within the literature. Acetylcholine, for example, is associated with many neurological processes, but has relatively high MIM scores with other objects related to capsaicin such as bradykinin, atropine, neuropeptide Y and substance P, which are all molecules that affect the transduction of sensory signals. Re-evaluating Swanson's original discoveries Finally, we also sought to re-evaluate some of Swanson's original hypotheses as has been done in other text-mining studies [ 10 , 11 , 30 ]. It makes less sense, however, to attempt to judge performance based upon whether or not Swanson's studies or hypotheses could be replicated, per se . To do so presumes that Swanson's initial study was the "correct" way of finding relevant implicit relationships and Swanson did not employ the open-discovery model in these examples anyway. It would be useful to know, however, where Swanson's predictions rank among others using models in which implicit relevance is judged by counting the number of shared relationships and where it is evaluated by the MMIM. Both Raynaud's and Migraine headaches were analyzed as starting objects (A), the goal being to find all C objects that share relationships and rank them by their relevance. Both known and implicit relationships were displayed. The top 10 results are summarized in Table 3 , and the entire dataset is available by request. When ranking implicit relationships by the number of shared relationships, fish oil scored #1025 in the Raynaud's list and magnesium (the link Swanson found with migraines [ 31 ]) scored #166 in the Migraine list. When ranked by MIM, fish oil scored #1512 and magnesium was ranked #458, lower in both cases. The scores for Raynaud's Syndrome<->Fish oil were lower than expected. Upon examination, Swanson's discovery of this link, although validated experimentally [ 6 ], has apparently not generated a lot of continued experimental research interest in this area in the 15 years since then. A search via Ovid on "(raynaud or raynauds or raynaud's) and (eicosapentaenoic or docosahexaenoic or fish oil)" yielded only 5 papers, three of which were text mining papers including Swanson's original study [ 5 , 30 , 32 ], the fourth was the 1989 validation study [ 6 ] and the fifth was a study showing that fish oil did not have a significant effect upon Raynaud's phenomenon in mixed cryoglobulinemia (a syndrome in which Raynaud's is one of many symptoms)[ 33 ]. Examining the relationships that tend to rank highly in both models it is apparent that, when ranking by the number of shared relationships, the higher-scoring entries tend to be more general and vague in nature (e.g., links to "blood", "development", "females" and "males"). When ranked by the MMIM, their relevance to the object in question is more readily apparent. For example, sumatriptan is a drug used to treat migraines and other items ranking highly on the list such as nausea , vomiting , and dizziness typically accompany migraines. Notably, one of the important links that Swanson used to surmise the role of magnesium is also on this list: Seizures , which cause migraines. Discussion Information retrieval (IR) methods are limited to querying what is known; yet often the most valuable information is what is not directly known. Mutual information measures have been used successfully in many IR applications, and a method has been presented here to extend it to inferable associations. We find that the normalized MMIM method of ranking inferences based upon their shared associations correlates best the level of currently established mutual information. A good correlation is suggestive that mutual information is being captured even though evaluation proceeds indirectly, through intermediates. For simplicity, we have used a cutoff of zero co-occurrences to suggest that no association between objects has been made, but it is quite possible that a number of co-occurrences could be noted between two objects yet no specific relationship between them documented. Or additionally, a certain relationship may be known between the two, but other important relationships still remain to be inferred. At this point, however, it is not clear how this would effectively and quantitatively be taken into account. The method reported was applied to biomedical research, but could ostensibly be applied to any domain in which the goal is to identify undiscovered relationships. Importantly, this method of automated inference ranking provides a quantitative way of prioritizing inferred associations when available literature is growing rapidly in size and scope. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526381.xml |
539276 | Optimized LOWESS normalization parameter selection for DNA microarray data | Background Microarray data normalization is an important step for obtaining data that are reliable and usable for subsequent analysis. One of the most commonly utilized normalization techniques is the locally weighted scatterplot smoothing (LOWESS) algorithm. However, a much overlooked concern with the LOWESS normalization strategy deals with choosing the appropriate parameters. Parameters are usually chosen arbitrarily, which may reduce the efficiency of the normalization and result in non-optimally normalized data. Thus, there is a need to explore LOWESS parameter selection in greater detail. Results and discussion In this work, we discuss how to choose parameters for the LOWESS method. Moreover, we present an optimization approach for obtaining the fraction of data points utilized in the local regression and analyze results for local print-tip normalization. The optimization procedure determines the bandwidth parameter for the local regression by minimizing a cost function that represents the mean-squared difference between the LOWESS estimates and the normalization reference level. We demonstrate the utility of the systematic parameter selection using two publicly available data sets. The first data set consists of three self versus self hybridizations, which allow for a quantitative study of the optimization method. The second data set contains a collection of DNA microarray data from a breast cancer study utilizing four breast cancer cell lines. Our results show that different parameter choices for the bandwidth window yield dramatically different calibration results in both studies. Conclusions Results derived from the self versus self experiment indicate that the proposed optimization approach is a plausible solution for estimating the LOWESS parameters, while results from the breast cancer experiment show that the optimization procedure is readily applicable to real-life microarray data normalization. In summary, the systematic approach to obtain critical parameters in the LOWESS technique is likely to produce data that optimally meets assumptions made in the data preprocessing step and thereby makes studies utilizing the LOWESS method unambiguous and easier to repeat. | Background DNA microarray technology has become a standard tool in biomedical research for large-scale transcriptional monitoring [ 1 ]. A growing number of microarray experiments seek to compare samples labeled with two different dyes, such as Cyanine5 (Cy5) and Cyanine3 (Cy3). However, several studies report that the dyes bind on a microarray slide differently due to the variations in their chemical characteristics [ 2 - 6 ]. In addition, the image scanner settings also affect dye intensity measurements. Should these discrepancies not be corrected, the resulting data may not be useful for analysis purposes. Thus, there is a need for dye normalization for the microarray slide prior to actual data analysis to reduce systematic variability. Microarray data preprocessing contains three phases: quality control, within-slide normalization, and between-slide normalization. Within-slide normalization aims to correct dye incorporation differences which affects all the genes similarly, or genes with the same intensity similarly [ 7 ]. One scatterplot-based normalization technique that is particularly suitable for balancing the intensities is called locally weighted scatterplot smoothing (LOWESS) and its original application was for smoothing scatterplots in a weighted, least-squares fashion [ 8 ]. This technique is typically chosen to calibrate microarray data because a popular, freely available implementation is available in the statistical software package R [ 9 ] and in many commercial microarray analysis software suites such as the Agilent Feature Extraction Software. Moreover, several other freely available microarray data handling packages have incorporated this normalization technique [ 10 , 11 ]. It is noted that many normalization studies simply call the function without rigorous consideration for the actual algorithmic parameters [ 12 , 13 ]. Our analysis reports that the choices of different parameter values drastically affect the quality of the normalization results. The original work on LOWESS clearly mentions the problem of obtaining parameter values and even offers some ideas for finding suitable data-dependent choices [ 8 , 14 ]. However, many microarray studies have omitted such rationale and made arbitrary selections for different experimental data sets [ 13 , 15 , 16 ] and some studies even failed to report their parameter assumptions in their methods [ 17 - 19 ]. Although this practice has not lead to significant consequences for most of the parameters in LOWESS, we show that the parameter that represents the fraction f of neighboring samples to be included in the weighted polynomial fit is particularly sensitive and its variation greatly affects the normalization results. This parameter should be carefully chosen through a systematic procedure where experimental assumptions are clearly specified. Benefits in the normalization process may be considered to be small in their own right, but these improvements are extremely meaningful in the context of searching for subtle biological differences in gene expression. We outline an optimization-based procedure for obtaining a systematic value for f in print-tip LOWESS normalization. Results are compared to common, arbitrary selections of f . The proposed procedure first examines a case study where we have utilized three quality filtered, self versus self hybridization experiments. With self versus self experiments, we are able to clearly detect normalization differences. Such analysis also verifies that the optimized method produces properly calibrated ratios. Our proposed technique is also demonstrated on a typical set of quality filtered microarray data. We utilize a set of breast cancer data that has replicated measurements for four different tumor cell lines [ 20 ]. In addition to visual comparisons, we quantitatively assess the performance of the different normalization procedures using a goodness-of-fit test. Our results demonstrate that arbitrarily selecting the LOWESS bandwidth parameter produces statistically different results for certain print-tips compared to the proposed optimized parameter selection formulation. Moreover, for genes that have been verified using reverse transcription-polymerase chain reaction (RT-PCR) experiments, we show that calibrated results are substantially affected by the choice of f . Our self versus self data, including the original TIFF images, are available online [ 21 ] and the replicated breast cancer data is posted by the original authors of that study [ 22 ]. Results and discussion Within-slide normalization Within-slide normalization is used to correct the dye intensity errors introduced across one microarray slide. The result of this step provides the normalized, calibrated ratios. Let denote the background corrected selection for the intensity of the j th gene of the Cy3 (green) colored sample. Similarly, let denote the j th gene of the Cy5 (red) colored sample. One key issue for the dyes is that they are consistently imbalanced [ 12 , 13 ]. Different labelling effciency between the two fluorescent dyes exists and in some labelling schemes Cy5 is systematically less intense than Cy3. Normalization techniques are needed in order to render the gene expression levels measured by the two different dyes comparable [ 23 , 24 ]. Dye biases can stem from a wide variety of factors, including physical properties of the dyes, effciency of dye incorporation, and processing errors. Such errors may be introduced by slight variations in the amount of mRNA used to create the target hybridized to each microarray or in the quantity of dye used to fluorescently label each target. For a single microarray experiment, there are n total gene expression ratios and we denote the observed vector of ratios for a single experiment as r ∈ ℝ n × 1 . The calibrated ratio of expression for each gene is obtained by dividing the test by the reference sample intensities with the proper normalization factor in the denominator, for i = 1, 2..., n , where n is the total number of spots on a microarray. The normalization factor, denoted by Φ(·), is a function of data-dependent variables. If the dyes are linearly dependent, it can be assumed that the normalization function is a constant, namely Φ(·) = φ . Many studies have looked at linear dependencies [ 25 ], as well as a generalized form of the normalization factor Φ(·) that is a function of an often times unknown number of experiment-specific parameters. Many studies perform within-slide normalization in a global manner by assuming the error effects are stationary across an entire slide. This is currently true for the cases of Affymetrix GeneChip or Agilent oligonucleotide microarrays. For cDNA microarrays, however, the sources of variation typically originate in a localized or spatial manner [ 13 ], mainly from the different print tips for each sub-array of the slide [ 26 ]. The process of determining the values for Φ(·) is highly dependent on the characteristics of the data for each print-tip [ 12 ]. For example, some print-tips have highly nonlinear effects, while other print-tips in the same experiment behave quite differently and may exhibit linear trends in dye bias. Furthermore, the systematic manner in which the experiment has been conducted also influences the results of different slides, but it is our intention that such effects will be satisfactorily captured in the behavior of the print-tip statistics. As a consequence, we omit global calibration considerations that neglect print-tip distinction and focus solely on scatterplot-based normalization in a termed localized manner. LOWESS method One of the most widely used nonlinear correction techniques is the LOWESS method, which was first applied to microarray data by Yang et al . [ 16 ]. The main idea behind LOWESS is to utilize a locally weighted polynomial regression of the intensity scatterplot in order to obtain the calibration factor. Compared to other techniques, like housekeeping-based normalization or dye-swap experiments, scatterplot-based normalization is more robust in many types of scenarios where assumptions of constantly expressed genes may break down [ 23 ]. Subsequent microarray studies have also chosen this method due to the robustness of fit in the presence of a few extreme outliers. Original studies have examined the ( I g , I r )-scatterplot in log 2 -space for determining the value of Φ(·). It has been suggested in separate works by Dudoit et al . [ 15 ] and Yang et al . [ 16 ] that a log 2 -based scatterplot of the average fluorescence intensity A versus the transformed ratio M should be used instead of a simple, log 2 -based intensity scatterplot. This type of scatterplot is commonly known as a Bland-Altman plot in the statistics literature. The values for A and M are given as, for i = 1, 2,..., n . Equations (2) and (3) are preferred over the original intensities because the ( A , M )-scatterplot may reveal artifacts that are not clearly visible in the ordinary intensity scatterplot. Such a transformation represents a scaled, 45° rotation of the ( I g , I r )-coordinate system [ 16 ]. The smoothing procedure has been designed to accommodate measured scatterplot data obeying the form M j = g ( A j ) + ε j , where the j th transformed ratio M j is a function of the corresponding overall intensity A j and a zero mean random variable ε j . The smoothed point at A j using LOWESS with a degree d polynomial is ( A j , ), where is the fitted value of the regression. The LOWESS estimate, , is a weighted linear combination of the M i where the h i ( A j ) depend on A i , ∀ i , but not on the M i . The LOWESS algorithm contains four data-specific parameters, namely the polynomial order d , the number of LOWESS algorithmic iterations t , the weight function w (·), and the fraction of the data points used in the local regression f . Consequently, these parameters all affect the values of the weights h i ( A j ) in Eq. (4). For a complete outline of the LOWESS algorithm, consult [ 8 , 14 , 27 ]. In practice, the polynomial order for DNA microarray data is usually selected as being either d = 0, 1, or 2, depending on the choice of ( I g , I r )- or ( A , M )-coordinate systems, the tri-cube weight function is quite standardized for all types of data [ 8 ], and the number of iterations is usually fixed at t = 3. The final parameter must be chosen where f ∈ (0, 1] and it is often times assigned an arbitrary value without any justification. However, since the choice of f ultimately determines the magnitude of calibration, it is essential to put heavy emphasis on choosing this parameter carefully. In the literature, many microarray studies neglect such concerns and arbitrarily select f for different experimental data sets [ 12 , 13 , 16 ]. Formal consideration of the parameter f is typically glossed over by simply stating that the larger the f value, the smoother the fit. Although this is a true statement, the consequences are deeper than the statement leads on. Different types of data may require smoother fits but DNA microarray data takes all shapes. Also, what defines a smoother fit is also highly subject to interpretation depending on the actual data. The optimized approach For a microarray experiment, there are a total of ℓ print-tips used on a single slide. In order to reliably determine the value of f for each print-tip group, we introduce an optimization approach based on the actual microarray data for each print-tip group. We slightly modify our notation to include print-tip indices as a subscript k for each transformed ratio. The goal is to select the appropriate values of f k that minimizes the mean squared difference between the LOWESS fit of the i th transformed ratio in the k th print-tip group, , and the corresponding normalization reference level, ψ k , i (·). The value of each ψ k , i (·) is a function of experiment-specific parameters such as temperature or other environment settings which may differ from sample to sample in a single experiment. Accordingly, the cost function to be minimized for the k th print-tip group across all transformed ratios is with the constraint that f k ∈ (0, 1]. Here, the value n k is the total number of ratios for the k th print-tip group. Correspondingly, for a total of ℓ print-tip groups, we have . For certain experiments, like self versus self hybridizations, the true expression value is known a priori . If ψ k , i (·) is unknown, reliable estimates that reflect experiment-specific assumptions may be used. Usually there are tens of thousands of genes in a microarray study and a plausible assumption is that the mean of the log 2 -transformed ratios after normalization is zero. Also, in a variety of experiments, platform-dependent control transcripts that are known to have certain expression at a constant level may be utilized in the optimized approach. Furthermore, in our breast cancer case study we show how to obtain statistically reliable estimates of ψ k , i (·) from replicate slides. We also show how our approach may be used if replicates are not available for typical microarray studies. Ultimately, the optimized approach requires experimenters to explicitly state their assumptions behind the study, which is systematically better than arbitrarily choosing parameter values. In addition, determining an experiment-specific f k by trial and error may be time consuming and will oftentimes lead to non-optimal results. The chosen optimization algorithm for minimizing the corresponding cost function is based on a combination of golden-section search and successive parabolic interpolation as outlined by Forsythe et al . [ 28 ]. This approach finds the best f k for minimizing δ k ( f k ) for each print-tip, k = 1,..., ℓ within a tolerance of ±0.01. Each print-tip, resultingly, may have a different, optimal bandwidth parameter. Normalization step After the estimates have been obtained, calibrating the intensities for all the A k , i is given as for i = 1,..., n k , and k = 1,..., ℓ. For the local LOWESS normalization within each print-tip group, the issue of how the total intensities are spread about the sample mean for the group becomes a factor to consider when normalizing the data [ 16 ]. After normalization, all the log 2 -ratios from the different print-tip groups are usually centered around zero. Some print-tips may have larger variances compared to others and an appropriate scale adjustment is needed to account for such differences. One proposed approach is to find the maximum likelihood estimate for the scale of the variance for each print-tip group [ 16 ]. This method assumes that all log 2 -ratios from the k th print-tip group follow a normal distribution with mean zero and variance σ 2 , where σ 2 is the variance of the true log 2 -ratios and is the estimated scale factor for the k th print-tip group. However, this is only valid for certain types of data that reasonably follow a normal distribution and in our work we observe that this assumption may often times lead to undesirable results. Refer to [ 16 ] for further details. Another approach proposed here that is able to deal with the variance scaling issue is to introduce a weighting factor in the calibration function that is of the form for i = 1,..., n k , k = 1,..., ℓ, and where the weight is given as . The bias-corrected sample variance for the k th print-tip is denoted by and is given as where denotes the sample mean for print-tip k . Furthermore, the minimum sample variance is given as Compared to the maximum likelihood method outlined by [ 16 ], this method stresses higher weighting on print-tip groups that exhibit less variance and lower weighting for highly variant print-tips. If such a weight is not introduced, the normalization may improperly calibrate highly variant print-tip groups that have extreme sample means and many genes may erroneously be considered as differentially expressed as a consequence. Other treatments, such as the one suggested by Quackenbush [ 12 ] examine the geometric mean of the tip variances as a scale factor for the normalization estimate. However, such a treatment may not always scale the tips properly since some tips may still be overly compensated. Our proposed scaling factor λ k takes values over (0, 1] while other scaling methods may have larger upper limits. By calibrating data using Eq. (9), we have obtained nearly identical sample means, but less total variance for the resulting data compared to previously published techniques. The computation of λ k is straightforward and easy to calculate but our novel variance stabilization procedure does not take into account any heteroscedasticity in the data, namely observed increasing ratio variance with decreasing measurement intensity A . A rigorous comparison of print-tip scaling is beyond the scope of this contribution, but it is noted that the different scaling procedures affect the overall calibration scheme. Case studies To demonstrate the utility of our optimized LOWESS normalization procedure, we first utilized a set of three self versus self experiments [ 21 ], BT-474, MCF-7, and HBL-100, which were obtained using the protocols delineated in the methods section. In addition, we calibrated a set of four breast cancer cell lines [ 22 ], BT-474, MCF-7, MDA-MB-436, and MDA-MB-361, each measured in comparison to the reference cell line HBL-100, which were obtained using the protocols outlined by Järvinen et al . [ 20 ]. For each cancer cell line, three replicate slide hybridizations were available. In order to reduce the effects of spots whose intensities are not reliable due to experimental or printing errors, we used two separate quality filtering methods and normalized the intensities after discarding values that were detected unreliable. The assessment of ratio quality was performed using the method proposed by Chen et al . [ 29 ] and the evaluation of spot quality was performed using the method of Hautaniemi et al . [ 30 ]. Optimized parameter selection for f k was performed and print-tip LOWESS normalization results are compared to the results using arbitrary choices of the parameter f k . The implementation took a few minutes to run on a standard desktop PC running MATLAB. Self versus self experiments Self versus self experiments provide a trivial application to test our method since the amount of mRNA in both the test and the reference samples is the same. Thus, the points of an intensity scatterplot in the log 2 - log 2 space should be distributed along a straight line that intersects zero with a slope of unity. In the ( A , M )-coordinate system, all values of M should lie on a straight line at M = 0 for all values of A ; this means that the calibrated ratios should ideally be unity for all variables. Correspondingly, the cost measure is given when ψ k , i (·) = 0, (∀ k , i ), in Eq. (5) for the ( A , M )-coordinate systems. Separate trials were conducted using weighted, zeroth-order ( d = 0), first-order ( d = 1), and quadratic ( d = 2) polynomial fits. For all trials, the number of print-tip LOWESS iterations was fixed at t = 3. The weight function used is given by Cleveland [ 8 ]. For each experiment, the local print-tip groups were separately normalized with their respective, optimized values of f k . As a comparison to arbitrary selections of f k , the print-tip normalization was also carried out using f k = 0.2, 0.4, 0.6, and 0.8 in separate trials. Figure 1 shows the ( M (Arb) , M (Opt) )-scatterplot comparison between the calibration results with d = 1 using optimal f k and arbitrary f k for the BT-474 self versus self experiment. The points that deviate from the blue line are the genes that are most affected by the choice of f k . The M ( Arb ) data in this figure was calibrated using f k = 0.4, ∀ k . In all three self versus self experiments, the global sample means of M were nearly the same after calibration, regardless of the choice of f k . However, the calibrations that used optimized selections of f k for each print-tip resulted in data that contained less overall variance compared to the arbitrary selections. The ultimate goal of calibration is to adjust the dynamic range for the transformed ratios and reduce the variability within the data. By using optimized selection of f k , we outperform all arbitrary formulations to achieve these goals. Typical microarray experiments One immediate concern for typical experimental microarray data is that many genes may be over- or under-expressed and the true, transformed gene expression ratio ψ k , i (·) surely will not be equal to zero for all genes. Accordingly, implementing the cost function in Eq. (5) becomes an immediate challenge since the normalization reference level of all the genes for a typical microarray experiment may be diffcult to determine with complete accuracy. We note that our cost function still may be used with the assumption that the sample mean for each print tip before log 2 -transformation is unity. In most microarray experiments, many genes may be assumed to have constant RNA concentrations while smaller numbers of genes may be over or under expressed, namely their sample mean over all the genes is zero, . Using this assumption in Eq. (5), our experiments show that by minimizing the cost function in this context, like in the self versus self case study, we are able to systematically choose f k and the only consequence is that the minimum of the cost will not be as low as in the self versus self scenario where all genes should be constantly expressed. The main benefit of utilizing LOWESS for microarray normalization is that it is robust to extreme outliers and the cost function implemented in this fashion further restricts the effects of such extreme points in the regression. Ultimately, this implementation results in reliably calibrated ratios compared to the arbitrary formulation where different choices of f k affect the resulting data. Since a single microarray experiment represents an observation, multiple observations would be needed to compute a reliable estimate of the true transformed ratio values. The use of only a small number of replicate slides may be satisfactorily used to determine reliable estimates of true gene expression and one study showed that three replicates suffce for significantly reducing experimental variability [ 31 ]. With the growing number of publicly available microarray data, conducting replicate experiments is becoming a popular solution to assess experimental errors and reduce noise bias in the measurements [ 32 ]. The advantages of replicate slides also greatly help the analysis of between-slide variability and help address formal statistical considerations when drawing biological conclusions. Here, we show that the optimized normalization approach may be directly extended in an iterative manner to use the estimates of the true ratio values for further specifying f k . After an initial round of optimized LOWESS normalization for each replicate slide with ψ k , i (·) = 0 in Eq. (5), the sample mean for each gene may then be calculated using the replicates. The normalization reference levels ψ k , i (·) were reassigned these average gene expression values in Eq. (5). Each experiment was then separately calibrated a second and final time using the optimization approach and the final results were noticeably different compared to the normalized data using f = 0.2 that Järvinen et al . posted on their website [ 20 ]. A noteworthy consideration to address here is the overall effect of an iterative calibration process on the underlying structure of the data. Experimentally, once the optimized LOWESS regression is computed using the average value for each gene and normalization is performed, subsequent calibration attempts using the cost function-based method do not result in drastically different data. The subsequent regressions are nearly constant lines near M = 0 in the ( A , M )-scatterplot if the cost function approach is used. Consequently, the calibrated data reach a stable domain with a small dynamic range. Empirically, we found that performing optimized normalization in an iterative manner will not propagate regression effects through to disrupt the underlying structure of the data. Figure 2 shows the scatterplot comparison between the calibration results using optimal and arbitrary selections of f k for the first replicate BT-474 hybridization. Some genes in this plot report 4-fold differences and ultimately these differences affect data analysis. Consequently, the errant choice of this parameter f k may have deleterious effects on different biological studies. To illustrate the differences for one representative print-tip in this breast cancer study for the first replicate of the BT-474 cell line, Figure 3 plots the regressions obtained by both methods. All the data points for this hybridization are shown as a two-dimensional histogram [ 33 ], while the spots given by print-tip k = 16 are highlighted in black. In this plot, we show that the regression obtained by the optimized choice of f 16 differs from the one obtained by arbitrarily selecting f 16 = 0.2 and the calibration results are thus affected. Figure 4 reports arbitrary calibration results and Figure 5 shows optimized results. The data in Figure 5 has less overall variance when calibrated with the optimized choices of f k . As further illustration of the calibration differences between the optimized and arbitrary calibration results, we employ a goodness-of-fit test [ 34 ]. We wish to make a direct test of the data, independent of any underlying parent distribution of the ratios, and we use the following statistic for the k th print-tip group where M (Arb) and M (Opt) are the arbitrary and optimized calibration results, and the denominator within the summation is simply the variance of the difference between M (Arb) and M (Opt) . The null hypothesis is defined to be H 0 : the normalized ratios using arbitrary f are comparable to ones using optimized f . We tested against p < 0.05 for the distribution and reported the alternative hypothesis for a few print-tip groups on almost all the slides. In this analysis, we compared optimized choices of f for each print-tip to the arbitrary choices f = 0.2, 0.4, 0.6, and 0.8. By looking across each replicate of the calibrated data for all four breast cancer cell lines, almost all slides in this study reported at least one print-tip to have statistically different calibration results based on the choice of f k . Often times a single slide would report two or three print-tip groups that had statistically different calibration results. In addition to statistical analysis, genes that exhibit known over-expression in the BT-474 cell line data [ 35 ] were selected here for more detailed analysis. In particular, genes that were verified experimentally using reverse transcription-polymerase chain reaction (RT-PCR) were of the highest interest. Comparing our optimized calibration results utilizing the replicate data to the normalized data by Järvinen et al . [ 20 ], our results conform strongly with most of the over-expressed genes given in a list from a parallel study [ 35 ]. Two genes in particular stand out to demonstrate the benefits of utilizing our proposed method: homeo box B7 , which was validated with RT-PCR [ 35 ], and v-erb-b2 , which is known to be over-expressed in the BT-474 cell line [ 35 ]. The results posted by Järvinen et al . [ 20 ] for calibrating the homeo box B7 gene shows that it falls within the top 18% of overall gene expression, but by using the optimized approach we report it to be within the top 13%. For the v-erb-b2 gene, both calibration techniques report that this gene falls within the top 1% of the genes in terms of expression. As a result, for the homeo box B7 gene, the calibration factor f k is responsible for about 5% change in the reported gene expression. This is a dramatic result that may influence how the expression for this gene may be interpreted in comparison to the accepted biological knowledge of a certain experiment. As public data from microarray experiments continues to become available, the knowledge of certain genes will undoubtedly be uncovered for well-studied cell lines and this information will help further assess normalization and microarray quality control tasks. Conclusions The LOWESS method has recently been applied in other applications for the biological sciences. Comparative genomic hybridization (CGH) is a molecular cytogenetic method of screening a tumor for genetic changes. The alterations are classified as DNA gains and losses and they reveal a characteristic pattern that includes mutations at chromosomal and subchromosomal levels. Our proposed optimized scheme is directly applicable to the application of calibrating CGH microarray experiments, as well as for data analysis aspects. For example, the work of Clark et al . [ 36 ] utilized the LOWESS method for identifying the regions where gene copy numbers were aberrantly high or low in prostate cancer using CGH microarray technology. The parameter f was chosen arbitrarily and its value was not reported in the study. Consequently, reproduction and verification of these results may be diffcult. For instance, some of the important biological findings, such as start and end points of amplifications and deletions, may be adversely affected by different choices of f . In addition to CGH analysis, LOWESS has found application in case-control studies where logistic regression has been used to model the relationship between binary responses and continuous predictor variables [ 37 ]. In these types of studies one may use LOWESS to remove systematic trends that contaminate the laboratory measurements of predictor variables. The analysis reported by Borkowf et al . [ 37 ] clearly shows that different choices of f result in noticeably different correction effects and the optimization method proposed here may be suitable for enhancing such a study. Adaptations to the cost function may be utilized to handle this type of data. In addition, analysis of other types of scatterplot data by utilizing the LOWESS method with an arbitrary choice for the bandwidth parameter is undoubtedly susceptible to varied interpretations or errant conclusions [ 38 , 39 ]. Another result of this optimized calibration study is that we uncovered a better understanding of choosing the parameter d in the weighted polynomial fit. A higher-order ( d > 2), weighted polynomial is rarely needed based on the argument that such an assumption is, to a certain extent, over-fitting the data. From the findings of our study, we find that it is better to use a linear estimate based on minimizing the estimate errors across ( A , M )-scatterplots. Consequently, different choices of d resulted in different optimized values for f . The reason is that for the higher-order polynomial, it is beneficial in general to retain a larger fraction of the values of A for the weight function in computing the polynomial coeffcients. It is very important to carefully select f since ultimately, the bandwidth is a function of the polynomial order. Here, we also reaffirmed the idea that the quality filtering of ratios and spots is a necessary step that should precede all experimental microarray data handling procedures, whether it is scatterplot-based normalization or any other normalization method, since errant ratios would surely have a deleterious affect on the calibration. For instance, in the BT-474 data, the first replicate slide had poor ratio quality for a handful of genes. Calibration without considering or removing these errant spots resulted in less reliable results. This study addresses the issue of locating sources of experimental error for print-tips that have high sensitivity for the parameter f . For one, print-tips are physically different and they are considered to have different types of errors introduced based on these properties. In the formulation of normalization, it is imperative to address such subtle issues when choosing and implementing any algorithm. The systematic choice of the parameters in the LOWESS algorithm has not been previously addressed in the microarray literature and the method proposed here may be utilized in different microarray platforms. Such a treatment is also important for a wide variety of applications that employ scatterplot-based regression. The findings of this study illustrate that by choosing different values of f for the LOWESS algorithm results in noticeably different normalization results. This proposed method requires the calibration step to clearly state the assumptions used for within-slide normalization. Our optimization algorithm is more systematic than simply choosing an arbitrary parameter value or through trial and error techniques since the optimized approach relies on the actual underlying structure of the data. We also stress that such an optimization algorithm may also be utilized for other studies in addition to DNA microarray normalization treatments. Proper changes need to be made to Eq. (5) to reflect the ideal model for the data captured in the function ψ k , i (·), but in some studies, such a function may be satisfactorily determined or estimated from the data. Methods Data resources For the self versus self hybridizations, custom cDNA microarray experiments proceed as follows. Altogether, three microarray hybridizations were performed using custom printed cDNA microarray slides from the same print batch. Labelling, hybridization and washing were done as described previously by Monni et al . [ 40 ] and Järvinen et al . [ 20 ]. Briefly, total RNA was extracted from cell lines BT-474, HBL-100, and MCF-7 and labelled with Cy3-dUTP and Cy5-dUTP (Amersham Biosciences, Piscataway, NJ). The custom printed cDNA microarrays comprised of 11,520 clones from Incyte Genomics IRAL cDNA library and 1,136 clones from Research Genetics library. Microarrays were printed on poly-l-lysine coated slides using an Omnigrid arrayer (GeneMachines) as described previously [ 20 ]. Microarrays were scanned with an Agilent laser confocal scanner (Agilent Technologies, Palo Alto, CA) and gridded using the DEARRAY software developed by Chen et al . [ 29 ]. For the four breast cancer cell lines, custom cDNA microarray experiments were provided in a separate contribution by Järvinen et al . [ 20 ] and detailed protocols are described in that work. The relevant genes in our study were verified using RT-PCR in a parallel study by Hyman et al . [ 35 ]. Data quality filtering All microarray experiments contained in this work were conducted and spotted using groups of ℓ = 32 print-tips, with each tip being responsible for either 384 or 420 spots in their respective subarray. In order to reduce the effects of spots whose intensities are not reliable due to experimental or printing errors, we used two separate quality filtering methods and normalized the intensities after discarding values that were detected unreliable. The assessment of ratio quality was performed using the method proposed by Chen et al . [ 29 ] and ratios that had a quality value below the threshold 0.5 were discarded from our analysis. This quality cutoff value has, in the past, been shown to represent less reliable cDNA microarray measurements due to either low signal intensity, high local background level, uneven distribution of the target intensity, and/or small target size. The evaluation of spot quality was performed using the method of Hautaniemi et al . [ 30 ]. In this Bayesian networks-based method, we utilized the following features in determining spot quality. Bleeding, spot roundness, and spot intensity were assessed for the Cy5 channel and bleeding, spot size, spot roundness, background intensity, and fitting error were evaluated for the Cy3 channel. These features were chosen since this set was found to result in the best classification accuracies [ 30 ]. The trained Bayesian network was applied to each slide in this study and all the spots having a quality value of zero were excluded from the subsequent analysis. Authors' contributions JAB developed the mathematical formulation of the problem, implemented the optimized normalization algorithm in MATLAB, developed the statistical analysis, and wrote the manuscript. SH developed the LOWESS normalization software in MATLAB, coordinated spot quality filtering, and assisted in drafting the manuscript. AKJ conducted the self versus self microarray experiments and performed ratio quality filtering for data analysis. HE assisted in data preparation and in drafting the manuscript. SKM participated in the design and coordination of the study and assisted in drafting the manuscript. JA reviewed the statistical analysis and participated in the design and coordination of the study. All authors read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539276.xml |
514496 | Malaria morbidity and immunity among residents of villages with different Plasmodium falciparum transmission intensity in North-Eastern Tanzania | Background The relationship between the burden of uncomplicated malaria and transmission intensity is unclear and a better understanding of this relationship is important for the implementation of intervention programmes. Methods A 6-month longitudinal study monitoring risk factors for anaemia and febrile malaria episodes was conducted among individuals aged below 20 years, residing in three villages of different altitude in areas of high, moderate and low malaria transmission intensity in North-Eastern Tanzania. Results The burden of anaemia and malarial fever fell mainly on the youngest children and was highest in the village with high transmission intensity. Although a considerable percentage of individuals in all villages carried intestinal worms, logistic regression models indicated that Plasmodium falciparum was the only significant parasitic determinant of anaemia. Interestingly, children who carried low-density parasitaemia at the start of the study had a lower risk of contracting a febrile malaria episode but a higher risk of anaemia during the study period, than children who were slide negative at this point in time. Conclusion Young children living in the high transmission village carried a very high anaemia burden, which could be attributed to malaria. The overall incidence of febrile malaria was also highest in the high transmission village particularly among those under five years of age. These data suggest that in rolling back malaria, available resources in prevention programmes should primarily be focussed on young children, particularly those residing in areas of high malaria transmission. | Background Plasmodium falciparum malaria remains an important public health problem in sub-Saharan Africa. To develop and assess the efficacy of control measures, it is important to obtain a better understanding of how the malaria disease burden is distributed among population groups and how this burden is affected by changes in malaria transmission intensity [ 1 ]. In areas of high malaria transmission infants and young children carry a very high disease burden [ 2 ], but protective immunity is developed in early childhood. Adults and older children are able to control parasitaemia and therefore only rarely suffer from mild malaria symptoms [ 3 , 4 ]. In areas of low malaria transmission, immunity develops slowly and malaria affect all age groups [ 5 , 6 ]. It has been suggested that the societal burden of malaria does not necessarily increase with transmission intensity, but peaks at a certain level of transmission after which it remains constant and may even decrease [ 7 - 9 ]. To address this issue we have compared the malaria situation in three communities situated North-Eastern Tanzania, which show differences in transmission intensity. In this area, transmission intensity is determined by altitude and large differences in transmission can be found within a limited geographical area [ 10 - 12 ]. This study reports the results of six months morbidity follow-up, during which the incidence of febrile malaria episodes and the prevalence of anaemia were assessed in cohorts of 0–19 year old individuals. Methods Study area The study was conducted in three villages in Tanga region, North-Eastern Tanzania. The three villages were Mgome (5°12'S, 38'51'E) at an altitude of approximately 200 meters, Ubiri (4°72'S, 38°29'E) at an altitude of approximately 1,200 meters, and Magamba (4°75'S, 38°29'E) at an altitude of approximately 1,700 meters. The climate in the area is characterized by variations in rainfall and temperature related both to season and altitude [ 12 ]. The long rainy period occurs during April-May, while short rains occur in November-December. Mean daily temperatures are highest in January and lowest in July. Generally, the malaria transmission season peaks just after the rainy seasons with most consistent transmission in lowland sites from April to July. Previous studies have reported parasite prevalence rates to be in the ranges of 79–90% in the lowlands, 27–46% at intermediate altitudes and 8–16% in the highlands [ 10 ]. Entomological surveys in the study areas have shown that Anopheles gambiae is the most prevalent vector in the lowlands, while Anopheles funestus predominates in the highlands [ 10 ]. The entomological inoculation rates (EIR) have been reported to be in the range between 91–405 in the lowlands, and between 1.8–34 at intermediate altitudes [ 10 ]. In the highlands, mosquito densities are too low to allow reliable EIR measurements, but an EIR of 0.03 has been extrapolated [ 10 ]. Villagers living at low and intermediate altitudes perceive malaria as a major problem among both children and adults, but at the highest altitudes villagers consider that malaria is not a major part of the disease burden in either adults or children. There is little difference in treatment seeking behaviour for febrile illness between the altitudes. Treatment is generally sought for symptoms rather than for the disease and first treatment is almost universally an anti-pyretic drug bought from local shops (Caroline Jones, unpublished data). For all three villages, the nearest health facility is located within a distance of 13 km. Mgome is served by Umba Dispensary (10 km), Masaika Dispensary (5 km), Mkuzi Health Centre (7 km) and Muheza Designated District Hospital (14 km). Ubiri village is served by Lushoto District Hospital at a distance of approximately 13 km. Magamba village has a government and a private missionary dispensary both within the village, and is served also by Lushoto District Hospital at a distance of about 15 km. At the time of the study, sulphadoxine-pyrimethamine (SP) was the first-line treatment for uncomplicated malaria in Tanzania. It has been documented that the level of SP resistance is high in the Mgome area [ 13 ], whereas the situation has not been monitored previously in Ubiri and Magamba. Land use in the lowland areas is characterized by subsistence farming of maize, rice, bananas, beans, cassava, coconuts, fruits and other crops, as well as large-scale production of sisal. In the highlands, there is subsistence farming, mainly of maize, beans, bananas, potatoes, cabbages, tomatoes and fruits, and also large-scale production of tea and coffee. Study population Prior to the study, census surveys were done in each village and study individuals randomly selected from a census list. Mgome village is inhabited mainly by the Bondei tribe (60%), while Ubiri and Magamba are inhabited by Sambaa at 97% and 57%, respectively. The aim was to recruit a total of 250 individuals below the age of twenty years from each village, distributed in different age groups as follows: 0–1 year: n = 25, 1 year: n = 25, 2 years: n = 25 3 years: n = 25, 4 years: n = 25, 5–6 years: n = 25, 7–9 years: n = 25, 10–14 years: n = 40 and 15–19 years: n = 40. Cross-sectional surveys Malariometric surveys were conducted in each village in April, July and September 2001. During the first survey, the purpose of the study was explained and consent to participate obtained from each study individual or their parents/guardians. Baseline demographic data were collected together with a history of migration and recent movements. The use of malaria preventive measures was also recorded. A history of recent illness was obtained, emphasizing symptoms suggestive of malaria. Physical examination on signs related to malaria such as temperature, pulse, spleen size, pallor and respiratory rate was conducted. Axillary temperature was measured using digital thermometers. Height, weight and upper-arm-circumference were recorded for estimation of nutritional status. For any individual diagnosed with mild disease, appropriate drugs were administered in the field. Individuals with symptoms of malaria were treated with SP. Participants with severe disease were referred to the nearby hospital. Five millilitres of venous blood were collected from study individuals aged three years and above into vacutainer tubes containing citrate buffer. For children below three years, 300–400 μl of capillary blood from a fingerprick were collected into eppendorf tubes containing EDTA. The haemoglobin (Hb) of each participant was measured from drops of blood using a HemoCue ® photometer (Ångelholm, Sweden). Whole blood was used to prepare thick and thin blood smears for malarial microscopy. These were stained with 10% Giemsa stain for 15–20 minutes after fixing thin smears with methanol. Asexual and sexual parasites were counted against 200 and 500 white blood cells, respectively. The differentiation of malaria parasite species was confirmed by microscopy of thin smears. A blood smear was declared negative only after examination of 200 high power fields. The density of asexual parasites was calculated assuming 8000 leucocytes per μl of blood and expressed as parasites per μl. During the first cross-sectional survey, study participants were asked to collect stool and urine specimens in special containers. Direct smear-technique was used to check for the presence of hookworm ova and other intestinal parasites. A pinhead of stool was collected, put on a slide and emulsified in a drop of normal saline. A cover slip was then applied and the slide examined using low-power microscopy. Longitudinal monitoring of febrile episodes Local village helpers (two community members per village) and health workers at nearby health facilities performed passive case detection during the 6-month study period. The village helpers were provided with first-line antimalarial drugs (SP), paracetamol, slides, blood lancets, treatment charts, febrile case detection forms and storage boxes. Villagers could seek treatment at any time from these helpers. Patients with symptoms of malaria were treated with first-line antimalarial drugs or, if they had severe symptoms or did not respond adequately to the first-line treatment, they were referred to a health facility. Prior to treatment the village helpers collected clinical information and a malaria blood smear. At each nearby health facility, two permanent staff members monitored study participants seeking medical treatment at the facility. If a study participant presented at the facility with history of fever and/or an axillary temperature ≥ 37.5°C, a form was completed and a blood smear collected. Once per month active febrile case detection was undertaken by the research team. During active case detection, each study participant was seen by a trained physician and a blood smear was taken from any study participant reporting a history of fever within two days and/or those who had an axillary temperature ≥ 37.5°C Case definitions Anaemia was defined as haemoglobin < 11.0 g/dl [ 14 , 15 ]. To adjust for the physiological effect of altitude on haemoglobin concentration, a correction factor was calculted with haemoglobin values being normalized to sea level for direct comparison between the study villages. The correction factor assumed a linear relationship between increasing altitude and haemoglobin, although the relationship may not necessarily always be exact [ 16 ]. For Mgome (200 m), the correcting factor was a reduction of 0.1 g/dl, for Ubiri (1,200 m) the factor was 0.8 g/dl and for Magamba (1,700 m) the factor was a reduction of 1.0 g/dl. Febrile malaria episodes were defined as an axillary temperature ≥ 37.5°C and /or a history of fever within the previous 48 hours in the presence of asexual P. falciparum parasites above a defined density cut-off level. Many individuals carried low density asymptomatic parasitaemia, and fever among parasitaemic individuals may also have been caused by other illness [ 17 ]. Thus, to account for the variation in levels and point prevalence of asymptomatic parasitaemia between study villages [18–20], as well as the different age groups involved in the study [ 21 ], different P. falciparum density cut off levels were applied in each village. To balance between sensitivity and specificity in diagnosing a febrile malaria episode, we aimed at a febrile malaria case specificity >80%. In Magamba (the low transmission village), a cut-off of 40 parasites/μl was applied, while cut-offs of 1000 parasites/μl and 5000 parasites/μl were used in Ubiri (the moderate transmission village), and Mgome (the high transmission village), respectively. Age-specific incidence rates of febrile malaria episodes were calculated as the number of episodes divided by the number of days that individuals in the age group were at risk during the follow-up. After a febrile malaria episode an individual was censored for 28 days [ 6 ]. The effect of using different parasite density cut-offs in the definition of a febrile episode was evaluated by not applying a cut-off in the definition or by applying age specific cut-offs [ 21 ]. Statistical methods All data were double-entered into a database in Epi-lnfo Version 6.04d (CDC, Atlanta, USA) and statistical analyses were performed with Stata version 8 (Stata Corporation, Texas, USA). Univariate analyses and multivariate logistic regression were performed to determine risk factors for anaemia and febrile malaria episodes. For Mgome village, a logistic regression model was developed to determine whether the result of the first slide reading in April could be used to predict the subsequent risk of developing anaemia or febrile malaria during the following six months of morbidity surveillance. In this model, P. falciparum parasitaemia was categorised as no parasitaemia if no parasites were detected microscopically, low-density if parasitaemia was between 40 parasites/μl and 4999 parasites/μl, and high-density if the level was above or equal to 5000 parasites/μl. Thus, the first slide reading of individuals who did not have fever/had normal haemoglobin levels at enrolment was used to predict the risk of developing a subsequent episode of malarial fever/anaemia. Ethical considerations Ethical clearance was granted by the Medical Research Co-ordinating Committee of the National Institute for Medical Research, Tanzania. Prior to the study, meetings were held with local authorities and with the villagers in each study village, during which the aims of the study were explained. Informed consent documents for the study were prepared in English and translated into Kiswahili before administration to both village leaders and participants. Written informed consent to participate was obtained from each study individual or from his/her parents or guardians. Study individuals were free to withdraw from the study at any time without giving any reasons, or being disqualified from any medical services that were provided to all villagers throughout the study period. At the end of the study, preliminary findings were presented at village meetings. Results Prevalence and densities of Plasmodium species and other infections Three study villages were selected to represent areas of markedly different malaria transmission intensity. In each village, approximately 250 individuals under the age of 20 years were recruited. Few individuals reported using anti-malarial preventive measures (Table 1 ). Repeat investigations on the same individuals were undertaken at enrolment in April 2001, and during subsequent cross-sectional surveys in July and September 2001. Only about 10% of the study participants were lost to follow-up in each village. Table 1 Baseline characteristics of the study villages Baseline characteristic Mgome Ubiri Magamba Altitude (m) [range] 196 [165, 208] 1216 [1174, 1262] 1585 [1659, 1751] Enrolled (0–19 years) (Male/Female) 254 (115/139) 250 (139/111) 255 (132/123) Use of preventive measures Nets (%) 18/254 (7.1) 5/250 (2.0) 14/255 (5.5) Burning coils (%) 8/254 (3.1) 1/250 (0.4) 0/255 (0) Neem (%) 0/254 (0) 1/250 (0.4) 0/255 (0) Spray (%) 0/254 (0) 0/250 (0) 3/255(1.2) Prophylaxis (%) 0/254 (0) 1/250 (0.4) 1/255 (0.4) As expected, P. falciparum prevalence and parasite densities (Figure 1 ) were higher in Mgome than in Ubiri and Magamba (trend test, z = 15.64, p < 0.001). In Mgome, the carrier rate was particularly high for children aged 1–9 years and then declined in the older age groups (trend test, z = -3.2, p < 0.001). In Ubiri, carrier rates were low in infants, peaked at the age of two years, but showed little variation in the age groups between 4 and 19 (Figure 1 ). Although carrier rates in Ubiri were slightly higher in April than in July and September, there were no marked seasonal changes in carrier rate by age in any of the villages. The parasite densities in those carrying parasites did not differ between villages after the age of six years. Among the under fives, children from Mgome and Ubiri carried higher levels of parasitaemia in July than in April and September 2001 surveys. Interestingly, between April and July surveys, there was a marked difference in the age-specific pattern of parasite density in Mgome. In April, the peak parasite density was noted in age group of 2 years, whereas the youngest had the highest parasite density in July surveys. Based on these findings, we categorised Mgome as a high transmission (holoendemic) village, Ubiri as a moderate transmission (mesoendemic) village, and Magamba as low transmission (hypoendemic) village. P. falciparum was the most predominant malarial parasite accounting for more than 95% of all malaria infections. The other malaria species were mainly found as mixed infections. The April prevalence rates of Plasmodium malariae in Mgome and Ubiri were 8.3 % and 3.9%, respectively, while these rates for Plasmodium ovale was 1.0% and 0%. In Magamba, only P. falciparum was found. Figure 1 Age-specific P. falciparum prevalence and geometric mean densities (positives only) by village and season. Panels A, B, and C show age-specific P. falciparum densities and prevalence in Magamba (low transmission), Ubiri (moderate transmission) and Mgome (high transmission), respectively. Lines indicate the prevalence rate for each survey. Solid lines with filled circle for April 2001, dotted lines with filled triangle for July 2001, and dashed lines with filled box for September 2001. Bars indicate P. falciparum densities (positives only) for each survey. Empty bars indicate the April 2001 surveys, hatched bars indicate the July 2001 surveys, and crossed hatched bars indicate the September 2001 surveys. Error bars indicate 95% confidence interval. A total of 492 individuals from the three villages submitted stool and urine samples, which were investigated for worms. Worms were found in 35.0% of study participants living in Mgome, and in 29.2% and 7.8% of individuals from Ubiri and Magamba, respectively. In Mgome, spleen enlargement was common (about 49.21%) and associated with age (Spearman rho = 0.238, p < 0.001) while in the two other villages the prevalence of splenomegaly was low and with no distinct age-pattern (data not shown). This distribution of splenomegaly remained stable during the study period. Haemoglobin levels and anaemia Haemoglobin levels were measured in all 759 individuals during enrolment and among those who reported for the subsequent cross-sectional surveys in July and September 2001. Regardless of the season, haemoglobin levels increased with both altitude and age (Figure 2 ). Univariate analysis indicated that age, altitude of residence, and presence of P. falciparum parasitaemia were associated with anaemia (Table 2 ) and this was supported by multivariate analyses in which P. falciparum was the only parasitic infection showing a statistically significant association to anaemia (Table 2 ). Figure 2 Age-specific anaemia prevalence and mean haemoglobin levels by village and season. Panels A, B, and C show results of surveys conducted in April, July and September 2001, respectively. The lines and symbols show patterns of anaemia prevalence for each village (Mgome: solid lines with filled circle, Ubiri: dotted lines with filled triangle, Magamba: dashed lines with filled square). Bars indicate mean altitude adjusted haemoglobin levels (g/dl) and 95% confidence intervals in each village (Mgome: empty bars, Ubiri: hatched bars, Magamba: crossed hatched bars). Table 2 Crude and adjusted odds ratios for risk factors for anaemia Explanatory variable Crude odds ratio (95% Cl) p-value Adjusted odds ratio (95% Cl) p-value Age group (years) 0–2 21.38 (8.36–54.65) <0.001 20.41 (7.44 – 56.0) <0.001 3–4 7.38 (2.82–19.25) <0.001 5.42 (1.93–15.20) <0.001 5–9 3.43 (1.36–8.66) 0.009 2.18 (0.82–5.81) 0.118 10–14 1.86 (0.70–4.99) 0.215 1.51 (0.54–4.27) 0.435 15–19 1 1 Sex Male 1.04 (0.70 – 1.54) 0.85 1.16 (0.71 – 1.90) 0.551 Female 1 1 Village Mgome 19.99 (7.08–56.42) <0.001 15.55 (4.78–50.65) <0.001 Ubiri 8.02 (2.80–23.82) <0.001 6.44 (2.08–19.92) 0.001 Magamba 1 1 Parasites P. falciparum 3.77 (2.44 – 5.83) <0.001 2.0 (1.11–3.62) 0.021 Hookworm 1.41 (0.83–2.39) 0.201 1.43 (0.76–2.69) 0.264 Ascariasis 0.86 (0.52–1.44) 0.564 1.04 (0.54–1.92) 0.952 Amoeba 0.16 (0.02–1.20) 0.074 0.16 (0.02–1347) 0.091 Schistosoma 0.42 (0.12–1.46) 0.174 0.28 (0.07–1.20) 0.086 Malaria morbidity during follow-up Of the 759 individuals enrolled in the study, 669 (88%) adhered to the follow-up scheme and were included in the analysis of febrile episodes. Loss to follow-up was due to death (three individuals) or emigration. Using the village-specific density cut-off described above, 54 individuals had febrile malaria episodes in Mgome, 10 in Ubiri and none in Magamba during the six-month follow-up period. The mean age of febrile malaria individuals was 1.97 years (95% Cl: 1.50, 2.59) and 3.23 years (95% Cl: 1.59, 5.86) for Mgome and Ubiri, respectively. Children below five years carried the major burden of febrile malaria episodes in Mgome (Figure 3 , panel A). The data was also analysed using age-specific parasite cutoffs [ 21 ] in the case definition (data not shown) and using a definition in which all fevers accompanied by a positive slide were considered a febrile malaria episode (Figure 3 , panel B). The latter definition increased the incidence rates in Mgome and Ubiri slightly, but the overall conclusion that the incidence rates were by far the highest in the children under five years living in Mgome was not affected by the case definition used. Figure 3 Incidence rates of febrile malaria episodes by age group. Panel A: Incidence rates calculated using village specific parasite density cut-offs in the definition of episodes. In Mgome (solid lines with filled circle) the cut-off was 5000 parasites per μl, in Ubiri (dotted lines with filled triangle) 1000 parasites per μl, and in Magamba (dashed lines with filled square) 40 parasites per μl. Panel B: Incidence rates calculated using a definition in which all fevers accompanied by positive slide were considered a febrile malaria episode. In Mgome, host age and the presence of low-density parasitaemia at the start of the study were consistently found to be associated with decreased risk of suffering a febrile malaria episode during the morbidity follow-up. Other variables such as sex, splenomegaly and use of a mosquito net did not contribute significantly to the model. In logistic regression models correcting for age, those who carried parasites at low densities in April had a four-fold lower risk (P < 0.03) of developing a febrile malaria episode during follow-up than those who were slide negative (Table 3 ). Table 3 Logistic regression model showing the risk of developing a febrile malaria episode during the 6 month morbidity surveillance according to age and the result of the slide at the initiation of the study in Mgome Explanatory variable Crude odds ratio (95% Cl) p-value Adjusted odds ratio (95% Cl) p-value Low parasite density 1 0.34 (0.15–0.78) 0.011 0.22 (0.06–0.89) 0.033 High parasite density 2 3.0 (1.09–8.29) 0.034 1.36 (0.27–8.82) 0.706 No parasitaemia 3 1 1 Age (years) 0.66 (0.58 – 0.76) <0.001 0.51 (0.332 – 0.772) 0.002 Age squared 0.98 (0.97 – 0.998) 0.007 1.03 (1.007–1.047) 0.007 1 Parasitaemia in April between 40 and 4999 parasites/μl 2 Parasitaemia in April >4999 parasites/μl 3 Slide negative in April In Mgome, 112 of the 254 individuals were not anaemic on enrolment in April 2001. Out of these, 68 (mean age (years) and 95% Cl: 11.2 [10.2, 12.3]) had normal haemoglobin levels during the July and September cross sectional surveys, while 44 developed anaemia during the study (mean age (years) and 95% Cl: 7.9 [6.4, 9.5]). Logistic regression models correcting for age showed that the risk of developing anaemia during the study was 4.4 times (p = 0.038) higher in individuals carrying low-density parasitaemia in April than in those who were slide negative (Table 4 ). Table 4 Logistic regression model showing the risk of developing anaemia during the 6 month morbidity surveillance according to age and the result of the slide at the initiation of the study in Mgome Explanatory variable 1 Crude odds ratio (95% Cl) p-value Adjusted odds ratio (95% Cl) p-value Low parasite density 3.98 (1.276–13.095) 0.02 4.38 (1.10–17.69) 0.038 High parasite density 3.17 (0.601–16.692) 0.17 2.43 (0.35–16.73) 0.369 No parasitaemia 1 1 Age (years) 0.86 (0.789 – 0.94) 0.001 0.51 (0.332–0.772) 0.002 Age squared 0.99 (0.99 – 0.998) 0.007 1.03 (1.007–1.047) 0.007 1 Refer to table 3 Discussion This prospective longitudinal study was designed to compare the burden of uncomplicated malaria in three similar villages situated in areas of markedly different transmission intensity. Not surprisingly, the study showed that individuals living in the village with very high malaria transmission carried a markedly higher burden of both anaemia and febrile malaria episodes compared to villagers at the sites with lower transmission. This result is in agreement with results from previous studies in the area [ 11 ]. The villages in the highlands are prone to malaria epidemics [ 12 ], but such epidemics did not occur during the study period. If they had, it is conceivable that the incidence of febrile malaria episodes at these sites would have reached or even exceeded the incidence found in the high transmission village [ 6 , 9 ]. Never the less, the anaemia burden in the high transmission village was very high among infants and young children. The burden among these children was much greater than among individuals of the corresponding age groups in the other two villages. During the first survey, villagers were also investigated for the presence of parasites in urine and faeces, but neither of these was shown to be a significant risk factor for anaemia in the multivariate logistic regression models (Table 2 ). Thus, the difference in anaemia burden between the sites appears likely to have been due to differences in malaria transmission intensity. The reason that hookworm infection did not constitute a risk factor for anaemia is likely to be a consequence of the fact that the villagers receive regular deworming medication as part of health promotion programmes, and therefore, the worm burden was rather low [ 22 ]. The heavy burden of anaemia carried in populations exposed to high malaria transmission has recently been highlighted [ 14 ]. The results of this study support the findings that malaria plays a major role in the burden of anaemia and these results are further corroborated by the fact that malaria interventions such as insecticide treated nets and intermittent preventive treatment in infants (IPTi) considerably reduce the incidence of anaemia [ 23 - 25 ]. From a public health perspective, our results reinforce the view that malaria prevention programmes should focus their attention on high-transmission areas and concentrate particularly on children under five years of age. Our study was not designed to compare incidences of severe disease and malaria deaths. It has previously been suggested that this malaria burden is in fact higher in populations exposed to moderate transmission than in populations living in areas of very high transmission [ 8 ]. The results of a large hospital-based study recently conducted in North-Eastern Tanzania over a wide range of transmission intensities suggested, however, that there was a positive correlation between severe malaria outcomes and intensity of transmission (Reyburn et al., submitted for publication). The longitudinal design of our study allowed an exploration of whether P. falciparum carriage at the beginning of the study influenced the risk of developing febrile malaria episodes or anaemia during the following study period. Interestingly, multivariate logistic regression models indicated that children carrying low-density parasitaemia during the first cross sectional survey were at a lower risk of developing a febrile malaria episode than children without detectable parasitaemia or children with higher levels of parasitaemia. This apparent protective effect of low-grade parasitaemia was recently also reported in a study from Ghana [ 26 ], but in our study this protection came with a price since children who controlled the parasite density at low levels were at markedly higher risk of developing anaemia. Conclusions The overall burden of malaria morbidity was found to be highest in the high-transmission area, where infants and children carried a very high malaria burden in the form of febrile episodes and anaemia. Populations in the areas of moderate and low transmission suffered a significantly lower morbidity. Therefore, in order to roll back malaria, available resources in malaria control programmes should focus on underfives residing in the high-transmission areas. Authors' contributions JPAL and LSV participated in the planning of the study, carried out field surveys, analysed the data and drafted the manuscript. BPM participated in designing the study, carried out field surveys and managed the data. CJD participated in study planning, in the fieldwork and in editing of the manuscript. CJ participated in the planning of the study and conducted the socio-economic analysis of study villages. JA and ZXS participated in the field surveys and performed microscopy of all blood smears. AYK, MML and TGT participated in study planning, coordination, and analysis of data. All authors participated in the paper writing and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC514496.xml |
490027 | The Kinetochore Is an Enhancer of Pericentric Cohesin Binding | The recruitment of cohesins to pericentric chromatin in some organisms appears to require heterochromatin associated with repetitive DNA. However, neocentromeres and budding yeast centromeres lack flanking repetitive DNA, indicating that cohesin recruitment occurs through an alternative pathway. Here, we demonstrate that all budding yeast chromosomes assemble cohesin domains that extend over 20–50 kb of unique pericentric sequences flanking the conserved 120-bp centromeric DNA. The assembly of these cohesin domains requires the presence of a functional kinetochore in every cell cycle. A similar enhancement of cohesin binding was also observed in regions flanking an ectopic centromere. At both endogenous and ectopic locations, the centromeric enhancer amplified the inherent levels of cohesin binding that are unique to each region. Thus, kinetochores are enhancers of cohesin association that act over tens of kilobases to assemble pericentric cohesin domains. These domains are larger than the pericentric regions stretched by microtubule attachments, and thus are likely to counter microtubule-dependent forces. Kinetochores mediate two essential segregation functions: chromosome movement through microtubule attachment and biorientation of sister chromatids through the recruitment of high levels of cohesin to pericentric regions. We suggest that the coordination of chromosome movement and biorientation makes the kinetochore an autonomous segregation unit. | Introduction The proper segregation of replicated chromosomes, or sister chromatids, to daughter cells during mitosis requires that sister chromatids establish stable attachments to microtubules emanating from opposite spindle poles, known as chromosome biorientation, and that sister chromatids move to opposite poles of the cell during anaphase. Chromosome biorientation is made possible by the cohesion of replicated sister chromatids, which occurs along the entire length of the chromosome and is especially robust in large centromere-flanking or “pericentric” domains ( Sumner 1991 ). The centromere is the site of assembly of the kinetochore, a protein complex that mediates the attachment and movement of chromosomes along the mitotic spindle. The centromere-flanking domains of cohesion are thought to play an important role in biorientation by constraining the kinetochores on paired sister chromatids in opposite directions. This orientation of sister kinetochores facilitates the capture of microtubules originating from different spindle poles, thereby ensuring that sister chromatids are segregated in opposition later in mitosis. A second function of pericentric cohesion and possibly cohesion along chromosome arms is to resist the poleward forces that are imposed by these bipolar spindle microtubule attachments. The resistance that is provided by cohesion prevents the premature dissociation of sister chromatids and contributes to a tension-based mechanism that stabilizes kinetochore–microtubule interactions ( Nicklas and Ward 1994 ). Given the importance of cohesion in pericentric regions, it is critical to understand how these large functional domains are assembled. An important clue came with the discovery of a group of proteins that mediate cohesion and are conserved in organisms from the yeasts to vertebrates ( Guacci et al. 1997 ; Michaelis et al. 1997 ; Furuya et al. 1998 ; Losada et al. 1998 ; Skibbens et al. 1999 ; Toth et al. 1999 ; Hartman et al. 2000 ; Tomonaga et al. 2000 ; Wang et al. 2000 ; Hanna et al. 2001 ; Losada and Hirano 2001 ). This group includes Pds5p and the members of a multisubunit “cohesin” complex, Mcd1/Scc1p, Irr1p/Scc3p, Smc1p, and Smc3p. Cohesins were shown to bind to specific regions of chromosomes ( Blat and Kleckner 1999 ; Megee et al. 1999 ; Tanaka et al. 1999 ; Hartman et al. 2000 ; Laloraya et al. 2000 ; Panizza et al. 2000 ). At centromere-distal locations, cohesin-binding sites span only approximately 0.8 kb ( Laloraya et al. 2000 ). In striking contrast, cohesin binding is highly enriched within an approximately 50-kb pericentric region of budding yeast Chromosome (CHR) III flanking the centromere–kinetochore complex, which occupies only a 0.25-kb nuclease-resistant region ( Bloom and Carbon 1982 ; Saunders et al. 1988 ; Blat and Kleckner 1999 ). It remains to be determined whether pericentric cohesin enrichment is unique to budding yeast CHRIII or is, in fact, a property of all pericentric regions. However, cytological observations support the notion that cohesin may be enriched in the centromere-proximal regions of all higher eukaryotic chromosomes, given that the chromosomes of mitotically arrested cells remain tightly associated in these regions ( Gonzalez et al. 1991 ). These observations suggest that these large domains of pericentric cohesion result from an enrichment of cohesin binding. How are these large cohesin domains assembled in kinetochore-flanking regions? Recent experiments in Schizosaccharomyces pombe have provided evidence for the role of repetitive heterochromatic DNA in pericentric cohesion. In this organism, the recruitment of high levels of cohesion factors to centromeric regions is dependent on Swi6 (SPAC664.01c), the fission yeast homolog of the heterochromatin protein HP1 ( Bernard et al. 2001 ; Nonaka et al. 2002 ). However, this is unlikely to be the only mechanism for generating large pericentric cohesin domains because the centromere on budding yeast CHRIII and neocentromeres in human cells are devoid of surrounding repetitive sequences characteristic of heterochromatin. One clue for a potential mechanism came from our observation that CEN3 placed ectopically on a minichromosome could direct the binding of cohesin to approximately 2 kb of centromere-flanking DNA even if that DNA did not normally associate with cohesin, establishing that centromeres could modulate cohesin binding on neighboring sequences ( Megee et al. 1999 ). The presence of these cohesin-enriched domains flanking CEN3 on both the endogenous chromosome and a minichromosome is intriguing, and their existence has raised many interesting questions. Here, we endeavored to determine whether large pericentric cohesin domains exist on all budding yeast chromosomes, and if so, whether these domains are equivalent in size and have a similar distribution of cohesin. Furthermore, we examined the possible roles of the centromere and centromere-flanking sequences in the assembly of the pericentric cohesin domains. Our results show that the budding yeast kinetochore behaves as an enhancer in the assembly of approximately 20-kb to 50-kb pericentric cohesin domains. Our observations suggest that the kinetochore mediates two essential segregation functions, coordinating not only the attachment of chromosomes to the mitotic spindle, but also the recruitment of sufficient levels of cohesin within pericentric regions to promote biorientation of sister kinetochores and to resist microtubule-dependent poleward forces. Thus, the kinetochore functions as a modular segregation unit. The integration of these functions by the kinetochore is therefore likely to play an important role in the maintenance of genomic integrity. Results Cohesin Is Enriched throughout Large Pericentric Domains Relative to Arm Sites Previous studies have demonstrated that cohesin binding is highly enriched in the centromere-proximal region of CHRIII in comparison to arm sites and also within the centromere-flanking region of a CEN3 -containing minichromosome in budding yeast arrested in mitosis using nocodazole, an inhibitor of microtubule assembly ( Blat and Kleckner 1999 ; Megee et al. 1999 ; Laloraya et al. 2000 ). To examine cohesin binding throughout centromere-proximal and -distal regions of the budding yeast genome, we have performed chromatin immunoprecipitation (ChIP) using epitope-tagged alleles of cohesin subunits (Mcd1-6HAp or Smc3-6Mycp) as markers for the cohesin complex. The immunoprecipitated (ChIP) and input DNA samples not subject to immunoprecipitation were then analyzed using two approaches ( Materials and Methods ). In some cases, PCR reactions that respond linearly to the amount of input DNA were performed for both ChIP DNA and diluted input DNA to determine the percentage of total chromatin bound by cohesin subunits in the ChIPs. Alternatively, input and ChIP DNA samples were labeled with aminoallyl dUTP conjugated to different fluorescent tags, and then hybridized competitively to DNA microarrays containing all budding yeast open reading frames (ORFs) and intergenic regions to analyze cohesin binding genome-wide. In both types of experiments, we examined the distribution (i.e., the locations of peaks and valleys) and the magnitude (peak height) of cohesin subunit binding. To determine whether the pericentric regions of all budding yeast chromosomes are similarly enriched for cohesin binding, input DNA and DNA crosslinked to Mcd1p in mitotically arrested cells were used to probe microarrays of the budding yeast genome. The profile of Mcd1p binding within each ORF and intergenic region was then superimposed on a map of the sixteen budding yeast chromosomes to visualize the distribution of Mcd1p association ( Figure 1 A). This analysis demonstrated dramatic differences in the distribution of Mcd1p in centromere-proximal and -distal regions of all sixteen budding yeast chromosomes, as had been observed previously in a study of CHRIII ( Blat and Kleckner 1999 ). While centromere-distal regions contained short, discrete foci (approximately 0.8 kb) of Mcd1p binding that were distributed at intervals of approximately 10 kb, centromere-proximal regions of all chromosomes contained large (approximately 20–50 kb) domains that were highly enriched for Mcd1p binding. Microarray analyses were also performed using ChIP DNA isolated from strains containing both Myc-tagged Smc3p (Smc3-6Mycp) and HA-tagged Mcd1p (Mcd1-6HAp), or Smc3-6Myc alone to confirm that another subunit of the cohesin complex showed a similar enrichment in pericentric chromatin. As was the case with Mcd1p, the magnitude of Smc3p association was higher in pericentric regions than at arm locations ( Figure 1 B). Furthermore, the pattern of Smc3p binding observed within pericentric regions in independent immunoprecipitations of chromatin isolated from the singly or doubly tagged strains was strikingly similar to that observed for Mcd1p ( Figure 2 ). Thus, the high level of concurrence in cohesin subunit binding patterns suggests that the association of the entire cohesin complex is enriched in pericentric chromatin compared to chromosome arm locations. In addition, we compared the datasets obtained by microarray analyses in this study to those obtained using Myc-tagged Mcd1p (Mcd1-18Mycp) in another commonly used budding yeast strain background as described in the accompanying report by Glynn et al. (2004) . These studies showed good agreement for the presence of enriched Mcd1p binding in pericentric regions and a similar distribution of peaks and valleys of binding at both pericentric and arm locations (correlation coefficient = 0.76; Glynn et al. 2004 ). Figure 1 Microarray Analyses of Mcd1p and Smc3p Binding DNA isolated from cohesin subunit ChIPs and control input DNA was labeled with aminoallyl dUTP, conjugated to Cy5 (red) or Cy3 (green) fluorescent tags, and then hybridized competitively to microarrays. Although the samples were labeled by different fluorescent tags depending on the experiment, for the purposes of analysis, the ratios are converted such that the ChIP signal is represented by red and control DNA by green. The red-to-green (R:G) ratio for each ORF and intergenic region was calculated for cells arrested in mitosis using the cdc16 mutation and assigned a color, and the median value obtained for each element was then plotted on a map of the sixteen chromosomes to determine the chromosomal distribution of Mcd1–6HAp. Regions with a R:G ratio less than 1.8 are shown in gray, and those with ratios of 1.8 or higher are shown in red. The red shading is an indicator of the intensity of cohesin subunit binding, such that regions with larger R:G ratios have lighter shades of red. Hybridization data are unavailable for regions shaded in blue (see Materials and Methods ), and genomic regions not present on the arrays are indicated in white. Centromere position is indicated by an asterisk. (A) Chromatin isolated from strains containing Mcd1-6HAp (1377A1-4B, 1829-15B, and PMY270) was immunoprecipitated using anti-HA antibodies. (B) Chromatin isolated from strains containing Smc3-6Mycp (PMY270 and 1839-3D) was immunoprecipitated with anti-Myc antibodies. Figure 2 Comparison of Mcd1p and Smc3p Binding Distributions in Centromere-Flanking Regions The log 2 of the median of the R:G ratios for Smc3-6Mycp (triangles) and Mcd1-6HAp (squares) binding within approximately 60-kb pericentric regions of CHRV, CHRVI, and CHRIX in cdc16 -arrested cells is plotted as a function of the indicated SGD coordinates. The relative position of the centromere within each pericentric region is indicated by the oval. To further investigate the nature of this pericentric cohesin enrichment, Mcd1p association profiles were examined within the approximately 50-kb pericentric regions of endogenous CHRI, CHRIII and CHRXIV using PCR analyses of ChIP DNA. For comparison, Mcd1p association was also examined within an approximately 37-kb centromere-distal region on the right arm of CHRIII ( Saccharomyces Genome Database [SGD] coordinates 242–279 kb) ( Figure 3 ). We observed that the magnitude of Mcd1p binding throughout these pericentric regions is on average 3- to 5-fold higher than the levels of association observed at the CHRIII centromere-distal location. Within each of the pericentric domains examined, the distribution of Mcd1p binding was not uniform, but instead consisted of peaks and troughs of association ( Figure 3 A– 3 C). These peaks of Mcd1p binding were much broader than those observed at the CHRIII centromere-distal location ( Figure 3 D) ( Blat and Kleckner 1999 ; Laloraya et al. 2000 ). The peaks of Mcd1p association within pericentric chromatin were separated by troughs having values approximately 0.5% of those of the input chromatin. Although significantly reduced in comparison to the peaks, Mcd1p association in these troughs reflects significant binding, given that other chromosomal regions such as ARS1 and ADE3 are absent from Mcd1p ChIPs (≤0.04% of input chromatin; unpublished data not shown). Furthermore, this binding within troughs is unlikely to reflect poor shearing of the pericentric chromatin by sonication, as ChIPs performed using antibodies specific for the kinetochore protein Mif2p showed an enrichment within centromeric DNA that decreased 5-fold in flanking regions only approximately 245 bp away from CEN3 ( Figure 4 ). Lastly, it is interesting to note that in many of the pericentric regions, cohesin subunit association in the interval that contains the centromeric DNA was reduced in comparison to the regions immediately flanking the centromere, possibly because the large complex of kinetochore proteins precludes the association of cohesin with the relatively small 120-bp centromeric DNA (see Figures 2 and 3 A– 3 C). Thus, microarray and PCR analyses of DNA crosslinked to cohesin subunits demonstrate that cohesin binding is highly enriched within large pericentric domains on all budding yeast chromosomes. Figure 3 Mcd1p Binding Profiles in Centromere-Proximal and -Distal Regions Cells containing Mcd1-6HAp were first staged in G1 using αF, and then released from G1 into medium containing nocodazole to arrest the cells in mitosis. For the centromere excision experiments (B–D), the cultures were divided in half after G1 arrest, and one half of each culture was treated with galactose for 2 h to induce centromere excision (see Materials and Methods ). Both the induced (acentric) and uninduced (centric) control cultures were then released from the G1 arrest into fresh medium and rearrested in mitosis. Once arrested in mitosis, cells were fixed in formaldehyde and then processed for ChIP using antiserum against epitope-tagged Mcd1p (Mcd1-6HAp) as an indicator of the cohesin complex. DNA isolated from the ChIPs and diluted input DNA not subject to immunoprecipitation were then subjected to PCR analysis using oligonucleotide primer pairs that amplify approximately 300-bp fragments within the indicated regions. Quantitation of DNA in the Mcd1p ChIPs, expressed as a percentage of the input DNA, is plotted as a function of the locations of the midpoints of those DNA fragments based on the SGD coordinates. Centromere position is indicated by an oval (not drawn to scale). (A) The Mcd1p association profile for the CHRI pericentric region in strain 1377A1-4B is shown. Mcd1p binding adjacent to CEN1 is difficult to assess fully because of the presence of a moderately repetitive Ty element in the region from approximately 160 to 166 kb, indicated with the dashed line. Similarly, the Mcd1p binding profiles in the pericentric regions of CHRIII (B) and CHRXIV (C) are shown in the presence (black squares) and absence (gray circles) of CEN3 and CEN14 using strains PMY185 and PMY206, respectively. (D) The Mcd1p binding profiles for a centromere-distal region of CHRIII are shown for comparison in the presence (black squares) and absence (gray circles) of CEN3 . Figure 4 Shearing of Centromere-Proximal Chromatin by Sonication As a control for the shearing of chromatin, a precipitation of chromatin was performed in each experiment using a polyclonal antiserum specific for the kinetochore protein Mif2p, which has been shown to interact with centromeric DNA ( Meluh and Koshland 1997 ). DNA crosslinked to Mif2 was then subjected to PCR analysis throughout a 2-kb region spanning CEN3, using a series of primer pairs that amplify 240 ± 21–bp fragments. CEN3 DNA spans SGD coordinates 114382 to 114498, indicated with the ovals. Data were plotted on a scale similar to cohesin subunit ChIP data for comparison, and the inset shows in detail the magnitude of binding within a 2-kb centromere-flanking region. While all pericentric regions were enriched for Mcd1p and Smc3p binding, we found that the pattern of cohesin subunit association was different within each pericentric region ( Figure 3 A– 3 C). For example, the CHRIII pericentric region had its highest levels of Mcd1p association within an approximately 2-kb region spanning the centromere, and this region was flanked symmetrically by peaks of Mcd1p association of lesser magnitude, located approximately 15 kb from the centromere. In contrast, the CHRI centromere was located in a local trough of Mcd1p association and was flanked by peaks of Mcd1p binding that were approximately 5 kb away. Furthermore, the pericentric region of CHRXIV contained a series of closely spaced peaks of Mcd1p association that were roughly equivalent in magnitude. Thus, although all pericentric regions were indeed enriched for cohesin association, the uniqueness of the distribution of cohesin within each pericentric region suggests that cohesin binding is influenced by local sequence characteristics. Enhancement of Pericentric Cohesin Binding Is Mediated by the Kinetochore Although pericentric cohesin recruitment in S. pombe likely occurs through an interaction with the heterochromatin constituent HP1 bound to repetitive DNA sequences, the absence of repetitive DNA flanking budding yeast centromeres indicates that some other mechanism is used for the recruitment of cohesin throughout large pericentric domains. One possibility is that the high density of cohesin within pericentric chromatin is dependent on the centromere–kinetochore complex. To test this possibility, we examined Mcd1p association in the centromere-flanking regions of chromosomes in the presence and absence of the centromere. For these experiments, the endogenous centromeres on CHRIII and CHRXIV were replaced by CEN3 and CEN14 sequences, respectively, flanked by site-specific recombination target sites for the R recombinase from Zygosaccharomyces rouxii ( Materials and Methods ). The 120-bp centromeric DNA and approximately 200 bp of flanking sequences, corresponding roughly to the nuclease-resistant region spanning budding yeast centromeres ( Bloom and Carbon 1982 ), were then excised from the chromosome in G1 cells by activating the expression of a galactose-inducible R recombinase, resulting in the generation of an acentric chromosome. The absence of Mcd1 protein in G1-staged cells and the interdependency of cohesin subunit association with chromosomes suggest that centromere excision occurs prior to the loading of the cohesin complex onto chromosomes in our experimental regimen ( Guacci et al. 1997 ; Toth et al. 1999 ). After centromere excision (≥95% efficiency; Figure 5 ), the cells were released from the G1 arrest, rearrested in mitosis using nocodazole, and then processed for ChIP to assess Mcd1p binding. Figure 5 Centromere Excision (A) CEN1 on CHRI was replaced with a CEN3-URA3 cassette flanked by head-to-tail-oriented site-specific recombination target sites (red arrows) for the R recombinase from Zygosaccharomyces rouxii, as described in Materials and Methods . This strain (1824-23B) contained the R recombinase under the control of a galactose-inducible promoter. Genomic DNA samples, taken prior to the addition of galactose to the culture medium (0) and at 0.5-h intervals for 4.5 h after the galactose addition, were digested to completion with PvuII (black arrows) and analyzed by Southern blot analysis using a 1.25-kb probe corresponding to CHRI SGD coordinates 151823 to 153080. (B) The percentage of centromere excision was determined for the timecourse shown in (A). Briefly, a phosphorimage of the Southern blot and ImageQuant software were used to determine the pixel intensities of the unexcised and excised bands (top and bottom bands, respectively). The percent excision was then calculated as the pixel intensity present in the excised band divided by the total pixel intensities of both bands at each timepoint. (C) A Southern blot analysis of centromere excision from CHRIII. The endogenous CEN3 on CHRIII was replaced by R-recombinase target-site-flanked CEN3 in strain 1829-15B, as described in Materials and Methods . The efficiency of centromere excision from CHRIII was determined by Southern blot analysis in two independent experiments using genomic DNA samples digested with SnaBI and a probe corresponding to CHRIII SGD coordinates 113799-114336. Lanes 1 and 3 represent uninduced controls, and lanes 2 and 4 represent the extent of centromere excision after 2 h of recombinase induction. The percent excision was determined as in (B). In nocodazole-arrested cells, the magnitude of Mcd1p binding within the approximately 50-kb region flanking the site of the excised CEN3 was reduced significantly compared to control cells that retained the centromere (see Figure 3 B). This reduction in Mcd1p binding occurred symmetrically throughout the entire 50-kb pericentric region flanking CEN3 . In contrast, the magnitude of Mcd1p binding within a centromere-distal location on the arm of CHRIII was similar in both the centric and acentric chromosomes, indicating that cohesin association on chromosome arms is unaffected by centromere excision (see Figure 3 D). Furthermore, the magnitude of Mcd1p association within the pericentric region of an endogenous chromosome (CHRI) that did not undergo centromere excision was also unaltered (unpublished data). In agreement with the results for CHRIII, we also observed a symmetrical reduction in Mcd1p binding throughout an approximately 50-kb pericentric region on an acentric CHRXIV when compared to control cells that retained CEN14 (see Figure 3 C). While the magnitude of Mcd1p association was dramatically reduced in the former pericentric regions following centromere excision, the relative positions of the peaks and troughs of Mcd1p binding were unaltered. These results suggest that the centromere is required for the enrichment of Mcd1p binding in pericentric regions, where it amplifies an intrinsic local pattern of Mcd1p association. This amplification appears to occur bidirectionally, even though the DNA and protein components of the centromere–kinetochore complex are inherently asymmetric ( Espelin et al. 1997 ). In addition, the loss of pericentric cohesin binding upon centromere excision occurred despite the presence of centromeres on the remaining chromosomes. Thus, the enhancement of cohesin binding in pericentric chromatin requires a centromere in cis . These observations suggest that the budding yeast centromere and its associated factors together behave as a bidirectional enhancer to increase cohesin binding throughout large pericentric regions in every cell cycle. While the removal of the centromere by site-specific recombination greatly reduced the levels of cohesin bound within pericentric DNA, we noted that the levels of binding within the valleys of the pericentric regions remained higher than the valleys observed in centromere-distal regions (see Figure 3 B and 3 D). This observation suggested that pericentric sequences might contribute to the enhancement of cohesin association independent of the centromere. However, when the CHRIII centromere was moved to an ectopic location on the right arm of the chromosome, the residual levels of Mcd1p binding within the troughs throughout the former pericentric region were further reduced ( Figure 6 ). In fact, this region now more closely resembled typical arm cohesin-association sites, where peaks of binding occur at approximately 10-kb intervals and are separated by regions with minimal or undetectable levels of cohesin association. Thus, this result suggests that cohesin enrichment within pericentric regions is mediated exclusively by the centromere–kinetochore complex and that a centromere-independent pathway does not contribute to the enhanced levels of cohesin binding in pericentric chromatin. The residual levels of cohesin that remain bound in the troughs immediately following centromere excision may reflect a difference in the timing of centromere loss. In the centromere excision experiment, Mcd1p association was examined during mitosis of the same cell cycle in which the centromere was removed, whereas in the ectopic centromere strain, Mcd1p association was examined many generations after centromere removal. Thus, the persistence of cohesin binding during the first cell cycle following centromere excision suggests the existence of an epigenetic component in the recruitment of cohesin to pericentric chromatin. Indeed, possible epigenetic contributions to kinetochore function have been suggested previously in budding yeast ( Mythreye and Bloom 2003 ). Figure 6 Mcd1p Binding within the Endogenous CHRIII Pericentric Region after Centromere Excision or Centromere Movement The Mcd1p binding profiles in the endogenous CHRIII pericentric region are shown in cells in which the centromere is absent, either because of centromere excision (gray circles, PMY185) or because of the movement of the centromere to an ectopic location on the right arm of CHRIII (black triangles, PMY318). PMY185 and PMY318 are highly related strains; PMY185 was one of the parental strains used to generate PMY318. In the centromere excision strain, Mcd1p binding was examined in the same cell cycle in which the centromere was lost, whereas in the ectopic centromere strain, Mcd1p association was determined many generations after centromere relocation (see text for further discussion). Mcd1p binding data from the centromere excision experiment are the same as those shown in Figure 2 B, but are replotted here for clarity. CEN3 normally occupies the interval between SGD coordinates 114382-114498. While the 120-bp centromeric DNA was the only conserved DNA sequence present within the excised regions of the two chromosomes, it was possible that some unidentified motif within the excised DNA was instead responsible for the enhancement of pericentric cohesin binding. To rule out this possibility, we tested whether enhancer activity was mediated specifically by the centromere and its associated factors by examining pericentric cohesin binding in cells lacking functional kinetochores. Kinetochore assembly was disrupted using a conditional mutant in the NDC10 gene. NDC10 encodes an essential subunit of CBF3, a complex of kinetochore proteins that binds to the conserved centromere DNA element CDEIII and nucleates kinetochore assembly ( Goh and Kilmartin 1993 ; Jiang et al. 1993 ). At the restrictive temperature of 37 °C, ndc10-42 mutants assemble a defective kinetochore, and consequently, arrest in G2/M due to the activation of the spindle assembly checkpoint ( Doheny et al. 1993 ). Cultures of the ndc10-42 mutant and an isogenic wild-type control strain were staged in G1 and then released from the G1 arrest at the restrictive temperature in medium containing nocodazole. After reaching a mitotic arrest, both the mutant and wild-type cultures were processed for ChIP to assess Mcd1p association in pericentric regions. We observed that Mcd1p binding was reduced 5-fold on average throughout an approximately 50-kb CHRIII pericentric region in the ndc10-42 cells at the restrictive temperature when compared to the isogenic wild-type control ( Figure 7 A). In fact, Mcd1p association was reduced throughout the same region that was affected by centromere excision. Similarly, Mcd1p binding throughout an approximately 40-kb pericentric region of CHRI was also reduced approximately 5-fold in the ndc10-42 mutant when placed at the restrictive temperature ( Figure 7 B). These results differ from those of a previous study which reported no change in cohesin association at an established endogenous centromere in ndc10-1 cells ( Tanaka et al. 1999 ). This difference is likely explained by the fact that the previous study examined Mcd1p binding within only one approximately 300-bp centromere-spanning region, whereas we examined Mcd1p association at multiple locations throughout 40-kb pericentric regions. To determine the extent to which kinetochore inactivation affected cohesin association at more centromere-distal locations, we examined global Mcd1p association in the ndc10-42 mutant at the restrictive temperature using the hybridization of ChIP DNA to microarrays. Consistent with the PCR quantitation of CHRI and CHRIII, we found that the magnitude of Mcd1p binding in the pericentric regions of all chromosomes was indeed reduced in the ndc10-42 mutant at the restrictive temperature when compared to the isogenic wild-type strain ( Figure 7 C; for brevity, only CHRV, CHRVI, and CHRX are shown). However, Mcd1p binding at centromere-distal locations was equivalent in wild type and ndc10-42 mutant cells, indicating that cohesin association in these regions is independent of the centromere–kinetochore complex ( Figure 7 C). These observations demonstrate that the centromere– kinetochore complex can increase the magnitude of cohesin association bidirectionally from the centromere over regions as large as 25 kb, thereby generating approximately 50-kb pericentric domains that are highly enriched for cohesin binding. Figure 7 A Functional Centromere–Kinetochore Complex Is Essential for Enhanced Pericentric Cohesin Association Cultures of isogenic wild-type (1846-15A) and ndc10-42 mutant (1846-15C) cells were arrested in αF at 23 °C and then released into fresh medium containing nocodazole at 37 °C. After the cells arrested in mitosis (approximately 3 h), the cultures were crosslinked with formaldehyde and processed for ChIP using a monoclonal antiserum against epitope-tagged Mcd1p (Mcd1-6HAp) as an indicator of the cohesin complex. The cohesin association profiles in the pericentric regions of CHRIII (A) and CHRI (B) are shown for NDC10 (black squares) and ndc10-42 (gray circles) cultures. The positions of the centromeres are indicated by ovals (not drawn to scale). The dashed line in (B) indicates a region containing a Ty element. (C) To identify chromosomal regions depleted for cohesin binding in the absence of a functional kinetochore, the Mcd1p-ChIP-to-input fluorescence ratio obtained for each ORF and intergenic region in genomic microarray analyses of CHRV, CHRVI, and CHRIX in ndc10-42 cells was divided by the ratio obtained for NDC10 cells and plotted on a map of the chromosomes. Regions that demonstrated 2.5-fold or greater reduction in Mcd1p binding in the ndc10-42 mutant are shaded dark green, while lighter green hues represent further fold reductions in Mcd1p binding. Regions where the magnitude of Mcd1p binding was similar in NDC10 and ndc10-42 cells are shown in gray. Gaps in the chromosomal maps are genomic regions not represented on the microarrays, while regions shaded blue were present on the arrays but gave no data during hybridizations for reasons described in Materials and Methods . The location of the centromere on each chromosome is indicated by an asterisk. Centromeric Enhancer Activity Is Context- and Orientation-Independent Our observations suggested that kinetochores mediate the enhancement of cohesin binding within pericentric regions, but that the distribution of cohesin-binding peaks and valleys is an intrinsic property of the flanking DNA. If correct, we reasoned that the distribution of cohesin within pericentric DNA would be unaltered by the replacement of centromeric DNA with centromeric sequences from a different chromosome. To test this hypothesis directly, we removed CEN1 and approximately 440 bp of flanking sequences from CHRI and replaced it with approximately 320 bp of pericentric DNA from CHRIII that contained CEN3 . The patterns of Mcd1p association within the pericentric regions of cells containing the altered or endogenous CHRI were then determined by ChIP in nocodazole-arrested cells. We observed that both the distribution and the magnitude of Mcd1p binding within the pericentric region of the modified CHRI were similar to those observed on the wild-type CHRI ( Figure 8 , compare with Figure 3 A). Furthermore, the excision of CEN3 sequences from CHRI using the same experimental regimen described above also resulted in a bidirectional reduction in Mcd1p association in the regions flanking the centromere at both high- and low-affinity regions, as observed previously for CHRIII and CHRXIV ( Figure 8 ). The finding that the magnitude of Mcd1p binding was equivalent within the altered and endogenous CHRI pericentric regions suggested that centromeric enhancers have similar abilities to mediate cohesin association within other pericentric regions. Furthermore, the replacement of CEN1 with CEN3 sequences was done in such a way that CEN3 was present in the opposite orientation with respect to the endogenous centromere, and the context of CEN3 within the CHRI pericentric region was further modified by the introduction of the URA3 gene immediately adjacent to the centromere. Thus, these results demonstrate that the centromeric enhancer can function in an altered chromosomal context and that the enhancement of cohesin binding in pericentric DNA is independent of both the primary sequence of the centromere and its orientation with respect to the pericentric sequences. Figure 8 Centromeric Enhancer Activity Is Context and Orientation-Independent Cells containing an endogenous CHRI (1377A1-4B) and those in which CEN1 was replaced with CEN3 marked with URA3 (1824-23B), as described in Materials and Methods , were staged in G1 using αF and then released into fresh medium containing nocodazole. In the case of strain 1824-23B, the G1-arrested culture was split in half, and one half was treated with galactose to induce excision of CEN3 from CHRI prior to release into medium containing nocodazole. After reaching a mitotic arrest, the cultures were crosslinked with formaldehyde and processed for ChIP using a monoclonal antiserum against epitope-tagged Mcd1p (Mcd1-6HAp). The cohesin association profiles for the modified CHRI with and without CEN3 are shown (squares and circles, respectively). The position of the centromere is indicated by the oval (not drawn to scale). The dashed line indicates the region containing a Ty element. See Figure 2 A for the Mcd1p association profile of the endogenous CHRI. Centromeric Enhancer Is Active at an Ectopic Location The importance of pericentric cohesion in the promotion of sister kinetochore biorientation may have maintained an evolutionary selection for pericentric sequences that favors higher levels of cohesin association. Consequently, these sequences may be particularly susceptible to centromeric enhancer activity. To test whether naive sequences that have never resided near a centromere can also respond to the presence of the centromere–kinetochore complex with increased cohesin association, we determined whether the movement of centromeric DNA to an ectopic location resulted in increased Mcd1p association in the flanking chromatin. In this experiment, Mcd1p association was examined within an approximately 37-kb region on the right arm of CHRIII spanning SGD coordinates 242–279 kb after the insertion of CEN6 DNA at SGD coordinate 260 kb. Endogenous CEN3 sequences were removed and the new centromere was inserted concurrently to prevent the production of a dicentric CHRIII ( Materials and Methods ). Cultures of cells containing the ectopic centromere or an isogenic control strain with the centromere at its endogenous location were staged in G1, and then released into fresh media containing nocodazole to arrest the cells in mitosis. Once the cells were arrested, they were processed for ChIP. Quantitation of the levels of Mcd1p-associated sequences in the wild-type cells revealed a major peak of Mcd1p binding at SGD coordinate approximately 277 kb and three minor peaks at 248 kb, 261 kb, and 272 kb ( Figure 9 A). In cells containing the ectopically placed centromere, the peaks of Mcd1p binding occurred at the same locations as those observed in the wild-type cells, but the magnitude of binding was significantly higher, most notably within the peaks at 248 kb and 273 kb. Moreover, the levels of Mcd1p binding in the troughs proximal to the ectopic centromere were also dramatically elevated. Indeed, when the fold increases in cohesin binding flanking the ectopic centromere were plotted, the analysis revealed that similar levels of enhancement are reached throughout the approximately 37-kb region examined (median fold increase of 6.0 ± 2.1), with the majority of the increase occurring in the trough regions ( Figure 9 B). Thus, the insertion of the centromeric enhancer resulted in the generation of a cohesin domain (≥37 kb) similar in size to that mediated by an endogenous centromere–kinetochore complex. In addition, cohesin binding at the ectopic location was not random, but instead, appeared to amplify the intrinsic pattern of association. Taken together, these observations indicate that the ability of the centromere to enhance cohesin binding in flanking DNA is not limited to endogenous pericentric sequences. Figure 9 The Centromeric Enhancer Is Active at an Ectopic Location The endogenous centromere on CHRIII was removed, and CEN6 was inserted at an ectopic location (SGD coordinate approximately 260 kb), producing yeast strain PMY318, as described in Materials and Methods . Cells containing the ectopic centromere and isogenic wild-type cells (1829-15B) were staged in G1 using αF, and then released into fresh medium containing nocodazole to arrest cells in mitosis. Cells were then fixed in formaldehyde and processed for ChIP using epitope-tagged Mcd1-6HAp as a marker for the cohesin complex. (A) The Mcd1p binding profiles at the ectopic location on endogenous CHRIII (black squares) and in the presence of the ectopic centromere (gray circles) are shown. The location of the ectopic centromere is indicated by the black oval. (B) The levels of Mcd1p binding in the region flanking the ectopically placed centromere were divided by those observed in the isogenic wild-type control strain to determine the fold increases in Mcd1p binding in the presence of the centromere. Data are plotted as a function of the SGD coordinates for this region. Discussion In this report we show that the approximately 120-bp point centromere of budding yeast increases the magnitude of cohesin association within large approximately 20–50-kb pericentric regions. The budding yeast centromere–kinetochore complex generates a specialized chromatin structure, consisting of an approximately 250-bp nuclease-resistant region flanked by several positioned nucleosomes that together span approximately 3 kb ( Bloom and Carbon 1982 ). Thus, the centromere-flanking cohesin domains are approximately 80–200 times larger than the nuclease-resistant region of the kinetochore. The ability of kinetochores to mediate increased cohesin association over large domains was not limited to endogenous pericentric sequences, but also occurred when the centromere was moved to an ectopic location. Although rare, other examples exist in which cis DNA elements have been shown to mediate the generation of large chromosomal domains, namely telomeric silencing and X chromosome inactivation ( Renauld et al. 1993 ; Hecht et al. 1996 ; Lee et al. 1999 ). Finally, the kinetochore enhanced cohesin association in pericentric regions only in cis and in an orientation-independent manner. Thus, the kinetochore functions as an enhancer of pericentric cohesin binding in addition to mediating the attachment to the mitotic spindle. The kinetochore-dependent generation of these extended pericentric cohesin domains may be the consequence of kinetochore-mediated de novo loading of cohesin, or, alternatively, the domains may reflect a role for the kinetochore in maintaining or protecting cohesin association in centromere-flanking regions. Evidence in support of both models exists. Cohesin association in the centromere-flanking region of a minichromosome was reduced upon centromere excision in M phase–arrested cells ( Megee et al. 1999 ). In addition, high levels of cohesin association were observed in the CHRIII pericentric region of cells arrested in S phase or mitosis, while cohesin binding along the arms was lower in mitotically arrested cells compared to cells arrested in S phase ( Blat and Kleckner 1999 ). These observations are consistent with a role for the kinetochore in maintaining cohesin association in pericentric regions. However, the elevated levels of pericentric cohesin binding were not identical in the S phase– and M phase–arrested cells, but were in fact higher in the mitotically arrested cells ( Blat and Kleckner 1999 ). Furthermore, we have found that the magnitude of cohesin binding within pericentric regions of mitotically arrested cells increases in response to environmental cues (P. Megee, unpublished data). These observations are consistent with the de novo loading of cohesins by the centromere–kinetochore complex. Thus, the kinetochore may mediate increased cohesin binding by multiple pathways. Our finding that kinetochores mediate cohesin binding throughout large approximately 50-kb pericentric domains may potentially reconcile seemingly paradoxical observations concerning the enrichment of cohesin in pericentric chromatin and the transient loss of cohesion between sister chromatids in these regions. Despite the presence of these large pericentric cohesin domains, sister chromatids undergo transient separations within pericentric regions shortly after the formation of bipolar spindle microtubule attachments ( Goshima and Yanagida 2000 ; He et al. 2000 ; Tanaka et al. 2000 ). However, the extent of sister chromatid separation observed in these studies is consistent with the deformation of only approximately 20 kb of pericentric DNA ( He et al. 2000 , 2001 ). Thus, our results indicate that pericentric cohesin domains extend well beyond the approximately 20 kb of chromatin undergoing significant microtubule-dependent stretching. These observations suggest that pericentric cohesion may play dual roles in the maintenance of genomic integrity. First, cohesin bound immediately adjacent to the kinetochore may sterically constrain the kinetochores on paired sister chromatids to face opposite poles, thereby facilitating the establishment of chromosome biorientation. Once biorientation is achieved, the poleward forces imposed by the microtubule attachments disrupt cohesin binding within approximately 20-kb pericentric regions, giving rise to transient sister chromatid separation. Because the kinetochore-mediated pericentric cohesin domain extends beyond the region of stretched chromatin, this domain may provide the resistance that is required for the production of tension between sister kinetochores ( Skibbens et al. 1995 ). This tension is thought to promote the stability of kinetochore–microtubule attachments ( Nicklas and Ward 1994 ). Our results also demonstrate that the distribution of cohesin within each pericentric domain differs between chromosomes and that the observed pattern is unrelated to the specific centromeric DNA sequence present on the chromosome. Instead, the distribution of cohesin in either endogenous or ectopic centromere-proximal locations is specified by an intrinsic property of the chromatin. Furthermore, although the ability of the centromeric enhancer to mediate cohesin binding extended over large domains, the loss of the kinetochore did not affect cohesin binding in centromere-distal regions (this study). This observation is consistent with two alternative explanations. First, the centromeric enhancer may form a gradient of activity that dissipates with distance from the kinetochore. This model is unlikely, however, since the fold increases in cohesin association adjacent to the ectopic centromere were similar throughout the flanking approximately 37-kb region. Alternatively, the length of pericentric chromatin that can be influenced by the centromeric enhancer may be constrained by cis factors, such as boundary elements or sequences nonpermissive for cohesin binding. Such a model is supported by the relatively small pericentric cohesin domain present on minichromosomes, where local sequence context was suggested to affect cohesin binding adjacent to the centromere ( Megee et al. 1999 ). Interestingly, during meiosis cohesins are removed from chromosome arms but remain bound in the large pericentric regions ( Klein et al. 1999 ; Watanabe and Nurse 1999 ). It is possible that the boundaries that dictate the pericentric region and constrain the centromeric enhancer are the same that modulate cohesin binding during meiosis. Moreover, since the distribution of cohesin association in pericentric regions appears to be unique to each chromosome, this variability in cohesin distribution may provide a basis for the range of nondisjunction frequencies associated with different chromosomes within an organism ( Campbell et al. 1981 ). In this report we have provided insights into the control of cohesin binding within pericentric chromatin in budding yeast. We have demonstrated that the centromere–kinetochore complex behaves as an enhancer for cohesin association in pericentric chromatin and appears to be largely responsible for the increased levels of pericentric cohesin association at both endogenous and ectopic locations. Presumably, the enhancer functions by activating a trans factor that either recruits cohesin to pericentric chromatin or maintains high levels of pericentric cohesin binding. One candidate for such a trans factor could be a histone-modifying enzyme. The centromeric enhancer would augment the recruitment of the histone-modifying enzyme and enhance cohesin association. Indeed, results from S. pombe indicate that a histone methyltransferase is targeted to centromere-proximal heterochromatin, and this modification is important for cohesin recruitment ( Bernard et al. 2001 ; Nonaka et al. 2002 ). Since budding yeast lacks centromere-proximal heterochromatin, the targeting method for pericentric cohesin recruitment is likely to be different. However, since pericentric cohesin domains are highly conserved, we suspect that the enhancer activity of the kinetochore may well be ubiquitous in all eukaryotes. Our results suggest that the coordination of microtubule attachment and pericentric cohesin recruitment by the budding yeast kinetochore generates an autonomous segregation unit that ensures sister kinetochore biorientation and, consequently, the maintenance of genomic integrity. The integration of these activities by the kinetochore was likely key for the first identification of budding yeast centromeric DNA, since microtubule attachment in the absence of pericentric cohesion is unlikely to have increased the mitotic stability of minichromosomes, the assay used for centromere DNA identification ( Clarke and Carbon 1980 ). Similarly, the coordination of cohesion and microtubule attachment by the kinetochore may also explain how neocentromeres in humans acquire full chromosome segregation capabilities in the absence of flanking repetitive DNA or cytologically detectable levels of pericentric heterochromatin ( Aagaard et al. 2000 ; Saffery et al. 2000 ; Amor and Choo 2002 ). Our observations suggest that kinetochore-mediated cohesin recruitment may compensate for the lack of heterochromatin-dependent cohesin recruitment, thereby promoting biorientation of sister kinetochores. Moreover, the ability of a neocentromere to generate new domains of cohesin binding is likely to have been instrumental for chromosome evolution by allowing some degree of flexibility for chromosomal rearrangements. Materials and Methods Yeast cell culture The genotypes of the yeast strains used in this study are listed in Table 1 . Cultures used for ChIP were synchronized in G1 using α-factor (αF) mating pheromone at final concentrations of 3 μM and 15 nM for BAR1 wild-type and mutant strains, respectively. Cells were then released from the G1 arrest by two washes in the appropriate growth medium containing 0.1 mg/ml Pronase (Sigma, St. Louis, Missouri, United States) to proteolyze the αF. Cells were then allowed to grow in medium containing 0.1 mg/ml Pronase and then arrested in mitosis using either 15 μg/ml nocodazole (Sigma) resuspended in 1% DMSO, or a temperature-sensitive cdc16 mutation, as indicated. Mitotic arrest, as determined by a large-budded cell morphology, was generally reached after 2.5–3 h of growth. The R-site-specific recombinase from Zygosaccharomyces rouxii used for centromere excision events was induced in G1 cells by the addition of galactose (2% final concentration) to rich medium containing 2% raffinose. Table 1 Saccharomyces cerevisiae Strains a Indicated strains contain two tandem copies of the Zygosaccharomyces rouxii recombinase under the control of the galactose promoter integrated at LEU2. b R signifies an R-recombinase target site c Strain background Centromere excision and strain construction Centromeres were excised from endogenous chromosomes by site-specific recombination events using a galactose-inducible R recombinase from Zygosaccharomyces rouxii ( Matsuzaki et al. 1990 ). A 320-bp BamHI CEN3 fragment was inserted between head-to-tail-oriented recombination target sites, as described previously ( Megee and Koshland 1999 ). URA3 was subcloned adjacent to the CEN3 centromere cassette outside the recombination target sites, and this construct replaced the endogenous centromere on CHRIII by a one-step replacement ( Rothstein 1983 ), selecting for uracil prototrophs. To restore the chromosomal context of the centromere on CHRIII, the one-step replacement was then repeated with a centromere cassette lacking URA3, and transformants were grown on 5-fluoroorotic acid to screen for the loss of URA3 . The constructs were then confirmed by Southern analysis of genomic DNA. Similarly, a 334-bp PCR fragment (SGD coordinates 628717-629051) containing CEN14 and approximately 200 bp of flanking DNA was inserted between head-to-tail-oriented recombination target sites. A URA3 -marked version of this construct was used to replace the endogenous centromere on CHRXIV by one-step gene replacement, and the chromosomal context of CEN14 was then restored using the same strategy outlined above for CEN3 . The construction of the CEN3 replacement of CEN1 DNA was performed similarly, except that URA3 was placed adjacent to the centromere between the recombination target sites. Replacement of CEN1 sequences with the CEN3-URA3 cassette resulted in the deletion of 558 bp containing CEN1 (CHRI, SDG coordinates 151263-151820), and was confirmed by Southern analysis of genomic DNA (unpublished data). To obtain PMY318, a strain having the centromere at an ectopic location on CHRIII, strain 1829-15B was transformed with DNA encoding a CEN6-URA3 cassette flanked by regions homologous to the TRX3 locus. The TRX3 ORF and a small amount of flanking sequences, encompassing SGD coordinates 259521-260061, were deleted by the integration of the CEN6-URA3 cassette. Prior to transformation, 1829–15B cells were grown in rich medium containing raffinose, and then plated on medium containing galactose to induce excision of endogenous CEN3 and approximately 200 bp of the flanking region by site-specific recombination, as described above. ChIP ChIP was performed as described ( Megee et al. 1999 ). A detailed protocol is available at http://www.uchsc.edu/sm/bbgn/megee.html (see also Protocol S1 in the accompanying paper by Glynn et al. [2004] ). Immunoprecipitations were done using 12CA5 anti-HA antibody (Roche, Basel, Switzerland), A-14 anti-Myc antibody (Santa Cruz Biotechnology, Santa Cruz, California, United States), or rabbit polyclonal anti-Mif2p antibody ( Meluh and Koshland 1997 ), as indicated. Immunoprecipitations of crosslinked chromatin prepared from strains lacking epitope-tagged versions of cohesin subunits did not precipitate centromeric DNA, as demonstrated previously ( Megee et al. 1999 ). In addition, mock immunoprecipitations were performed to exclude the possibility that centromere-flanking chromatin was nonspecifically enriched in the immunoprecipitates obtained from epitope-tagged strains. We observed that centromeric sequences were enriched over those observed in the mock immunoprecipitations at least 500-fold and 11-fold using HA-tagged or Myc-tagged cohesin subunits, respectively. This difference likely reflects the observation that for any cohesin-associated region, a larger percentage of total chromatin is precipitated with the HA epitope tag than with the Myc tag (Mcd1-6HAp compared to Mcd1-18Mycp) (P. Megee, unpublished data). Lastly, analysis of the input DNA showed that our samples were routinely sheared to an average size of 500 bp, with a range of 200–1,000 bp. In all experiments, duplicate immunoprecipitations were performed for cohesin subunits and subjected to a preliminary PCR analysis using centromere-specific or -proximal oligonucleotide primer pairs. The duplicates had values that were routinely within 10% of one another, and, thus, one sample was chosen randomly for further analysis. The linearity of PCR under our experimental conditions was tested empirically. Briefly, input DNA was diluted 90-fold relative to immunoprecipitated DNA before PCR analysis. Increasing amounts of input DNA were then used to program PCR reactions, and the resulting products were quantitated to determine whether the amount of product responded linearly to the levels of input DNA (unpublished data; Megee et al. 1999 ). This relationship was determined empirically in experiments initiated with approximately 1.65 × 10 9 cells, and all subsequent experiments were performed with the same cell number to maintain linearity of PCR. PCR fragments were separated on 2.5% NuSieve (Cambrex, East Rutherford, New Jersey, United States) gels containing 0.15 μg/ml ethidium bromide. Digital images of ethidium bromide–stained gels were quantitated using ImageQuant software (Molecular Dynamics, Sunnyvale, California, United States). The details of oligonucleotide primer pairs used for PCR analyses are available upon request. In general, PCR primer pairs amplified approximately 300-bp sequences and were separated from neighboring pairs by an average of approximately 1.5 kb. In the CHRIII pericentric region, the mean size of the 37 PCR products used in the ChIP analysis was 304 ± 68 bp. For CHRXIV, the mean size of the 41 PCR products used in the analysis was 309 ± 36 bp. In the CHRI pericentric region, the mean size of the 22 PCR products used in the analysis was 306 ± 31 bp. ChIP experiments were performed at least twice, and representative data from one experiment are presented. Microarray analyses of ChIP DNA Preparation of Cy5- and Cy3-labeled DNA, hybridization, and analysis were performed as previously described ( Gerton et al. 2000 ). Briefly, the recovered ChIP DNA was randomly PCR amplified in the presence of amino-allyl dUTP ( Bohlander et al. 1992 ; Gerton et al. 2000 ; see also Protocol S2 in the accompanying paper by Glynn et al. [2004] ), which was then coupled to a fluorescent dye (e.g., Cy5) and competitively hybridized to a polyL-lysine-coated spotted glass DNA microarray in the presence of total genomic DNA similarly labeled with a second fluorescent dye (e.g., Cy3). Hybridizations were performed at 63 °C overnight under standard conditions and slides were washed successively with 0.6X SSC/0.03% SDS and then 0.06X SSC prior to scanning (see also http://microarrays.org ). The fluorescence in each spot on the microarray was detected using an Axon 4000B laser scanner/detector (Axon Instruments, Union City, California, United States), and the ratio of the two signals was determined using GenePix 4.0 software (Axon Instruments). This ratio indicates the enrichment for a given sequence in the ChIP. The AMAD database was used to normalize and store microarray data. The microarrays used in this study contained all the ORFs and intergenic regions in the yeast genome as individual spots ( Iyer et al. 2001 ). For each experimental condition, a minimum of two hybridizations from two independent immunoprecipitations was performed. Data were normalized and then filtered in the following ways: the spot intensity was required to be 200 or greater; flagged spots were not used; and spots were required to have a correlation coefficient of 0.5 or greater. For analysis purposes, any feature with less than two measurements was excluded, and excluded regions are colored blue in the maps of cohesin binding. The raw data and datasets for the microarrays presented herein are provided as Datasets S1–S22 , and can also be viewed at http://research.stowers-institute.org/jeg/2004/cohesin/index.html . Supporting Information Datasets S5–S22 correspond to the individual GenePix results (GPR) files for each array performed. For each dataset, the Cy3 and Cy5 channel samples and the method of mitotic arrest are listed. Dataset S1 Dataset for Smc3-6Myc ChIPs Performed in cdc16 -Arrested Cells File cdc16_Smc3-6Myc_A364a. (701 KB TXT). Click here for additional data file. Dataset S2 Dataset for Mcd1-6HA ChIPs Performed in cdc16 -Arrested Cells File cdc16_Mcd1-6HA_A364a. (957 KB TXT). Click here for additional data file. Dataset S3 Dataset for Mcd1-6HA ChIPs Performed in Nocodazole-Arrested NDC10 Cells File Mcd1-6HA_S288c_NZ. (449 KB TXT). Click here for additional data file. Dataset S4 Dataset for Mcd1-6HA ChIPs Performed in Nocodazole-Arrested ndc10-42 Cells File Mcd1-6HA_ncd10-42_S288c_NZ. (418 KB TXT). Click here for additional data file. Dataset S5 SIMRUP2_147 Cy3 = cdc16 -ts Mcd1-6HA ChIP in A364a; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S6 SIMRUP2_170 Cy3 = cdc16 -ts Mcd1-6HA ChIP in A364a; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S7 SIMRUP2_171 Cy3 = cdc16 -ts Mcd1-6HA ChIP in A364a; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S8 SIMRUP2_178 Cy3 = genomic DNA; Cy5 = cdc16 -ts Mcd1-6HA ChIP in A364a. (4.6 MB XLS). Click here for additional data file. Dataset S9 SIMRUP2_180 Cy3 = genomic DNA; Cy5 = cdc16 -ts Mcd1-6HA ChIP in A364a. (4.6 MB XLS). Click here for additional data file. Dataset S10 SIMRUP2_183 Cy3 = Mcd1-6HA ChIP in nocodazole-arrested S288C background; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S11 SIMRUP2_184 Cy3 = ndc10-42 Mcd1-6HA ChIP in nocodazole-arrested S288C background; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S12 SIMRUP2_185 Cy3 = Mcd1-6HA ChIP in nocodazole-arrested S288C background; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S13 SIMRUP2_187 Cy3 = cdc16 -ts Smc3-6Myc ChIP in A364a; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S14 SIMRUP2_190 Cy3 = ChIP of Mcd1-6HA in cdc16 -ts Smc3-6Myc Mcd1-6HA A364a strain; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S15 SIMRUP2_191 Cy3 = ChIP of SMC3-6Myc in cdc16 -ts SMC3-6Myc MCD1-6HA A364a strain; Cy5 = genomic DNA. (4.6 MB XLS). Click here for additional data file. Dataset S16 SIMRUP2_226 Cy3 = genomic DNA; Cy5 = cdc16 -ts Smc3-6Myc ChIP in A364a. (4.6 MB XLS). Click here for additional data file. Dataset S17 SIMRUP2_228 Cy3 = genomic DNA; Cy5 = Mcd1-6HA ChIP in nocodazole-arrested S288C background. (4.6 MB XLS). Click here for additional data file. Dataset S18 SIMRUP2_229 Cy3 = genomic DNA; Cy5 = ndc10-42 Mcd1-6HA ChIP in nocodazole-arrested S288C background. (4.6 MB XLS). Click here for additional data file. Dataset S19 SIMRUP2_230 Cy3 = genomic DNA; Cy5 = Mcd1-6HA ChIP in nocodazole-arrested S288C background. (4.6 MB XLS). Click here for additional data file. Dataset S20 SIMRUP2_231 Cy3 = genomic DNA; Cy5 = ndc10-42 Mcd1-6HA ChIP in nocodazole-arrested S288C background. (4.6 MB XLS). Click here for additional data file. Dataset S21 SIMRUP2_254 Cy3 = genomic DNA; Cy5 = ChIP of SMC3-6Myc in cdc16 -ts Smc3-6Myc Mcd1-6HA A364a strain. (4.6 MB XLS). Click here for additional data file. Dataset S22 UP6_205 Cy3 = genomic DNA; Cy5 = Mcd1-6HA ChIP in nocodazole-arrested S288C background. (4.4 MB XLS). Click here for additional data file. Accession Numbers The Saccharomyces Genome Database ( http://www.yeastgenome.org/ ) accession numbers for the genes and gene products discussed in this paper are Irr1p/Scc3p (SGDID S0001288), Mcd1/Scc1p (SGDID S0002161), Mif2p (SGDID S0001572), Pds5p (SGDID S0004681), Smc1p (SGDID S0001886), and Smc3p (SGDID S0003610). The S. pombe Genome Project ( http://www.sanger.ac.uk/Projects/S_pombe/ ) accession number for Swi6 is SPAC664.01c. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC490027.xml |
548672 | Dutch women with a low birth weight have an increased risk of myocardial infarction later in life: a case control study | Background To investigate whether low birth weight increases the risk of myocardial infarction later in life in women. Methods Nationwide population-based case-control study. Patients and controls: 152 patients with a first myocardial infarction before the age of 50 years in the Netherlands. 568 control women who had not had a myocardial infarction stratified for age, calendar year of the index event, and area of residence. Results Birth weight in the patient group was significantly lower than in control women (3214 vs. 3370 gram, mean difference -156.3 gram (95%CI -9.5 to -303.1). The odds ratio for myocardial infarction, associated with a birth weight lower than 3000 gram (20 th percentile in controls) compared to higher than 3000 gram was 1.7 (95%CI 1.1–2.7), while the odds ratio for myocardial infarction for children with a low birth weight (< 2000 g) compared to a birth weight ≥ 2000 g was 2.4 (95%CI 1.0 – 5.8). Both figures did not change after adjustment for putative confounders (age, education level, body mass index, waist-hip ratio, hypertension, diabetes, hypercholesterolemia, smoking, and family history of cardiovascular disease). Conclusions Low birth weight is associated with an increased risk of myocardial infarction before age of 50 in Dutch women. | Background Intrauterine malnutrition, as reflected by birth weight and abnormal thinness at birth, has been associated with an increased incidence of risk factors for arterial disease, i.e. hypertension, impaired glucose tolerance, diabetes and to a lesser extent hyperlipidemia and body fat distribution in adulthood [ 1 - 10 ]. This observation has become known as the 'fetal origins of adult disease' or 'Barker hypothesis', which suggests that several of the major diseases of later life, including coronary heart disease, stroke and cardiovascular death, originate in impaired intrauterine growth and development [ 11 , 12 ]. In cohort studies, Barker [ 13 - 15 ] in England and Finland, Rich-Edwards et al . [ 16 ] as part of the Nurses' Health Study in the USA and Leon et al. [ 17 ] from Uppsala in Sweden showed an inverse relationship between birth weight and the clinical endpoint ischemic heart disease. Leon et al. [ 17 ] found a significant relationship only among male singletons and adjusted their results for gestational age and socioeconomic confounding. The association was not found in a cohort study from Gothenburg [ 18 ]. Our aim was to investigate the association in a case control study among Dutch women. Methods The RATIO (Risk of Arterial Thrombosis In relation to Oral contraceptives) study is a population-based case-control study on myocardial infarction in relation to oral contraceptive use among women aged 18 to 49 years in the Netherlands [ 19 ]. An additional standardized questionnaire was sent to all 218 patients and 769 controls from whom also blood samples had been taken for determination of metabolic risk factors (diabetes and hypercholesterolemia). Questions elicited information on birth weight, waist and hip circumference and data on the menstrual cycle. For 13 women no current address could be found (12 patients, 1 control). Four women had died since the index date (2 patients, 2 controls), which was the date of the first myocardial infarction for the patients and the midyear for the controls. One hundred and fifty two patients (71%) and 568 controls (75%) responded to the questionnaire and women were asked to measure their waist and hip circumference. Body mass index (BMI) was calculated as body weight (kg) divided by height squared (m 2 ). Waist-hip-ratio was calculated as waist circumference divided by hip circumference. Multiple linear and unconditional logistic regression were used to analyze the data. Odds ratios for the relationship between birth weight and myocardial infarction were calculated and 95% confidence intervals (95%CI) were derived from the models. Birth weights were categorized according to quintiles in control women in order to investigate an association between birth weight and the risk of myocardial infarction later in life. These were <3000 g, 3000 to 3199 g, 3200 to 3499 g, 3500 to 3883 g, and >3884 g, respectively. To determine whether women with a lower birth weight had a higher risk for a myocardial infarction, patients were divided in a group with a birth weight equally or higher than 2000 g and a group with a birth weight lower 2000 g [ 20 ]. Odds ratios were adjusted for age, education level, body mass index, waist-hip ratio, hypertension, diabetes, hypercholesterolemia, smoking, and family history of cardiovascular disease, when appropriate. Interaction between low birth weight and low education level was investigated by computing a dummy variable. Results The characteristics of 152 women with myocardial infarction and 568 control women at the index date are shown in Table 1 . At the moment of completing the questionnaire, patients were aged 32–59 years (mean 50), and control women 25–60 years (mean 47). The mean body mass index was 25.1 kg/m 2 for the patients and 23.4 kg/m 2 for control women, mean difference 1.76 kg/m 2 (95%CI 1.05–2.47), p < 0.001. Ninety-seven patients (64%) and 415 (73%) controls could give their birth weight. Compared with control women, patients had a significantly lower mean birth weight (3214 vs. 3370 g, mean difference -156.3 g (95%CI -9.5 to -303.1). The odds ratio for myocardial infarction for children with a low birth weight (< 2000 g) compared to a birth weight ≥ 2000 g was 2.4 (95%CI 1.0 to 5.8). After adjustment for putative confounders (age, education level, body mass index, waist-hip ratio, hypertension, diabetes, hypercholesterolemia, smoking, and family history of cardiovascular disease) the odds ratio did not change. Odds ratios for myocardial infarction in different categories of birth weight as compared to the reference category (birth weight higher than 3884 g) were 1.3 (95%CI 0.5–3.3) for a birth weight 3500 to 3883 g, 1.4 (95%CI 0.6–3.4) for a birth weight 3200 to 3499 g, 1.7 (95%CI 0.6–5.1) for a birth weight 3000 to 3199 g, and 2.3 (95%CI 1.0–5.4) for a birth weight lower than 3000 g (Table 2 ). The risk of myocardial infarction was 6.2 fold increased (95%CI 2.7–13.9) among women with low birth weight and a low educational level compared to women with a high birth weight and a high educational level (reference category). Table 1 Characteristics of patients with a first myocardial infarction and control women Characteristic Patients (N = 152) Control women (N = 568) Age – yr (SD) 42.1 (0.5) 38.6 (0.3) Caucasian ethnicity (%) 142 (93) 538 (95) Educational level - Primary school or less (%) 83 (55) 160 (28) - Secondary school (%) 52 (34) 257 (45) - Higher education or university (%) 17 (11) 149 (26) Current smokers (%) 128 (84) 218 (39) History of hypertension (%) 35 (23) 35 (6) History of hypercholesterolemia (%) 16 (11) 14 (3) History of diabetes (%) 8 (5) 7 (1) Family history of cardiovascular disease (%) 98 (66) 194 (36) Birth weight-gram - Mean (SD) 3214 (676) 3370 (659) - Median (range) 3150 (1500–5010) 3500 (1500–5800) Body Mass Index – kg/m 2 – Mean (SD) 25.1 (0.4) 23.4 (0.2) Waist circumference – cm (SD) 89.5 (13.1) 83.0 (10.2) Waist/hip ratio (SD) 0.85 (0.006) 0.81 (0.008) Premenopausal (%) 132 (87) 480 (85) Table 2 Odds ratios (95%CI) for myocardial infarction in quintiles of birth weight as compared to the reference category Birth weight (g) Odds Ratio (95% CI) > 3884 1* 3500–3883 1.3 (0.5–3.3) 3200–3499 1.4 (0.6–3.4) 3000–3199 1.7 (0.6–3.4) < 3000 2.3 (1.0–5.4) * Reference category Discussion In this case control study we have found that women with a low birth weight had a higher risk of myocardial infarction than women with a higher birth weight. The data confirm the association found in cohort studies [ 13 - 17 ] and as such support the 'fetal origins of adult disease' or 'Barker hypothesis'. A limitation of the study is that we have no information on gestational age. However, recently it has been demonstrated that both children who had been born prematurely and children who are small for gestational age had a reduction in insulin resistance [ 21 , 22 ]. The self-report of birth weights as well as the rather high percentage of missing values for birth weight may also limit the study, but the random events among cases and controls cannot explain our results. Socio-economic factors associated with low birth weight are also associated to risk factors for arterial disease later in life. As pointed out by several others, it will be nearly impossible to disentangle these effects [ 16 , 17 ]. In the present study we confirmed an interactive effect between low birth weight and a low educational level on the risk of myocardial infarction in women. However, when we adjusted for age, education level, body mass index, waist-hip ratio, hypertension, diabetes, hypercholesterolemia, smoking, and family history of cardiovascular disease, factors of which some may partly been seen as (the result of) socio-economic/environmental and genetic factors, we still observed an association between low birth weight and a higher risk of myocardial infarction. This is in agreement with others who also found that genetic and socio-economic circumstances at birth and in adult life can not completely explain the association between low birth weight and disease late in life [ 10 , 16 , 17 , 23 - 25 ]. Among the proposed underlying biological mechanisms to explain the association is impaired endothelial development. Already at very young age, individuals with low birth weight exhibit endothelial dysfunction that persists into childhood and adult life, suggesting that endothelial dysfunction precedes the development of vascular related diseases later in life and represents the link between low birth weight and these diseases [ 26 - 32 ]. Furthermore, Smith et al . [ 33 ] found that mothers, who once gave birth to thin babies, have a higher risk of developing ischemic heart disease later in life. Therefore, these mothers seem to have, just as their children, an impaired endothelial function. If the impaired endothelial function is already manifest during the process of implantation, it might lead to inadequate development of the vasculature in the maternal part of the placenta, which enfeebles the function of the placenta resulting in low birth weight. Conclusions In conclusion, our study shows that a low birth weight (<2000 g) is associated with a 2.4 fold higher risk of myocardial infarction before the age of 50 as compared with a birth weight ≥ 2000 g. Because the risk of cardiovascular disease is known to increase with an increasing number of risk factors, women with a low birth weight should try to avoid acquired risk factors, like smoking and obesity. In addition, extra attention should be given in detecting diabetes or hypertension at a later stage in life and in detecting exaggerated growth during childhood, as these individuals seem most prone to develop disease later on in life [[ 34 - 36 ]]. Competing interests The author(s) declare that they have no competing interests. Authors' contributions BT participated in the design, execution and analysis of the RATIO- study and drafted the manuscript. KK analyzed the data and drafted the manuscript. RH collected the data and performed statistical analyses. FR initiated the study and helped to draft the manuscript. FH participated in the design of the study and the writing of the paper. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548672.xml |
529298 | Decrease of resistance to air flow with nasal strips as measured with the airflow perturbation device | Background Nasal strips are used by athletes, people who snore, and asthmatics to ease the burden of breathing. Although there are some published studies that demonstrate higher flow with nasal strips, none had directly measured the effect of the strips on nasal resistance using the airflow perturbation device (APD). The APD is an inexpensive instrument that can measure respiratory resistance based on changes in mouth pressure and rate of airflow. Method This study tested forty-seven volunteers (14 men and 33 women), ranging in age from 17 to 51. Each volunteer was instructed to breathe normally into the APD using an oronasal mask with and without nasal strips. The APD measured respiratory resistance during inhalation, exhalation, and an average of the two. Results Results of a paired mean t-test comparing nasal strip against no nasal strip were statistically significant at the p = 0.05 level. The Breathe Right™ nasal dilator strips lowered nasal resistance by an average of 0.5 cm H 2 0/Lps from an average nasal resistance of 5.5 cm H 2 0/Lps. Conclusions Nasal strips reduce nasal resistance when measured with the APD. The effect is equal during exhalation and during inhalation. | Background Nasal dilator strips (NDS) are used by athletes, people who snore, and asthmatics to ease the burden of breathing. The nasal strips are used as a mechanical means of reducing nasal airflow resistance [ 1 ]. By lowering nasal resistance, they reduce the work of breathing and the supply of oxygen into the body could increase [ 2 , 3 ]. The size of the nostril limits the amount of air entering into the body. The NDS is placed along the nasal valve of the nose. The adhesiveness of the strip binds to the creases of the nasal valve to prevent the outer wall tissue of the nose from collapsing inward during nasal breathing. This mechanism thus dilates the nose and allows more air to flow into the nose [ 3 ]. The primary effect of the NDS could be either to dilate the air passage of the nose or to stiffen the nasal wall. Either mechanism would reduce nasal resistance and allow higher flow of air, but they can be distinguished over a range of air flows. Stiffening the nasal wall would have its most profound effect at higher flows where the Bernoulli effect would decrease internal nasal pressures and tend to constrict nasal passage diameter. Air passage dilation would tend to decrease nasal resistance more uniformly over a range of air flows. Recent studies on the effectiveness of Breathe Right™ nasal strips tested participants under various rest and exercise conditions. Some found that the strips neither improve or diminish airflow [ 2 , 4 - 9 ], which contradict results found by others [ 1 , 3 , 10 - 14 ]. Various techniques were used to assess NDS effectiveness. Some measured the amount of airflow, others the area of the nostrils, and still others the nasal airflow resistance. The Airflow Perturbation Device (APD) is a small, light weight, and easy to use instrument that measures respiratory resistance [ 15 ]. A segmented rotating wheel in the air flow path changes air flow and mouth pressure as the wheel momentarily partially obstructs the flow passage (Figure 1 ). The magnitude of these perturbations depends on the resistance of the wheel and respiratory resistance. Measurement of wheel resistance allows respiratory resistance to be calculated directly (Figure 2 ). Resistance values appear on a computer screen within a minute from starting the measurement. Thereafter they are updated as they occur. Figure 1 The APD Sensor consists of a rotating wheel in the air path, a pneumotach, and pressure transducers [15] Figure 2 The APD system consists of pressure and flow transducers, analog-to-digital conversion, and a computer display of results [1]. People breathe normally through the APD. No special breathing maneuvers are required. For this reason, the APD can be used with young children, older adults, unconscious patients, and animals. Respiratory resistance can be separated into inhalation and exhalation components, and resistance can be displayed against lung volume and air flow rate. Respiratory resistance is normally measured through the mouth, with a nose clip and hands pressed against the cheeks. An oronasal mask may be used to obtain combined mouth and nose resistance, or nose resistance by itself if the mouth is closed. The APD used with an oronasal mask should be an ideal instrument to assess the effect of NDS. This study is as much a test of the capabilities of the APD as it is a study of NDS. Objectives of this study were to: 1) determine if APD measurements of respiratory resistance measured with an oronasal mask matched those with breathing through a mouthpiece, and 2) measure the effects of NDS on respiratory resistance made with the APD. This is not a clinical study. Methods This study tested forty-seven volunteers (14 men, 33 women; age 17–51 yr; height 147–188 cm; weight 38–105 kg). Some had symptoms of nasal congestion such as asthma, allergies, and snoring. A written informed consent was obtained from each subject and the protocol was approved by the University of Maryland Institutional Review Board (IRB). The nasal strips used in this study were a commercial product called "Breathe Right" (CNS, Inc., Minneapolis, MN, clear, medium/large nasal strips). According to the manufacturer's instructions, the nasal strips should be placed halfway down the nose along the nasal valve. The two end regions of the nasal strips should cover the left and right nasal creases. This study contained three phases: 1) to determine the ability of the APD to measure oral resistance using either a mouthpiece or an oronasal mask, 2) to determine the ability of the APD to measure nasal resistance, and 3) to determine the effect of nasal strips on nasal resistances. Phase I consisted of two tests that measured oral breathing resistance. In the first test, the subject's nose was occluded with two layers of Durapore surgical tape (3 M, St. Paul, MN) while the subjects were sitting in an upright position. The subject was instructed to breathe normally into a cardboard mouthpiece. The second test repeated the same procedure as the first test, except the subject was breathing into an oronasal mask (Adult Mask 4–5 + , Laerdal Medical, Wappingers Falls, NY). The subject was instructed to press their face against the mask while he/she breathed normally. The size of the mask was large enough that it contacted only the hard tissue on the bridge of the nose and did not compress the soft nasal septum. The second phase of this study consisted of a test to measure nasal breathing resistance. The subject was instructed to breathe normally through the nose with the mouth closed and with no NDS. For the third phase, the subject placed a NDS across the nasal valve on his/her nose as shown on the instructions provided by the manufacturer. In both of these tests, the oronasal mask was used. Air flow perturbations with the APD occur at a rate of about 10 per second [ 15 ]. Measurements were obtained in these experiments over approximately 100 perturbations. It has been previously found that measurements made over that time are relatively stable and reproducible [ 15 ]. Several time-averaged resistance values are displayed: resistance during inhalation, 2) resistance during exhalation, and 3) the average of inhalation and exhalation resistances. All three of these have been found to be useful. Primary comparisons for this study were made using the average respiratory resistance. Secondary comparisons in the second phase of this study investigated the effects of NDS on inhalation and exhalation respiratory resistances. Statistical comparisons were made using a paired mean t-test with significance at the p = 0.05 level. Results Subject data appear in Table 1 . Average resistance measured during mouth breathing with mouthpiece ranges from 1.90 to 5.03 cm H 2 O/Lps. In the past, average respiratory resistances for healthy adults have generally fallen in the range of 2.5 - 3.5 cm H 2 O/Lps, and such is the case here. Also, as expected, most respiratory resistance values during exhalation exceed those measured during inhalation. Table 1 Subject data for APD Measurements of Respiratory Resistance when Measured Through the Mouth and Nose. Resistances are given in cm H 2 O/Lps. Subject No. Sex Mouth Piece Mask Mouth Mask Nose Mask NDS Inh Avg Exh Inh Avg Exh Inh Avg Exh Inh Avg Exh 1 F 3.15 3.06 2.97 2.86 3.08 3.30 4.65 5.13 5.61 4.08 4.13 4.18 2 F 3.00 3.26 3.52 2.98 3.29 3.59 5.84 5.72 5.59 3.64 3.47 3.30 3 F 2.69 3.19 3.68 2.84 3.38 3.92 4.94 5.73 6.52 4.55 5.26 5.97 4 F 4.00 4.62 5.25 3.97 4.46 4.95 5.40 5.87 6.31 5.11 5.45 5.79 5 F 3.13 4.04 4.95 3.58 4.11 4.63 5.33 5.59 5.85 4.62 5.30 5.98 6 M 2.14 2.41 2.69 2.10 2.34 2.58 3.76 4.15 4.53 3.71 4.16 4.61 7 F 2.73 3.15 3.85 2.43 3.11 3.79 3.92 4.16 4.39 3.54 3.62 3.69 8 M 2.03 2.33 2.64 2.12 2.24 2.36 4.32 4.50 4.69 3.65 3.85 4.06 9 M 1.87 1.96 2.06 1.83 1.82 1.80 3.16 3.53 3.91 3.26 3.47 3.68 10 M 3.49 3.95 4.41 3.27 3.90 4.52 5.96 6.27 6.57 6.13 6.20 6.28 11 F 2.89 2.97 3.04 2.76 2.91 3.07 7.59 7.33 7.07 7.48 7.05 6.61 12 F 3.67 4.24 4.82 4.18 4.38 4.58 6.36 7.14 7.92 6.75 6.82 6.90 13 M 2.20 2.54 2.89 2.13 2.49 2.84 5.27 5.50 5.73 4.80 5.04 5.27 14 F 3.08 2.98 2.88 2.98 3.07 3.17 5.12 5.25 5.38 5.34 5.36 5.38 15 M 2.37 2.45 2.53 2.64 2.77 2.90 5.44 5.63 5.82 5.33 5.57 5.81 16 M 3.48 3.58 3.68 3.16 3.47 3.77 5.71 5.30 4.90 4.71 4.88 5.04 17 M 2.40 2.85 3.30 2.22 2.74 3.26 4.54 4.90 5.25 4.38 4.70 5.07 18 F 2.96 2.96 2.95 2.95 3.04 3.12 6.84 7.19 7.54 6.71 6.89 7.07 19 M 2.00 2.30 2.59 2.54 2.10 1.67 3.82 4.25 4.68 3.20 3.39 3.57 20 F 2.40 2.85 3.30 2.81 2.77 2.73 4.12 4.16 4.20 3.61 3.73 3.85 21 F 4.22 4.20 4.17 4.34 4.42 4.49 6.76 6.97 7.10 6.36 6.75 7.14 22 F 2.73 3.12 3.50 3.02 3.12 3.22 6.60 6.74 6.89 5.90 6.08 6.26 23 F 4.87 5.03 5.19 4.49 4.96 5.42 5.96 6.23 6.70 5.06 5.30 5.54 24 F 4.38 4.79 5.20 4.41 4.64 4.87 6.35 6.68 7.01 5.70 6.05 6.39 25 F 2.61 3.02 3.44 2.51 2.94 3.36 3.98 4.03 4.08 3.56 3.42 3.28 26 M 2.12 2.42 2.71 1.80 2.26 2.71 5.87 5.84 5.81 5.31 4.38 5.44 27 F 3.22 3.77 4.31 3.48 3.58 3.68 4.51 4.76 5.01 4.24 4.58 4.93 28 F 2.85 3.01 3.17 2.74 2.94 3.14 4.71 4.64 4.56 3.57 4.17 4.76 29 F 2.84 3.26 3.67 3.04 3.39 3.75 4.94 5.56 6.18 4.40 5.04 5.68 30 F 2.76 3.08 3.40 2.81 3.05 3.29 6.24 6.40 6.56 5.57 5.76 5.94 31 F 2.90 3.21 3.51 2.68 2.88 3.09 4.71 4.88 5.04 5.03 5.16 5.29 32 M 1.76 1.90 2.03 1.54 1.87 2.20 4.02 4.49 4.97 4.00 4.18 4.35 33 F 3.65 3.94 4.23 3.76 3.92 4.09 4.90 5.30 5.71 4.90 5.26 5.61 34 M 1.94 2.15 2.36 1.97 2.17 2.36 5.19 4.90 4.61 4.37 4.36 4.36 35 F 2.78 2.93 3.09 2.81 3.05 3.29 3.81 4.45 5.10 3.75 4.20 4.65 36 F 5.21 5.09 4.97 4.46 4.81 5.71 4.81 5.19 5.56 4.00 4.04 4.08 37 F 3.18 3.44 3.69 3.02 3.34 3.66 5.49 5.70 5.91 5.50 5.58 5.65 38 F 2.77 3.48 4.20 2.62 3.31 4.00 6.70 6.60 6.51 5.97 6.12 6.26 39 F 3.66 3.57 3.48 3.53 3.54 3.55 7.62 7.73 7.83 6.94 7.33 7.72 40 F 2.57 3.05 3.52 2.58 2.97 3.36 3.64 4.13 4.63 3.58 3.78 3.98 41 F 3.76 3.74 3.71 3.17 3.68 4.18 4.70 5.11 5.51 4.13 4.54 4.95 42 M 3.27 3.39 3.58 3.43 3.61 3.80 6.08 6.44 6.79 5.90 6.25 6.60 43 F 3.27 3.65 4.03 3.89 3.68 4.47 4.47 4.78 5.09 3.69 3.98 4.27 44 M 2.23 2.51 2.79 2.17 2.34 2.51 6.48 6.42 6.35 5.87 6.01 6.15 45 F 2.09 2.95 3.02 2.98 3.07 3.16 6.10 6.65 7.20 5.86 6.20 6.55 46 F 2.98 3.12 3.26 3.15 3.23 3.30 5.70 5.65 5.60 4.57 4.68 4.78 47 F 2.67 2.64 2.61 2.56 2.56 2.57 4.35 4.74 5.12 3.29 3.64 3.99 Average 2.96 3.24 3.51 2.96 3.21 3.48 5.24 5.50 5.74 4.80 5.00 5.25 Std dev 0.76 0.75 0.82 0.73 0.76 0.88 1.09 1.02 1.03 1.11 1.10 1.12 Breathing through the mouth into the oronasal mask yielded almost the same values. Means of values with the mouthpieces and oronasal mask are 3.24 and 3.21, respectively. The difference was not statistically significant using a paired-t test at p = 0.05. Figure 3 shows the graph of average respiratory resistance of mask vs. mouthpiece while breathing through the mouth. The graph has a slope of nearly 1.0 and an intercept of nearly 0.0, indicating a nearly perfect correspondence between the two methods of measurement. Both slope and intercept were tested statistically and the line was found to be identical to y = x at the p = 0.05 level. Comparison of inhalation resistance between mouthpiece and oronasal mask yielded the following equation: Figure 3 Average oral respiratory resistance measured with a mouthpiece and an oronasal mask. There is almost perfect agreement between the two methods. y = 0.9624 + 0.1041 R 2 = 0.8435 (1) where y = mask value of resistance and x = mouthpiece value of resistance This equation was tested to be statistically equivalent to y = x. This indicates that the oronasal mask had no effect on the inhalation values. A similar comparison of exhalation resistances gave the following: y = 0.874bx + 0.4594 R 2 = 0.8828 (2) This equation did not pass the statistical test for equivalence y = x. The oronasal mask may have affected the measurement of respiratory resistance in the exhalation direction. Figure 4 shows the relationship of average respiratory resistance when breathing through the nose measured with and without the nasal strips. The NDS data have a slope of nearly 1.0 and a y-intercept is approximately -0.4. This signifies a reduction of nasal breathing resistance using the nasal strip. Nasal resistances with no nasal strip range from 3.53 to 7.73 cm H 2 O/Lps, while nasal resistance with the NDS ranges from 3.39 to 7.33 cm H 2 O/Lps. This demonstrates the expected resistance reduction with NDS. Figure 4 Average respiratory resistance while breathing through the nose in 47 subjects. Nasal strips showed a decrease of nasal resistance of 0.43 cm H 2 O/Lps. All subjects except three showed a decrease in nasal resistance when breathing with the NDS. Average value of nose breathing without NDS was 5.50; with NDS it was 5.00. These means were highly statistically significantly different. The effect of NDS on resistance during exhalation was also statistically highly significant. There was an average reduction of 0.45 cm H 2 O/Lps in resistance, and only six out of 47 subjects failed to demonstrate a decrease in resistance with NDS. The effect of NDS on resistance during inhalation tested to be statistically highly significant, as well. The average resistance reduction was 0.49 cm H 2 0/dps. Again, six subjects failed to demonstrate a decrease in resistance with NDS. These were not the same subjects that increased resistance in the exhalation direction. Resistance differences with and without NDS in the inhalation and exhalation directions were tested to determine if NDS had a larger effect while breathing in one direction or the other. Means of the differences for inhalation and exhalation directions were tested with a paired t-test, and found to be statistically nonsignificant. It appears, therefore, that NDS affect nasal resistances equally during inhalation and exhalation. Discussion This study confirmed the results of other studies that showed a reduction of about 10% in nasal breathing resistance, as well as supported the claim of the manufacturer that the nasal strips provide nasal relief. Several subjects who had nasal congestion reported some relief in nasal breathing when using the nasal strip. Exactly which subjects these were was not recorded. There was one surprise, though, in the results. The reduction in respiratory resistance due to NDS was a constant amount and not proportional to the resistance level present without NDS. This result was not expected, and no adequate explanation for it can be given at this time. It is not clear why this should be so, but we do not doubt that measurements made with the APD are correct, based on previous studies [ 15 , 16 ]. We cannot comment on the clinical significance of the resistance reduction with NDS use. It seems likely that some benefit could be obtained from such a resistance change, but whether it is actually detectable is not clear. Other reports in the literature [ 17 , 18 ] have concluded that the minimum detectable external resistance is about a constant 25–30% proportion of the resistance already present. The resistance change measured in this study is about 10% of the baseline resistance. If the use of NDS does result in a detectable change, then it may be that a different detection mechanism is operating. It is possible that the subject could detect nasal resistance only, rather than total respiratory resistance. Based on that supposition, NDS reduce nasal resistance by about 17%. The APD has been shown to be able to measure respiratory resistance with either a mouthpiece or an oronasal mask. This may be a significant advantage of the instrument, especially because respiratory resistance measurement on unconscious or uncooperative patients would be much more easily made with a mask than with a mouthpiece. Equations (1) and (2) show the close correspondence between measurements made with both techniques, although the presence of an intercept and a slope different from unity indicate that the correspondence between mask and mouthpiece is not perfect. Resistances with a mask are both higher than resistances with a mouthpiece. The reason for this seems to be different mouth positions in both cases. We have laboratory experiences (not published) that demonstrate that tongue position can influence measured resistance. Breathing through the mask is probably done with the mouth closed more than when breathing through the mouthpiece. The measured difference between inhalation and exhalation resistances could reflect the effect of a pressure difference across the distensible smaller airways, which is greater inside than outside during inhalation, but smaller inside than outside during exhalation. This would lead to a dynamic compression of the small intrathoracic airways during exhalation. Another possible explanation is natural movement of the vocal chords such that they are closer during exhalation than during inhalation. This study was a good test of the capabilities of the APD measuring device. Testing confirmed that the APD can detect resistance changes, and that measurements are easy to obtain. Results in this study are generally more consistent than other studies using other techniques [ 1 - 11 ]. The fact that the APD directly measures respiratory resistance, and is not an indirect measurement may be one reason for this consistency. Then, again, our subject population exhibited some homogeneity in age, social class, and racial makeup. Conclusions Nasal strips reduce nasal resistance by about 0.5 cm H 2 O/Lps. Thus, nasal strips do have a measurable effect on nasal resistance. The effect of NDS appears to be equal during exhalation and during inhalation. The APD can be used to measure nasal resistance, and can detect resistance levels. The APD can consistently measure oral resistance with either a mask or a mouthpiece. Authors' Contributions LW conducted the testing as an undergraduate student. ATJ provided the APD and mentored LW. All authors have read and approved this manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529298.xml |
539262 | Chloride equilibrium potential in salamander cones | Background GABAergic inhibition and effects of intracellular chloride ions on calcium channel activity have been proposed to regulate neurotransmission from photoreceptors. To assess the impact of these and other chloride-dependent mechanisms on release from cones, the chloride equilibrium potential (E Cl ) was determined in red-sensitive, large single cones from the tiger salamander retinal slice. Results Whole cell recordings were done using gramicidin perforated patch techniques to maintain endogenous Cl - levels. Membrane potentials were corrected for liquid junction potentials. Cone resting potentials were found to average -46 mV. To measure E Cl , we applied long depolarizing steps to activate the calcium-activated chloride current (I Cl(Ca) ) and then determined the reversal potential for the current component that was inhibited by the Cl - channel blocker, niflumic acid. With this method, E Cl was found to average -46 mV. In a complementary approach, we used a Cl-sensitive dye, MEQ, to measure the Cl - flux produced by depolarization with elevated concentrations of K + . The membrane potentials produced by the various high K + solutions were measured in separate current clamp experiments. Consistent with electrophysiological experiments, MEQ fluorescence measurements indicated that E Cl was below -36 mV. Conclusions The results of this study indicate that E Cl is close to the dark resting potential. This will minimize the impact of chloride-dependent presynaptic mechanisms in cone terminals involving GABA a receptors, glutamate transporters and I Cl(Ca) . | Background Regulation of intracellular chloride levels results in a chloride equilibrium potential (E Cl ) that is hyperpolarized with respect to the resting potential in many nerve cells, but depolarized in others [ 1 - 5 ]. For example, E Cl in salamander rod photoreceptors is 25 mV more positive than the dark resting potential [ 6 ]. The resting potential of cone photoreceptors in darkness is around -42 to -47 mV and estimates of E Cl in cones have ranged from -65 mV to -36 mV [ 7 - 11 ]. Cone photoreceptors possess a number of Cl - conductances that help to shape their responses and synaptic output. As discussed below, the value of E Cl in cones is an important parameter for determining the strength and polarity of these effects. It has been suggested GABA a receptors in the terminals of cones may mediate inhibitory synaptic feedback from horizontal cells to cones [ 8 ]. Under this hypothesis, the light-evoked hyperpolarization of horizontal cells causes a cessation of GABA release and this disinhibition leads to a "feedback depolarization" in cones. There is evidence both for [e.g., [ 8 ]] and against [e.g., [ 12 , 13 ]; see review in ref. [ 14 ]]) this hypothesis. However, one prediction of the hypothesis is that the Cl - equilibrium potential (E Cl ) must be negative to the resting potential in order for GABA disinhibition to depolarize a cone. Cones possess prominent Ca 2+ -activated Cl - currents (I Cl(Ca) ) [ 15 - 17 ] activated by the influx of Ca 2+ through voltage-gated Ca 2+ channels as well as by release of Ca 2+ from intracellular stores [ 16 ]. Cl - flux through I Cl(Ca) can be substantial: during a 1.4 sec depolarizing step, the charge movement accompanying activation of I Cl(Ca) is estimated to be 8.5 times that produced by activation of I Ca alone [ 16 ]. These large membrane currents can strongly influence photoreceptor responses, but the nature of these effects depends on the value of E Cl . If E Cl is positive to the resting potential, activation of I Cl(Ca) can boost depolarizing feedback responses from horizontal cells onto cones and produce prolonged, regenerative depolarizing responses lasting many seconds [ 9 , 18 , 19 ]. On the other hand, if E Cl is negative to the resting potential, activation of I Cl(Ca) can operate as a negative feedback mechanism to limit regenerative activation of Ca 2+ channels [ 15 , 17 ]. In addition to altering membrane potential, depletion of intracellular Cl - can directly inhibit the open channel probability of single Ca 2+ channels, presumably by modifying an anion binding site on the intracellular surface of the channel [ 11 ]. In rods, where E Cl is positive to the resting potential, there is evidence for a negative feedback pathway between I Ca and I Cl(Ca) in which activation of I Ca stimulates I Cl(Ca) leading to a Cl - efflux that in turn inhibits Ca 2+ channel activation [ 6 , 20 ]. If, however, E Cl in cones is negative to the membrane potential, then activation of I Cl(Ca) would stimulate an influx of Cl - that would be expected to enhance Ca 2+ channel open probability [ 11 ]. Cone photoreceptors have presynaptic glutamate transporters that are coupled to Cl - channels [ 21 - 23 ]. The transporters in cones have been shown to respond to glutamate released from their own terminals [ 24 ]. Whether synaptically released glutamate causes cones to hyperpolarize or depolarize depends on E Cl . Furthermore, analogous to the negative feedback from I Cl(Ca) onto I Ca described above, the chloride current produced by activation of glutamate transporters in rods can cause a Cl - efflux that inhibits I Ca [ 25 ]. As with the feedback between I Cl(Ca) and I Ca , the strength and polarity of this potential interaction in cones depends on E Cl . Given the importance of E Cl in determining the impact of various feedback mechanisms in the photoreceptor terminal, we determined E Cl in cone photoreceptors of the salamander retina using a combination of imaging with a chloride-sensitive dye and electrophysiological approaches. Results In control superfusate, dark resting potentials of cones from slices prepared under visible light averaged -46.0 ± 2.00 mV (n = 9) after correcting for the liquid junction potential. This is nearly identical to the dark resting potentials of salamander cones prepared under infrared illumination (-46.8 ± 2.03 mV, n = 18). To measure E Cl , I Cl(Ca) was recorded using gramicidin perforated patch whole cell recordings and activated by applying a 500 ms step from -78 to -8 mV. This depolarizing step typically evoked a sustained inward tail current arising largely from activation of I Cl(Ca) [ 19 ]. Only cells that exhibited an inward tail current were used for analysis. As shown in the example of Fig. 1A , the current/voltage relationship of a cone cell was assessed during the tail current by using a ramp voltage protocol (1 mV/ms from -98 to +52 mV) begun 25 ms after the end of the depolarizing step. The same protocol was then repeated after applying niflumic acid (0.1 mM; Fig. 1B ). At this concentration, niflumic acid is a selective inhibitor of I Cl(Ca) in cones [[ 19 ]; niflumic acid may not be as selective in rods: [ 20 , 26 ]]. Subtracting the control ramp-evoked current from that obtained in the presence of niflumic acid yields the current/voltage profile for I Cl(Ca) (Fig. 1C ). In the example shown in Fig. 1C , the difference current reversed around -46 mV. The reversal potential of the niflumic acid-sensitive difference current determined from 8 cones averaged -45.5 ± 2.5 mV. As a control for the possible perturbation of intracellular Cl - by possible patch rupture, we repeated the same experiment using a pipette solution with only 3.5 mM Cl - . E Cl was not significantly different when measured using the low Cl - pipette solution (-50.4 mV ± 3.4 mV; n = 7; p = 0.49, unpaired t-test). If patch rupture had occurred, E Cl would be expected to attain -89 mV with the low Cl - pipette solution and -20 mV with the original pipette solution. Bath application of GABA evoked small reversible inward currents that averaged -3.4 ± 0.3 pA at the holding potential of -78 mV (not shown). The small size of these currents may be due to receptor desensitization [ 27 ]. Consistent with results obtained from measurements of I Cl(Ca) , difference currents calculated from ramps applied before and during GABA application indicate that the GABA-evoked current reversed at -46.4 ± 2.7 mV (n = 9). In a complementary approach for measuring E Cl , we used a Cl-sensitive dye, MEQ, to examine the Cl - flux that accompanied cone depolarization evoked by bath application of various high K + solutions (12, 22, 31, 41, 50 and 70 mM K + ). In a separate set of experiments, we used gramicidin-perforated patch recording methods to measure the membrane potentials produced in cones by application of the different high K + solutions. Slices used for MEQ experiments and for measurement of membrane potentials in different solutions were prepared using similar techniques under visible illumination; control experiments showed that the fluorescent illumination used during MEQ experiments did not produce any further changes in the cone resting membrane potential (n = 3). An example of a retinal slice loaded with MEQ is shown in Fig. 2A . Measurements of MEQ fluorescence were made from the cone soma (circle, Fig. 2A ). For a single wavelength dye such as MEQ, the change in fluorescence relative to basal fluorescence (ΔF/F) can be used as a measure of the change in ion concentration [ 28 ]. In the cone in Fig. 2B , bath application of 12 mM K + , which depolarized cones to -36 mV, produced a 1.2% decrease in MEQ fluorescence. Since MEQ fluorescence is quenched by Cl - ions this indicates that depolarization to -36 mV stimulated an influx of Cl - ions. Application of a solution with 70 mM K + , which depolarizes cones to -7 mV, produced a greater influx of Cl - as evidenced by the 10% decrease in MEQ fluorescence seen in a different cone (Fig. 2C ). Fig. 2D shows the average change in ΔF/F (x100) plotted as a function of the membrane potential evoked by the different high K + solutions. The finding that 12 mM K + consistently stimulated an influx of Cl - indicates that the reversal potential must be below -36 mV. Discussion The main finding of this study is that E Cl in salamander cones is close to the dark resting potential (~-46 mV). E Cl was found to be -46 mV from block of I Cl(Ca) by niflumic acid; small GABA-evoked currents reversed around the same potential. MEQ fluorescence changes produced by depolarization support these electrophysiological measurements by indicating that E Cl is below -36 mV. There can be local variations of E Cl within cells [ 4 ]. Large single cones in the salamander retina do not have a distinct axon and terminal; synaptic proteins are instead located at the base of the soma [ 29 ]. MEQ measurements were made in the cell soma from a region adjacent to the synaptic ending (see Fig. 2 ). I Cl(Ca) is localized to the terminal region in rods (30) and these channels are probably also localized to the terminals of cones. Thus, the measurements in the present study are likely to provide estimates of E Cl in the synaptic terminal and adjacent regions of the cone cell. Measurements of intracellular Cl - levels suggest that E Cl in the inner segment is not significantly different from that measured in the soma [ 11 ]. The finding that the Cl - equilibrium potential is close to the resting potential does not necessarily mean that Cl - is passively distributed. Electrophysiological experiments required that cells be voltage clamped at -70 mV for many minutes. Nonetheless, the value of E Cl determined from these electrophysiological experiments in which cells were voltage clamped at -70 mV was similar to the value estimated from MEQ studies in which cells were not voltage clamped and thus at their resting membrane potential. Results from experiments on the prolonged depolarization in cones also suggest that E Cl can be maintained indefinitely at a value above the membrane potential. The plateau phase of the prolonged depolarization, which largely reflects I Cl(Ca) activation [ 9 , 19 ], could remain above the membrane potential established by an adapting background for hours [ 9 ]. The ability of cones to maintain E Cl above the membrane potential may arise from activity of the Na/KCl cotransporter as shown in rods (20) as well as from other mechanisms (e.g., CLC-2) [ 2 , 31 ]. Comparisons with other studies E Cl in cones has been estimated in a number of previous studies. The most positive value for E Cl of -36 mV comes from calibration of MEQ fluorescence levels to determine the resting intracellular Cl - concentrations in cones isolated from the salamander retina (11). However, these measurements showed a large variability (range of S.E.M.: -26.5 to -46.6 mV). The most negative estimate of E Cl comes from a study by Attwell et al [ 7 ] showing that the sign-reversing pathway from rods to cones reversed around -65 mV. Based on the presumption that this pathway involved disinhibition of GABAergic inputs into cones, this study has been interpreted as suggesting that E Cl is around -65 mV. However, more recent evidence questions whether the horizontal cell to cone feedback pathway thought to underlie this sign-reversing pathway from rods to cones is truly GABAergic [ 9 , 13 , 14 ]. Other studies have arrived at values for E Cl similar to those found in the present study. 1) By examining the polarity of GABA-evoked currents after patch rupture with either 12 or 24 mM Cl - in the recording pipette, Kaneko and Tachibana [ 8 ] estimated E Cl to be around -47 mV in isolated turtle cones. 2) Based on the membrane potential attained by the plateau phase of the prolonged depolarization in turtle cones from the eyecup slice preparation, E Cl was estimated to be at or slightly above the dark resting potential of -42 mV [after correction for a liquid junction potential of -2 mV; ref. [ 9 ]]. 3) In a single recording from a salamander cone obtained with a Cl-sensitive electrode, Miller and Dacheux [ 32 ] found that E Cl was 2 mV more positive than the dark resting potential. 4) A slightly more negative value for E Cl was found in ruptured patch recordings from goldfish cones by examining the voltage dependence of the I Cl(Ca) tail current [ 10 ]. By extrapolating measurements back to the time of patch rupture, Kraaij et al [ 10 ] concluded that E Cl was ~-55 mV. Functional implications I Ca in cones, like that of rods, can be inhibited by lowering extracellular Cl - [ 33 ]. The inhibition of I Ca produced by lowering extracellular Cl - appears to result from a reduction in intracellular Cl - which in turn causes a reduction in the open probability of single Ca 2+ channels [ 11 ]. In rods, where E Cl is positive to the resting potential, activation of Cl - channels leads to a Cl - efflux thereby producing an inhibition of Ca 2+ channels [ 6 , 11 , 20 ]. The present results indicate that activation of Cl - channels when the cell is at its resting potential would produce minimal changes in intracellular Cl - in cones. Therefore, the feedback between I Ca and I Cl(Ca) postulated for rod photoreceptors [ 6 , 20 ] would be expected to be minimal in cones in darkness. Another implication of the finding that E Cl is close to the dark resting potential is that the stimulation of Cl - channels associated with glutamate transporters by glutamate released from cone terminals [ 24 ] would tend to stabilize the cell membrane potential near the dark potential. In rods, the Cl - efflux accompanying activation of glutamate transporters appears to contribute to a glutamate-mediated inhibition of I Ca [ 25 ]. As with the feedback between I Ca and I Cl(Ca) considered in the previous paragraph, the finding that E Cl is near the resting potential leads to the prediction that in darkness there would be no Cl - efflux accompanying glutamate transporter activation and therefore glutamate would not be expected to inhibit I Ca . Cones hyperpolarize to light, although with prolonged illumination the membrane potential recovers to near the dark resting potential. The impact of chloride-dependent negative feedback between I Cl(Ca) and I Ca or the glutamate transporter chloride current and I Ca would be expected to increase as a cone hyperpolarizes in response to light. By reducing glutamate release, these chloride-dependent negative feedback mechanisms might thus contribute to making post-synaptic responses more transient. The finding that E Cl is near the resting potential of cones indicates that GABAergic disinhibition near the dark potential should produce little membrane potential change. This result is inconsistent with the postulated role for GABA in generating the feedback depolarization [ 8 ] and supports other studies suggesting that GABA is not directly responsible for horizontal to cone feedback [ 9 , 13 , 14 ]. Conclusions Electrophysiological measurements, supported by experiments using chloride-sensitive dyes, indicate that E Cl in salamander cones is close to the dark resting membrane potential. By minimizing the trans-membrane flux of chloride, this will minimize the presynaptic impact of GABA a receptors, I Cl(Ca) , and glutamate transporter chloride channels. Methods Tissue preparation E Cl is positive to the resting potential of many neurons in the immature brain [ 5 ]. Based on their size, the neotenous tiger salamanders ( Ambystoma tigrinum , 15–25 cm) used in these experiments are thought to be 2–7 years old out of a life span of ~12 years (34). Salamanders were handled humanely in accordance with protocols approved by the Institutional Animal Care and Use Committee at the University of Nebraska Medical Center. Chilled salamanders were rapidly decapitated, an eye was enucleated, and the front of the eye was removed. The resulting eyecup was cut into three or four pieces and a single piece was placed vitreal surface down onto a piece of filter paper (2 × 5 mm, Millipore type AAWP, 0.8 μm pores). After adhering to the filter paper, the retina was isolated under chilled amphibian superfusate and cut into 125 μm slices using a razor blade tissue chopper (Stoelting Co., Wood Dale, IL). The slices were rotated 90° to view the retinal layers when placed under a water immersion objective (60X, 1.0 NA) on an upright fixed stage microscope (EF 600, Nikon Inc., USA). Slices were prepared under visible light but recordings were performed in darkness. All experiments were done using red-sensitive large single cones selected by anatomical criteria [ 35 ]. Solutions and perfusion Solutions were applied with a single-pass, gravity-feed perfusion system (1 ml/min). The normal amphibian superfusate contained (in mM): 111 NaCl, 2.5 KCl, 1.8 CaCl 2 , 0.5 MgCl 2 , 10 N -2-hydroxyethylpiperazine- N ' 2-ethanesulfonic acid (HEPES), and 5 glucose (pH 7.8). The osmolarity was measured with a vapor pressure osmometer (Wescor, Logan, UT) and adjusted, if necessary, to 242 ± 5 mOsm. For high K + solutions, various quantities of NaCl were replaced with equimolar KCl. Niflumic acid was diluted (1:10,000) from DMSO stock solutions. Unless otherwise specified, chemicals were obtained from Sigma/Aldrich/RBI (St. Louis, MO). Electrophysiology Patch pipettes were pulled on a PP-830 vertical puller (Narishige USA, New York) from borosilicate glass pipettes (1.2 mm O.D., 0.95 mm I.D., with internal filament) and had tips of ~1 μm outer diameter with resistances of 10 to 15 MΩ. To maintain endogenous levels of intracellular Cl - , we obtained perforated patch whole cell recordings using the cation channel, gramicidin [ 36 ]. Gramicidin was dissolved in ethanol (5 mg/ml) and then added to the pipette electrolyte solution to achieve a final concentration of 5 μg/ml. For current clamp measurements of membrane potentials, the pipette electrolyte solution contained (in mM): 54 KCl, 61.5 KCH 3 SO 4 (Pfaltz and Bauer, Waterbury, CT), 3.5 NaCH 3 SO 4 , 10 HEPES. The pH was adjusted to 7.2 with KOH. The liquid junction potential (LJP) of this solution was estimated to be -7 mV using the junction potential calculator of PClamp (Axon Instruments). Membrane potential values reported throughout this manuscript were corrected for the LJP. For experiments with niflumic acid or GABA, pipettes were typically filled with a solution containing (in mM): 54 CsCl, 61.5 CsCH 3 SO 3 , 3.5 NaCH 3 SO 4 , 10 HEPES (LJP = -8 mV). In some experiments, a low Cl - pipette solution was used containing: 115.5 mM CsCH 3 SO 3 , 3.5 NaCl, 10 HEPES (LJP = -10 mV). The pH of both solutions was adjusted to 7.2 with CsOH. The osmolarity of pipette solutions were also adjusted, if necessary, to 242 ± 5 mOsm. Recordings were made using an Axopatch 200B amplifier (Axon Instruments Inc., Union City, CA) and PClamp 8 software (Axon Instruments). Cell input resistance calculated using a step from -70 to -90 mV averaged 695 ± 111 MΩ. Access resistance estimated from the peak of the capacitative transient averaged 30.7 ± 4.8 MΩ (n = 24). Imaging experiments Digital fluorescent images were obtained with a cooled CCD camera (SensiCam, Cooke Corp., Auburn Hills, MI). Axon Imaging Workbench (AIW 2.2, Axon Instruments Inc., Union City, CA) was used to control the camera, filter wheel, and image acquisition. Pixel binning (2 × 2) of the images was used to decrease acquisition time to ≤1 s. Images were acquired at 5 to 10 s intervals during experimental trials. For measurements of [Cl - ] i we used the dye, 6-methoxy-N-ethylquinolinium iodide (MEQ, Molecular Probes, Eugene, OR) [ 37 ]. MEQ was loaded into cells after reducing it to DiH-MEQ by adding 30 μM sodium borohydride (100 μl) to MEQ (5 mg) under a continuous stream of nitrogen gas [ 38 ]. DiHMEQ enters cells during the incubation period (15 min) where it is oxidized and retained in the form of MEQ. Fluorescence emission decreases as Cl - quenches MEQ. The slow exponential decay in MEQ fluorescence due to dye leakage and bleaching was determined from a 3 min. series of control measurements prior to drug application and subtracted before analysis [ 11 , 20 ]. Variance is reported as ± S.E.M. Authors' contributions WT conceived the study and drafted the manuscript. WT and EB participated in all aspects of the experiments but most recordings were performed by EB. Both authors have read and approved the manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539262.xml |
526183 | Genes, Genomes, and the Road to Diversity: How Regulatory Networks Evolve | null | Evolutionary biologists have long been interested in understanding the molecular basis for the great diversity in size, shape, and behavior seen in life on earth. Recent attention has focused on the role that gene expression changes play in organismal evolution. Tracing the evolution of gene regulation, however, has proved difficult. This is in large part due to the difficulty in identifying and comparing the regulatory elements that control gene expression in different species. Gene expression depends on cis -regulatory elements, short sequence motifs embedded in the DNA that flank a gene's coding region. Regulatory proteins bind to specific cis -regulatory sequences, and command the activation or repression of the corresponding gene. The challenge in studying the evolution of cis -regulatory elements lies in identifying those elements in multiple species. Unlike protein sequences, which are typically a few hundred amino acids long and relatively straightforward to identify in related organisms, cis -regulatory elements are often short and can have variations in sequence. This makes it very difficult to distinguish the regulatory elements from the nonfunctional DNA that surrounds them. It is even harder to identify corresponding regulatory elements across species. As the evolutionary distance between species increases, so, too, does the difficulty in identifying corresponding cis-elements in those species. In this issue of PLoS Biology , Audrey Gasch and her colleagues describe a comparative genomics approach that allows them to identify potential cis -regulatory elements in thousands of genes across 14 ascomycete fungi whose diversity represents the effects of several hundred million years of evolution. Ascomycetes are a large class of fungi with extremely diverse morphologies, reproductive strategies, and habitats. A divergence dating back 500 million to 1 billion years ago gave rise to three groups: Archaeascomycetes, Euascomycetes, and Hemiascomycetes. The genome of the brewer's yeast, Saccharomyces cerevisiae , a hemiascomycete, was completely sequenced in 1995, and that of fission yeast, Schizosaccharomyces pombe , an archaeascomycete, in 2002. Since that time, complete genome sequences have been released for more than nine additional hemiascomycetes and three euascomycetes. This gives the authors an opportunity to compare regulatory systems among progressively more distantly related species, on a genomic scale. Phylogeny of fungi used to study evolution of gene regulation Genome-wide expression studies in the yeast S. cerevisiae have revealed groups of genes whose expression levels vary simultaneously under varying experimental conditions. Such co-regulated genes, the authors reasoned, must harbor common regulatory elements that coordinate their response to experimental triggers. Gasch and colleagues looked for such cis -elements and found 35 groups of co-regulated S. cerevisiae genes with at least one shared cis -element. The authors then argued that co-regulation may reflect selection pressures that also apply to other ascomycetes, and so they identified the equivalent of the 35 co-regulated gene groups in each of the 13 other species. They then looked for shared cis -elements within each group and in each species independently, and compared the regulatory systems across the species. The results of this study show that the majority of cis -elements first identified in yeast are retained in the equivalent gene groups in other species, in a manner that reflects the species' evolutionary distance from yeast. One cis -element, in a group of co-regulated genes that control the cell cycle, is found all the way from budding yeast to fission yeast, suggesting a selection pressure on the co-regulation of these genes that has withstood greater than 500 million to 1 billion years of evolution. In contrast, there were other examples in which the same gene groups contained different putative cis -elements in each species, suggesting that the regulation of those genes has evolved. In the case of cis -elements found in genes controlling protein degradation, a related element was identified in all of the hemiascomycetes, whereas the euascomycetes appear to have adopted a novel cis -element for this gene group. Interestingly, the hemiascomycete element displays a sequence variation in Candida albicans that is not found in S. cerevisiae . The two species diverged 200 million years ago. Gasch and colleagues showed that the protein that binds to the hemiascomycete element has evolved to have slightly different DNA interactions in the two species, allowing the C. albicans protein to bind the novel sequence found only in the C. albicans genes. This provides evidence for co-evolution between a transcription factor and its target cis -element. Overall, this analysis has uncovered striking cases of conservation and innovation of gene regulatory systems, and therefore provides important insight into the evolutionary forces that have shaped the evolution of gene regulation. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC526183.xml |
449785 | A Chromosomal Memory Triggered by Xist Regulates Histone Methylation in X Inactivation | We have elucidated the kinetics of histone methylation during X inactivation using an inducible Xist expression system in mouse embryonic stem (ES) cells. Previous reports showed that the ability of Xist to trigger silencing is restricted to an early window in ES cell differentiation. Here we show that this window is also important for establishing methylation patterns on the potential inactive X chromosome. By immunofluorescence and chromatin immunoprecipitation experiments we show that histone H3 lysine 27 trimethylation (H3K27m3) and H4 lysine 20 monomethylation (H4K20m1) are associated with Xist expression in undifferentiated ES cells and mark the initiation of X inactivation. Both marks depend on Xist RNA localisation but are independent of silencing. Induction of Xist expression after the initiation window leads to a markedly reduced ability to induce H3K27m3, whereas expression before the restrictive time point allows efficient H3K27m3 establishment. Our data show that Xist expression early in ES cell differentiation establishes a chromosomal memory, which is maintained in the absence of silencing. One consequence of this memory is the ability to introduce H3K27m3 efficiently after the restrictive time point on the chromosome that has expressed Xist early. Our results suggest that this silencing-independent chromosomal memory has important implications for the maintenance of X inactivation, where previously self-perpetuating heterochromatin structures were viewed as the principal form of memory. | Introduction In mammals, dosage differences of X-linked genes between XX female and XY male cells are adjusted by transcriptional inactivation of one of the two female X chromosomes. X inactivation is a multistep process, in which the cell counts the number of X chromosomes, chooses one to be active, and silences all others. Initiation of silencing is triggered by accumulation of the 17-kb noncoding Xist RNA ( Borsani et al. 1991 ; Brockdorff et al. 1991 ; Brown et al. 1991 ). Remarkably, Xist RNA attaches to chromatin and spreads from its site of transcription in cis over the entire inactive X chromosome (Xi), mediating transcriptional repression. Xist is essential for initiation of silencing, but not for the maintenance of transcriptional repression on the Xi at later stages of cellular differentiation ( Penny et al. 1996 ; Marahrens et al. 1998 ; Csankovszki et al. 2001 ). Presently, the molecular nature of the silencing mechanism is not known. Previous studies have shown that X-chromosome inactivation involves the progressive recruitment of a variety of different factors and posttranslational modifications of lysine residues in the amino termini of histones (reviewed in Brockdorff 2002 ). The current view is that Xist expression initiates the formation of heterochromatin on the Xi, which can be perpetuated by redundant silencing mechanisms at later stages. Consistent with this view, it has been shown that the Xi in mouse embryonic fibroblasts is kept inactive in the absence of Xist by redundant mechanisms, including DNA methylation and histone H4 hypoacetylation ( Csankovszki et al. 2001 ). The Polycomb group proteins Ezh2 and Eed localise to the Xi in embryonic and extraembryonic tissues early in mouse development ( Wang et al. 2001 ; Mak et al. 2002 ; Plath et al. 2003 ; Silva et al. 2003 ). The human EZH2/EED and its homologous E(z)/ESC complex in Drosophila melanogaster show intrinsic histone H3 lysine 9 (H3-K9) and lysine 27 (H3-K27) methyltransferase activity ( Cao et al. 2002 ; Czermin et al. 2002 ; Kuzmichev et al. 2002 ; Muller et al. 2002 ). Interestingly, H3-K27 methylation is one of the earliest chromosomal modifications on the Xi ( Plath et al. 2003 ), and the requirement of Eed for histone methylation on the Xi has been demonstrated ( Silva et al. 2003 ). However, analysis of Eed mutant embryos suggests that Eed is not required for initiation of silencing in trophoblast cells but is required for the maintenance of the Xi at later stages ( Wang et al. 2001 ). Although data are consistent with the interpretation that Xist RNA recruits the Ezh2/Eed complex, thereby introducing histone H3 methylation, the significance of H3-K27 methylation for chromosomal inactivation is unclear. In flies, methylation on H3-K27 facilitates the binding of Polycomb to amino-terminal fragments of histone H3 ( Cao et al. 2002 ; Min et al. 2003 ). Polycomb recruitment to the Xi has not been observed, and current models suggest that H3-K27 methylation in X-chromosome inactivation is indepen-dent of classical Polycomb silencing ( Mak et al. 2002 ; Silva et al. 2003 ). We have previously shown that chromosomal silencing can be recapitulated in embryonic stem (ES) cells by expressing Xist RNA from cDNA transgenes integrated into autosomes and the X chromosome ( Wutz and Jaenisch 2000 ), and this allowed for an uncoupling of Xist regulation from cellular differentiation. In this transgenic system, Xist expression is under the control of a tetracycline-responsive promoter, which can be induced by the addition of doxycycline to the culture medium. We showed that Xist RNA localisation and silencing can be separated by introducing specific mutations in Xist RNA ( Wutz et al. 2002 ). Initiation of silencing depends on the repeat A sequence at the 5′ end of Xist. Deletion of this element results in an RNA that localises to chromatin and spreads over the chromosome, but does not trigger transcriptional repression. Initial silencing in ES cells is reversible and dependent on Xist expression. At a later stage in differentiation this silent state becomes irreversible and independent of Xist, corresponding to the maintenance phase of X inactivation. We also showed that Xist expression must be induced early in ES cell differentiation to cause transcriptional repression ( Wutz and Jaenisch 2000 ). Therefore, establishment of silencing is restricted to an initiation window in ES cell differentiation, and induction of Xist expression at a time point later than 24 h in differentiation no longer causes silencing. We found that Xist RNA loses its potential to initiate transcriptional repression roughly 24 h earlier in differentiation than the point at which silencing becomes irreversible. Notably, this left a gap of approximately one cell cycle in length between the initiation and maintenance phases. How silencing is maintained during this period and how the silent state becomes irreversible remained previously unexplained. In this report we perform kinetic measurements and quantification of histone H3 lysine 27 trimethylation (H3K27m3), revealing a novel chromosomal memory that is established by Xist expression at an early time point in ES cell differentiation independent of transcriptional repression. Our analysis suggests that this chromosomal memory might have an important role in the transition from the initiation phase to the maintenance phase of X inactivation. Results Profiling Histone Modification States at the Initiation of X Inactivation We have previously reported that the initial steps of chromosomal silencing in mammalian X inactivation can be recapitulated in transgenic undifferentiated male ES cells ( Wutz and Jaenisch 2000 ). Such ES cells are useful for studying the function of Xist RNA in the initiation of chromosomal silencing and for analysing the kinetics and relevance of chromosomal modifications. We aimed to delineate a pattern of histone methylation states that define the initial decision for facultative heterochromatin. To achieve this we performed immunofluorescence staining against the various modification states on histone H3 and H4 lysine residues in clone 36 ES cells, in which Xist expression can be induced from a transgene integrated on Chromosome 11 by addition of doxycycline to the culture medium ( Wutz and Jaenisch 2000 ). We used highly specific antisera for a defined methylation state (mono-, di-, or tri-) at a particular lysine residue in the amino terminus of histone H3 and H4 ( Peters et al. 2003 ; Perez-Burgos et al. 2004 ). Some cross reactivity of the H3K27m2 antiserum with H3K27m1 and H3K27m3, of the H3K4m3 antiserum with H3K4m2, and of the H4K20m2 antiserum with H4K20m1 and H4K20m3 was detected on peptide blots ( Figure S1 ), but does not affect the conclusions drawn in this study. Our cytological experiments show a focal signal for H3K27m3 in the interphase nuclei of clone 36 ES cells upon Xist expression, which colocalises with Chromosome 11 in metaphase spreads and Xist RNA in interphase nuclei ( Figure 1 ). In cells grown in the absence of doxycycline, a diffuse nuclear signal was observed. H3-K27 mono- and dimethylation were equally present on the inactivated chromosome and other autosomes ( Table 1 ). Notably, we did not observe any specific enrichment for the H3K9m1, H3K9m2, or H3K9m3 signal on Chromosome 11 upon Xist induction (Figures 1 C, 1 G, and S2 ). H3K9m3 and H3K27m1 colocalised strictly with constitutive heterochromatin at pericentric regions and the Y chromosome ( Figure 1 G and 1 H). H3K4m2 and H3K4m3 gave banded signals on chromosome arms that were reduced but not entirely erased on the transgenic chromosome, when Xist expression was induced (Figures 1 E and S2J ). The heterochromatic Y chromosome completely lacked both H3K4m2 and H3K4m3 in the same metaphase spread. Thus, we conclude that the reduction of H3K4m2 and H3K4m2 on the Xist -expressing chromosome is consistent with a state of transcriptional repression ( Santos-Rosa et al. 2002 ) and with earlier reports that implicate H3-K4 hypomethylation early in X inactivation ( Heard et al. 2001 ; O'Neill et al. 2003 ). H3K4m1 was equally present on the Xist -expressing chromosome and other autosomes. Using antisera specific for methylation states of H4K20, we observed that H4K20m1 decorated Chromosome 11 upon Xist induction in undifferentiated clone 36 ES cells (in 46% of interphase nuclei; Figure 1 B). H4K20m2 and H4K20m3 were not enriched on the Xist -expressing chromosome (Figure S2H and S2I ; G. Schotta and M. Lachner, unpublished data). We also investigated the acetylation state of histone H4 in these cells using a sheep polyclonal antiserum that preferentially recognises multiply acetylated H4 ( Morrison and Jeppesen 2002 ). Using this antiserum, we detected partial hypoacetylation of Chromosome 11 in metaphase spreads of clone 36 ES cells that were induced to express Xist (Figures 1 F, S2K , and S2L ). This observation is different from the global chromosome-wide hypoacetylation of H4 that was reported on the Xi later in differentiation ( Keohane et al. 1996 ) and might reflect the absence of active promoters. We also detected a degree of hypoacetylation when a silencing-defective Xist RNA was expressed ( Figure S3 ), making it likely to be the consequence of cross talk with H4-K20 methylation, which is mutually exclusive at least with H4-K16 acetylation ( Nishioka et al. 2002 ). In conclusion, H3K27m3, H4K20m1, reduction of H3K4m2 and H3K4m3, and reduced multiple-lysine acetylation of histone H4 correlate with the inactive state of the chromosome in undifferentiated ES cells ( Table 1 ). Figure 1 Epigenetic Imprints at the Initiation of X Inactivation (A–H) Indirect immunofluorescence and subsequent DNA FISH analysis on mitotic chromosomes prepared from undifferentiated clone 36 ES cells after 3 d of Xist induction. H3K27m3 (A), H4K20m1 (B), and Ezh2 (D) are enriched on the arms of Chromosome 11 upon ectopic Xist expression. H3K9m2 (C) is not enhanced upon Xist expression. H3K4m2 (E) is reduced on Chromosome 11 upon Xist induction (green box) and absent from pericentric heterochromatin and the Y chromosome (orange arrow). (F) Histone H4 multiple-lysine acetylation is partially reduced (green box, left panel). Hypoacetylation (red) is restricted to chromosomal regions which show high levels of H3-K27 trimethylation (green, right panel). H3K9m3 (G) and H3K27m1 (H) are enriched at constitutive heterochromatin of pericentric regions and the Y (orange arrows). (I–K) Indirect immunofluorescence (upper panels) and subsequent Xist RNA FISH (red, Xist RNA; blue, DAPI) analysis of H3K27m3 (I), H4K20m1 (J), and Ezh2 (K) in interphase nuclei of undifferentiated clone 36 ES cells expressing Xist for 3 d. Table 1 Histone Lysine Methylation States as Epigenetic Imprints during X Inactivation +, chromosome-wide mitotically stable methylation marks recruited by Xist RNA; −, decreased levels due to initiation of X inactivation; 0, abundance and distribution independent of Xist (equal on all chromosomes); 0 a , small regional increase during differentiation revealed by ChIP (see text); 0 b , identified as epigenetic imprints of constitutive heterochromatin Further confirmation of the cytological findings comes from chromatin immunoprecipitation (ChIP) experiments using antibodies specific for H3K27m3, H4K20m1, H3K4m3, H3K4m2, and H3K9m2 in both undifferentiated and differentiated clone 36 ES cells in the presence or absence of doxycycline ( Figure 2 ). We observed enhanced H3K27m3 and H4K20m1 in the cells expressing Xist regardless of the differentiation state on three microsatellite sequences on Chromosome 11 ( Figure 2 ). A control microsatellite on Chromosome 15 did not show this effect ( Figure 2 F and 2 L). Upon Xist expression, we also observed H3K27m3 on the puromycin marker gene cointegrated with the Xist transgene on Chromosome 11, compared to nearly undetectable levels in the uninduced sample ( Figure 2 B). This increase in H3K27m3 was paralleled by a marked decrease in H3K4m2 and H3K4m3, but no increase in H4K20m1 could be observed at this locus in undifferentiated ES cells. Upon differentiation, an increase in the H4K20m1 signal was observed when Xist was expressed on all sequences on Chromosome 11. A control tubulin gene located on Chromosome 15 showed no significant change upon Xist induction ( Figure 2 E and 2 K). These data show that H3K27m3 and H4K20m1 are elevated by Xist RNA expression on the transgenic chromosome, in agreement with our cytological analysis. However, regional differences are revealed by the higher resolution of the ChIP experiment, showing that the two modifications do not display a completely overlapping distribution on the chromosome. Differentiation of the ES cells resulted in increased H4K20m1 signals dependent on Xist expression. H3K9m2 was also elevated on two loci on Chromosome 11. Figure 2 ChIP Mapping of H3K27m3, H4K20m1, H3K9m2, H3K4m3, and H3K4m2 on the Xist -Expressing Chromosome 11 during Differentiation of Clone 36 ES Cells A genetic map of Chromosome 11 indicating the loci analysed is given on the left ( Xist -TG, approximate integration site of Xist transgene; puro, PGKpuromycin marker). (A to F) Chromatin was prepared from undifferentiated clone 36 ES cells grown for 3 d in the presence (light bars) or absence (dark bars) of doxycycline. H3K27m3 and H4K20m1 were enriched at three intergenic microsatellite sequences at 18.0 (A), 45.5 (C), and 75.2 (D) cM. (B) H3K27m3 was established over the coding sequence of PGKpuromycin in doxycycline-induced cells, which was accompanied by a loss of H3K4m2 and H3K4m3. (E) Tubulin control. (F) Control microsatellite located on Chromosome 15. (G–L) Analysis of H3K27m3, H4K20m1, and H3K9m2 in clone 36 ES cells differentiated for 9 d with (light bars) or without (dark bars) doxycycline. Histone methylation marks were monitored. Experiments were performed in duplicate, and the standard error is indicated in the graphs. H3K27m3 and H4K20m1 Are Triggered by Xist RNA Localisation and Are Independent of Silencing In agreement with earlier studies ( Plath et al. 2003 ; Silva et al. 2003 ), our results indicate that chromosome-wide histone H3K27m3 is efficiently triggered in undifferentiated ES cells and therefore is an early mark of X inactivation. We measured the kinetics of H3K27m3 following induction of Xist RNA expression in undifferentiated clone 36 ES cells ( Figure S3A ). At 6, 12, and 24 h after induction 0%, 12%, and 37% of the cells, respectively, showed a signal, and by 48 h a maximum of 70% was reached. Furthermore, the recruitment of Ezh2 protein to the transgenic Chromosome 11 upon Xist expression (see Figure 1 D and 1 K) is consistent with the idea that the Ezh2/Eed complex contains the enzymatic activity causing H3K27m3 in X inactivation ( Mak et al. 2002 ; Plath et al. 2003 ; Silva et al. 2003 ). To identify the Xist sequences that are required for the binding of the Ezh2/Eed complex and to trigger H3K27m3, we examined a panel of Xist RNA mutations ( Figure 3 A). In an earlier study we inserted Xist cDNA transgenes containing defined deletions into the Hprt gene locus on the single X chromosome in male mouse T20 ES cells and measured their ability to cause silencing ( Wutz et al. 2002 ). We used deletions spanning the entire RNA that eliminate relatively large parts of Xist to analyse H3K27m3 by immunofluorescence in ES cells after induction of transgenic Xist expression ( Figure 3 ). H3K27m3 staining was observed for all Xist mutations tested, with the exception of the ΔXSa deletion, where sequences required for localisation are deleted. The resulting XistΔXSa RNA did not localise well to chromatin and showed consequently greatly diminished potential to silence ( Figure S4 ). We interpret the absence of detectable H3K27m3 in this case as a consequence of the failure of the RNA to localise. All other mutants analysed, including that containing a ΔXN deletion spanning a similar region, gave rise to RNA that localised well to chromatin and caused H3K27m3. A mutant with a deletion of repeat A (T20:ΔSX ES cells; Figure 3 ), which localises to chromatin but does not cause silencing, was able to induce H3K27m3, suggesting that methylation can be established independent of silencing, a finding consistent with the results of an earlier study ( Plath et al. 2003 ). The expression of the silencing-deficient Xist RNA led to a significantly lower percentage of cells with H3K27m3 foci in interphase nuclei (3- to 4-fold reduction compared to wild-type Xist RNA; Figure 3 D). Moreover, on metaphase chromosomes methylation appeared mostly as a single band (only 5% showed a wild-type pattern; Figure 3 C). Since the transgene is integrated in the Hprt locus on the X chromosome and the endogenous Xist gene is still present in this cell line, the possibility exists that the transgenic RNA might have stabilised the endogenous Xist RNA or vice versa to effect H3K27m3. To address this point we made use of another cell line in which repeat A was deleted from the endogenous Xist gene and an inducible promoter was inserted by homologous recombination (J1:XistΔSX-tetOP; Wutz et al. 2002 ). Induction of Xist RNA expression caused H3K27m3 on the single X chromosome in these cells, confirming that H3K27m3 can be established by Xist expression in complete absence of repeat A sequences. However, in undifferentiated ES cells, expression of the silencing-deficient Xist RNA led consistently to lower numbers of cells (30%–35%) showing H3K27m3 staining compared to the wild-type Xist RNA (80%; Figure 3 B and 3 C). Mono- and dimethylation of H3-K27 were not visibly elevated in J1:XistΔSX-tetOP cells at the expense of the H3K27m3 signal (data not shown), suggesting that recruitment of the Ezh2/Eed complex was impaired in the absence of repeat A, and ruling out the possibility that repeat A would change the specificity of the complex to induce trimethylation activity. Consistent with this interpretation, Ezh2 was observed in only 9% of the J1:XistΔSX-tetOP ES cells compared to 76% of the clone 36 ES cells (see Figures 1 K and S3D ). We note that the lower methylation potential of Xist RNA lacking repeat A sequences was only observed in undifferentiated ES cells. When the cells were differentiated, methylation levels were elevated (see Figure S3C ). We further determined the role of H4K20m1 in silencing. We detected H4K20m1 upon induction of Xist expression in 14% of the interphase nuclei in undifferentiated J1:XistΔSX-tetOP ES cells, showing that H4K20m1 can be established in the absence of repeat A (see Figure S3D ). We conclude that H3K27m3 and H4K20m1 are independent of and not sufficient for silencing. Figure 3 Sequences of Xist RNA Required for H3K27m3 Establishment (A) Schematic representation of the Xist cDNA (top) indicating repeats A to E, restriction sites, and the locations of deletions (coloured bars) relative to the location of sequences required for localisation (black and hatched boxes; Wutz et al. 2002 ). (B) Analysis of H3K27m3 on metaphase chromosome spreads from undifferentiated ES cells after 3 d of Xist induction (see text). The staining patterns ( n > 100) were scored as chromosome-wide dense methylation (black), reduced methylation (grey), and a single band (open). (C) Pattern of H3K27m3 triggered by different Xist mutants on metaphase chromosomes after 3 d of induction. Enlarged view of Chromosome 11 (clone 36) or the X chromosome (T20 lines, J1 knock-in line). (D) Focal H3K27m3 staining in interphase nuclei (percentage given; n > 100) of undifferentiated ES cells expressing Xist constructs. Efficient H3K27m3 Is Restricted to Early Stages of Differentiation Xist -mediated transcriptional silencing is restricted in ES cell differentiation in that the potential of Xist to initiate repression diminishes 48 h after differentiation ( Wutz and Jaenisch 2000 ). We investigated whether the ability to establish H3K27m3 would be restricted to this initiation window in clone 36 ES cells. These cells carry a puromycin resistance gene (puro), which is cointegrated with the Xist cDNA transgene on Chromosome 11 and can be silenced by transgenic Xist expression. Xist expression was induced either from the beginning or at 24, 48, 72, 96, or 120 h after the onset of differentiation. The ability of Xist to initiate silencing at various time points was monitored by measuring puro expression, and H3K27m3 was analysed in parallel in all cultures at 12 d after differentiation ( Figure 4 A). When Xist was induced within 24 h of differentiation, H3K27m3 was observed in a large fraction of the cells. Induction of Xist after 24 h led to significantly lower methylation levels (10%–15% of cells; Figure 4 A). The efficiency in H3K27m3 pattern establishment correlated at all time points with the potential of Xist to initiate silencing and Ezh2 recruitment ( Figure 4 B). Hence, an efficient H3K27m3 pattern was established in a time window that overlapped with the window for the initiation of Xist -mediated repression. We also determined the levels of Eed and Ezh2 protein during ES cell differentiation ( Figure 4 C). Our analysis shows that Eed levels are significantly reduced at day 3 of differentiation and Ezh2 levels diminish more gradually towards even later time points. This demonstrates that the ability of Xist to induce efficient H3K27m3 is restricted at a time when both Eed and Ezh2 proteins are detected in similar amounts, as in undifferentiated ES cells, suggesting that the efficiency of methylation is not a function of the protein levels. Figure 4 Restriction of H3K27m3 Establishment and Transcriptional Silencing in Differentiation (A) Initiation of H3K27m3 during clone 36 ES cell differentiation. Xist expression was induced at the beginning (+) or at various time points (24 to 120 h) after the start of differentiation, or not induced (−). The percentages of interphase cells showing H3K27m3 (black bars; n > 700) and Ezh2 (grey bars; n > 200) staining were determined at day 12 of differentiation. (B) Initiation of transcriptional silencing during differentiation was assessed by Northern blot analysis of PGKpuromycin (puro) and Gapd as a loading control in parallel cultures as described for (A). (C) Western analysis of Ezh2 and Eed protein levels during differentiation of clone 36 ES cells after induction with retinoic acid. Histones H3 and H4 were used as a loading control. (D) Establishment of H3K27m3 during embryonic development. Xist expression was induced from the single X chromosome of male Xist-tetOP embryos (see text) for 3 d (E9.5–12.5 and E13.5–16.5). The percentage of cells with H3K27m3 staining in interphase (left) and clusters of Xist RNA (right, open bars) are given ( n > 300). Grey areas indicate the proportion of H3K27m3-positive cells to Xist -positive cells. (E) Xist RNA FISH (top) and H3K27m3 (bottom) staining of histological sections prepared from neck connective tissue of embryos described in (C). To confirm this finding, we assayed the effect of induction of Xist expression on H3K27m3 in embryonic fibroblasts. Fibroblasts were isolated from male, day 13.5 embryos carrying an insertion of the doxycycline-inducible promoter in the endogenous Xist locus (Xist-tetOP allele) and a homozygous insertion of the tetracycline-responsive transactivator in the ROSA26 locus (ROSA26-nlsrtTA allele; Wutz et al. 2002 ; F. Savarese, unpublished data). In these fibroblasts, expression of the endogenous Xist RNA from the single male X chromosome could be induced in 80% of the cells by addition of doxycycline (data not shown). In uninduced cultures and control male fibroblasts no H3K27m3 foci were detected by immunofluorescence in interphase nuclei. However, upon Xist induction 5% (after 48 h of Xist induction) or 15% (after 72 h) of the cells showed focal H3K27m3 staining (H4K20m1 was established, as well; see Figure S3G ). In control female fibroblasts H3K27m3 staining was detected in 85% of the cells. This shows that Xist induction in embryonic fibroblasts leads to H3-K27 methylation in a low percentage of cells. We further examined histological sections of male embryos carrying the inducible Xist-tetOP allele and the ROSA26-nlsrtTA allele. Xist expression was induced by feeding doxycycline in drinking water to the mothers for 3 d starting either from day 9.5 or day 13.5 of gestation. Embryos were dissected 3 d later, on day 12.5 and 16.5, respectively. In the sections, 74% (day 12.5 embryos) and 52% (day 16.5 embryos) of the cells expressed Xist, as determined by RNA fluorescent in situ hybridization (FISH) analysis ( Figure 4 D and 4 E). Focal H3K27m3 staining was detected in 61% of the cells in sections of the day 12.5 embryos but in only 18% of the day 16.5 embryos ( Figure 4 D and 4 E), demonstrating a clear reduction in the number of cells showing H3K27m3 staining in response to Xist expression in the later-stage embryos. In summary, our data demonstrate that Xist has been able to effect H3K27m3 in all cell types tested. However, the efficiency of methylation is regulated in cellular differentiation and development. Our experiments show that Xist is not sufficient for efficient establishment of the H3K27m3 pattern in differentiated cells. Reversibility of H3K27m3 Once efficient H3K27m3 is established by Xist expression in early ES cell differentiation, it can be maintained throughout differentiation. This would be consistent with the view that lysine methylation is a permanent epigenetic mark. To test whether H3K27m3 is stably maintained in the absence of continuous Xist expression, we tested H3K27m3 reversibility in undifferentiated clone 36 ES cells. Xist expression was induced from the transgenic Chromosome 11 in these cells for 3 days, and then the cells were washed and split into medium without doxycycline to shut off Xist expression. H3K27m3 levels and Xist RNA were determined by combined immunofluorescence RNA FISH at consecutive time points at 6, 12, 24, and 48 h. High levels of H3K27m3 persisted until 24 h after Xist was turned off, but H3K27m3 disappeared by 48 h ( Figures 5 A and S3B ). Our data show that the Xist RNA signal disappeared by 12 h after the withdrawal of doxycycline, demonstrating that H3-K27 methylation is reversible in undifferentiated ES cells and is removed after a period of approximately two cell divisions following the turning off of Xist expression. We also analysed the reversibility of H4K20m1 and Ezh2 in undifferentiated clone 36 ES cells. The percentage of cells showing a signal went from 46% and 70% initially to 5% and 11% at 48 h after withdrawal of doxycycline for H4K20m1 and Ezh2, respectively. Figure 5 Kinetic Study of H3K27m3 Stability (A) The percentage of interphase nuclei ( n > 100) showing H3K27m3 staining and Xist RNA was analysed for undifferentiated clone 36 ES cells, which expressed Xist for 3 d (+) or were further grown without inducer for 6, 12, 24, or 48 h. (B) Representative images of the time points analysed in (A) are shown. (C) Reversibility of H3K27m3 in differentiating clone 36 ES cells. The percentage of interphase cells showing H3K27m3 staining ( n > 100) was determined for cells differentiated for 4 d in the presence of doxycycline (+) or further differentiated for 48, 72, or 96 h in the absence of inducer. To test whether H3K27m3 would become irreversible during ES cell differentiation, we turned off Xist expression in clone 36 ES cells at progressively later time points up to 6 d after initiation of differentiation. The H3K27m3 pattern was analysed in all cultures at day 12 of differentiation. In cells continuously expressing Xist during differentiation, methylation was detected in 60% of the cells at day 12. If Xist expression was turned off at any time points in the course of differentiation, the percentage of cells showing H3K27m3 was reduced to less than 10%, suggesting that methylation was reversible throughout differentiation and not stabilised (data not shown). We then analysed the kinetics of loss of methylation in differentiated ES cells. Clone 36 ES cells differentiated for 4 d in the presence of doxycycline were differentiated for 24, 48, and 72 more hours in the absence of doxycycline, and H3K27m3 was measured ( Figure 5 C). Focal H3K27m3 staining was initially observed in 97% of interphase nuclei and was reduced to 50% and 25% at 3 and 4 d, respectively, after Xist had been turned off. This shows that the decay of focal H3K27m3 was slower than in undifferentiated ES cells, possibly reflecting the slower cell cycle of differentiating cells. Early Xist Expression Triggers a Chromosomal Memory Independent of Silencing Detection of focal H3K27m3 staining persisted throughout ES cell differentiation when Xist was continuously expressed. Yet the methylation mark was reversible throughout ES cell differentiation, and Xist RNA could only establish an efficient methylation pattern during the initiation window early in ES cell differentiation. These observations could indicate that silencing enhances histone methylation in ES cell differentiation. To address this interpretation, we analysed the H3K27m3 pattern caused by expression of a mutant Xist RNA lacking repeat A, which cannot initiate silencing, in differentiating J1:XistΔSX-tetOP ES cells. When these cells were differentiated in the presence of doxycycline, focal H3K27m3 staining was observed in 78% of the cells at day 12 ( Figure 6 ). This clearly indicated that methylation was maintained in a high number of these cells. Silencing is therefore dispensable for methylation in ES cell differentiation. Notably, we observed H3K27m3 staining in a high percentage of differentiated ES cells but in a significantly reduced percentage of undifferentiated ES cells expressing a silencing-defective Xist RNA (see Figure 3 B, 3 C, and S3C ). Silencing or repeat A sequences are therefore required to sustain high H3K27m3 levels specifically in undifferentiated ES cells but are dispensable upon differentiation. Figure 6 Early Xist Expression Imparts a Chromosomal Memory Independent of Silencing Transgenic Xist expression was induced from Chromosome 11 in clone 36 ES cells (black bars) or a silencing-deficient Xist RNA from the X in J1:XistΔSX-tetOP ES cells (open bars) at time points during differentiation (see text). The percentage of cells showing H3K27m3 staining is plotted ( n > 250). Below, a scheme of Xist induction is given for all cultures, with arrows representing time of analysis. To test whether continuous Xist expression was required for maintenance of efficient H3K27m3, we induced Xist expression from the transgenic Chromosome 11 in undifferentiated clone 36 cells and from the X chromosome in J1:XistΔSX-tetOP ES cells for 3 d. The cells were then differentiated for 5 d in the presence of doxycycline followed by 5 and 7 d, respectively, without the inducer. At the end of this period H3K27m3 was analysed and could be detected in less than 20% and 10% of the cells, respectively ( Figure 6 ). Parallel cultures were differentiated for 5 d in the presence of doxycycline followed by 4 d in the absence of doxycycline, and then doxycycline was added back for 1 or 3 more days. In these cells, in which Xist had been induced early, H3K27m3 was restored and detectable in 50%–55% of all cells. This level is significantly higher than the level in control cultures that had been induced de novo at day 6 or day 9 of differentiation (10% of all cells). In cells that had been continuously differentiated in the presence of doxycycline, methylation was detected in 73%–78% of the nuclei. Our data show that efficient methylation at late time points in differentiation did not require continuous Xist expression. Efficient remethylation occurred on a chromosome that had been exposed to Xist in early ES cell differentiation, consistent with the idea that Xist triggers a chromosomal change in early differentiation that is remembered until later time points to enhance H3K27m3 reestablishment. Importantly, the silencing-deficient Xist mutant RNA in J1:XistΔSX-tetOP ES cells gave identical results, showing that this memory is established independent of silencing. We further determined at which time point in differentiation the chromosomal memory is established. For this, clone 36 ES cells were differentiated for 0, 12, 24, 36, 48, 60, or 72 h in the presence of doxycycline. Then Xist was turned off until day 8 of differentiation, when doxycycline was added back, and remethylation was assayed by immunofluorescence at day 13 in differentiation ( Figure 7 ). In this experiment a transition occurred in a 24-h interval around 60 h if Xist was expressed for more than 48 h early in differentiation, allowing for efficient remethylation, a result consistent with the establishment of the memory in this interval. When Xist was turned off earlier than 60 h, remethylation was observed in only 10%–30% of the cells, demonstrating that the memory was not established. Turning off Xist at 72 h or later allowed remethylation in 85% of the cells. We also analysed the transition from Xist -dependent reversible to irreversible silencing in this experiment by Northern analysis of puro expression from the transgenic chromosome in differentiated clone 36 ES cells ( Figure 7 B). These data show that irreversible silencing was established in an interval between 48 and 72 h in ES cell differentiation, with puro expression levels dropping from 60% to 15% of the level in uninduced samples ( Figure 7 C), in agreement with our initial report ( Wutz and Jaenisch 2000 ). The 24-h intervals for the transition can be explained by the asynchronous cell cycle states in the ES cell culture (doubling time, 21.4 h) at the time when differentiation was induced. We conclude that the establishment of the chromosomal memory is silencing independent and occurs at the time when X inactivation becomes irreversible and Xist independent. Figure 7 Establishment of Chromosomal Memory during ES Cell Differentiation (A) Clone 36 ES cells were differentiated for 13 d in the presence of doxycycline (lane 1) or in the absence of inducer (lane 2) and the percentage of cells with H3K27m3 staining was determined ( n > 800). At the beginning of differentiation, parallel cultures received either no Xist induction (lane 3) or a pulse of doxycycline for 24 h (lane 4), 36 h (lane 5), 48 h (lane 6), 60 h (lane7), or 72 h (lane 8) followed by withdrawal of inducer and concerted late induction from day 8 to day 13. A dashed red line indicates the 24-h interval of the transition when the chromosomal memory is recruited. (B and C) Establishment of irreversible transcriptional silencing during differentiation. (B) Ectopic inactivation of Chromosome 11 caused by Xist induction in differentiating clone 36 ES cells was assessed by Northern blot analysis of PGKpuromycin (puro) and Gapd as a loading control. Lanes were aligned electronically for better readability. ES cells were differentiated for 13 d in the presence of doxycycline (lane 1) or in the absence of inducer (lane 2). At the start of differentiation, parallel cultures received a Xist pulse for 24, 36, 48, or 60 h followed by withdrawal of inducer for the rest of the time (lanes 3 to 7) or followed by reinduction of Xist at day 8 of differentiation (lanes 8 to 11). All cells were analysed at day 13 of differentiation. (C) A quantitation of the puro expression relative to Gapd was derived from two independent Northern blots using tnimage software. A dashed red line indicates the 24-h interval in which the transition from reversible to irreversible silencing occurs. Discussion Our results identify H3K27m3 and H4K20m1 as specific modifications that mark the Xist -expressing chromosome in undifferentiated ES cells and contribute to the epigenetic histone code of the Xi ( Table 1 ). We did not observe an enrichment of H3K9m2 or H3K9m3 signals on the Xist -expressing chromosome, which has been reported by other studies. This could be a shortcoming of our transgenic system, but we also did not detect the H3K9m2 or H3K9m3 signals in female mouse primary embryonic fibroblasts (less than 2% of the cells). We attribute the different observations in other studies to the various antisera used. We supply peptide blot analysis for our antisera that suggest that the antibodies are highly specific (see Figure S1 ). This is also supported by the specific staining patterns in immunofluorescence experiments. The lysine 9 methylation signal observed in other studies could potentially be a result of cross reactivity with H3K27m3, a fact we can exclude for our H3K9 antibodies based on the staining pattern and peptide blots. Alternatively, our antibody might not recognise the H3K9m2 modification in the context of the chromosome. However, this is unlikely since the H3K9m3 signals for the pericentric regions and the Y chromosome are clearly identified. The H3K9m2 antiserum has been successfully used in ChIP analysis of the minor centromeric repeats ( Yan et al. 2003 ) and reacts with these repeats in immunofluorescence, but does not show cross reactivity to H3K27m3. This suggests that our reagent is able to detect the modification in both ChIP and immunofluo-rescence experiments. Using highly specific antisera, we failed to see a strong signal for H3K9m2 in either ChIP or immunofluorescence experiments (see Figures 2 and S2E ). In our ChIP analysis two chromosomal loci showed an increase for H3K9m2 upon Xist expression in differentiated ES cells, suggesting some enrichment for H3K9m2. We take these data to indicate that H3K9m2 is not a prominent mark of X inactivation but might be enriched locally to some degree upon differentiation. Using Xist alleles that express a mutated version of Xist, which has a deletion of repeat A sequences and is unable to cause silencing, we showed that both H3K27m3 and H4K20m1 were established in the absence of transcriptional repression. This demonstrates that neither modification is sufficient to trigger silencing. Xist expression led to rapid H3K27m3, which was complete after 1 to 2 d of Xist expression in both ES cells and differentiated cells (see Figure S3A and Figure 6 , columns 1 and 2). This kinetics follows the localisation of Xist RNA, which accumulates between 4 and 12 h after doxycycline addition in ES cells ( Wutz and Jaenisch 2000 ), suggesting that H3K27m3 is an immediate effect. We have further shown that in undifferentiated ES cells no progressive accumulation of the histone modifications occurs over time by comparing the percentage of cells showing H3K27m3, H4K20m1, and Ezh2 staining after 3 and 10 d expressing either full-length Xist RNA or a silencing-deficient mutant lacking repeat A (see Figure S3C ). We have shown that H3K27m3 is a reversible modification throughout ES cell differentiation and depends at all stages on Xist expression. In undifferentiated ES cells H3K27m3 disappeared 48 h after Xist expression was turned off, corresponding to about two cell divisions. The kinetics would be consistent with the idea that replication is involved in the replacement of methylated histones, albeit our data do not rule out an active enzymatic process of demethylation. Importantly, we have observed nearly unchanged methylation levels 24 h after Xist expression has been turned off (see Figure S3B ). This could reflect the intrinsic stability of the trimethylation mark or the persistence of the Eed/Ezh2 complex, which can stably associate with metaphase chromosomes from which Xist RNA is displaced (see Figure 1 C; Mak et al. 2002 ). The transient maintenance of H3K27m3 might be significant for the mechanism of X inactivation. It could explain our observation that the inactive state will be “locked in” roughly 24 h after Xist loses its ability to initiate silencing, it will be locked in at 72 h of ES cell differentiation ( Wutz and Jaenisch 2000 ). Efficient methylation is established only when Xist expression is induced early in ES cell differentiation. The window in which Xist causes efficient methylation overlaps precisely with the initiation window, in which transcriptional silencing can be initiated. Yet methylation is independent of initiation of silencing. This would be consistent with the notion that H3K27m3 is necessary but not sufficient for silencing. However, this is unlikely, as a previous report has shown that in Eed mutant embryos, initiation of silencing is normal, but a defect in the maintenance of the inactive state leads to reactivation at later stages ( Wang et al. 2001 ). Lower levels of Ezh2 and Eed could explain the restriction on the ability of Xist to induce H3K27m3 efficiently in differentiated ES cells ( Silva et al. 2003 ). We do not favour this interpretation, as this restriction is observed at day 2 in differentiation, when Ezh2 and Eed protein levels are still high (see Figure 4 C). Our data further show that the ability to efficiently methylate a chromosome late in ES cell differentiation is a feature of the chromosome and not a function of the protein levels of Eed and Ezh2. This is also in line with our observation that chromosome-wide H3K27m3 in clone 36 ES cells, in which Eed messenger RNA was reduced to 10%–15% of wild-type levels by stable RNAi, was still detected in 45%–60% of cells compared to 80% in control clone 36 cells (data not shown). Therefore, less abundant levels of Eed are sufficient to achieve efficient methylation. Xist induction later in ES cell differentiation or in cells of embryonic origin establishes H3K27m3 in only a small percentage of cells. The significance of H3K27m3 in this small number of cells is unclear at present. The restriction of efficient methylation to early ES cell differentiation and the finding that methylation is reversible logically require that a chromosomal memory exists that enables H3K27m3 maintenance during differentiation. Previous models have suggested that a lock-in of X inactivation is based on chromosomal silencing, arguing that self-maintaining heterochromatin structures establish the principal form of memory. Our data clearly demonstrate that H3K27m3 is maintained in the absence of transcriptional repression, suggesting a chromosomal memory independent of silencing on the Xi. Using the inducible Xist expression system we have directly demonstrated the chromosomal memory (see Figure 6 ). A chromosome that had been exposed to Xist and been H3-K27 trimethylated early could be remethylated later in differentiation, after a period where Xist was turned off and methylation decayed, with significantly greater efficiency than a chromosome that had not expressed Xist early (see Figure 6 ). We have further determined the time point in ES cell differentiation when the chromosomal memory is established and found that it overlaps with the transition from Xist -dependent and reversible silencing to irreversible silencing. These data place the establishment of the memory in a critical phase of X inactivation. We note that the establishment of efficient H3K27m3 in the initiation window and the implementation of the memory are separated by a gap of approximately one cell division in ES cell differentiation. This parallels the gap between initiation of silencing and the maintenance of the silenced state independent of Xist. Our kinetic measurements indicate that H3K27m3 would decay from the Xist -expressing chromosome after two cell divisions; therefore, H3K27m3 could bridge the gap (critical window). We suggest that Xist expression and H3K27m3 might be the signal to recruit a chromosomal memory mediating the lock-in of X inactivation ( Figure 8 ). In this model, silencing would be specified by separate signals depending on repeat A of Xist, which we predict would interact with the memory at the transition from reversible to irreversible and Xist -independent repression. In this regard we note that silencing or repeat A sequences enhance the efficiency of H3K27m3 in undifferentiated ES cells (see Figure 3 B). However, there is no requirement for repeat A when ES cells are induced to differentiate (see Figures 6 and S3C ). This could point to interactions between the silencing machinery and the Ezh2/Eed methylation complex specifically in ES cells. Figure 8 Model for the Transition from Initiation to Maintenance of X Inactivation Phases of X inactivation are given relative to days of ES cell differentiation (bottom). (A) In undifferentiated ES cells, efficient chromosome-wide H3K27m3 depends on both Xist RNA localisation to the chromosome in cis and initiation of transcriptional silencing via the A repeat (black triangles). (B) Early in differentiation, silencing becomes dispensable for high-level H3K27m3 (dotted arrow). (C) The beginning of the critical window is specified in that Xist loses its potential to trigger H3K27m3 (dotted arrow) and transcriptional silencing. The critical window is negotiated by sustaining high levels of H3K27m3, which is thought to constitute—together with Xist RNA—the signal for the recruitment of the chromosomal memory (black oval). The memory is established on the Xi exactly when silencing becomes irreversible and Xist independent. (D) During the maintenance phase of X inactivation the chromosomal memory allows Xist RNA to establish H3K27m3 efficiently. The molecular basis for the chromosomal memory is presently unknown. Our data rule out the possibility that continuous Xist RNA expression or silencing is required for maintenance of the chromosomal memory and suggest that H3K27m3 is also not involved. The latter interpretation has to be treated cautiously, as it depends on the sensitivity of our assay to detect H3K27m3. Formally it is conceivable that low levels of H3K27m3 undetected by our assay could remain on the chromosome. Presently, it is also unclear what the role of H4K20m1 is and to what extent it interacts with H3K27m3. A H4-K20–specific histone methyltransferase has been identified ( Fang et al. 2002 ; Nishioka et al. 2002 ; Rice et al. 2002 ), and we have performed in vitro functional analysis of the mouse Pr-Set7 protein ( Figure S5 ; Protocol S1 ). Our results indicate that Pr-Set7 is a monomethylase for H4-K20. Its involvement in X inactivation and the function of H4K20m1 remain unclear at present. Future work is needed to identify the components of the memory configuration and to determine its precise function in X inactivation. Materials and Methods Cell lines, culture conditions, and histological sections. Clone 36 ES cells ( Wutz and Jaenisch 2000 ) and J1:XistΔSX-tetOP, T20:Xist, and ES cells expressing Xist deletions ( Wutz et al. 2002 ) were cultured in DMEM (Biochrome, Berlin, Germany), 15% fetal calf serum (Euroclone, Milan, Italy), and 250 U of LIF/ml as described in those references. ES cells were induced to differentiate in ES medium without LIF by addition of all- trans -retinoic acid to 100 nM as described previously ( Wutz and Jaenisch 2000 ). Primary mouse embryonic fibroblasts were derived from day 13.5 embryos and grown in DMEM (Biochrome) and 10% fetal calf serum as described previously ( Wutz and Jaenisch 2000 ). Xist expression was induced by the addition of 1 μg/ml of doxycycline to the culture medium or was administered in drinking water (100 mg and 100 g of sucrose per liter). For sections, embryos were sexed ( Lambert et al. 2000 ) and fixed, and 10-μm-thick frozen sections were prepared. Mice were handled according to institutional guidelines. Immunostaining and Western blot. For metaphase chromosome spreads, cells were incubated for 15 min at 37 °C in RBS solution (10 mM Tris-HCl [pH 7.5], 10 mM NaCl, 5 mM MgCl 2 ), centrifuged for 10 min at 1,200 rpm onto Menzel SuperFrost slides (Roth, Karlsruhe, Germany) using a Cytospin 3 centrifuge (Thermo Shandon, Pittsburgh, Pennsylvania, United States). Staining was performed as described previously ( Peters et al. 2003 ). Briefly, slides were extracted for 10 min at room temperature (RT) in KCM (10 mM Tris [pH 8.0], 120 mM KCl, 20 mM NaCl, 0.5 mM EDTA, 0.1% [vol/vol] Tween-20) containing 0.1% (vol/vol) Triton-X100, fixed for 10 min at RT in 2% PFA/PBS, washed in KCM/0.1% Tween-20, and blocked for 30 min at RT in KCM containing 2.5% (wt/vol) BSA, 0.1% Tween-20, and 10% normal goat serum (Jackson ImmunoResearch, West Grove, Pennsylvania, United States). Primary antibodies were diluted in blocking solution and incubated overnight at 4 °C. After washes in KCM/0.1% Tween-20, slides were incubated with secondary antibodies for 1 h at RT, washed, and mounted (Vectashield; Vector Laboratories, Burlingame, California, United States). For analysis of interphase nuclei, differentiated ES cells were grown on Roboz slides (CellPoint Scientific, Gaithersburg, Maryland, United States) and undifferentiated cells were attached to poly- l -lysine coated coverslips or cytospun as described above. Immunostaining was performed as described previously ( Peters et al. 2003 ). Briefly, cells were fixed for 10 min at RT in 2% PFA in PBS, permeabilized for 5 min at RT in 0.1% Na Citrate/0.1% Triton-X100, blocked for 30 min at RT in PBS containing 2.5% (wt/vol) BSA, 0.1% Tween-20, and 10% normal goat serum, and processed as described above. Antibodies for histone lysine methylation states are described elsewhere ( Peters et al. 2003 ) and were used as follows (metaphase spreads/interphase): α-H3-K9m1 (IgG fraction of α-2x-monomethH3-K9, #4858, 1.7 mg/ml), 1:200/1:500; α-H3-K9m2 (IgG fraction of α-2x-dimeth H3-K9, #4679, 1.7 mg/ml), 1:100/1:200; α-H3-K9m3 (IgG fraction of α-2x-trimeth H3-K9, #4861, 1.3 mg/ml), 1:300/1:500; α-H3-K27m1 (IgG fraction of α-2x-monometh H3-K27, #8835, 0.7 mg/ml), 1:500/1:1,000; α-H3-K27m2 (IgG fraction of α-2x-dimeth H3-K27, #8841, 0.6 mg/ml), 1:500/1:1,000; α-H3-K27m3 (IgG fraction of α-2x-trimeth H3-K27, #6523, 1.1 mg/ml), 1:300/1:500. Additional antibodies were as follows: α-H3-K4m1 (α-monomethyl-Histone H3 [Lys4], #1799; Upstate Biotechnology, Lake Placid, New York, United States), 1:400/1:1,000; α-H3-K4m2 (α-dimethyl-Histone H3 [Lys4], #07-030; Upstate), 1:400/1:1,000; α-H3-K4m3 (α-trimethyl-Histone H3 [Lys4], #1819; Upstate), 1:700/1:1,000; α-H4-K20m1 (α-monomethyl-Histone H4 [Lys20], #07-440; Upstate), 1:100/1:200; α-H4-K20m2 (α-dimethyl-Histone H4 [Lys20], #07-367; Upstate), 1:200/1:200; α-H4-K20m3 (α-trimethyl-Histone H4 [Lys20], #07-463; Upstate), 1:350/1:500; polyclonal sheep α-H4Ac ( Morrison and Jeppesen 2002 ), 1:500/1:1,000; polyclonal rabbit α-Ezh2 ( Sewalt et al. 1998 ), 1:100/1:200. Secondary antibodies (Molecular Probes, Eugene, Oregon, United States) were as follows: Alexa A-11034 Fluor 488 goat antirabbit IgG (H+L), Alexa A-11036 Fluor 568 goat antirabbit IgG (H+L), and Alexa A-21099 Fluor 568 donkey antisheep IgG (H+L), all at 1:500. For Western blots, total nuclear extract was separated by SDS PAGE, blotted onto a PVDF membrane (Immobilon-P; Millipore, Bedford, Massachusetts, United States), blocked in blocking solution (PBS, 3% [wt/vol] BSA), and incubated with primary antibodies for 3 h. After washing three times for 10 min in TBST (50 mM Tris-HCl [pH 8.0], 100 mM NaCl, 0.1% Tween 20) and incubation with secondary antibodies (HRP; Jackson Laboratory, Bar Harbor, Maine, United States), detection was performed using ECL reagent (Amersham Pharmacia Biotech, Little Chalfont, United Kingdom). Rabbit polyclonal α-Eed (1:3,500), rabbit polyclonal α-Ezh2 (1:1,000), goat polyclonal α-histone H3 (1:800, #sc-8654; Santa Cruz Biotechnology, Santa Cruz, California, United States), and rabbit polyclonal α-histone H4 (1:300, #07–108; Upstate) were used. DNA FISH and RNA analysis. For DNA FISH analysis, biotin-labelled STAR*FISH mouse whole chromosome-specific probes (1187-YMB-02, 1187–11MB-01; Cambio, Cambridge, United Kingdom) were detected with streptavidin, Alexa Fluor 633 conjugateS-21375 (Molecular Probes). RNA FISH probes were generated by random priming (Stratagene, La Jolla, California, United States) using Cy3-dCTP (Amersham). Hybridisation and washing were carried out as described previously ( Wutz and Jaenisch 2000 ). Specimens were analysed using a fluorescence microscope (Zeiss Axioplan, Oberkochen, Germany) equipped with a CCD camera and the MetaMorph image analysis software (Universal Imaging, Downingtown, Pennsylvania, United States). Northern analysis was performed using 20 μg of RNA (Trizol; Invitrogen, Carlsbad, California, United States) as described previously ( Wutz et al. 2002 ). ChIPs. Cells were cross-linked with 1% formaldehyde for 10 min at RT and quenched with 125 mM glycine, and whole-cell extracts were prepared. ChIPs were performed in duplicates as described previously ( Martens et al. 2003 ). Briefly, 400 μg of fragmented chromatin (between 400 and 1000 base pairs) was used for immunoprecipitation, and DNA was extracted from the precipitates and analysed by real-time PCR using a Lightcycler (Roche Diagnostics, Basel, Switzerland). Results were corrected for nonspecific binding to the beads and presented as a percentage of the input DNA (4 μg of fragmented chromatin, 100%). Primers sequences were as follows: tubulin, CCTGCTGGGAGCTCTACT and GGGTTCCAGGTCTACGAA; puromycin, GCTGCAAGAACTCTTCCTC and GCCTTCCATCTGTTGCTG; d11mit117, AAAAGACCCTATTTACAATACAACTGA and TGTCATTTTTGATTAATCGCTCC; d11mit108, GGCACAAGAAAGACACAGCA and AAAGAGAAACCCCAGAGGGA; d11mit102, CCAGGAGAGCAGGAAGGTC and TCCTTCTGGGTGCTGCAT; d15mit15, AGCATACACTCTTGTTCCTGCT and AATAAATACCAGAGAAGCACCGTG. Supporting Information Figure S1 Specificities of H3-K9, H3-K27, H4-K20, and H3-K4 Mono-, Di-, and Trimethyl Antibodies Immunodotblot analysis ( Peters et al. 2003 ) of the antisera used to detect specific methylation states of histone H3 on Lysine 9 (A), H3 on Lysine 27 (B), H4 on Lysine 20 (C), and H3 on Lysine 4 (D). IgG fractions of the methyl-lysine histone antibodies were tested at various dilutions, with the most optimal dilution being displayed. Dotblots contain 0.4, 2, 10, and 50 pmol of linear H3 (amino acids 1–20; amino acids 19–34; amino acids 25–45; amino acids 72–91) and peptides, either unmodified or mono-, di-, or trimethylated at the K4, K9, K27, K36, or K79 positions. In addition, a linear H4 (amino acids 12–31) peptide, mono-, di-, or trimethylated at the K20 position, was also used. (611 KB PDF). Click here for additional data file. Figure S2 Histone Modification Pattern of the Inactive X Chromosome Immunofluorescence staining of metaphase spreads of clone 36 ES cells induced to express Xist for 3 d using H3K27m1 (A), H3K27m2 (B), H3K27m3 (C), H3K9m1 (D), H3K9m2 (E), H3K9m3 (F), H4K20m1 (G), H4K20m2 (H), H4K20m3 (I), H3K4m3 (J), and H4Ac (K) antisera. Chromosome 11 was identified by a DNA FISH probe (red; blue, DAPI) in (J) and (K). Clone 36 ES cells grown in the absence of doxycycline are used as a control for the H4Ac staining without Xist expression (L). (4.8 MB TIF). Click here for additional data file. Figure S3 Initiation and Maintenance of Histone Methylation during Differentiation (A) The kinetics of H3K27m3 was measured in undifferentiated clone 36 ES cells. The number of cells showing H3K27m3 staining 6, 12, 24, and 48 h after induction of Xist expression is shown. (B) The stability of H3K27m3 was determined in undifferentiated ES cells. The percentage of metaphase chromosome spreads ( n > 150) showing H3K27m3 staining was analysed in undifferentiated clone 36 ES cells, which expressed Xist for 3 d (lane 1) or were further grown without inducer for 24 h (lane 2) or 48 h (lane 3). This experiment complements data presented in Figure 5 A and 5 B providing a ‘cell cycle synchronous' view of the H3K27m3 decay kinetics. (C) Levels of H3K27m3 were measured in undifferentiated ES cells after 3 and 10 d of Xist expression. No progressive accumulation over time was observed, indicating that the steady state of H3K27m3 has been reached at 3 d Xist expression. However, a marked increase in methylation is observed in J1:XistΔSX-tetOP ES cells upon differentiation for 2 d (hatched bar). (D) Combined Xist RNA FISH (red) immunofluorescence analysis of Ezh2 and H4K20m1 in undifferentiated J1:XistΔSX-tetOP cells expressing Xist for 3 and 10 d (percentage of nuclei showing a staining is given). Analysis of H3K27m3 and H4 acetylation using an antiserum specific for multiply acetylated forms of H4 in clone 36 and J1:XistΔSX-tetOP ES cells that were grown for 4 d in the presence of doxycycline and then shifted to differentiation conditions for 2 d more in the presence of doxycylcine. (E) Male primary mouse fibroblasts (PMEFs) hemizygous for the inducible Xist-tetOP allele and homozygous for the tetracycline-inducible transactivator were induced with doxycycline for 2 d (lane 1) or 3 d (lane 2), and the number of cells showing H3K27m3 staining in interphase was analysed. Control female PMEFs showed a methylation signal in the large majority of cells (lane 3); uninduced male PMEFs were always negative. (F) Representative indirect immunofluorescence of uninduced (top) and induced (bottom) male Xist-tetOP PMEFs. The inducible Xist RNA triggers less pronounced and less dense foci of H3-K27 trimethylation (green) compared to the female wild-type control. (G) Upon Xist expression, H4-K20 monomethylation (green) is observed in interphase Xist-tetOP PMEFs (left). Focal enrichment colocalises with the site or Xist RNA clusters (red) on the X chromosome. Female wild-type PMEFs (right). (3.0 MF TIF). Click here for additional data file. Figure S4 Analysis of the XistΔXSa Mutation The XistΔXSa transgene was integrated by Cre-mediated recombination into the Hprt locus on the single X chromosome in T20 ES cells ( Wutz et al. 2002 ). A schematic representation of the Xist cDNA in given (top): repeats A to E are indicated by arrays of triangles, sequences mediating localisation to chromatin are indicated by boxes underneath (degree of hatching represents importance), and the location of the deletion is indicated by a coloured box. RNA localisation was analysed by FISH (lower left), showing that the RNA localises in small clusters in some cells. The ability of the RNA to induce silencing was measured by cell survival of differentiating cultures under induced versus uninduced conditions (lower right). Controls are cells either having a fully functional Xist cDNA transgene (Xist) or a cDNA lacking repeat A that is incompetent to induce silencing (ΔSX). The ΔXSa RNA shows poor silencing activity, presumably as a consequence of its failure to localise well to the chromosome. (1.5 MB TIF). Click here for additional data file. Figure S5 Selective H4-K20 Monomethylation Activity of Mouse Pr-Set7 In Vitro (A) Schematic presentation of full-length mouse PR/SET domain-containing protein 07 (Pr-Set7), indicating SET domain in black (gi:38080595). Below, region tested for histone methyltransferase (HMTase) activity. (B) Coomassie stain (left) shows purified recombinant GST-tagged Pr-Set7 (arrow), H4 peptides (arrowhead), and histones used for in vitro reactions with S-adenosyl-[methyl- 14 C]-L-methionine as methyl donor. Fluorography (right) indicates HMTase activity on the unmodified H4 peptide comprising residues 12–31 of the histone H4 N-terminus. Notably, no further methyl groups could be transferred to the same peptide if it had been synthetically monomethylated at residue H4 lysine 20 (K20m1) before usage in the in vitro reaction. Free histones are not accepted as substrate. (C) Fluorography indicates histone H4 HMTase activity of GST–Pr-Set7 selective for the unmodified histone H4 peptide (12–31). H4-K20 monomethylation obviously is the terminal state for Pr-Set7, because synthetically mono- (K20m1), di- (K20m2), and trimethylated H4 peptides (K20m3) could not be significantly methylated. (948 KB TIF). Click here for additional data file. Protocol S1 Supplementary Methods (22 KB DOC). Click here for additional data file. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC449785.xml |
545801 | MONKEY: identifying conserved transcription-factor binding sites in
multiple alignments using a binding site-specific evolutionary model | MONKEY is a new method for identifying conserved transcription-factor binding sites from multiple-sequence alignments. | Background Different types of genomic features have characteristic patterns of evolution that, when sequences from closely related organisms are available, can be exploited to annotate genomes [ 1 ]. Methods for comparative sequence analysis that exploit variation in rates and patterns of nucleotide evolution can identify coding exons [ 1 , 2 ], noncoding sequences involved in the regulation of transcription [ 3 , 4 ] and various types of RNAs [ 5 - 7 ]. While most of these methods have been developed for and applied to pairwise comparisons, sequence data are increasingly available for multiple closely related species [ 8 ]. It is therefore of considerable importance to develop sequence-analysis methods that optimally exploit evolutionary information, and to explore the dependence of these methods on the evolutionary relationships of the species in comparison. Sequence-specific DNA-binding proteins involved in transcriptional regulation (transcription factors) play a central role in many biological processes. Despite extensive biochemical and molecular analysis, it remains exceedingly difficult to predict where on the genome a given factor will bind. Transcription factors bind to degenerate families of short (6-20 base-pairs (bp)) sequences that occur frequently in the genome, yet only a small fraction of these sequences are actually bona fide targets of the transcription factor [ 9 ]. A major challenge in understanding the regulation of transcription is to be able to distinguish real transcription factor binding sites (TFBSs) from sequences that simply match a factor's binding specificity. Because the evolutionary properties of TFBSs are expected to be different from their nonfunctional counterparts, comparative analyses hold great promise in helping to address this challenge. In the past few years, several methods have been introduced to identify conserved (and presumably functional) TFBSs for a factor of known specificity (in contrast to the larger set of methods that use comparative data in motif discovery or to otherwise identify sequences likely to be involved in cis -regulation). Each of these methods explicitly or implicitly adopts one of several distinct definition of a conserved TFBS. These include a binding site in a reference genome that is perfectly or highly conserved [ 8 , 10 - 12 ]; a binding site in a reference genome that lies in a highly conserved region [ 4 ]; or a position at which the binding model predicts a binding site in all species [ 13 - 18 ]. In a previous study we characterized the evolution of experimentally validated TFBSs in the Saccharomyces cerevisiae genome, finding that functional TFBSs evolve more slowly than flanking intergenic regions, and more strikingly, that there is considerable position-specific variation in evolutionary rates within TFBSs [ 19 ]. We further showed that evolutionary rate at each position is a function of the selectivity of the factor for bases at that position. Our goal here is to incorporate these specific evolutionary properties of TFBSs into the search for conserved TFBSs. Or, more precisely, to develop a method that, given the specificity of a transcription factor, identifies conserved binding sites in multiple alignments by taking into account the sequence specificity and patterns of evolution expected for TFBSs, while still fully exploiting the phylogenetic relationships of the species being compared. In addition to developing new methods, there are several hypotheses regarding the comparative annotation of TFBSs that we are interested in testing. It has been noted that the effectiveness of such analyses will depend critically on the evolutionary distance separating the species used. At very close distances TFBSs will appear conserved because there has been insufficient time for substitutions to occur. As distance increases, and substitutions occur most rapidly at nonfunctional positions, our ability to detect constrained binding sites should improve until we are no longer able to reliably assign orthology based on sequence alignment. To overcome this problem of divergence distances exceeding what can be aligned, the sequences of multiple closely related species can be used to span the same evolutionary distances (and presumably provide the same discriminatory power) as fewer more distantly related ones. However, aside from these qualitative expectations, the dependence of the ability to identify conserved TFBSs on evolutionary distance and tree topology has not been rigorously investigated. Because the software MONKEY can be applied to multiple alignments of varying numbers of species and produces scores that can be meaningfully compared across different sets of species, we are now able to address these issues. Results Overview We developed an approach to identify conserved TFBSs that combines probabilistic models of binding-site specificity [ 20 - 22 ] with probabilistic models of evolution [ 23 , 24 ]. Starting with an alignment of sequences from multiple related species, we use the known sequence specificity for a transcription factor to compare the likelihood of the sequences under two evolutionary models - one for background and one for TFBSs. The central feature of this method that underlies its ability to identify conserved TFBSs is that it uses a specific probabilistic evolutionary model for the binding sites of each transcription factor. The evolutionary model we use for TFBSs [ 25 ] assumes that sites were under selection to remain binding sites throughout the evolutionary history of the species being studied. This model uses the sequence specificity of the factor to predict patterns and rates of evolution that recapitulate the patterns and rates observed in real TFBSs [ 19 ]. MONKEY: scanning alignments to identify conserved transcription factor binding sites MONKEY, our tool for identifying conserved TFBSs, takes as input a multiple sequence alignment, a tree describing the relationship of the aligned species, a model of a transcription factor's binding specificity and a model for background noncoding DNA. It returns, for each position in the alignment, a likelihood ratio comparing the probability that the position is a conserved binding site for the selected factor compared to the probability that the position is background. Extending matrix searches to multiple sequence alignments For the model of binding specificity, we use a traditional frequency matrix [ 20 - 22 ]. The values in the matrix - f ib - represent the probability of observing the base b (A, C, G or T) at the i th position in a binding site of width w . For the model of the background, we use a single set of base frequencies g b . A widely used statistic for scoring the similarity of a single sequence to a frequency matrix is the log likelihood ratio comparing the probability of having observed a sequence X of width w under the motif model (a frequency matrix, designated as motif ) to the probability of having observed X under the background model (designated by bg ), which can be easily reduced to: where X ib is an indicator variable which equals 1 if base b is observed at position i , and zero otherwise. This classifier can be motivated by the approximation that the data are distributed as a two-component mixture of sequences matching the frequency matrix and sequences drawn from a uniform background. In practice, we compute this score using a position-specific scoring matrix (PSSM) with entries, M ib = log( f ib / g b ), and find S for a particular w -mer by adding up the entries that correspond to the bases in the query sequence. In extending this to a pair of aligned sequences X and Y , we want to perform the same calculation on their common ancestor A . Since A is not observed, we consider all possible ancestral sequences by summing over them, weighting each by their probability given the data ( X and Y ), the phylogenetic tree ( T ) that relates the sequences, and a probabilistic evolutionary model [ 23 ]. We can write a new score representing the log-likelihood ratio that compares the hypothesis that X and Y are a conserved example of the binding site represented by the frequency matrix to the hypothesis that they have been drawn from the background: where R motif and R bg are rate matrices describing the substitution process of the binding site and background respectively. Using the conditional independence of the sequences X and Y on the ancestor, A , and writing T AX for the evolutionary distance separating sequence X from A , this becomes: The class of evolutionary models used by MONKEY define a substitution matrix, p ( X i | A i , t ) = e Rt , that represents the probability of observing each base at position i in the extant sequence ( X ) given each base in the ancestral sequence ( A ) after t units of evolutionary time or distance, given some rate matrix, R [ 23 ]. Since these models retain positional independence, we can rewrite this as: This can be extended to more than two sequences, that is, ( X , Y , ..., Z ), by replacing the probabilities of X and Y with the probability with the left and right branches of the tree below, and performing the calculation at the root. The probabilities of the left and right branches of the tree can be calculated recursively as has been described previously [ 23 ]. Once again, for practical purposes we can convert these scores to a PSSM, whose entries are given for the pairwise case by: where at each position we now index by the bases a and b in the two sequences. For multiple alignments of n species, each position requires 4 n entries. Evolutionary models The use of evolutionary models is critical to the function of MONKEY. Myriad of such models exist, and in principle all can be used in MONKEY. For the background, it is natural to use a model appropriate for sites with no particular constraint, such as the average intergenic or synonymous rates. MONKEY allows the use of the JC [ 26 ] or HKY [ 27 ] models, and here we use the latter with the base frequencies, rates and transition-transversion rate-ratio estimated from noncoding alignments assuming a single model of evolution over the noncoding regions (see details in Materials and methods). It is also possible to estimate the evolutionary model separately for each intergenic alignment, although the small size of yeast intergenic regions leads to variable estimates. In principle, the JC and HKY models can also be used for the motif, with rates set according to our expectation of the overall rate of evolution in functional binding sites, which has been estimated as two to three times slower than the average intergenic rate [ 19 ]. However, we have previously shown that there is position-specific variation in evolutionary rates within functional transcription factor binding sites [ 19 ] and that positions in a motif with low degeneracy in the binding-site model evolve more slowly than positions with high degeneracy; this relationship between the equilibrium frequencies and the position-specific evolutionary rates is accurately predicted by an evolutionary model from Halpern and Bruno (HB model) [ 25 ]. In using this model, we assume that sequences evolve under constant purifying selection to maintain a particular set of equilibrium base frequencies. The use of this model corresponds to a definition of a conserved TFBS as a sequence position where there has always been a binding site for the transcription factor. Although the model does not strictly require that a binding site be present in each of the observed species, positions lacking such sites will have lower probabilities as they require the use of less probable substitutions. The rate of change from residue a to b at position i in the motif is given by: where Q is the (position independent) underlying mutation matrix, which we set equal to the background model ( Q = R bg ), and f is the frequency matrix describing the specificity of the factor. Thus, for each position in the motif, the HB model predicts the rates of each type of substitution as a function of the frequency matrix, and the background model. Comparing hits for different factors and evolutionary distances: computing the null distribution To compare scores from different evolutionary distances and different factors, it is critical that we are able to assign significance to a particular value of the score. To do so, we need to compute the distribution of the score under the null hypothesis that the sequence is part of the background. Calculating a p -value for a score S in a single sequence requires the enumeration of all possible w -mers that have a score S or greater under the background model. For n aligned sequences this requires the enumeration all 4 wn possible sets of aligned w -mers with scores S or greater under the background model. While the number of possible alignments of n w -mers can be unmanageably large for even small values of n and w , because we treat each position independently we can enumerate these possibilities efficiently using an algorithm developed for matrix searches of single sequences [ 28 , 29 ]. Every observed score is a sum of w numbers, one from each column of the matrix. The probability of observing exactly score S is the number of paths through the matrix whose entries add up to S , weighted by the probability of the path. By converting the matrix to integers, we can compute this probability for all values of S recursively. We initialize P i ( S ) (the probability of observing score S after i columns in the matrix) by setting P 0 ( S ) = 1 for S = 0, and P 0 ( S ) = 0 for S ≠ 0. We then compute the values of the function for i = [1, w ] as follows: For aligned sequences, c represents a column in the alignment, and the sum is over all 4 n possible columns an alignment of n sequences. The probability distribution function (PDF) of scores is P w ( S ), and from this the cumulative distribution function (CDF), the probability of observing a score of S or greater, can be directly computed. Although in principle we can compute the probabilities to arbitrary precision, because the time complexity increases with the number of possible scores, we limit the precision to within approximately 0.01 bits. Figure 1 compares empirical p -values from 5,000 pairs of sequences evolved in a simulation (see Materials and methods) with those computed by this method, and shows that they agree closely. We have used this method to compute the CDFs for alignments of up to six species, and therefore can apply our method to most comparative genomics applications. We note, in addition, that the likelihood ratio scores are approximately Gaussian (data not shown). As the means and variance of the scores under each model can be computed efficiently (see Materials and methods) we can estimate p -values using a Gaussian approximation (Figure 1 ) when the number of sequences in the alignment is large. Heuristics for alignments with gaps The treatment of alignment gaps in identifying conserved TFBSs is somewhat problematic. One the one hand, nonfunctional sequences may be inserted and deleted over evolution more rapidly than functional elements [ 30 - 32 ], and thus the presence of a gap aligned to a predicted binding site could indicate that it is nonfunctional. On the other hand, alignment algorithms are imperfect, and must often make arbitrary decisions about the placement of gaps. We sought to design a heuristic that accommodated both these aspects of genomic sequence data by locally optimizing alignments for the purpose of comparative annotation of regulatory elements. The idea is to assign a poor score to regions of the alignment with a large number of gaps, but to locally realign regions with a small number of gaps to identify conserved but misaligned binding sites. To do this, we scan along the ungapped version of one of the aligned sequences - the 'reference' sequence. For each position in the reference sequence p r , we define a window in each other sequence around p s , the position in sequence s aligned to position p r . The window runs from p s - ( a + b ) to p s + w + ( a + b ), where a and b are the number of gaps in the aligned versions of sequences r and s in position p to p + w , where p is the position in the alignment of p r . For each subsequence of length w in the window, we calculate the percent identity to the reference sequence, and create an alignment of p r to p r + w (in the reference sequence) to the most similar word in the window of each other sequence. This locally optimized alignment is then scored. Note that if a and b are zero (meaning there are no gaps in the aligned sequence), no optimization is done. If a is too large (in most contexts greater than five) we exclude that region of the alignment from further. This heuristic encapsulates the idea that too many gaps are indicative of lack of constraint, but conservatively allows for a few gaps due to alignment or sequence imperfections. Application to Saccharomyces The genome sequences of several species closely related to the budding yeast Saccharomyces cerevisiae have recently been published and become models for the comparative identification of transcription factor binding sites [ 8 , 11 ]. We aligned the intergenic regions of S. cerevisiae genes to their orthologs in S. paradoxus , S. mikatae , S. bayanus and S. kudriavzevii genomes using CLUSTALW (see Materials and methods) and sought to evaluate the effectiveness of MONKEY under different evolutionary models and distances. Ideally, we would use several diverse transcription factors with known binding specificity, where the set of matches to the factor's matrix in the S. cerevisiae genome could be divided into two reasonably sized sets: those known to be bound by the factor (positives) and those known not to be bound by the factor (negatives). Unfortunately, even in yeast, the number of such cases is limited. For many factors we can identify true positives by combining high- and low-throughput experimental data that supports the hypothesis that a particular position in the genome is bound by a given factor. A true negative set, however, must be constructed on the basis of lack of evidence that a sequence is functional, as the interpretation of negative results almost always is ambiguous. In the case of transcription factor binding sites this is particularly problematic, because DNA-binding proteins have overlapping specificity, and we may therefore observe conservation of a binding site because it is bound by another factor with similar specificity. After evaluating all factors with binding specificity in Saccharomyces cerevisiae Promoter Database (SCPD) [ 33 ], we focus on Gal4p and Rpn4p for further analysis (see Table 1 for properties of these factors, and Materials and methods for a description of the selection of positive and negative sets). The effects of evolutionary models on the discrimination of functional binding sites To evaluate the performance of our evolutionary method in correctly identifying bona fide binding sites, we calculated the p -values of the positive and negative sites for each factor, using MONKEY on alignments of all five genomes for Rpn4p and four species (with S. kudriavzevii excluded because too few sequences were available) for Gal4p. We compared the performance of MONKEY with the HB model to scores from S. cerevisiae alone and to a 'simple' score (equal to the average of the single sequence log likelihood ratios) that utilizes all the comparative data without an evolutionary model. The results are summarized in Table 2 . An ideal scoring method would assign low p -values to real sites (positives) and high p -values to spurious sites (negatives), and we therefore compared the p -values assigned by monkey based on the HB model to those based on the 'simple' score. Not surprisingly, both methods were a great improvement over searching in S. cerevisiae alone. Overall, when compared to each other, the HB score assigned lower p -values to the binding sites more often in the positive sets (90% for Gal4p and 80% for Rpn4p) and less often in the negative sets (20% for Gal4p and 25% for Rpn4p) than did the simple score. We note that some of the supposedly functional Rpn4p sites were assigned higher p -values in S. cerevisiae alone, suggesting that they are not in fact conserved; these will be discussed below. The effect of evolutionary distance on the discrimination of functional binding sites As evolutionary distance increases, we expect fewer matches to the matrix to be conserved by chance, which implies that the probability of observing matches as highly conserved as the functional sites should decrease. Similarly, we expect the nonfunctional sites to show many substitutions and their p -values to increase over evolution. To explore the change in p -values over evolutionary distance, we scored the functional and nonfunctional sets of binding sites at a variety of evolutionary distances by creating alignments of different combinations of species (see Materials and methods). The median p -value of the positive set of TFBSs decreases monotonically with evolutionary distance, with the rate of decrease an approximately constant function of evolutionary distance (see Figure 2 ). The median p -value for the binding sites in the negative set increases with evolutionary distance, although somewhat erratically. This demonstrates that MONKEY effectively exploits evolutionary distance, and confirms our intuition that as evolutionary distance increases, functional elements should be increasingly easy to distinguish from spurious predictions. To test this hypothesis on a more quantitative level we sought to compare the observed scores with the expected scores assuming that binding sites evolved precisely according to the evolutionary models used by MONKEY. Briefly, given a binding-site model and a phylogenetic tree, we assume we have observed a binding site in the reference genome, and that this site evolves along the tree under either the motif model (HB) or background model (HKY), representing functional and nonfunctional binding sites, respectively (see Materials and methods for details). The expected p -values associated with the functional binding sites (Figure 2 , solid lines) showed reasonable agreement with the models, consistent with previous observations that they are evolving under constraint that is well modeled by the purifying selection on the base frequencies in the specificity matrix [ 19 ]. Pairwise versus multi-species comparisons The comparisons at the different evolutionary distances used in Figure 2 employed variable numbers of species, with the shorter distances representing primarily pairwise comparisons and the longer distances comparisons of three or more species. While we expect the variation in p -values with different combinations of species to be primarily a function of the evolutionary distance spanned by these species, there will also be effects related to the number of species and the topology of the three. For example, in the limit of very long branch lengths, the evolutionary p -values are on the order of the power of the number of species and are independent of evolutionary distance. In contrast, in the limit of very short branch lengths, the evolutionary p -values depend only on the distance spanned by the comparison, as most of the information provided by additional species is redundant. However, because most comparisons that are actually carried out are far from either of these extremes, we sought to evaluate the effects of species numbers and tree topology for the Saccharomyces species analyzed here. First, we recomputed the expected p -values for all the distances analyzed in Figure 2 , except that instead of using the real tree topology, we used a single pairwise comparison at the same evolutionary distance (Figure 2 , dotted lines). For example, for the Rpn4p analyses using all five species we assumed a pairwise comparison at an evolutionary distance of around 1.1 substitutions per site. Note that this is considerably more distant than any of the pairwise comparisons available among these species. The predictions for the pairwise and multi-species comparisons are very similar, suggesting that at the evolutionary distances spanned by these species there is little difference in using multiple species alignments relative to a pairwise alignment that spans the same evolutionary distance. Only at the longest distances considered (greater than 0.8 substitutions per site) does the power of the pairwise comparison begin to level off, although there are other reasons that multiple species comparisons might still be preferred (see Discussion). To complement this theoretical analysis, we were interested in using empirical data to compare pairwise and multi-species analyses. Fortuitously, the evolutionary distance between S. cerevisiae and S. kudriavzevii is almost exactly equal to the evolutionary distance spanned by S. cerevisiae , S. paradoxus and S. mikatae (median tree length approximately 0.5 substitutions per site; see Figure 3a ). Because our models predict that we are in a regime where evolutionary distance is the primary determinant of the p -values, we expect searches using these different sets of species to yield similar results. We tested this hypothesis by calculating the p -values associated with the Rpn4p-binding sites using the sequences from these two comparisons. The median p -values in both the positive and negative sets are very similar (Figure 3b ), confirming that at these relatively short evolutionary distances, the power of the comparative method is independent of the number of species considered (see Discussion). Taken together, these results strongly support the idea that when appropriate methods are used, data from multiple species can be combined effectively to span larger evolutionary distances. Note that this in no way implies that the addition of extra species to an existing pairwise comparisons is not useful - such additions will always increase the evolutionary distance spanned by the species and thus will increase the power of the comparison. Testing the power of comparative annotation of transcription factor binding sites At the distances spanned by all available sequence data, the p -values are so small that we no longer expect to find matches of the quality of those in the positive set by chance, especially for Rpn4p. To test this further, we scanned both strands of all the available alignments of all five sensu stricto species (around 2.7 Mb) to identify our most confident predictions of conserved matches to the Rpn4p matrix. We chose the p -value cutoff of 1.85 × 10 -8 , which corresponds to a probability of 0.05 of observing one match at that level over the entire search (using a Bonferroni correction for multiple testing). After excluding divergently transcribed genes, there were 56 genes that contained putative binding sites at that p -value. Of 32 genes in our positive set that had sequence available for all five species, 30 had binding sites below this p -value. Of the 28 genes in the negative set for which sequences were available, only three had binding sites below this cutoff. In this (nearly ideal) case we have ruled out nearly 90% of the negative set at the expense of less than 10% of the positives. Examining the expression patterns of these genes (Figure 4a ) allows them to be divided into three major classes. The first is a group (indicated by a blue bar) containing 30 genes (28 of which were in our original positive set and two other genes) that show a very similar pattern over the entire set of conditions. The second group (indicated by a green bar) contains 11 genes (of which only one was in our original positive set) that show uncoordinated gene expression changes in some conditions in addition to the stereotypical Rpn4p expression pattern. It is possible that these genes' regulation is controlled by multiple mechanisms under different conditions [ 34 ], and regulation by Rpn4p is one contribution to their overall pattern of expression. Further supporting this hypothesis, only one of these genes ( UFD1 ) is annotated as involved in protein degradation, and three ( YBR062C , YOR052C and YER163C ) have unknown functions. Finally, and most surprising from the perspective of comparative annotation, is a third set of 14 genes, including one from our original positive set and three from our negative set, most of which show no evidence of the proteasomal expression pattern associated with Rpn4p (Figure 4b ). It is extremely unlikely that these sequences have been conserved by chance, and we suggest that they represent matches that are conserved for reasons other than binding by Rpn4p (see Discussion). Nonconserved binding sites in regulated genes Having identified examples of conserved binding sites whose nearby genes showed no evidence of function, we decided to examine the converse: binding sites near regulated genes, and therefore presumably functional, that are not conserved. Figure 5 shows the p -values of individual positive Rpn4p sites at different evolutionary distances. While most of the sites follow the trajectory predicted for sites evolving under the HB model, the p -values for four of the positive sites seem to be well-modeled by the 'background' or unconstrained model. This is surprising because we expect these binding sites to be functional, and therefore under purifying selection. One explanation is that some of these sites may have been misannotated as functional. For example, in addition to a nonconserved positive site, the upstream region of REH1 contains another binding site that is a weaker match to the Rpn4p matrix (Figure 5b ) and did not pass our threshold for inclusion in the positive set (see Materials and methods). This weaker match is more highly conserved and may represent the functional site in this promoter. In the case of PTC3 , however, we can find no other candidate binding sites nearby (Figure 5c ). This represents a possible example of binding-site gain, a proposed mechanism of regulatory evolution at the molecular level (see Discussion). Different factors have different relationships between significance and evolutionary distance The optimal selection of species for comparative sequence analysis remains an open question. To analyze this question for transcription factor binding sites, we examined the relationship between evolutionary distance and the MONKEY p -values for several S. cerevisiae transcription factors (Figure 6 ) for which sufficient characterized binding sites were available in SCPD [ 33 ]. We find that while all factors show the tendency for p -values to decrease with evolutionary distance, the p -values for each factor remain very different. For example, with alignments of four species spanning about 0.8 substitutions per site, we expect a conserved match to the Gcn4p matrix as good as the median functional binding site (Figure 6a , red triangles) approximately every million bases of aligned sequence. This in contrast to Rpn4p, for which in the same alignments we expect such a match (Figure 6a , violet crosses) only once in about 1 billion base pairs. Thus, the evolutionary distance required to achieve a desired p -value is different for different factors. Understanding the relationship between a frequency matrix and the behavior of its p -values is an area for further theoretical exploration. We note that, once again, we can predict the behavior of these p -values (Figure 6b ), and that while our predictions agree qualitatively, there is considerable variability. Software MONKEY is implemented in C++. It is available for download under the GPL and can be accessed over the web at [ 35 ]. Discussion By formulating the problem of identifying conserved TFBSs in a probabilistic evolutionary framework, we have both created a useful tool (MONKEY) for comparative sequence analysis capable of functioning on relatively large numbers of related species, and enabled the examination of several important questions in comparative genomics. While most previous approaches to this problem have used heuristics to define conserved and nonconserved TFBSs, with the probabilistic scores and p -value estimates presented here the assumptions underlying our approach can be made explicit, and where those assumptions hold we can be assured the reliability of our method. In addition, the probabilistic framework allows us to estimate the amount of evolutionary distance required to achieve a certain level of significance. Evolutionary models The score based on the evolutionary model proposed by Halpern and Bruno [ 25 ] effectively discriminated the functional and nonfunctional Gal4p- and Rpn4p-binding sites in S. cerevisiae (Table 2 ). We believe the success of the HB model in predicting position-specific rates of evolution [ 19 ] and identifying conserved TFBSs reflects its encapsulation of a model of binding sites evolving under constant purifying selection. Although not every functional binding site will remain under purifying selection, as a result of either functional change or binding-site turnover (see below), a large subset of functional binding sites do remain under purifying selection, and for these, the 'HB' score performs better than the 'simple' score. It is interesting to note, however, that the simple score, which is not based on an evolutionary model and does not take into account the relationships of the species used in the comparison, still shows great improvement over one genome alone, highlighting the value of comparative sequence data even when used suboptimally. Effects of evolutionary distance An important hypothesis of the comparative genomics paradigm is that as evolutionary distance increases, observing a match with a given level of conservation should become less and less likely by chance - the p -values for functional sites that are conserved are expected to decrease. We confirm this hypothesis for a small number of factors from S. cerevisiae . In addition, our probabilistic models allow us to quantify this relationship. We can directly measure the confidence that a specific site is a conserved binding site, and we can predict the evolutionary distance needed to achieve a desired level of significance. Typical p -values for functional binding sites scored by matching a matrix to a single genome are on the order of 10 -4 to 10 -6 . Even in a relatively small genome like yeast, with roughly 12 million bases, we expect many matches at this significance level to occur by chance. Adding four closely related species that span a total evolutionary distance of approximately one substitution per site reduces these p -values by approximately three orders of magnitude to the range 10 -7 to 10 -9 . In the yeast genome we expect few, if any, matches to occur at this level of significance by chance. When we search the alignments of these species with the Rpn4p matrix with a low enough p -value that we expect a match at that significance to occur only once in a random 50 Mb genome, we recover nearly the entire positive set of Rpn4p-binding sites while excluding most of the negative set, highlighting the utility of MONKEY and the statistics we have developed. As a measure of the improvement over searching a single genome alone, we note that even the best possible match to the Rpn4p matrix in one genome does not meet this significance criterion. The expected relationship between evolutionary distance and p -value can, in principle, be used to guide to choice of species to be sequenced for comparative analyses. However, the dependence of p -values on evolutionary distance is not the same for all factors (Figure 6 ). This suggests that our ability to annotate functional sequences by comparative methods will depend on the type of sequences that we are trying to annotate, and that there is no single evolutionary distance sweet-spot for identifying TFBSs. Pairwise versus multiple species comparisons In theory, for a given reference genome it should be possible to pick a single comparison species at an evolutionary distance sufficient to identify any conserved feature of interest. Our results suggest that at distances of up to approximately 0.6 substitutions per site, pairwise alignments provide essentially the same amount of resolving power as multiple comparisons spanning the same evolutionary distance. We showed that S. cerevisiae and S. kudriavzevii span almost exactly the same evolutionary distance as S. cerevisiae , S. paradoxus and S. mikatae , and that that distance is well below 0.6 substitutions per site. Consistent with this, MONKEY produces nearly identical p -values for conserved binding sites from these two sets of species. Thus, our results suggest that from a theoretical perspective, if the goal of comparative analysis is to identify conserved binding sites for factors like the ones considered here, it is not necessary to sequence species much more closely related than this limit. We note, however, that there are myriad practical reasons other than evolutionary resolving power (the only factor considered in our models) for sequencing multiple closely related sequences. First, there may simply be no extant species at the exact evolutionary distance desired. Second, the quality of DNA alignments is expected to be much higher for multiple closely related species than for more distant pairwise alignments - if alignment errors prevent correct assignment of orthology, conserved binding sites will not be identified. For the factors considered here, the pairwise comparison performed nearly as well as the multiple species comparison well beyond the evolutionary distances at which pairwise alignments are reliable [ 36 ], suggesting that the necessity of alignment will limit the maximum distance between species. Finally, and perhaps most important, is the assumption that our models make about constant functional constraint over evolution. To illustrate this, consider the binding sites for Gal4p used in the analysis in Figure 2a . These binding sites could not be included in Figure 3 because S. kudriavzevii orthologs for these genes were not available in SGD, apparently because of the degeneration of the galactose-utilization pathway in this species [ 37 ]. Sequencing multiple closely related species provides insurance against such functional changes, because they are less likely to have occurred in all the lineages. Conserved sites and binding-site turnover MONKEY was very effective in identifying functional Rpn4p-binding sites from the alignment of five Saccharomyces species. In our search, 41 of 56 (73%) predicted sites were found near genes showing the expected expression pattern, and are therefore likely to be functional. Even at this level of stringency, however, there are highly conserved sequences that match the matrix, but do not appear to be near genes that are regulated by Rpn4p. It is very unlikely that these sites are conserved by chance. One possible explanation for this high degree of conservation is that these are functional sites, but that the expression of these genes is not accurately detected in high-throughput assays, or their function has not been accurately determined. A more likely possibility is that these sites are conserved because they perform other, unknown functions. Consistent with this hypothesis is the fact that many of these matches fall near other highly conserved sequences (Figure 4b ), suggesting that they may be parts of larger conserved features. In addition to the conserved sequences that are unlikely to represent bona fide binding sites, we also found examples of binding sites associated with properly regulated genes that do not seem to be conserved (Figure 5 ). Once again there are several possible explanations for this observation. First, these binding sites may not actually be functional and may have been included in our positive set erroneously. While this is a possible explanation for the case of the Rpn4p-binding sites shown in Figure 5 (and may be likely in the case of REH1 , where we could identify another apparently conserved binding site in the region) we have also found nonconserved examples among the TFBSs in the SCPD database (approximately 20% of TFBSs we examined, see Additional data file 1), all of which have at least some direct experimental support. Another potential explanation is that these binding sites are actually conserved, but were not aligned correctly. While this is difficult to rule out in general, in the few nonconserved cases for Rpn4p at least we could not find (by eye) errors in the alignments. Most interesting, of course, would be the situation where these nonconserved binding sites are not due to some error on our part, but rather represent a biological change in the functional constraints on these sequences, possibly resulting in a change in the regulation of the expression of these genes. Our results represent an upper bound on the number of TFBSs for which this has occurred. Cis -regulatory changes have been proposed to be an important source of genetic variation [ 32 ]. Gains and losses of functional binding sites represent an important class of these changes [ 38 , 39 ], and an important area for future computational and experimental analysis, particularly as the genome sequences of closely related metazoans become available. We expect MONKEY to be a useful tool in the comparative analysis of these genomes, and we have found comparable increases in the significance of functional binding sites in alignments of Drosophila melanogster and D. pseudoobscura (see Additional data file 2). Conclusions We have developed a method to identify conserved TFBSs in sequence alignments from multiple related species that provides a quantitative framework for evaluating results. The method - implemented in the open-source software MONKEY - extends probabilistic models of binding specificity to multiple species with probabilistic models of evolution. We have found that a probabilistic evolutionary model [ 25 ] that assumes binding sites are under constant purifying selection performs effectively in discriminating functional binding sites. We have developed methods to assess the significance of hits, and have shown that the significance of functional matches increases while the significance of spurious matches decreases over increasing evolutionary distance. We can explicitly model the relationship between the significance of a hit and evolutionary distance, allowing the assessment of the potential of any collection of genomes for identifying conserved binding sites. Applying MONKEY to a collection of related yeast species we find that most functional binding sites are highly significantly conserved, but also find evidence for conserved sites that are not functional and vice versa . Our results suggest that development of methods that model the evolutionary relationships between species and the evolution of the genomic features of interest yield insight into the challenges for comparative genomics. Materials and methods Simulating pairs of sequences To generate the empirical p -values shown in Figure 1 , random sequences of length w were generated according to the average intergenic base frequencies of the S. cerevisiae genome. These were then evolved according to the Jukes-Cantor substitution model, to a specified evolutionary distance. Likelihood ratio scores and p -values were then calculated for each of the pairs of sequences using the method implemented in MONKEY. Finally, all pairs of sequences were ranked by their scores, and the rank divided by the total number of pairs was taken as the empirical p -value. Preparation of alignments for different groups of species We aligned the upstream regions of all S. cerevisiae genes to their orthologs in S. paradoxus , S. mikatae , S. bayanus and S. kudriavzevii by taking the 1,000 bp upstream of each gene, identifying the corresponding region from the other species using data in the Saccharomyces Genome Database [ 40 ], aligning them with CLUSTAL W [ 41 ] and trimming them to remove regions corresponding to S. cerevisiae coding sequence. We used this strategy rather than simply aligning intergenic regions to control for differences in alignments that might arise from the use of variably sized regions. To obtain estimates of the evolutionary distance spanned by each comparison, we ran PAML [ 24 ] on the entire set of intergenic alignments, using the HKY model [ 27 ], with constant rates across sites. We used the median PAML estimate of kappa (the transition-transversion rate ratio) of 3.8, the S. cerevisiae background frequencies (ACGT) = (0.3, 0.2, 0.2, 0.3) and the median of the branch lengths estimates as the 'background' evolutionary model. The trees with these branch lengths were used as input to MONKEY to calculate p -values. The distances in Figure 4 represent the sum of the median branch lengths in each comparison. The subsets (with evolutionary distances in parentheses) were as follows: S. cerevisiae and S. paradoxus (0.194); S. cerevisiae and S. mikatae (0.403); S. cerevisiae , S. paradoxus S. mikatae (0.477); S. cerevisiae and S. bayanus (0.559); S. cerevisiae , S. paradoxus, S. mikatae and S. bayanus (0.816); S. cerevisiae , S. paradoxus, S. mikatae, S. bayanus and S. kudriavzevii (1.090). Definition of Rpn4p and Gal4p matrices and positive and negative sets Rpn4p: we used Rpn4p sites in proteasomal genes [ 42 , 43 ] to build an Rpn4p specificity matrix (using a pseudocount of 1 per base per position). To identify additional likely targets, we obtained expression data from public sources [ 30 , 31 ] and compared the expression patterns of all genes to the average expression pattern of proteasomal genes using the following metric: where θ is the 'uncentered correlation', a commonly used distance metric for gene-expression data [ 44 ]. Our score adds a correction for the number of datapoints, n , that are available for each gene. All matches to the Rpn4p matrix ( S. cerevisiae likelihood ratio score > 9) in the upstream region of a gene that matched the proteasomal expression pattern ( t > 8) were considered to be true Rnp4p sites. The negative set consists of all sites that matched the Rpn4p matrix with a score greater than 9, and excluded sites in genes with even weak similarity to the proteasomal expression pattern ( t > 0) or that were annotated [ 40 ] as involved in protein processing or degradation. Gal4p: we used the matrix from SCPD [ 33 ] (with a pseudo count of 1 per base per position). To define a positive set we used the binding sites in SCPD and systematic studies of this Gal4p regulatory system [ 45 , 46 ], and used matches near additional genes that we identified in these studies with scores above the lowest score in the SCPD set. To define a negative set, we again scanned the S. cerevisiae genome with a cutoff equal to the lowest score in the positive set and then eliminated any binding sites near genes that showed evidence for regulation in the systematic studies. It is important to note that our categorization of sequences as positive and negative is done independently of the comparative sequence data, thus avoiding potential circularity. Calculations of expected scores Because our methods employ explicit probabilistic models for the evolution of noncoding DNA, it is possible to compute the expected scores under various assumptions. The expectation of the log likelihood ratio for examples of the motif is the 'information content' and its calculation has been addressed [ 47 ]. We can extend this to calculation to our evolutionary scores, as follows. Using the fact that all the scores treat the positions of the matrix independently, and the linearity of the expectation, we write: where E [ x ] denotes the expectation of the random variable x , m denotes a frequency matrix and a corresponding evolutionary model, either { motif , R motif } or { bg , R bg }. p ( X i , Y i | m , T ) is calculated as above, and we define: We can write a similar expression for the variance, V : In order to predict the scores for the genes in our positive and negative sets, we are interested in the case were we have observed a match to the motif in one species, but the constraints on its evolution are either those of the background or the motif. We can compute the expected scores under these assumptions as follows: where p ( X i | motif ) is the single species probability of observing the base X i at position i in the specificity matrix ( f ), and using Bayes' theorem: This calculation can be extended to the multiple species case, by replacing the distributions p ( X i , Y i ) and p ( Y i | X i ) with p ( X i , Y i , ..., Z i ) and p ( Y i , ..., Z i | X i ) and changing the sum over Y i to a sum over all the other leaves in the tree except the reference, in this case, X i . For the functional set, we assumed the binding sites were evolving under the HB model [ 25 ], and for the nonfunctional set we assumed evolution under the HKY background model described above. To model the sequence-specificity matrices most accurately, we reduced the pseudocount (equal to the background probability of observing each base). Additional data files Additional data file 1 shows the fraction of binding sites that are not conserved for several different S. cerevisiae transcription factors. Additional data file 2 shows the conservation p-values of predicted binding sites in high-density binding site clusters in the Drosophila melanogaster genome, with the binding sites grouped according to whether the cluster has regulatory activity. Supplementary Material Additional data file 1 The fraction of binding sites that are not conserved for several different S. cerevisiae transcription factors Click here for additional data file Additional data file 2 The conservation p-values of predicted binding sites in high-density binding site clusters in the Drosophila melanogaster genome, with the binding sites grouped according to whether the cluster has regulatory activity Click here for additional data file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC545801.xml |
387270 | The Cytoskeleton In Vivo | A comprehensive understanding of the cytoskeleton can only be achieved by the combination of biochemical, cellular, and whole organism studies | As a student I always marvelled at the sight of single cells in culture moving over artificial surfaces and exhibiting membrane ruffles and protrusions. However, while I found cultured cells fascinating I always wondered how cells are able to move and regulate their shape in the context of a whole organism where so many space constraints exist and where all cellular processes have to be tightly regulated. Some answers to my questions began to emerge in a paper written by Baum and Perrimon (2001) , in which the authors showed the expression and regulation of the actin cytoskeleton and of actin binding proteins in a real epithelium. The cytoskeleton is a meshwork of protein polymers extending throughout the cytoplasm. It not only provides structural support for the cell but also plays a central role in a range of dynamic processes from signalling to endocytosis and intracellular trafficking. A particularly clear example of this is the use of actin cytoskeleton as a “wool” for knitting multiple dynamic structures such as lamellae, filopodia, and stress fibres. These structures determine cell shape and also produce the driving force accompanying many types of cellular movements including muscle contraction and cell division. We know many details about some of the proteins that modulate the dynamics of actin in these structures. However, most of them have been found biochemically and their function has been elucidated primarily using in vitro and cell culture assays of actin assembly. What about these proteins in the context of a developing organism? How do cells generate a spatially and temporally ordered network of actin filaments represented at the tissue level? To answer these questions, we need to move to experimentally accessible multicellular organisms, such as Drosophila , which offers virtually unlimited possibilities as a model system for the genetic and molecular analysis of biological processes. Baum and Perrimon (2001) analyzed the function of a number of proteins involved in actin dynamics within the context of a developing epithelium—the follicle cells that surround the germ line cyst during Drosophila oogenesis. These cells have a simple polarised arrangement of actin filaments, which provides a useful system to study the spatial organisation of the actin cytoskeleton. Taking advantage of the ability to generate clones of cells lacking specific proteins, the authors identified new functional roles for actin regulators such as CAP (a Drosophila homologue of adenylyl cyclase-associated proteins), Enabled (Ena) and Abelson (Abl). These proteins had been well characterized in cell culture and in vitro studies, but little was known about their function in a developing organism. Clones of cells lacking CAP ( Figure 1 ), a protein known to inhibit actin polymerisation, maintained their epithelial polarity but had higher levels of actin and defects in the apical actin organisation. This result indicates that the inhibitory activity of CAP is restricted to one side of the cells, thus demonstrating that actin dynamics can be independently modified at opposite poles of an epithelium. Ena, a member of the Ena/VASP family proteins that catalyse filament formation, and Abl, a protein kinase that binds CAP in mammalian cells, were found to work with CAP in this process. The authors proposed that CAP, Ena, and Abl regulate the level and spatial organization of actin in the follicle cells. Figure 1 CAP Mutant Clones Follicle cells lacking CAP accumulate actin (red) in their apical region. Ena (blue in the bottom panel), also accumulates apically in the mutant cells (looks pink in the clone of cells due to overlap with F-actin in red). The mutant cell clones are identified by the absence of GFP (green in the top panel). Using this technique the cytoskeleton of mutant cells can be analysed in the context of a wild type epithelium. (Image kindly supplied by Buzz Baum.) In contrast to the spatially restricted functions of CAP, Ena, and Abl, profilin and cofilin were shown to regulate actin filament formation throughout the cell cortex, a more global function that matches the results obtained in cell culture experiments. In summary, this study showed how proteins can organise actin in space and began to highlight some of the differences and similarities between cells in culture and in vivo. The functions revealed in the follicular epithelium were consistent with the roles previously shown in mammalian systems, but the experiments on intact tissue began to reveal a spatial and temporal functional dimension that could not have been observed in cell culture. These experiments could be expanded to large-scale screens ( St Johnston 2002 ), but this would be time consuming and could encounter the problem that some genes will be cell lethal, preventing the analysis of their function in actin dynamics. However, two more recent reports ( Kiger et al. 2003 ; Rogers et al. 2003 ) describe a complementary and exhaustive search for regulators of cytoskeletal dynamics by taking advantage of genomic resources and the powerful RNA interference (RNAi) technique ( Hutvágner and Zamore 2002 ). RNAi allows individual genes to be knocked out in a simple and controlled fashion. Kiger et al. (2003) used RNAi in two different cell lines of Drosophila to screen a number of genes involved in signalling and cytoskeletal dynamics. They targeted 994 genes, of which 160 produced phenotypes in the experiment. The range of phenotypes varied from specific defects in the actin and tubulin cytoskeleton to others affecting cell cycle progression, cytokinesis, and cell shape. They also showed that only about 40% of the genes had similar loss-of-function phenotypes in both cell lines. This alone indicates an important limitation of many tissue culture experiments, since the same protein can have different effects depending on the cell type. Another valuable element of this work is that clustering of genes with similar phenotypes leads to the identification of pathways and networks of genes that are involved in cytoskeletal function. Rogers et al. (2003) , using only one Drosophila cell line, studied the effects of proteins involved in the formation of lamellae. The authors looked at the effects of loss of function in 90 genes known to be involved in actin dynamics and the formation and activity of the lamella. As well as confirming the function of many proteins already known to play a role in this process, this analysis allowed them to find interactions between genes and to build genetic pathways. Together these two studies reveal that RNAi screens in tissue culture can be a powerful tool for finding new functions of known and uncharacterized genes, and new relationships between genes. However, this is only the beginning, and the genes identified in this manner will have to be tested in vivo, in systems like that of Baum and Perrimon, where specific functions can be assessed in time and space within the confines of real organisms. The focus must be to understand how all these molecular events and regulation cascades operate in individual cells to contribute to the generation of changes in a whole individual. Increasingly, the attention of developmental biologists is being drawn from genes and their products towards cells ( Kaltschmidt and Martinez Arias 2002 ). The future, it seems to me, lies in the combination of in vitro systems, cell culture, and in vivo studies. I hope to apply this view in my analysis of the process of dorsal closure in Drosophila embryos, as an example of how signalling pathways coordinate and regulate the activity of the cytoskeleton in the generation of shape and morphogenetic movements ( Jacinto et al. 2002 ). | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC387270.xml |
539048 | Pancreatic Islet Transplantation | Islet transplantation offers hope to many patients with diabetes, who envision a life free of glucose checks and insulin injections. What are the barriers to its widespread implementation? | Diabetes: Epidemiology and Complications Treatment for, and the prognosis of, type-1 diabetes mellitus (T1DM) has progressed dramatically during the last century, but the disease remains a major cause of morbidity and mortality. Although precise figures are not available, over 1 million United States citizens currently live with the disease, with approximately 30,000 new cases diagnosed in the US each year. The total number of people with diabetes worldwide is expected to rise to 366 million in 2030, up from 171 million in 2000 [1] . The exact etiology of the disease remains uncertain, but extensive research suggests an interaction between genetic predisposition and environment. In fact, for unknown reasons, the incidence of T1DM is increasing [2] . Diabetes continues to have a tremendous societal impact; it is both difficult and expensive to treat and is associated with a number of long-term complications, including kidney failure, blindness, nerve damage, and premature mortality (predominately due to cardiovascular problems). Insulin's Impact Banting and Best's discovery of insulin in the early 1920s revolutionized diabetes treatment and greatly improved the prognosis for what had previously been a rapidly fatal disease. As shown by the Diabetes Control and Complications Trial and the more recent Epidemiology of Diabetes Interventions and Complications trial, insulin therapy has made such considerable advances (with better insulin formulations and delivery systems) that many patients can maintain their blood sugar levels within a tight range and thereby reduce their risk for the disease's long-term complications [ 3 , 4 , 5 ]. In addition, improved treatment of other associated conditions such as hypertension and hyperlipidemia have helped reduce, or at least delay, many of the long-term sequelae of diabetes [6] . However, problems with insulin-based treatment regimens persist. For the patient, treatment is expensive and difficult, requiring strict attention to blood glucose monitoring, insulin dosing, diet, and exercise. Further, good glycemia control is not easily achieved by all patients, and even for those able to achieve this goal, the treatment is not always completely effective. Promising Directions Just as financial investors balance a portfolio, with some risky investments and others that are more secure, researchers will undoubtedly continue to further refine “secure” insulin-based regimens to help patients achieve even better glycemia control. At the same time, scientists are pursuing more high-risk, high-payoff approaches to revolutionize diabetes care. One such approach is the closed-loop insulin pump (i.e., a pump that continuously monitors blood glucose and concurrently converts that data into appropriate insulin dosing), which offers the potential to serve as a mechanical pancreas. However, such a mechanical system would need be fail-safe in order to avoid devastating effects (e.g., if the monitor were to register a falsely elevated blood glucose and thereby trigger an inappropriately high insulin dose). In other, similar scenarios with no tolerance for error, NASA (for instance) sets up systems in which two independent monitoring systems must come up with similar measurements before an action is taken. Perhaps the engineering obstacles that currently limit the closed-loop insulin pump can be overcome. Other research groups are investigating whether the insulin-producing cells within the pancreas (so-called ß cells), might be promoted to regenerate (in vitro or in vivo) to replace the pool of insulin-producing cells reduced by autoimmune destruction. Another promising approach for creating cells capable of physiologically regulated insulin secretion is to “coax” stem cells—undifferentiated cells with self-regenerative capacity—to differentiate into ß-like cells. Gene therapy approaches may overcome present obstacles and result in cells capable of physiologically regulated insulin secretion [7] . Lastly, the recent completion of the Human Genome Project suggests that the genetics of diabetes may eventually become clearer and may direct appropriate preventative approaches. While such potential therapies remain experimental, pancreas transplantation is currently performed in patients with complicated diabetes. However, a recent report that shows benefit for patients with both diabetes and kidney failure who receive a combined pancreas and kidney transplant also found that an isolated pancreas transplant (for patients with preserved kidney function) actually worsened survival [8] . The main point is that as we develop new therapies, we must maintain humility and recognize that newer approaches may have great promise, but they also have the potential for harm. History of Islet Transplantation Islet transplantation has recently received considerable interest as a potentially definitive treatment for diabetes. The concept of islet transplantation is not new—investigators as early as the English surgeon Charles Pybus (1882–1975) attempted to graft pancreatic tissue to cure diabetes. Most, however, credit the recent era of islet transplantation research to Paul Lacy's studies dating back more than three decades. In 1967, Lacy's group described a novel collagenase-based method (later modified by Dr. Camillo Ricordi, then working with Dr. Lacy) to isolate islets, paving the way for future in vitro and in vivo islet experiments [9] . Subsequent studies showed that transplanted islets could reverse diabetes in both rodents and non-human primates [ 10 , 11 ] ( Figure 1 ). In a summary of the 1977 Workshop on Pancreatic Islet Cell Transplantation in Diabetes, Lacy commented on the feasibility of “islet cell transplantation as a therapeutic approach [for] the possible prevention of the complications of diabetes in man” [12] . Improvements in isolation techniques and immunosuppressive regimens ushered in the first human islet transplantation clinical trails in the mid-1980s. Yet despite continued procedural improvements, only about 10% of islet recipients in the late 1990s achieved euglycemia (normal blood glucose). In 2000, Dr. James Shapiro and colleagues published a report describing seven consecutive patients who achieved euglycemia following islet transplantation using a steroid-free protocol and large numbers of donor islets, since referred to as the Edmonton protocol [13] . This protocol has been adapted by islet transplant centers around the world and has greatly increased islet transplant success. Figure 1 Central Concepts Underlying Islet Transplantation The main idea of islet transplantation is to process the organ donor's pancreas so as to remove the 95% of the gland responsible for its exocrine functions (secretion of digestive enzymes) and isolate the 5% of the gland responsible for the endocrine hormone secretion— the so-called pancreatic islets. Once isolated, the medical team can infuse the insulin-producing islets through a thin tube, placed in the main vein that transports blood from the intestines to the liver. Once infused, the islets are transported by the bloodstream into the liver, where they lodge, take up residence, and begin making the right amount of insulin to regulate the blood sugar. (Illustration: Giovanni Maki) Current Limitations of Islet Transplantation While significant progress has been made in the islet transplantation field [14] , many obstacles remain that currently preclude its widespread application. Two of the most important limitations are the currently inadequate means for preventing islet rejection, and the limited supply of islets for transplantation. Current immunosuppressive regimens are capable of preventing islet failure for months to years, but the agents used in these treatments are expensive and may increase the risk for specific malignancies and opportunistic infections. In addition, and somewhat ironically, the most commonly used agents (like steroids, calcineurin inhibitors, and rapamycin) are also known to impair normal islet function and/or insulin action. Further, like all medications, the agents have other associated toxicities, with side effects such as oral ulcers, peripheral edema, anemia, weight loss, hypertension, hyperlipidemia, diarrhea, and fatigue [15] . Perhaps of greatest concern to the patient and physician is the harmful effect of certain widely employed immunosuppressive agents on renal function. For the patient with diabetes, renal function is a crucial factor in determining long-term outcome, and calcineurin inhibitors (tacrolimus and cyclosporin) are significantly nephrotoxic. Thus, while some patients with a pancreas transplant tolerate the immunosuppressive agents well, and for such patients diabetic nephropathy can gradually improve, in other patients the net effect (decreased risk due to the improved blood glucose control, increased risk from the immunosuppressive agents) may worsen kidney function. Indeed, Ojo et al. have published an analysis indicating that among patients receiving other-than-kidney allografts, 7%–21% end up with renal failure as a result of the transplant and/or subsequent immunosuppression [16] . Looked at another way, patients with heart, liver, lung, or kidney failure have a dismal prognosis for survival, so the toxicity associated with immunosuppression is warranted (the benefits of graft survival outweigh the risks associated with the medications). But for the subset of patients with diabetes and preserved kidney function, even those with long-standing and difficult-to-control disease, the prognosis for survival is comparatively much better. In addition to the immunosuppressive toxicities, other risks are associated with the islet transplant procedure itself, including intra-abdominal hemorrhage following the transplant, and portal vein thromboses. The fact that there is already a good alternative to islet transplantation (i.e., the modern intensive insulin regimen) forces us to regard any newer, riskier interventions with a critical eye. Like all transplantation therapies, islet transplantation is also handicapped by the limited donor pool. The numbers are striking; at least 1 million Americans have T1DM, and only a few thousand donor pancreata are available each year. To circumvent this organ shortage problem, researchers continue to look for ways to grow islets—or at least cells capable of physiologically regulated insulin secretion—in vitro, but currently only islets from cadaveric donors can be used to restore euglycemia. Further exacerbating the problem (and unlike kidney, liver, and heart transplants, where only one donor is needed for each recipient) most islet transplant patients require islets from two or more donors to achieve euglycemia. Lastly, the current methods for islet isolation need improvement, since only about half of attempted isolations produce transplant-ready islets. While islet transplantation research has made important progress and the success stories are encouraging, the long-term safety and efficacy of the procedure remain unclear. Other concerns relating to the field include questions about the impact of having insulin-producing foreign cells within the hepatic parenchyma, the long-term consequences of elevated portal pressures resulting from the islet infusion, and the fact that islet recipients can be sensitized against donor tissue types, making it more difficult to find a suitable donor should another life-saving transplant be required in the future. Also, very few islet transplant recipients have remained euglycemic without the use of any exogenous insulin beyond four years post-transplant. Thus, while most islet recipients achieve better glycemia control and suffer less serious hypoglycemia, islet transplantation continues to fall short of the definitive diabetes cure. Is Islet Transplantation Ready for Widespread Use? While no one suggests that the therapy is ready for widespread clinical application, another way of highlighting current problems is to focus on cost. Assuming present hurdles were cleared, islet transplantation costs approximately $150,000 per patient per transplant. With over 1 million Americans dealing with T1DM, it would cost over $100 billion to give each patient a single islet transplant, with little assurance as yet of any long-term benefit. In contrast, the annual direct cost of a proven therapy like intensive insulin treatment is about $3,500 per patient [17] . The limitations of islet transplantation force us to recognize that the therapy remains experimental, and that many questions must be answered before it is incorporated into general clinical practice. At the present time, we urge a focus on the selection of only those patients for whom this procedure offers the greatest likelihood of benefit. Most people with diabetes can, with diligence and perseverance, implement an insulin regimen that maintains tight glucose control while avoiding dangerous hypoglycemia. However, there are some patients who continue to have tremendous difficulty managing their disease despite optimal care and effort. Even the statement “despite optimal care and effort” is difficult to define, and we advocate that all patients being considered for an islet transplant first be referred for several months to specialty teams that are committed to diabetes care. Since such patients whose diabetes is the most difficult to control have a poor quality of life, islet transplantation offers potential benefit. Even a low baseline level of insulin production by the transplanted islets may lower the amount of insulin required, while reducing the number and severity of hypoglycemic events. We also believe the islet transplant risk-benefit ratio is favorable for those with both T1DM and kidney failure who are listed for a life-preserving kidney transplant; such patients will have to take immunosuppressive agents after transplant to preserve the kidney allograft function, so the islets can be added without too much additional risk. Where do we go from here? Just as early studies showed islet transplantation's promise, research must now overcome the hurdles revealed by the recent islet transplant experience. New immunomodulatory agents offer the greatest hope of revolutionizing the field. New drug regimens capable of inducing tolerance to the transplanted islets would allow recipients to maintain their grafts without general immunosuppression and its associated toxicities. While many targets are currently under investigation, none are ready for clinical use. We advocate that such immunomodulatory approaches be tested first in controlled models where the results can be appropriately attributed to the agent itself. Conclusion Less than a century ago, T1DM was invariably a fatal disease. With the advent of insulin, the prognosis changed overnight, and we have continued to witness improvements in diabetes care and outcomes. Pancreatic islet transplant has offered renewed hope to many patients with diabetes, who envision a life free of glucose checks and insulin injections. Some transplanted patients have enjoyed “success” and are pleased with their decisions; unfortunately these results are not universal. Researchers must continue to look for ways to improve the procedure while protecting the welfare of each individual patient. The field has come a long way, but we must remain cautious, as we are treating a non-fatal disease for which there is a very effective standard therapy. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC539048.xml |
548666 | No influence of the P-glycoprotein polymorphisms MDR1 G2677T/A and C3435T on the virological and immunological response in treatment naïve HIV-positive patients | Background In a retrospective study of HIV-infected patients, we investigated the influence of the MDR1 genotype (G2677T/A and C3435T) on the virological and immunological response of treatment naïve patients. Methods The MDR1 genotype was analysed from 72 patients in whom antiretroviral therapy was initiated between 1998 and 2004. Data were obtained at week 4, 12, 24 and 48 and were analysed by the Kruskal-Wallis test. Results During the first 48 weeks of antiretroviral therapy, there were no significant differences in the virological and immunological response with respect to the MDR1 2677 and 3435 genotypes and the 2677/3435 haplotype. Conclusions In view of different results from several studies concerning the influence of MDR1 polymorphisms on the immunological and virological response to antiretroviral therapy, further studies with larger patient groups and longer follow-up are necessary in order to resolve conflicting issues. | Background The P-glycoprotein (P-gp) is an ATP-dependent efflux transporter (ABCB1) encoded by the multidrug resistance gene (MDR1) which extrudes large lipophilic, positvely charged molecules from cells, among them HIV-1 protease inhibitors [ 1 , 2 ]. P-gp is expressed on a variety of cells including human lymphocytes, the target cells of HIV and of antiretroviral substances [ 3 ]. 48 single nucleotide polymorphisms (SNP) have been described so far for the MDR1 gene [ 4 ]. Though there is still controversy about the biological relevance of these SNPs, there appears to be an association between specific genotypes and mRNA expression, P-gp expression and/or P-gp function (reviewed in [ 5 , 6 ]). Recently, Fellay et al . found lower nelfinavir and efavirenz plasma levels associated with the TT genotype of the SNP C3435T in exon 26 and a greater rise of the CD4 cell count 6 months after initiation of antiretroviral therapy in patients with this genotype [ 7 ]. However, the mechanisms underlying this observation remain unclear. In contrast to protease inhibitors (PI), non-nucleoside reverse transcriptase inhibitors (NNRTI) such as efavirenz are not substrates of the P-gp. Therefore, it has been speculated that the P-gp may modulate the clinical course of HIV infection independent from its role in drug transport. Indeed, there have been reports showing inhibition of apoptosis and decreased HIV production in cells overexpressing P-gp [ 8 - 12 ]. However, these observations from in vitro studies have not been confirmed in vivo when the disease progression before treatment was assessed in HIV infected individuals with different MDR1 genotypes [ 13 ]. The genetic variant C3435T in exon 26 is a synonymous polymorphism that does not alter the amino acid sequence. How this variant could affect P-gp expression is still unknown [ 6 ]. According to one hypothesis, functional effects of the C3435T SNP may not be genotype- but haplotype-dependent. The exon 26 C3435T polymorphism is in linkage disequlibrium with the polymorphism G2677T in exon 21, which results in the amino acid change Ala893Ser [ 6 ]. 2677A leading to Ala893Thr is an infrequent third allel of this SNP. The results of several studies on the functional effects of mutations at position 2677 in exon 21 have shown conflicting results [ 6 ]. Because of the unresolved issues surrounding the potential effects of MDR1 polymorphisms and P-gp function in HIV infection, we investigated whether there was an association between the MDR1 polymorphisms 3435 and 2677 and the immunological response in HIV infected individuals after initiation of antiretroviral therapy. Methods Patients Of the HIV 1 infected patients seen at the Department of Infectious Diseases of the University of Würzburg, Germany, 72 patients (18 women, 54 men; mean age 39.5 years, range: 26 – 59 years; 64 Caucasians, 5 African, 3 Asian), started antiretroviral therapy between 1998 and 2004. The therapy consisted of three nucleoside reverse transcriptase inhibitors (NRTI) (n = 12), two NRTIs plus at least one PI (n = 40; including RTV n = 26) or two NRTIs plus one NNRTI (n = 20). HIV load and CD4 cell count were determined four weeks after initiation of therapy and approximately every three months thereafter. In patients who received NNRTI or PI drugs as part of their therapy, plasma levels were monitored and adjusted to therapeutic drug levels when necessary as previously described [ 14 , 15 ]. The treatment was based on current international treatment guidelines [ 16 ], taking into account individual circumstances of each patient (e.g. known intolerabilities, side-effects of previous therapies, concomitant medication). The study was in accordance with the Helsinki Declaration and was approved by the local ethics committee. Patients gave informed consent for the study. Genotype analysis DNA was extracted from 200 μl blood and the MDR1 3435 genotype was determined with genotype specific hybridisation probes and melting curve analysis on the LightCycler (Roche, Mannheim, Germany) as previously described [ 17 ]. The 2677 genotype was determined in a similar fashion [ 18 ]. Briefly, DNA was amplified on the LightCycler with the Quantitect Probe PCR Kit (Qiagen, Hilden, Germany) by using the primers 5'-gcaggagttgttgaaatgaaaatg-3' (forward) and 5'-cgcctgctttagtttgactca-3' (reverse). Hybridization probes were added to the master mix to a final concentration of 0.05 μM (sensor probe: 5'-ttcccagTaccttct-fluorescein; locked nucleic acid base in upper case letter) and 0.15 μM (anchor probe: 5'-LC Red640-ctttcttatctttcagtgcttgtcc-p). Primers and probes were obtained from TIB Molbiol (Berlin, Germany). The melting points were 41°C, 47°C, and 52°C for the T-, G-, and A-alleles, respectively. The results of 30 samples were confirmed by sequencing. Statistical analysis Data were analysed by the Kruskal-Wallis test, which as a nonparametric test compairs three or more unpaired groups. Because of the small number of samples a nonparametric test was needed. SPSS version 12.0 (SPSS GmbH, Munich, Germany) was used for statistical analysis. A P -value of 0.05 was considered to be significant. Analogous to [ 13 ], the SNPs 3435 and 2677 were analyzed both separately and in combination as composite genotypes: H1: 2677GG and 3435CC (wild type); H2: 2677GT or TT and 3435CT or TT (2677T/3435T haplotype carrier); H3: 2677GG and 3435CT or TT; H4: 2677GT or TT and 3435CC; and H5: 2677AG or AT and any 3435 genotype (2677A carrier). Results At initiation of antiretroviral therapy 8 patients had viral loads of <10.000 copies/ml, 24 patients between >10.000 and <100.000 copies/ml, and 40 patients >100.000 copies/ml. 39 patients had a CD4 cell count of <200 cells/μl, 16 patients between >200 and <350 cells/μl, and 17 patients >350 cells/μl. Determination of NNRTI- and PI drug levels indicated compliance of the patients with a NNRTI or PI containing regimen, because the measured drug levels were in the therapeutic range. Genotype analysis of the MDR1 gene at position 3435 in exon 26 revealed 20 patients with the CC genotype, 33 with the CT genotype and 19 with the TT genotype (5 patients of African origin: 4 with CC and 1 with CT genotype; 3 patients o f Asisan origen: 2 with CT and 1 with TT genotye). Analysis of the 2677 polymorphism in exon 21 demonstrated that 24 patients had the GG-, 23 the GT-, 18 the TT-, 4 the AG-, and 3 the AT genotype; the AA genotype was not found in this group (5 patients of African origin: 4 with GG and 1 with GT genotype; 3 patients of Asian origin: 3 with TT genotype). Detailed genotype and haplotype results with respect to the initial viral load and CD4 cell count are presented in table 1 . Table 1 Genotype analysis of 72 treatment naive HIV-positive patients with respect to baseline CD4 cell count and viral load. CD4 [cells/μl] VL [copies/μl] n = 2677 3435 2677/3435 $ GG GT TT AG AT AA CC CT TT H1 H2 H3 H4 H5 <200 >10 5 27 7 7 8 3 2 - 7 10 10 5 15 2 - 5 10 4 – 10 5 8 4 3 1 - - - 2 6 - 2 4 2 - - <10 4 4 2 1 1 - - - 1 2 1 1 2 1 - - 200 >10 5 6 2 3 1 - - - 1 4 1 1 4 1 - - -350 10 4 – 10 5 9 3 3 3 - - - 3 3 3 3 6 - - - <10 4 1 1 - - - - - 1 - - 1 - - - - >350 >10 5 7 1 4 2 - - - 1 3 3 1 6 - - - 10 4 – 10 5 7 1 2 2 1 1 - 2 4 1 1 4 - - 2 <10 4 3 3 - - - - - 2 1 - 2 - 1 - - total 72 24 23 18 4 3 0 20 33 19 17 41 7 - 7 $ composite genotypes: H1: 2677GG and 3435CC (wild type); H2: 2677GT or TT and 3435CT or TT (2677T/3435T haplotype carrier); H3: 2677GG and 3435CT or TT; H4: 2677GT or TT and 3435CC; and H5: 2677AG or AT and any 3435 genotype (2677A carrier) Table 2 and 3 show the median and mean values of the viral load decline and the CD4-cell increase, respectively, determined at week 4, 12, 24 and 48 after initiation of therapy. There were no significant differences of the viral load decline neither between patient groups with different genotypes at positions 2677 or 3435 nor between patient groups with different 2677/3435 haplotypes. As to the CD4-cell response, there were no significant differences between the different genotypes and haplotypes, either. There was a trend of a more pronounced mean CD4-cell increase at week 12 and 24 in patients with the 3435TT genotype. However, this trend did not persist at week 48 and was not confirmed by the corresponding median values. A graphical presentation of the mean viral load decrease and CD4-cell increase with respect to the 2677/3435 haplotypes is given in fig. 1 . Analysis of patients with different therapy-regimens (only NRTI, NRTI + NNRTI, or NRTI + PI) revealed no differences in the virological and immunological response between the different genotypes either, but was limited by small patient numbers in the subgroups (data not shown). Table 2 VL-decrease [log copies/ml] at week 4, 12, 24 and 48 after initiation of antiretroviral therapy with respect to the MDR1 2677 and 3435 genotypes. Statistical analysis was done with the Kruskall-Wallis test. week 4 week 12 week 24 week 48 MDR1 n = median [log/ml] mean [log/ml] SA p = n = median [log/ml] mean [log/ml] SA p = n = median [log/ml] mean [log/ml] SA p = n = median [log/ml] mean [log/ml] SA p = 2677 GG 22 -2,301 -2,210 0,590 >0.3 22 -3,048 -3,287 0,682 >0.05 20 -3,943 -3,876 0,748 >0.9 15 -3,716 -3,835 0,777 >0.5 GT 21 -2,602 -2,537 0,566 19 -3,786 -3,794 0,635 17 -3,929 -3,933 0,499 16 -3,942 -3,946 0,488 TT 17 -2,495 -2,372 0,651 16 -3,739 -3,602 0,595 17 -4,000 -3,850 0,708 14 -3,977 -3,795 0,615 AG 3 -2,301 -2,689 0,606 § 4 -3,952 -4,167 0,884 § 4 -4,500 -4,537 0,702 § 4 -4,151 -4,362 0,726 § AT 3 -1,875 -2,032 0,493 § 3 -4,176 -3,752 0,690 § 3 -4,176 -4,085 0,223 § 3 -4,176 -4,085 0,223 § 3435 CC 17 -2,301 -2,382 0,501 >0.6 19 -3,301 -3,416 0,699 >0.4 17 -4,000 -3,974 0,685 >0.7 13 -3,716 -3,818 0,709 >0.6 CT 32 -2,247 -2,363 0,640 29 -3,778 -3,742 0,737 28 -3,812 -3,830 0,704 25 -3,929 -3,984 0,614 TT 17 -2,398 -2,348 0,668 17 -3,602 -3,530 0,595 17 -4,176 -4,054 0,571 15 -4,114 -3,849 0,609 2677/3435 † H1 15 -2,398 -2,398 0,531 >0.8 16 -3,199 -3,361 0,724 >0.08 14 -3,943 -3,926 0,715 >0.8 10 -3,659 -3,774 0,787 >0.4 H2 38 -2,548 -2,463 0,611 35 -3,778 -3,706 0,624 34 -3,954 -3,891 0,614 30 -3,954 -3,875 0,556 H3 7 -1,778 -1,806 0,502 § 6 -2,827 -3,089 0,506 § 6 -3,827 -3,761 0,807 § 5 -4,176 -3,958 0,741 § H5 6 -2,261 -2,361 0,642 § 7 -4,176 -3,989 0,833 § 7 -4,301 -4,343 0,594 § 7 -4,176 -4,243 0,584 § total 66 64 61 52 † composite genotypes: see methods; § excluded from statistical analysis Table 3 CD4-cell increase [cells/μl] at week 4, 12, 24 and 48 after initiation of antiretroviral therapy with respect to the MDR1 2677 and 3435 genotypes. Statistical analysis was done with the Kruskall-Wallis test. week 4 week 12 week 24 week 48 MDR1 n = median [cells/μl] mean [cells/μl] SA p = n = median [cells/μl] mean [cells/μl] SA p = n = median [cells/μl] mean [cells/μl] SA p = n = median [cells/μl] mean [cells/μl] SA p = 2677 GG 24 56,5 69,8 81,5 >0.3 22 97,0 130,6 132,4 >0.5 21 107,0 136,1 120,5 >0.8 15 244,0 235,7 132,1 >0.1 GT 23 24,0 37,8 82,5 20 84,5 144,3 157,2 17 132,0 180,8 178,0 16 161,5 238,3 218,6 TT 18 9,5 67,4 117,7 16 70,0 143,9 222,3 16 95,0 154,5 162,0 14 107,5 143,9 159,1 AG 4 70,5 85,8 68,5 § 4 82,5 74,8 34,4 § 4 54,5 80,3 81,3 § 4 131,5 126,5 71,0 § AT 3 112,0 105,7 13,4 § 3 176,0 180,7 63,8 § 2 154,5 154,5 54,5 § 3 172,0 248,0 118,2 § 3435 CC 20 76,0 84,8 78,6 >0.1 19 91,0 123,0 113,0 >0.9 18 106,5 132,4 112,4 >0.5 13 191,0 220,1 121,5 >0.6 CT 34 25,5 47,2 76,1 30 85,0 123,7 135,3 28 131,0 134,2 106,8 25 159,0 187,7 118,4 TT 18 12,0 61,6 119,1 17 74,0 175,9 228,1 16 134,0 212,2 214,7 15 122,0 223,9 259,3 2677/3435 † H1 17 72,0 83,1 78,7 >0.1 16 115,0 132,1 119,8 >0.8 15 107,0 142,2 113,3 >0.9 10 194,0 239,6 128,4 >0.5 H2 41 20,0 50,8 100,6 36 74,5 144,1 188,9 33 130,0 168,1 170,9 30 134,5 194,2 198,8 H3 7 19,0 37,3 79,1 § 6 92,0 126,7 161,2 § 6 122,0 120,8 135,7 § 5 302,0 227,8 138,9 § H5 7 87,0 94,3 53,5 § 7 105,0 120,1 71,9 § 7 100,0 131,0 98,7 § 7 172,0 178,6 111,7 § total 72 65 61 52 † composite genotypes: see methods; § excluded from statistical analysis Figure 1 Composite genotypes (see methods) and response to antiretroviral treatment. Suppression of viraemia (lower panel) and CD4-cell count increase (upper panel) are shown. Data are mean values ± standard error. Discussion In an analysis of the virological and immunological response of treatment naïve patients with respect to the MDR1 G2677T/A and C3435T polymorphisms, virus load and CD4 cell count were assessed longitunally after initiation of antiretroviral therapy. We did not find an association between the CD4-cell increase or the HIV load decline and the MDR1 2677 or 3435 genotype during the first year of therapy. Our data are in contrast to a report of Fellay et al . that showed a significantly greater mean CD4-cell rise in patients with the MDR1 3435TT genotype during an observation period of 24 weeks [ 7 ]. The number of patients in the previous report (n = 96) was slightly larger than our population (n = 72), but the number of homozygotes of the 3435 polymorphism was very similar (22CC/20TT in [ 7 ], 20CC/19TT this study). Though we saw a trend towards a greater mean CD4-cell rise in patients with the 3435TT genotype as well, we doubt that this is indication of a real difference, because this trend was not seen when median values were considered. In our study, neither median values nor mean values were significantly different by Kruskal-Wallis test and ANOVA, respectively. For genotype subgroup sizes as in this study, analysis of the mean values is probably less robust. If there was a real difference between the immunological response of patients with the 3435CC and TT genotype, it did not persist at week 48. At this time point, mean and median values in the two groups were almost identical. A possible explanation for the diverse results is the selection of the study patients. While we included all treatment naïve patients in whom therapy was started between 1998 and 2004, Fellay et al . selected part of the patients on the basis of long-term viral suppression. In both studies, a variety of antiretroviral regimens was used. While all of the patients in the study of Fellay et al . received either nelfinavir or efavirenz, the treatment regimens in our study were more heterogenous and did not allow a meaningful separate analysis. The results of our study are supported by Nasi et al . [ 19 ] who analysed data of 149 treatment naïve patients who were treated with a PI-containing regimen (n = 106) or a NNRTI-containing regimen (n = 46) and found no association between the MDR1 genotype at position 3435 and the CD4 cell increases or plasma viral load decreases during the first six months of treatment among individuals with different genotypes. Likewise, in 142 patients enrolled in an open-label, randomized phase IIIb study comparing nelfinavir and efavirenz for treatment of HAART-naïve individuals the CD4 cell counts did not increase to a higher level in individuals with the homozygous variant genotype (TT) at the MDR1 C3435T locus in either the nelfinavir or the efavirenz treatment groups [ 20 ]. The MDR1 3435 polymorphism is a synonymous polymorphism. The TT genotype may have a reduced translation efficiency or lead to post-translational differences [ 21 - 23 ]. It has been shown, that a linkage disequilibirum exists between the exon 26 C3435T and the exon 21 G2677T/A polymorphism [ 6 ]. Therefore, we investigated both SNPs in our study. Neither the 2677 polymorphism alone nor the 2677/3435 haplotype was associated with differences in the virological or immunological response of treatment naïve patients. These data are in agreement with those of Haas et al . [ 24 ], who showed that the phase 1 and phase 2 viral decay as well as changes in CD4- and CD8-T-cells during triple therapy with ritonavir, zidovudine, and lamivudine in 31 treatment naïve patients were not associated with the MDR1 2677 and 3435 genotypes. In a retrospective study of 455 treatment naïve patients initiating antiretroviral therapy with 40 months of follow-up [ 25 ], there was a trend to earlier virological failure in the 3435CC group ( p = 0.07) with no effect of the C3435T polymorphism in the MDR-1 gene on immunological failure. However, the difference in the virological response was not observed during the first 10 months. Further follow-up of our patient group is ongoing in order to detect long-term effects that may not have been apparent during the observation period analysed in this report. Conclusions During a follow-up of 48 weeks, we found no evidence for an association between the MDR1 G2677T/A and C3435T polymorphisms and the virological and immunological response to therapy in HIV-positive drug-naïve patients. The individual response to antiretroviral therapy is a complex phenomenon, which is influenced by a large number of biological variables. Further studies on the role of polymorphisms of the MDR1 and other transporters and enzymes involved in drug metabolism are necessary in order to elucidate the role of pharmacogenetic effects in HIV therapy. Competing interests None declared. Authors' contributions RW participated in the design of the study and the molecular genetic studies, performed the statistical analysis and drafted the manuscript. PL participated in the design of the study, obtained and reviewed clinical data and helped in the data analysis. MZ participated in the clinical management of the study patients. FT and JS participated in the molecular genetic studies and the virological and immunological analysis. HK participated in the study design and coordinated clinical management of the study patients. BW participated in the study design and coordination and made contributions to the manuscript. All authors read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548666.xml |
549550 | Structural similarity of loops in protein families: toward the understanding of protein evolution | Background Protein evolution and protein classification are usually inferred by comparing protein cores in their conserved aligned parts. Structurally aligned protein regions are separated by less conserved loop regions, where sequence and structure locally deviate from each other and do not superimpose well. Results Our results indicate that even longer protein loops can not be viewed as "random coils" and for the majority of protein families in our test set there exists a linear correlation between the measures of sequence similarity and loop structural similarity. Results suggest that distance matrices derived from the loop (dis)similarity measure may produce in some cases more reliable cluster trees compared to the distance matrices based on the conventional measures of sequence and structural (dis)similarity. Conclusions We show that by considering "dissimilar" loop regions rather than only conserved core regions it is possible to improve our understanding of protein evolution. | Background Globular proteins are considered to be structurally similar if their regular secondary structure elements can be superimposed well and are connected in the same order. The loop regions connecting secondary structures demonstrate less regularity in their conformations even though short loops linking specific secondary structures can be classified into distinct classes [ 1 - 6 ]. The structures and sequences in loop regions may deviate from each other so that they do not superimpose well and as a result loops are very often not aligned by structure-structure or sequence alignment methods. Loops apparently do not contribute much to protein stability but may be quite important for protein specific function and for the interaction with other components of the cell. In our previous work we showed that a measure derived from the loop regions can distinguish homologous from analogous proteins with the same or higher accuracy compared to the conventional measures which are based on comparing proteins in structurally aligned regions only [ 7 ]. Recently it has been observed that structural variation in the core of homologous proteins is linearly correlated with sequence changes [ 8 , 9 ]. As was also shown several years ago, the probability of insertion and deletion events, which occur predominantly in the loop regions, strongly depends on the evolutionary distance between two homologous proteins [ 10 , 11 ]. Based on these observations one might argue that more closely related proteins may exhibit more similarity in the structure of their loop regions compared to distantly related proteins and the structural loop (dis)similarity should correlate with evolutionary distance. To check this hypothesis we performed an analysis of structural variation in the loop regions within different homologous protein families using a recently introduced new measure of loop similarity [ 7 ]. This new measure is based on the concept of the Hausdorff metric, which is used in mathematical topology to define a distance between two point sets of a metric space. It does not require an alignment or one to one correspondence between two point sets. We show that there exists a linear correlation between the average structural change in the loop regions and the evolutionary distance, which allows us to use the loop (dis)similarity measure for inferring the phylogenetic history of homologous protein families. Methods Test set To select sets of homologous proteins the Conserved Domain Database (CDD) version 1.62 was taken, which can be accessed at [ 12 ]. The CDD collection of protein domain alignments included curated CDDs [ 13 ] and preprocessed domain families imported from SMART and PFAM, altogether 6222 protein domain families[ 14 ]. Upon import, the sequences from SMART/PFAM alignments with more than 75% identity with known structures were substituted by the most similar structures from the Protein Data Bank [ 15 ]. Each CDD family was decomposed into a set of pairwise structure-structure alignments. Structural alignments were computed by the VAST algorithm [ 16 ] and only those structures which had more than 80% mutual overlap between the VAST alignment footprint and CDD footprint were considered in the analysis. The footprint for a given sequence was defined as a region between the first and the last residues aligned by VAST or CDD. Those families containing short sequence repeats and having average alignment length less than 50 residues were excluded from the test set. The structural pairs within the remaining CDD families were disregarded if at least one of the following conditions held true: - at least one structure in a pair had X-ray resolution of greater than 3.0 Å - the Blast E-value calculated for the VAST alignment exceeded 0.01 - at least one structure in a pair contained a chain discontinuous domain inconsistently aligned between VAST and CDD - at least one structure in a pair contained more than 25% of its nonaligned loops with missing residues. To ensure that protein families span a wide range of sequence similarity, all families were examined and those having less than 30% sequence identity span were not considered in further analysis. The redundancy between protein families was checked by using the procedure implemented in the CDART algorithm [ 17 ] and not more than 2 protein families from the same CDD cluster were retained in the final test set. At the end, the test set comprised 59 CDD families with more than 10 structurally aligned pairs of homologs. This test set covered a wide range of functional and structural classes and the list of test families together with their length, number of protein pairs and correlation coefficients is shown in Table 1 . Table 1 List of the names of 59 test protein families together with their CDD accession names, lengths, number of protein pairs, Pearson correlation coefficients between LHM (AHM) and normalized Blast bitscore. The families are ordered with respect to decreasing quality of LHM correlation. The supplementary table is available at [27]. Family name CDD acc Length #Obs AHM LHM Xylose_isom pfam00259 381 28 -0.99 -0.98 MHC_I pfam00129 175 28 -0.95 -0.96 PTPc smart00194 248 25 -0.92 -0.96 IPT smart00429 97 21 -0.90 -0.94 ZnMc_1 smart00235 137 34 -0.83 -0.94 RNAse_Pc cd00163 99 25 -0.82 -0.94 gpdh_C pfam02800 153 39 -0.72 -0.93 Aamy_C smart00632 81 31 -0.94 -0.90 peroxidase pfam00141 240 48 -0.90 -0.90 copper-bind pfam00127 81 87 -0.84 -0.89 CBM_20 pfam00686 94 15 -0.91 -0.89 RnaseA pfam00074 98 44 -0.48 -0.87 IGv cd00099 105 133 -0.78 -0.86 ADH_zinc_N pfam00107 337 64 -0.93 -0.86 ldh_C pfam02866 143 29 -0.93 -0.86 RIP pfam00161 232 28 -0.87 -0.85 Peptidase_C1 pfam00112 200 55 -0.82 -0.85 ZnMc_2 cd00203 134 23 -0.87 -0.85 PROF cd00148 120 15 -0.90 -0.85 plant_peroxidase cd00314 236 76 -0.90 -0.83 alpha-amylase_C pfam02806 78 39 -0.93 -0.82 sodcu pfam00080 139 15 -0.98 -0.81 fer2_1 cd00207 78 38 -0.86 -0.80 Pept_C1 smart00645 202 90 -0.86 -0.79 ferritin pfam00210 152 19 -0.94 -0.79 ldh pfam00056 135 44 -0.82 -0.78 SH2 pfam00017 86 21 -0.48 -0.78 flavodoxin pfam00258 143 26 -0.88 -0.78 EFh cd00051 57 59 -0.75 -0.77 rhv_1 cd00205 195 71 -0.86 -0.76 LYZ1_1 smart00263 116 67 -0.66 -0.75 aldo_ket_red pfam00248 277 28 -0.93 -0.73 COesterase pfam00135 485 28 -0.80 -0.72 TIG pfam01833 89 39 -0.90 -0.72 fer2_2 pfam00111 69 73 -0.77 -0.70 beta-lactamase pfam00144 264 45 -0.90 -0.70 rhv_2 pfam00073 216 95 -0.86 -0.70 GLECT cd00070 124 28 -0.80 -0.67 globin pfam00042 133 96 -0.74 -0.66 GST_C pfam00043 107 77 -0.77 -0.63 LYZ1_2 cd00119 109 24 -0.43 -0.61 PA2c smart00085 102 210 -0.29 -0.57 lipocalin pfam00061 131 55 -0.62 -0.56 phoslip pfam00068 102 102 -0.21 -0.54 proteasome pfam00227 189 56 -0.80 -0.51 UBCc smart00212 141 45 -0.79 -0.50 Sm smart00651 63 30 -0.54 -0.49 Tryp_SPc smart00020 208 561 -0.55 -0.46 CLECT_1 smart00034 90 35 -0.59 -0.44 crystall pfam00030 81 10 -0.76 -0.41 CLECT_2 cd00037 93 263 -0.45 -0.36 RHO smart00174 173 10 -0.52 -0.36 IGc1 cd00098 88 85 -0.65 -0.32 Tryp_SPc cd00190 211 378 -0.55 -0.31 MHC_II_beta pfam00969 86 32 -0.52 -0.26 ADK pfam00406 174 28 -0.37 -0.19 Rho cd00157 172 66 -0.20 -0.16 Phycobilisome pfam00502 148 15 -0.85 -0.10 ADF smart00102 116 10 -0.85 0.34 Measures of structural and sequence similarity To measure the sequence similarity between homologous proteins from the same family we used a Blast bitscore normalized by the alignment length. Among structure similarity measures used in this paper, two of them, RMSD and alignment-based Hausdorff measure (AHM) were computed by comparing the proteins in structurally aligned regions, while the loop-based Hausdorff measure (LHM) quantified the difference in the loop regions. The root mean squared deviation (RMSD) was calculated using the superposition algorithm due to McLachlan [ 18 ]. The AHM and LHM measures were based on the mathematical concept of Hausdorff distance[ 19 ]. Let A = { a 1 ,..., a m } and B = { b 1 ,..., b n } be finite point sets in a Euclidean space. The Hausdorff distance between the sets A and B is then defined by: d H ( A , B ) = max {min j d ( a 1 , b j ),..., min j d ( a m , b j ), min i d ( a i , b 1 ),..., min i d ( a i , b n )} (1) Here the terms d ( a i , b j ) denote the usual Euclidean distance between the points. In other words, the Hausdorff distance between the sets A and B is the smallest distance such that every point a i ∈ A is within this distance of some point b j ∈ B and vice versa. Hausdorff distance can be calculated under the assumption that the C α atoms for both structures are in a common coordinate frame which is defined by the structural alignment between two domains. The Hausdorff measure for loops (LHM) was calculated as an average of Hausdorff distances over all loops in the protein pair, where n s is the number of aligned secondary structure elements: The "loop" was defined as a region between two consecutive aligned secondary structure elements and: h i = 0, if the i -th loop regions do not have any unaligned residues; h i = d H ( A i , B i ), where A i contains the set of C α coordinates of non-aligned residues in the i -th loop of the first structure in a pair, the last aligned residue from the preceding aligned region and the first aligned residue from the following aligned region. Similarly, B i is defined for the second structure in a pair. The sets ( A i , B i ) are defined to include two aligned residues so that the measure can be defined even if one of the sets of non-aligned residues is empty. The Hausdorff measure for the structurally aligned regions (AHM) was defined similarly. In this case, instead of the sets that contain the coordinates for the C α atoms in the loops, we use the coordinates for the C α atoms in the aligned segments and average over the number of aligned segments. The correlation analysis between the measures of sequence and structural similarity, linear/nonlinear regression analyses and cluster analysis were performed using Splus version 6. Pearson (ρ) and Spearman correlation coefficients were calculated to quantify the accuracy of linear correlation. The P-value under the null hypothesis that the correlation coefficient between two variables is equal to zero has been estimated and those families with the P-values less than 0.01 were considered as having statistically significant correlation. The cluster analysis was done using the complete linkage clustering [ 20 ] where the distance between two clusters was measured as a maximum distance between a point in one cluster and a point in another cluster. The cluster trees based on p-distance and LHM were compared using the Phylip program [ 21 ] by generating 1000 bootstrap alignments from the structural alignments of a protein family and by calculating p-distance based cluster trees from the bootstrap alignments. The bootstrap support for the LHM based tree or different partitions of this tree was calculated by counting how many times the LHM topology occurs among the bootstrap cluster trees. Results and discussion Tables 1 and 2 show the accuracy of correlation obtained between the various measures of structural similarity (RMSD, AHM and LHM). As can be seen from these tables, the correlation quantified by the Pearson correlation coefficient is quite high for most of the families and half of the families have coefficients between -0.76 and -0.81 depending on the structural similarity measure used (Spearman rank correlation coefficients were shown to be very close to those reported in Tables 1 and 2 ). This result is consistent with the studies of Wood and Pearson who showed on a smaller test set of 35 protein families that half of them have correlation coefficients greater than 0.878 [ 8 ]. In their case the sequence-structure correlation was quantified, however, by using only the measures based on the structurally aligned regions of the proteins. The dependence of structural similarity on sequence similarity in some cases can be more accurately described by the nonlinear regression model taking into account higher order quadratic terms. To quantify how much the nonlinear terms improve the data fitting, we use the ratio of squared correlation coefficient for linear ( ) and nonlinear ( ) models ( ). In the overall test set only 12 families have r 2 – ratio smaller than 0.9 (with LHM used as a structural similarity measure) indicating that for these cases adding the non-linear term improves the performance of modeling by about 10%. As was shown previously, the evolutionary relatedness between proteins can be successfully gauged from the comparison of their loop regions [ 7 ]. Indeed, Table 2 and Figure 1 show that within the families of homologous proteins, the structural changes in loops are strongly coupled with evolutionary distance, which in the first approximation can be estimated using normalized Blast score. The structural-sequence dependence in loop regions for 71% of our protein families can be well described by a linear model and for 88% of protein families the linear correlation coefficients are found to be statistically significant. Comparing different measures of structural similarity one can see that AHM performs somewhat better than other quantities yielding 90% of families with statistically significant linear correlation coefficients (with P-value < 0.01) and 80% of families with r 2 > 0.9. Table 2 Table shows the median of Pearson correlation coefficients, fraction of families with statistically significant correlation (P-value less than 0.01) and the fraction of families with the ratio r 2 higher than 0.9 for each measure of structural similarity used in the study. Median correlation coefficient % families with significant correlation % families with r 2 > 0.9 RMSD -0.81 90 71 AHM -0.82 90 80 LHM -0.76 88 71 Figure 1 Hausdorff measure (in Angstroms) for loop (LHM) and aligned (AHM) regions is plotted versus the normalized Blast bitscore for three families: Pancreatic ribonucleases (RnaseA), Ig-like plexins/transcription factors (IPT) and Trypsin-like serine proteases (Tryp_SPc). Solid line shows the linear regression fit of the data. However, not all families exhibit such good correlation. One example of a protein family showing particularly low LHM correlation is the family of Actin depolymerisation factor/cofilin-like domains (ADF). The sequence-structure correlation for loop regions of this family is not statistically significant (the Pearson correlation coefficient is close to zero) whereas the sequence-structure correlation for the protein core is very high (ρ = -0.85 with AHM). Indeed, different proteins of this family show distinctly different loop conformations and evolutionary analysis of ADF family argued that the insertions present in the vertebrate ADF/cofilins (and not present in non-vertebrate cofilins) might be important for nuclear function of mammalian cofilins [ 22 ]. Therefore, in this case the structural heterogeneity of loop regions can be explained by the acquisition of a new distinct function by some members of this family. For some families, for example, Trypsin-like serine protease (Tryp_SPc), neither LHM (ρ = -0.31) nor AHM (ρ = -0.55) similarity measures exhibit a good sequence-structure correlation (Figure 1(c) ). Among families with particularly high LHM correlation are the families of Xylose isomerase (Xylose_isom), Class I Histocompatibility antigen (domains alpha 1 and 2, MHC_I), Protein tyrosine phosphatase (PTPc) and others. Figure 1 shows two families with high sequence-structure correlation using the LHM measure: Ig-like plexins (IPT) and Ribonucleases A (RnaseA). The IPT family is characterized by high sequence-structure correlation for both core (ρ AHM = -0.90) and loop regions (ρ LHM = -0.94). On the other hand, the protein core structure of the RnaseA family changes very little with sequence whereas the loop structure gradually diverges as sequence becomes more and more dissimilar (ρ AHM = -0.48, ρ LHM = -0.87). To understand whether significant sequence-structure correlation for loop regions has an underlying biological meaning, we performed a cluster analysis of proteins from two diverse families, Ribonuclease A (RnaseA), and SH2 domain (SH2, ρ AHM = -0.48, ρ LHM = -0.78), using different measures of sequence and structural similarity. Figure 2 depicts the cluster trees constructed using distance/similarity matrices which were based on the fraction of non-identical residues (p-distance), RMSD and LHM for these two families. Figure 2 Complete linkage cluster tree produced using fraction of non-identical residues (p-distance), RMSD (Å), and LHM (Å) is plotted between proteins from Pancreatic ribonuclease family (RnaseA). Five major groups of RnaseA family according to Rosenberg et al [23] are: eosinophil ribonucleases (ER), pancreatic ribonucleases (PR), angiogenins (ANG), Rana ribonucleases (RR) and ribonuclease 4 (R4). The maximum parsimony tree described by Rosenberg et al [23] is given in the Phylip format: (RR, ((ANG, R4), (PR, ER))). The RnaseA family represents a very interesting example to study as it is characterized by considerably different catalytic efficiency and substrate preferences among family members and the different aspects of its activity is not well understood. Although cysteines that form disulfide bonds, catalytic histidines and lysine residues are mostly structurally and sequence conserved, there is a great variability in sequence between other regions of RnaseA proteins [ 23 , 24 ]. We compared the obtained cluster trees (Figure 2 ) with the maximum-parsimony phylogenetic tree derived by Rosenberg et al [ 23 ], the Phylip format of this tree is given in the captions of Figure 2 . As shown in this figure, the RMSD-based tree divides pancreatic ribonucleases (PR) into two groups and puts together two very different proteins: angiogenin (ANG) and Rana ribonuclease (RR) although angiogenin has a very weak enzymatic activity and is a tumor-growth promoter while Rana ribonuclease P-30 has ribonuclease activity and antitumor effects. In contrast to the RMSD cluster tree, distance matrices based on the loop (dis)similarity measure correctly cluster the representatives of the five major groups of the Ribonuclease family as per Rosenberg et al [ 23 ]. Although the topology of the p-distance based cluster tree is somewhat different from the topology of the LHM based tree (with bootstrap support less than 0.001), it also produces a biologically meaningful clustering as judged from Rosenberg et al [ 23 ]. SH2 domains represent phosphor-tyrosyl peptide binding modules which are found in many signaling proteins. The specificity of phosphate interaction with a protein has been attributed to the hydrophobic pocket which is mostly formed by two loop regions [ 25 ]. Our analysis shows that indeed the loop regions have a much higher accuracy in clustering of functional subfamilies of SH2 domains. Comparing our cluster trees with the classification of Songyang et al [ 26 ] and cluster trees of SH2 phosphotyrosyl binding sites [ 25 ] we can see from Figure 3 that p-distance based and RMSD based distance matrices cluster correctly two representatives of the "1A" subfamily (vsrc, hck), but separate proteins from subfamily "1B" (csk, csk, syk) and "4" (shptp2 and shc). In contrast, these subfamilies ("1B" and "4" [ 26 ]) are very well supported by the cluster tree which is based on the LHM measure. The bootstrap calculations (see Methods) show that the LHM based topology is supported by the p-distance based clustering algorithm at less than the 0.001 level. Different partitions of this tree are supported at higher but still non-significant levels, namely 0.11 for the "1B" subfamily (csk, csk, syk) and 0.01 for the subfamily "4" (shptp2 and shc). This in turn indicates that the two cluster trees can be considered statistically different. Figure 3 Complete linkage cluster tree produced using fraction of non-identical residues (p-distance), RMSD (Å), and LHM (Å) is plotted between proteins from SH2 family (SH2). The classifications of SH2 domains according to [25, 26] are given in the parentheses: syk (1B, B), shptp2 (4, C), vsrc (1A, A), hck (1A, A), csk (1B, B), P85a (3, D) and shc (4,). Conclusions Here we have presented an analysis of how the structure of protein loops changes in evolution as homologous proteins diverge from each other. We showed that for the majority of protein families there exists a statistically significant linear correlation between measures of sequence similarity and average loop structural similarity. This in turn suggests that loops change in evolution via a stepwise insertion or deletion process and clearly one can not portray even longer loop regions as "irregular conformations" or "random coils". Indeed, our results imply that, in general, loops are under constant evolutionary constraints which, apparently, are weaker than those for a protein core but still strong enough to preserve the loop overall structure. Since loops do not contribute much to the protein core stability, these constraints predominantly arise from the importance of loops in interacting with ligands, other proteins and cells, as well as a possible role of loops in protein folding. Modeling of insertion and deletion events in evolution poses a lot of difficulties and protein evolution is usually reconstructed based only on the aligned regions of proteins. We demonstrated that loop regions which usually correspond to the non-aligned protein regions can be very important in inferring the phylogenetic history of a protein family. Moreover, it was shown, that sometimes sequence and structure similarity measures comparing proteins in their core are not sensitive enough to detect subtle (dis)similarities between the subfamilies. Loop-based measures which emphasize the dissimilarities between different protein members can shed light on the evolutionary relationships between homologous proteins. Authors' contributions AP and TM contributed equally to this paper. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549550.xml |
548128 | Using large-scale perturbations in gene network reconstruction | Background Recent analysis of the yeast gene network shows that most genes have few inputs, indicating that enumerative gene reconstruction methods are both useful and computationally feasible. A simple enumerative reconstruction method based on a discrete dynamical system model is used to study how microarray experiments involving modulated global perturbations can be designed to obtain reasonably accurate reconstructions. The method is tested on artificial gene networks with biologically realistic in/out degree characteristics. Results It was found that a relatively small number of perturbations significantly improve inference accuracy, particularly for low-order inputs of one or two genes. The perturbations themselves should alter the expression level of approximately 50–60% of the genes in the network. Conclusions Time-series obtained from perturbations are a common form of expression data. This study illustrates how gene networks can be significantly reconstructed from such time-series while requiring only a relatively small number of calibrated perturbations, even for large networks, thus reducing experimental costs. | Background Recent technological advances have led to an explosive growth in high-throughput genomic and proteomic data such as DNA microarrays. The rapid growth in available data has led in turn to a need for novel quantitive methods for analysis. As a consequence of this need, the reconstruction of gene network architectures from DNA microarray expression data has become a major goal in the field of systems biology. An increased understanding of the network architectures and their respective dynamics will enable novel approaches to disease treatments by allowing us, for example, to identify drug targets in silico which manipulate the functional outputs of these networks. This process is expected to lead to novel classes of drug based on a network approach to cellular dynamics. Frequently, the gene expression data itself is derived from perturbation experiments such as stress conditions, temperature shifts, and chemical treatments; for example, the widely used yeast cell-cycle datasets of Cho [ 1 ] and Spellman [ 2 ]. Although these global perturbations are carried out in order to reveal causality between genes, it is not always clear how experiments should be designed so as to reveal as much causality as possible, while both minimising costly experimentation and remaining computationally tractable. A range of computational and mathematical techniques have been adopted in the effort to find a successful gene network reconstruction technique. Reconstruction methods often have to negotiate a tradeoff between intensive (often intractable) computations, and having to perform a large number of costly experiments. Certain progress can be achieved by making simplifications, such as imposing a limit on the number of inputs to each gene, or making steady state assumptions about the system [ 3 , 4 ]. Some techniques described in the literature offer efficient algorithms, but require a large number of experiments, perhaps as many as there are genes [ 5 - 7 ]. On the other hand, theoretical work on Boolean models has shown [ 8 ] that perhaps as few as O ( log ( n )) experiments (input/output pairs) might be required for n genes, but that to infer these relationships requires the use of computationally costly enumeration methods. In this paper, we propose to explore the issue of how perturbation microarray experiments might be designed, and to suggest how such experiments might be optimised so as to maximize inference capability. Logical gene network models have previously been used to investigate gene network robustness [ 9 ], perturbation dynamics [ 10 ] and evolutionary potential [ 11 ], and form the basis of the inference method used in this study. This inference method [ 11 ] is similar to others in which networks with a minimal number of connections are reconstructed through enumeration [ 12 , 13 ]. Given the significant speed advantage of integer computation over floating point computation, and that most genes are expected to have few inputs (93% have between 1 and 4 [ 14 ]), the method is considered to be adequate for this investigation. In this study, exhaustive evaluation was performed up to a maximum of 4 inputs of both positive and negative sign (see Methods). Enumeration is computationally feasible on an ordinary desktop computer for medium-sized networks ( n ~ 100), and still tractable for large networks ( n ~ 1000), though this would require some parallelisation. The global perturbations themselves are simulated by changing the state of each gene at random. A perturbation intensity measure q , defines the probability that each gene will change state (see Methods). Results and discussion A limited number of perturbations significantly improve accuracy A discrete dynamical model was used to generate time series data from random networks (see Methods). To measure the effect of adding perturbations on inference ability, inference sensitivity (defined as true positives/true positives + false negatives, see Methods) was measured against P , the number of additional perturbations. Figure 1 shows the results for predicted solutions with one and two inputs, as well as overall sensitivity. The top graph in figure 1 shows that overall sensitivity is clearly enhanced by including more perturbation experiments, with lower order solutions (one and two inputs) reaching higher levels of sensitivity. The bottom graph shows the corresponding inverse relationship for the standard deviation of the sensitivity (lower for higher P ). It should be noted that the algorithm tends to underestimate the number of inputs a gene may have. This is to be expected in genes for which dynamics cannot be informative: for example, consider a gene i which has one or more negative inputs, as well as having default value OFF. Since the discrete dynamics for this gene will be the same as if it had no inputs at all (i.e. zero gene expression for t > 0), the presence of the inputs is impossible to infer. This underestimation effect is clear in table 1 , which compares the distribution of inferred solution set sizes (| Y i |, see Methods) with the actual solution sizes (i.e. the indegree distribution), and shows that the method is only able to produce roughly half the number of one and two input solution sets that actually exist. The increase in sensitivity with P can be explained at least partially, in the following way. Since the time series are discrete, many of the genes may have identical behaviour over time despite having different inputs (i.e. s i ( t ) = s j ( t ) for two different genes i and j ). If we define a "concatenated" time series vector for gene i , and then map each gene i onto S i , we obtain a many-to-one mapping. As we increase the number of perturbations, we might expect the number of distinct time series also to increase. We define a simple measure to quantify this mapping, M = n '/ N where n ' is the number of distinct vectors S i , and N is the number of genes. The maximum value of M = 1 indicates that the mapping of genes to time series is one-to-one, whereas lower values indicate degenerate mappings. The manner in which M increases with the number of perturbations is shown in figure 2 , and shows how the increase in M reflects the corresponding increase in sensitivity (figure 1 ). Network size and optimal perturbation intensity The experiments described above were repeated to consider variations in two other parameters: the network size N , and the perturbation intensity parameter q (roughly, the proportion of genes whose initial expression level is changed in each perturbation experiment – see Methods). To consider the first case, the minimum number of perturbations P * required to reach a given high accuracy criterion was measured for different values of the network size N . The high accuracy criterion was defined as average sensitivity = 0.95 for one-input solution sets (average sensitivity is found using a default value q = 0.5 and averaging for all the sensitivity measurements obtained from 250 random networks). To find P *, we first find the number of perturbations P + , such that average sensitivity P + ≥ 0.95, and average sensitivity ( P + - 1) < 0.95. If average sensitivity P + > 0.95, we use simple linear interpolation to find the (real) value of P * between P + and ( P + - 1) for which average sensitivity = 0.95. The resulting values for P * are shown in figure 3 . Since the relationship is expected to be logarithmic [ 8 ], the plot shows log ( N ) against P * (logarithms used are base 10). A least squares best fit gives P * ≃ 1.75 log ( N ) + 7.02, which, for N = 1000, gives P * ≃ 12.26. In order to obtain a measure of variance for P *, we would need to calculate P *-equivalent values for many individual networks separately, then consolidate these values to obtain the relevant statistics. However, because it was only feasible to consider medium-sized networks (20 ≤ N ≤ 70), and for any such network we often find only a small number of one-input solution sets, such statistics were found to be unreliable. The second case (varying perturbation intensity) suggests an optimal range for q . Figure 4a shows the inference sensitivity over a range of values for q , and figure 4b shows the corresponding standard deviation. Again, inference sensitivity for one-input solutions is higher than for two-input solutions, which in turn is higher than overall sensitivity. For one-input solutions, the results show a clear peak for sensitivity close to the range 0.5 < q < 0.6. Together with a corresponding minimisation of the standard deviation in this interval (though it still remains fairly high in absolute terms), these results suggest that perturbation intensity should be close to this range to optimise inference accuracy. Conclusions A recent analysis of the yeast genetic network has shown that 93% of genes are regulated by between 1 and 4 genes [ 14 ]. This suggests that enumerative network reconstruction methods can be useful within computationally feasible limits. Experiments involving large-scale perturbations (such as temperature shifts, chemical stress) are a standard way of obtaining time-series of gene expression data [ 1 , 2 ]. A key result of [ 14 ] is that indegree appears to follow an exponential distribution, whereas outdegree follows a scale-free distribution, which has enabled the generation of realistic artificial gene networks used here. A logical model [ 11 ] was used to simulate the perturbed expression data. Subsequently, experimental parameters were considered in relation to inference accuracy, namely: a) number of perturbations required, P , and b) perturbation intensity, q . The inference method itself is most useful for low order inputs, with inference accuracy maximized for predicted single input genes. More accurate methods have been proposed, though these generally require a much larger number of experiments [ 5 , 15 ]. Methods such as the one proposed here, which infer relationships from expression data may well be more successful when used in conjunction with other methods such as promoter analysis [ 16 , 17 ], or when used to drive experimental procedure [ 18 ]. Here, the results show that only a relatively small number of perturbations are necessary in order to achieve a substantial inference accuracy, even for large N . These relatively modest experimental requirements would presumably imply lower experimental costs. The results also suggest that the perturbations should be calibrated (by changing stress intensity, for example), so as to alter the expression levels of approximately half the genes in each experiment. Generating perturbations which alter the expression level of half the genes at random may be difficult to achieve in practice, though experiments can be designed to come as close to this goal as possible. Even in the absence of optimal perturbations, we hope the simulation approach described here will still serve as a useful tool for planning experiments. Methods Discrete dynamical model For a system of N genes, the state of each gene s i ( i = 1, .., N ) is represented by the binary values 0(OFF) and 1(ON). Additionally, each gene is assigned a default ON/OFF state θ i ∈ {0, 1}. The gene interactions are described by an ( N × N ) matrix C , composed of elements C ij ∈ {-1, 0, +1}, representing the positive(+1), zero(0) or negative(-1) influence of gene j on gene i . State transitions are calculated as follows: The state of the i th gene at the next timestep, s i ( t + 1), is therefore determined by the balance of positive versus negative inputs which are ON at the previous timestep t . If the balance is positive, then u i ( t ) > 0 and the next state will be 1(ON). Similarly, if the balance is negative, then u i ( t ) < 0 and the next state will be 0(OFF). If u i ( t ) = 0 (indicating either that there are no active input connections, or that they balance out), then the default value θ i determines the next state. This default value needs to be given a priori , and for the purpose of this study will be random. Network inference method Assuming we are given the state dynamics s ( t ) and the default vector θ , the problem is to find the necessary model parameters ( C ) which will reproduce these dynamics. Specifically, a system initialised at s (0) should reproduce the given dynamics s ( t ) for t > 0. Note that multiple s ( t ) expression patterns may be defined, which will be denoted as s r ( t ) for r = 0, .., P , corresponding to time series with different initial states s r (0). Our problem is to find at least one interaction matrix that will reproduce all given dynamics s r ( t ). The problem of finding an appropriate matrix C may be broken up into N sub-problems, since in this system, each gene i may be solved independently from the others. More precisely, the inputs to gene i (i.e. C i , the i th row of C ), can be found independently of the other genes. This reduces the search space from down to O ( N 3 N ). Each input z i to gene i is represented as an ordered pair ( j , g ), j ∈ {1, .., N }, g ∈ {± 1}, indicating an input from gene j of sign g . A solution y ( i ) for gene i is a set of K inputs (with y ( i ) = φ if K = 0). For K inputs there are solutions to evaluate. Starting with K = 0 (no inputs), we progress up to a maximum of K = 4, exhaustively evaluating all possible solutions for each K . However, making a parsimony assumption, if solutions are found for some K s < 4, the method no longer evaluates for K > K s . Note that the method does not stop as soon as a solution is found, but evaluates all possible solutions for K s . The failure rate (percentage of genes for which no solution was found for K ≤ 4) never exceeded 3% of the genes in any single network for which reconstruction was attempted. Global perturbations and the perturbation intensity measure The control time series s 0 ( t ) is generated by setting s 0 (0) = θ . The other time series s r ( t ), r > 0 are obtained from initial conditions which are perturbations of θ , and correspond to standard experiments such as stress conditions, or chemical treatments. Since, experimental perturbations can usually be modulated in intensity (for example, a temperature shift), this was represented using modulated artificial perturbations. Perturbed initial states s r (0) were generated by randomly changing each state s 0 (0) with probability q . This means that, on average, there will be qN random state differences between each perturbed initial state s r (0), and θ . Measuring inference accuracy Assuming one or more solutions y 1 ( i ), y 2 ( i ), ... are found for gene i , these are consolidated into a solution set, Y i = ∪ l { y l ( i )}. Note that some information about the solutions has been lost using this approach. For example, a solution set obtained from a single two-input ( K = 2) solution: , may be equal to another solution set resulting from two single-input ( K = 1) solutions: with and . However, this consolidation is convenient in that the solution set is easily compared with the known network structures using standard accuracy measures such as sensitivity and specificity . Here, accuracy was measured using sensitivity , defined as true positives / (true positives + false negatives). The relatively large number of true negatives, makes specificity , defined as true negatives / (true negatives + false positives), an uninformative statistic. Here, true positives are members of the solution set Y i which are also true inputs (since the networks will be generated artificially, true inputs are known), and false negatives are those true inputs which are not members of the solution set Y i . Accuracy statistics were gathered from inferences performed on a large number of medium-sized random networks (20 ≤ N ≤ 70). Inferences on R random networks (each with N genes), will produce approximately RN sensitivity measurements (slightly fewer due to the nonzero failure rate). Artificial gene network generation It appears to be the case in gene networks that indegree follows an exponential distribution, whereas outdegree appears to follow a scale-free distribution. More specifically, for the yeast network, the probability distribution for indegree k follows p k ~ C in e - βk with β ~ 0.45, whereas the distribution for outdegree follows p k ~ C out k - τ , with τ ~ 1 ( C in , C out constants) [ 14 ]. Here, artificial gene networks [ 19 ] were created using the algorithm for generating directed graphs with arbitrary in/out degree distributions described in [ 20 ]. The exponential probability distribution for indegree k is given by: p k = (1 - e - β ) e - βk , where β = 0.45 is a constant. Similarly, the power law distribution (including an exponential cutoff term which is both biologically realistic and necessary analytically when τ < 2 [ 20 ]) for outdegree k is described by: p k = Ck - τ e - γk , where C , γ , and τ = 1 are constants. Since the algorithm begins by generating in/out-degree pairs for each node, we require equal means for both indegree (< k in >) and outdegree (< k out >). Following [ 20 ], we obtain expressions for the mean in/out degree: Since β is given, we obtain a value < k in > = 1.76, and fit the free parameter γ = 0.436 to obtain < k out > = < k in > Since the resulting networks are unweighted, non-zero weights ( C ij ∈ {-1, +1}) are assigned at random with probability 0.5, as in [ 19 ]. It should be noted that autoregulatory interactions can be (and indeed were) generated, and that these present no particular problem for the inference method. An example of a network which was used in the analysis is shown in figure 5 . Authors' contributions TM devised and implemented the experiments and drafted the manuscript. RS and AP supervised the project. All authors read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC548128.xml |
516040 | Potential mechanism of phytochemical-induced apoptosis in human prostate adenocarcinoma cells: Therapeutic synergy in genistein and β-lapachone combination treatment | Background Prostate cancer is the second leading cause of male death in the United States. The incidence increases most rapidly with age, and multiple genetic and epigenetic factors have been implicated in the initiation, progression, and metastasis of the cancer. Nevertheless, scientific knowledge of the molecular mechanisms underlying the disease is still limited; and hence treatment has only been partially successful. The objective of the current studies was to examine the role of caspase 3 (CPP32) and NAD(P)H:quinone oxidoreductase (NQO1) in the signaling of genistein-and β-lapachone (bLap)-induced apoptosis in human prostate carcinoma cells PC3. Results Both genistein and bLap produced dose-dependent growth inhibition and treatment-induced apoptosis in PC3. Treatment with caspase 3 inhibitor, DEVD-fmk before exposure to genistein, significantly inhibited caspase 3 expression and treatment-induced apoptosis; implicating CPP32 as the main target in genistein-induced apoptosis in PC3. Contrary to this observation, inhibition of CPP32 did not significantly influence bLap-induced apoptosis; implying that the major target of bLap-induced apoptosis may not be the caspase. Treatment with NQO1 inhibitor, dicoumarol (50 μM), prior to exposure of PC3 to bLap led to significant decrease in bLap toxicity concurrent with significant decrease in treatment-induced apoptosis; thus implicating NQO1 as the major target in β-lapachone-induced apoptosis in PC3. In addition, the data demonstrated that NQO1 is the major target in bLap-genistein (combination)-induced apoptosis. On the contrary, blocking NQO1 activity did not significantly affect genistein-induced apoptosis; implying that NQO1 pathway may not be the main target for genistein-induced apoptosis in PC3 cells. Furthermore, blocking NQO1 and CPP32 did not confer 100% protection against genistein-induced or bLap-induced apoptosis. Conclusion The data thus demonstrate that both genistein-and bLap-induced apoptosis are mostly but not completely dependent on CPP32 and NQO1 respectively. Other minor alternate death pathways may be involved. This suggests that some death receptor signals do not utilize the caspase CPP32 and/or the NQO1 death pathways in PC3. The demonstrated synergism between genistein and bLap justifies consideration of these phytochemicals in chemotherapeutic strategic planning. | Background Prostate cancer is the most common non-skin malignancy and the second leading cause of male death in the United States [ 1 ]. The incidence of prostate cancer increases most rapidly with age, and multiple genetic and epigenetic factors have been implicated in the initiation, progression, and metastasis of prostate cancer. Nevertheless, scientific knowledge of the molecular mechanisms underlying the disease is still limited. The problem often faced with the clinical management of prostate cancer is derived not only from the fact that no single gene or molecule can serve as a reliable marker [ 2 , 3 ], but also that there is still no effective therapeutic regimen available without serious, sometimes fatal side effects. Unfortunately, at the time of clinical diagnosis, human prostate cancers mostly present themselves as heterogeneous entities – hormone-dependent and hormone-independent, and proliferating and non-proliferating. The tumor re-growth that occurs after post-treatment remission is largely due to progression of initially androgen-dependent to androgen-independent cancer cell [ 4 ] and/or non-proliferating to proliferating tumor cells. Therefore chemotherapeutic strategies should focus on eradicating all cancer cells irrespective of state of growth or sensitivity to hormone. This calls for a search for drug combination treatment that works through different mechanism of action. The facts that prostate cancer cells retain the inherent apoptotic machinery potentially subject them to an appropriate efficacious chemotherapeutic intervention. The molecular mechanism(s) and intracellular mediators of both spontaneous-and treatment-induced apoptosis are not fully elucidated. However, evidence from several research investigations seem to indicate that a variety of stimuli, including physiological, pathologic, environmental or cytotoxic, can trigger the process of apoptosis in many mammalian cells [ 5 , 6 ], and that both apoptosis and necrosis may share some upstream events in the molecular pathways that lead to induction of apoptosis [ 7 - 11 ]. An emerging strategy for cancer chemotherapy is the choice of drugs that induce apoptosis and/or disruption of angiogenesis with eventual elimination of the cancer. It is suggested that blocking of caspase activation in an apoptotic process may divert apoptotic cell death to a necrotic demise [ 10 ]; implying that apoptosis and necrosis may share some upstream events in the molecular pathways of apoptosis induction. Among the dietary phytochemicals of potential therapeutic significance, are genistein isoflavone and β-lapachone, both of which induce apoptosis and also inhibit angiogenesis (genistein) in an array of cancer cells [ 6 , 12 , 13 ]. Genistein isoflavone [4',5',]-trihydroxyisoflavone) is a metabolite of soy [ 14 ] and has a heterocyclic, diphenolic structure similar to estrogen [ 14 ]. The phytochemical isoflavonoid family to which genistein belongs is a group of plant chemicals that resemble steroid estrogens and mimic their biological reactions [ 15 , 6 , 16 ]. Several clinical studies indicate that genistein has some chemoprotective and chemotherapeutic potential against many tumors, including prostate, breast, and colon cancers through several mechanisms of action including: apoptosis induction; modulation of cell cycle activity by arresting cell cycle at the G 2 -M stage [ 17 ]; inhibition of DNA topoisomerase-II and tyrosine protein kinase [ 18 ]; competitive inhibitor of ATP binding to the catalytic domain of tyrosine kinase [ 14 , 18 ]; stimulating the production of sex hormone-binding globulin (SHBG), which may lower the risk of hormone related cancers by decreasing the amount of free and active hormones in the blood [ 19 , 20 ]. The other phytochemical of potential therapeutic significance is β-lapachone [3, 4-dihydro-2, 2-dimethyl-2H-naphtol (1,2-b) pyran-5,6-dione], a simple plant product with a chemical structure different from currently used anti-cancer drugs. It has been previously demonstrated that the primary mode of cytotoxicity of β-Lapachone is through the induction of apoptosis [ 21 , 22 ]. Structural similarities between β-lapachone and other members of the naphthoquinone family, such as menadione, suggest that the enzyme, NAD(P)H:Quinone oxidoreductase enzyme (NQO1) may be involved in the activation or detoxification of β-lapachone [ 23 - 25 ]. While a number of in vitro effects of β-lapachone and genistein have been described, knowledge of the key intracellular targets of β-lapachone and genistein is limited. Recent reports have suggested that by β-lapachone-induced apoptosis is non-caspase mediated in breast [ 26 ] and prostate cancer cells [ 21 , 34 ], and that the cytotocxicity of this compound is dependent on the activity of NAD(P)H:Quinone oxidoreductase enzyme (NQO1/xip3) [ 26 , 27 ]. B-Lapachone has been shown to be an inhibitor of DNA repair that sensitizes cells to DNA-damaging agents [ 28 , 29 ]. It directly inhibits DNA topoisomerase I and II [ 30 - 32 ] and induces a cell-cycle delay in G 1 and/or S phase followed by apoptotic and/or necrotic cell death [ 13 ]. The apoptosis induced by β-lapachone is p53 independent [ 21 ], and has been associated with upregulation of Bak as well as cleavage of caspase-7 [ 33 ] and caspase 3 in a variety of mammalian cells [ 13 , 33 ]. The objective of this study was to determine the potential chemosensitivity of human prostate adenocarcinoma, PC3 cells to β-lapachone and Genistein and the role of caspase 3 (CPP32) and NAD(P)H:Quinone oxidoreductase enzyme (NQO1) in the signaling of β-Lapachone and genistein-induced apoptosis in PC3 cells. The hypothesis is that combination treatment with the two phytochemicals will be strongly preventive and/or interceptive against prostate cancer by modulating epigenetic events (apoptosis) associated with the progression of active and latent cancer cells to clinical malignancy. Results Genistein and β-lapachone inhibit growth and proliferation of human prostate carcinoma cells, PC3 Human prostate carcinoma cells PC3, was used to determine the chemosensitivity of prostate cancer to genistein isoflavone and β-lapachone in vitro using Trypan blue exclusion, LDH and MTS bioassays. In single and combination treatments, both genistein and β-lapachone inhibited cell growth and decreased cell survival through induction of cell death [Figures. 1 , 2 , 3 ]. The data indicated that PC3 sensitivity to both single and combination treatment is dose-dependent, and that PC3 was significantly more sensitive (P < 0.05) to the combination treatment than to the single treatment; indicating a potential synergism between genistein and β-lapachone [Figures 2 , 3 ]. Figure 1 β-lapachone-induced growth inhibition in PC3. PC3 cells (1 × 10 4 cells/well) were cultured in 24-well plates for 48 hr to allow 85–90% confluence; treated with varyingconcentrations of bLap and assessed for post-treatment viabilitywith the MTS assay. Note the dose-dependent growth inhibitionin PC3. Data points represent means ± SEM of three independentexperiments performed in triplicates Figure 2 Genistein (Gn)/β-Lapachone combination treatment of PC3. Cells were treated as described in the methods and subjected to post-treatment viability with MTS colorimetric assay. Data points represent the means ± SEM of three independent experiments performed in triplicates. Figure 3 Single and combination of PC3 cells with genistein (Gn) and β-lapachone (bLap) βLap. Briefly, PC3 cells were seeded at 1 × 10 4 cells/well in 48-well MTP and co-cultured with Gn 0-70 with/without bLap (1.2 μM); followed by determination of treatment-induced cytotoxicity as described in the methods. Data points represent means ± SEM of three experiments performed in triplicates Genistein and β-lapachone induce apoptosis in human prostate cancer cells Extensive cell death was observed in proliferating human prostate cancer cells after treatment with β-lapachone and genistein isoflavone. To determine if the treatment-induced cell occurred through cytotoxic necrosis and/or apoptosis, cells were harvested and assayed for apoptosis induction with Annexin V-FITC and TUNEL apoptosis assays to detect early and late apoptosis respectively. Aliquots of cells were also stained with acridine orange/ethidium bromide nuclear stain to distinguish between apoptotic and necrotic cells. The results revealed that in both single and combination treatments, cell death was mostly through apoptosis in a dose-dependent manner [Figures 4 , 5 ]. With increasing concentration of the agents, cell death through necrosis increased correspondingly. Furthermore, combination treatment induced significantly more apoptosis in PC3 (p <0.01) than individual treatment with either agent. Figure 4 βLap)-induced cell death in PC3 cells. PC3 cells were co-cultured with varying concentrations of bLap; and and the degree of treatment-induced apoptosis and/or necrosis assessed with the Annexin V-FITC assay, as described in the methods. Data points are means ± SEM of three independent experiments performed in triplicates. Figure 5 Combination treatment-induced cell death in PC3 cells. PC3 cells were co-cultured with varying concentrations of Gn (Gn 0-70 ) with or without bLap (1.2 μM), and apoptotic/necrotic cell death assessed with theAnnexin V-FITC assay as described in the methods. Data points are means ± SEM of three independent experiments performed in triplicates Dicoumarol enhanced the survival of human prostate cancer cells (PC3) following single treatments with β-lapachone (bLap) and Gn/bLap combination but not in PC3 cells treated with genistein alone To determine the potential role of the enzyme NAD(P)H:quinone oxidoreductase (NQO1) in β-lapachone (bLap)-and genistein (Gn)-induced apoptosis in PC3, the cells were exposed to Gn and bLap in the presence or absence of dicoumarol in single and combination treatments; and then assayed for apoptosis by the Annexin V-FITC and TUNEL assays. Dicoumarol is a specific inhibitor of NQO1. The results revealed that blocking NQO1 activity with dicoumarol (50 μM) significantly reduced bLap-induced apoptosis [Figure 6 ]; indicating that bLap-induced apoptosis requires involvement of NQO1 target. However, dicoumarol did not appear to have significant effect on Gn-induced apoptosis [Fig 7 ]; indicating that NQO1 did not play significant role in Gn-induced apoptosis. [Figures 6 , 7 ]. The degree of apoptosis induction was highest in the Gn-bLap combination treatment without inhibiting NQO1 activity with dicoumarol [Figure 7 ]; implying that a possible synergy between Gn and bLap may be due to NQO1 activity. Figure 6 Role of NQO 1 in β-Lapaphone-mediated apoptosis in PC3 cells. PC3 cells were treated with bLap alone or in combination with 50 μM dicoumarol (NQO 1 inhibitor) as described in the methods; and TUNEL assays performed to monitor apoptosis. Data points represent means + SEM of three independent experiments performed in triplicates. Figure 7 NQO1 is the main target in bLap/Gn-induced apoptosis in PC3 cells. Cells were treated with genistein (Gn) and Gn/bLap combination with or without 50 μM dicoumarol as described in the methods; and TUNEL assays performed to monitor apoptosis. Data points represent means ± SEM of three independent experiments performed in triplicates. Activation of CPP32 in genistein-induced apoptosis in PC3 but not in β-lapachone-induced apoptosis in PC3 To determine if apoptosis induced by β-lapachone and/or genistein involved activation of caspase 3 protease (CPP32), the PC3 cells, were subjected to treatments with Gn and/or bLap co-administered with or without CPP32 inhibitor (DEVD-fmk), and then cultured as previously described. Post-treatment apoptosis was determined as previously described. As shown in Figures 8 and 9 , blocking the release of caspase 3 significantly decreased genistein induced apoptosis but not bLap-induced apoptosis; indicating the significant role of CPP32 in the molecular pathway of Gn-induced apoptosis; and minor involvement of CPP32 in bLap-induced apoptosis in PC3. Furthermore, blocking CPP32 activity did not significantly affect combination treatment-induced apoptosis (Figure not shown). Figure 8 Caspase-3 (CPP32) activity in genistein-induced apoptosis in PC3 cells. PC3 cells (2.5 × 10 3 cells/well) were cultured; then treated with/without 100 μM caspase inhibitor (zVAD-fmk) for 2 hr; and then with 10–70 μg/mL genistein for 4 hr as described in the methods. Cells were thenanalyzed for caspase (CPP32) activity and corresponding apoptosis in the cells. Data points were the means ± SEM of two independent experiments performed in triplicates. Figure 9 CPP32 is the major pathway in genistein-induced apoptosis in PC3 cells. PC3 cells (2.5 × 10 3 cells/well) were cultured in 48-well culture plates; treated with/without 100 μM caspase inhibitor (zVAD-fmk) for 2 hr; then with 1–8 μM β-Lapachone (bLap) for 4 hr as described in the methods. Cells were then analyzed for caspase (CPP32) activity and corresponding apoptosis. Data pointsare the means ± SEM of two independent experiments performed in triplicates Discussion and Conclusions In this study, we determined the role of caspase 3 (CPP32) and the enzyme NAD(P)H:quinone oxidoreductase (NQO1) in the signaling of β-lapachone (bLap)-and genistein (Gn)-induced apoptosis in human prostate adenocarcinoma, PC3 cells. Data from this study demonstrate significant inhibition of cell growth and proliferation in PC3 cells, with significant difference in chemosensitivity of PC3 to genistein and β-lapachone (P < 0.01). Furthermore, growth inhibition of PC3 cells strongly correlated with the MTS and LDH assay results. The pattern of response and percent post-treatment live cells was consistent with previous results [ 6 , 37 - 39 ]. The genistein-and bLap-induced morphological changes observed in the cells were identical in pattern but differed in severity at a given exposure time; indicative of differences in chemosensitivity of PC3 to genistein and β-lapachone. These observations were consistent with previous results [ 5 , 6 ]. Furthermore, previous studies have shown that β-lap [ 22 ] induces morphologic changes indicative of apoptosis in human breast cancer cells. Similar alterations in morphology including cell shrinkage and chromatin condensation in the PC3 cells following single and combination treatment with β-lapachone and genistein isoflavone. The present data also implicates caspase-3 protease, CPP32, in the molecular pathway of genistein-induced apoptosis in prostate PC3 cancer cells, consistent with previous investigations [ 10 , 11 , 39 ]. Using the caspase inhibitor DEVD-fmk, caspase activity was arrested concurrent with significant decrease in genistein-induced apoptosis in PC3 cells. However, it is noteworthy that inhibition of caspase did not confer 100% protection against genistein-induced apoptosis; implying alternative death pathways, which suggests that some death receptor signals do not utilize the caspase CPP32 death pathways in PC3. We have previously demonstrated the significant role of caspase-3 protease in the genistein-induced apoptosis pathway in both testes and prostate (PC3) cancer cells [ 38 , 39 ]. The present data indicate a possible alternate CPP32 pathway in bLap-induced apoptosis in PC3. However, unlike the observation in genistein-induced apoptosis, blocking the CPP32 activity with the specific caspase inhibitor, DEVD-fmk, did not significantly change the percentage of bLap-induced apoptosis in PC3 cells; indicating that CPP32 many not be the main death pathway of bLap-induced apoptosis in PC3 cells. Activation of the caspase 3 in bLap-induced apoptosis has been reported in previous studies [ 10 ]. The potential role of NAD(P)H:quinone oxydoreductase (NQO1) activity in genistein-and bLap-induced apoptosis in PC3 was investigated. Co-culture of PC3 cells with dicoumarol, a specific inhibitor of NQO1 activity, significantly reduced the cytotoxicity of β-Lapachone in PC3 cells, as reflected in the significant reduction in the percentage of treatment-induced apoptosis. Dicoumarol increased cell survival. These results implicate NQO1 as the main target in bLap-induced apoptosis in PC3, consistent with previous observations [ 26 , 34 , 40 ]. However, the fact that blocking of NQO1 did not confer 100% protection against induction of apoptosis indicates a possible alternate pathway in bLap-induced apoptosis. The present data indicate some involvement of caspase protease CPP32, though not with the same significance as NQO1. The activation of cysteine protease has been observed after bLap treatment [ 26 ]. Pink et al [ 26 ] reported activation of the cysteine protease in MCF-7 and T4D breast cancer cells in bLap-induced apoptosis. Contrary to the observation in bLap-induced apoptosis, blocking NQO1 activity did not significantly influence genistein-induced apoptosis in PC3 cells; implying NQO1 may not be the major target in genistein-induced apoptosis in PC3 cells. However, the overall data indicate a synergistic effect of genistein-bLap combination treatment of PC3 and, that the major target in the combination treatment-induced apoptosis in PC3 cells is NQO1. Investigation into genistein/bLap synergism in a number human cancer cells is on-going. Conclusion It is concluded from the data obtained that: i) both genistein and bLap exert growth inhibition effects in PC3 cells, with significant differences in chemosensitivity of PC3 to the two agents; ii) there is a manifestation of synergism between genistein-bLap combination treatment; iii) both genistein and bLap induce apoptosis in PC3 cells; iv) the major target in genistein-induced apoptosis in PC3 cells seems to be CPP32; v) the major molecular target in bLap-induced apoptosis in PC3 cells appear to be NQO1; vi) the major target in the genistein-bLap combination treatment-induced apoptosis appears to be NQO1; and vii) combination treatment appears significantly more efficacious than single treatments. More extensive studies are ongoing to delineate and clarify the molecular mechanisms underlying the combination effects. Materials and Methods Chemicals Genistein isoflavone (Indoline Chemical Co. Summerville, NJ, USA) was constituted in DMSO (dimethylsulfoxide) solvent as 10, 20, 30, 50 and 70 μg/ml solutions (G 10-70 ) and frozen at -37°C until used. β-lapachone (Sigma Scientific (St. Louis, MO, USA) was constituted in DMSO solvent as 1, 2, 3, 5 and 8 μM/ml solutions (bL 1-8 ) and stored at -37°C. Dicoumarol (Sigma Scientific St. Louis, MO, USA) was constituted in DMSO as 50 μM/ml and frozen at -37°C until used. The caspase 3 inhibitor, DEVD-fmk and substrate DEVD-afc were purchased from Biovision (Palo Alto, CA). Culture media (RPMI 1640), antibiotics, trypsin-EDTA, and other reagents were purchased from Sigma scientific (St. Louis, MO, USA). Cell lines Human prostate adenocarcinoma (PC3) was a generous donation from Rambaugh-Goodwin Cancer Research Institute, Plantation FL. Cell Culture To assess the chemosensitivity of human prostate cancer cells to single and combination treatment with genistein (Gn) and β-lapachone (bLap), cells were sub-cultured under 5% CO 2 at 37°C for 48 hrs to reach 80–90% confluence. All cells were grown and maintained as monolayers in 25 m 2 tissue culture flasks (Sigma Scientific, St. louis, MO, USA) in RPMI 1640 containing 15 mM HEPES, and supplemented with 0.45% glucose (w/w), 5.0% FBS and 100 U . mL -1 penicillin + 100 mg . ml -1 streptomycin. The cells were then harvested by gentle scraping with a cell scraper. The cell suspensions were then grown at a density of 2.5 × 10 3 cells/well in 24-well microtiter plates (MTP) for 36 hr to allow adherence. The supernatants were discarded and the agents (Gn or bLap) were added over a range of 5 cytotoxic concentrations in single and combination treatments. In preliminary studies with bL 1-8 , the IC 50 ranged between 1.8–3 μM for a number of cells. Therefore in the present studies, 2.0 μM (bL 2 ) was used in the combination studies with varying concentrations of genistein. All treatments were in triplicates. Dicoumarol was added to the cells and incubated for 4 hr before treatment with either genistein or β-lapachone alone or in combination. All plates were then incubated at 37°C in a humidified atmosphere of 5% CO 2 in air for a maximum of 72 hr. At 12, 24 and 36 hr of incubation, 100 μl of the supernatant from each well was gently aspirated and replenished with 100 μl of fresh media. The supernatants were stored at -37°C until assayed for lactate dehydrogenase (LDH) enzyme activity. At 36 hr incubation, the cells were harvested by trypsinization with trypsin-EDTA, and processed for post-treatment metabolic activity using cell viability and apoptotic assays as described. A. Cell Viability Assays A1.1 MTS assay MTS assay depends on mitochondrial enzyme reduction of MTS solution to detect and determine cell viability. The MTS cell proliferation assay is a colorimetric method for determining the number of viable cells in proliferation. It is composed of solutions of a tetrazolium compound [3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium, inner salt; MTS] and an electron coupling reagent (phenazine ethosulfate; PES). MTS is bioreduced by the cells into formazan product that is soluble in cell culture medium. Following cell culture as described above, 100 μL of cells were harvested from each treatment group and added to a 96 MTP followed by addition of 20 μl of MTS (2.5 mg/mL: Sigma Chemical Co) stock solution to each well. After 2 hr incubation under standard conditions of 5% CO 2 and 37°C, the purple formazan product (indicative of reduction of MTS) was visible. The absorbance was read on Multiskan biochromatic automated microplate reader (Multiskan, DC) at 490 nm. The signal generated (color intensity) is directly proportional to the number of viable (metabolically active) cells in the wells. Relative cell numbers can therefore be determined based on the optical absorbance (optical density, OD) of the sample. The blank values were subtracted from each well of the treated cells and controls; and the mean and standard error for each treatment (singles and combination) were calculated relative to the control: where A C = absorbance of the control (mean value): A T = absorbance of the treated cells (mean value) A B = absorbance of the blank (mean value) A1.2 Trypan Blue exclusion assay For the Trypan blue exclusion test, cells were treated and cultured as described. They were harvested and Trypan blue dye solution was added to the cell suspensions. Total cell counts and viable cell number (survival rate) were determined by a standard hemocytometer procedure. Live-viable cells were seen as colorless (impermeable to the dye due to intact cell membrane) and dead cells were seen as blue (permeable to dye due to disruption of cell membrane): A2. LDH assay Lactase dehydrogenase activity was measured by a non-radioactive protocol using the LDH cytotox kit (Cat. #. 1644 793: Boehringer-Mannheim GmbG, Bochemica). The previously frozen supernatants were thawed for LDH determination. Briefly, 100 μL/well of each cell-free supernatant was transferred in triplicate into wells in a 96-well microtiter plate (MTP) and 100 μL of LDH-assay reaction mixture (Kit: dye-catalyst mixture) added to each well. After 90 min incubation under standard conditions the absorbance of the color generated was read on Multiskan biochromatic automatic microplate reader at 490 nm. The mean absorbance/optical density (OD) for each treatment group was calculated. The blank values were subtracted from each well and the mean percent treatment-induced cytotoxicity for each cell line and treatment type (single and combination) was calculated as: Where: ABS expt = mean absorbance from the treated cells: ABS low = mean absorbance from controls (untreated cells)(spontaneous release of LDH) ABS hi = mean absorbance from Triton X-100 treated cells (standard/maximum LDH release)(positive control). B. Detection of Treatment-induced Apoptosis Treatment-induced apoptosis was assessed by two independent assays, Annexin V-FITC assay and the DNA Fragmentation (TUNEL) assay. PC3 cells were treated and co-cultured with the test agents as previously described in this study; and the subjected to the apoptosis determination assays as below: B1.1 Annexin V-FITC assay Apoptosis-associated translocation of phosphatidylserine from the inner to the outer leaflet of the plasma membrane in GC27 and K833 cells was assessed with the use of FITC-labeled Annexin V, a calcium-dependent phospholipid-binding protein with a high affinity for phosphatidylserine; using AnnexinV-FITC Staining Kit (Boehringer Mannheim). Briefly, 100 μl aliquots of the previously prepared cell suspensions were centrifuged, and the cell pellets re-suspended in Annexin binding buffer, incubated with AnnexinV-FITC substrate; then cells were smeared onto microscope slides and either evaluated immediately with fluorescence microscope, or smears were fixed with 4% depolymerized paraformaldehyde and stored at -40°C for later examination as previously described [ 35 ]. Percentage of apoptosis in the cells was quantified based on morphological and fluorescence characteristics of apoptotic cells as previously described [ 5 , 35 , 36 ]. All tests were run in triplicates. B1.2 DNA Fragmentation (TUNEL) assay The presence of apoptosis was determined by terminal deoxynucleotidyl transferase (TdT)-mediated dUTP nick end labeling (TUNEL), using the ApopTag R kit (Boehringer Mannheim Co, Indianapolis, IN) as previously described [ 37 ]. The kit reagents detect apoptotic cells in situ by specific end labeling and detection of DNA fragments produced by the apoptotic process. To perform the TUNEL assay, slides of the PBS suspended cells were fixed with 4% paraformaldehyde for 30 minutes. The cells (slides) were then permeabilized with Triton X-100 at 4°C for 2 min; then flooded with TdT enzyme and digoxigenin-dUTP reaction buffer (TUNEL) reagent for 60 min in a humidity chamber at 37°C, washed with distilled water, incubated for 10 minutes with streptavidin-horseradish peroxidase complex. The stained mounted cells were examined at 100×, 200× and 400× magnification of the microscope (Olympus BH-2). Cell death was quantified by counting 150 cells in 5–7 separate fields of view per slide, and noting the percentage of apoptotic cells based on morphological appearance, as previously described [ 5 , 36 ]. C Potential mechanism(s) of action The potential involvement of caspase-3 protease (CPP32) and/or the enzyme NQO1 [NAD(P)H:quinone oxidoreductase] in the molecular pathways of β-lapachone-and/or genistein-induced growth inhibition and apoptosis in PC3 cells were determined, after treatment of the cells as already described. C1.1 Caspase-3 expression/activity in treatment-induced apoptosis In order to determine the potential role of caspase-3 proteases (CPP32) in the common pathways of β-lapachone and genistein-induced growth inhibition and apoptosis, human prostate cancer cell lines were treated as previously described above. The activity of caspase 3 was determined using a the fluorometric substrate DEVD-afc and caspase 3 inhibitor DEVD-fmk according to the protocol of the Caspase Activity Assay kit. Briefly, PC3 cells were treated and incubated as previously described. At 24, 48 and 72 hr cells were scrapped into suspension and centrifuged at 10,000 rpm for 10 min. The pellet was resuspended in 100 μl of lysis buffer and incubated at 4°C for 10 min, followed by centrifugation at 10,000 rpm for 10 min. Fifty μl aliquots of the supernatants were removed and placed in a 96-well microtiter plate (MTP) containing reaction buffer. The DEVD-afc substrate was added and the MTP was incubated at 37°C for 30 min. Activity was monitored with the linear cleavage and release of the afc side chain; and compared with a linear standard curve generated by the controls on the same MTP. C1.2 NQO1 Activity in treatment-induced apoptosis In order to determine the potential role of enzyme NQO1 [NAD(P)H:quinone oxidoreductase] in the molecular pathways of β-lapachone-and genistein-induced growth inhibition and apoptosis in human prostate cancer, PC3 cell lines were treated as previously described. Dicoumarol (3-3'-methylene-bis (4-hydroxycou-marin) is a commonly used inhibitor of NQO1, which competes with NADH or NADPH for binding to the oxidized form of NQO1. Dicoumarol thereby prevents reduction and activation of various target quinines like β-lapachone. The cells, cultured as previously described, were treated concomitantly in single and combination treatments of varying concentrations β-lapachone (bLap), genistein (gen), and bLap-Gn combination with and without 50 μM dicoumarol as previously described. The treated cells were harvested and tested for treatment-induced apoptosis by the methods previously described in this study. Authors' contributions JKD contributed 50% in all aspects of the research. All other authors contributed equally-50% All authors read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC516040.xml |
549544 | Psychometric properties and clinical utility of the Scale for Suicidal Ideation (SSI) in adolescents | Background Accurate assessment of suicidality is of major importance in both clinical and research settings. The Scale for Suicidal Ideation (SSI) is a well-established clinician-rating scale but its suitability to adolescents has not been studied. The aim of this study was to evaluate the reliability and validity, and to test an appropriate cutoff threshold for the SSI in a depressed adolescent outpatient population and controls. Methods 218 adolescent psychiatric outpatient clinic patients suffering from depressive disorders and 200 age- and sex-matched school-attending controls were evaluated by the SSI for presence and severity of suicidal ideation. Internal consistency, discriminative-, concurrent-, and construct validity as well as the screening properties of the SSI were evaluated. Results Cronbach's α for the whole SSI was 0.95. The SSI total score differentiated patients and controls, and increased statistically significantly in classes with increasing severity of suicidality derived from the suicidality items of the K-SADS-PL diagnostic interview. Varimax-rotated principal component analysis of the SSI items yielded three theoretically coherent factors suggesting construct validity. Area under the receiver operating characteristic (ROC) curve was 0.84 for the whole sample and 0.80 for the patient sample. The optimal cutoff threshold for the SSI total score was 3/4 yielding sensitivity of 75% and specificity of 88.9% in this population. Conclusions SSI appears to be a reliable and a valid measure of suicidal ideation for depressed adolescents. | Background Accurate assessment of suicidality is of major importance in both clinical and research settings. Adolescent suicide occurs usually in the context of an active, often treatable, but unrecognized or untreated mental illness [ 1 - 3 ]. The increase in the antidepressant treatment of adolescents in the USA [ 4 ] may partly explain the decline in the incidence of youthful suicide [ 5 ], though recently some reports have connected SSRI-treatment in adolescents to an increase in suicidality [ 6 ]. Suicide attempts are complex acts for which no single set of clinical features can be expected to be a good predictor [ 7 ]. Suicidal ideation, self-harming, suicide attempts and completed suicides are different forms of suicidality. Although the domain of suicidal behavior probably is multidimensional [ 8 ], a continuum from suicide ideation to suicide attempts has been reported in youthful clinical populations [ 9 , 10 ]. Thus, although most patients with suicidal ideation do not attempt suicide, identification and assessment of severity of suicidal ideation is of major importance. The Scale for Suicidal Ideation (SSI) [ 11 ] was designed to measure the intensity, pervasiveness, and characteristics of suicidal ideation in adults. It also aims to assess the risk of later suicide attempt in individuals who have thoughts, plans, and wishes to commit suicide [ 12 ]. It is a well-established clinician-rating scale and is presented in a semi-structured interview format. The psychometric properties of the SSI have been evaluated in adult population and in inpatient children. Both in a sample of adult psychiatric inpatients and in a sample of inpatient children the internal consistency of the scale was good [ 11 , 13 ]. SSI reportedly has three dimensions [ 11 ], which have been only partly replicated in some factor analytical studies [e.g. [ 13 , 14 ]]. SSI has been found to converge with scales measuring related constructs e.g. hopelessness and depression in adults, and hopelessness, depression and self-harm in children [ 11 , 13 ]. The predictive validity of the SSI has been studied in a sample of hospitalised patients, where the SSI scores of those who committed suicide were not significantly higher than the scores of inpatients that did not [ 15 ]. In a sample of 3701 adult outpatients those who scored over a SSI threshold value had 5.42 times higher odds of committing suicide than those who scored under [ 16 ]. The threshold value was derived from a receiver operating characteristic (ROC) analysis that yielded optimal threshold of 1/2 for predicting future suicide. In the same study, SSI-scores inquiring the worst point in life (SSI-W) yielded an odds ratio of 13.84 for predicting suicide. A recent study that inquired retrospectively records of suicide victims to find communications that fit the SSI-items found no suicide-predicting power for the instrument [ 17 ]. Some instruments have evolved from the SSI, for example the Modified scale for suicidal ideation (MSSI) [ 18 ] that was designed to suit paraprofessionals and the Beck scale for suicidal ideation (BSS) [ 19 ] that is a self-report scale. The SSI has been used widely in adult psychiatric populations [e.g. [ 20 , 21 ]], but its psychometric properties have not been evaluated in adolescents. According to a recent comprehensive review "despite its potential utility, the SSI's suitability to adolescents... remains to be elucidated" [ 22 ]. Rating scales should be validated in each patient population in which they are used. The aim of this study was to evaluate the reliability and validity of the SSI and test an appropriate cutoff threshold for clinically significant suicidal ideation in an adolescent population. Methods Sample The study population consisted of two samples; a psychiatric outpatient sample of 218, and an age- and sex-matched control sample of 200 school-attending adolescents. The outpatients suffered from depressive mood disorder, were of ages 13 through 19, and took part in the Adolescent Depression Study (ADS). They were recruited between 1.2.1998 and 31.12.2001 from a consecutive sample of patients attending the outpatient clinics of the Department of Adolescent Psychiatry of Peijas Medical Health Care District covering approximately 210,000 inhabitants and comprising the cities of Vantaa and Kerava in the Helsinki metropolitan area, southern Finland. Of the eligible (appropriate age, knowledge of Finnish language and adequate cognitive capacity) 660 outpatients, 624 (94.5%) were screened during their first consultation visit by the Beck Depression Inventory (BDI) [ 23 ] and the General Health Questionnaire-36 (GHQ-36) [ 24 , 25 ]. Those 373 (59.8%) with scores of 10 or more and 5 or more, respectively, were considered screen positives, and were asked to participate in the study. 118 (31.6%) outpatients refused and 34 (9.1%) dropped out at this stage. 221 (33.5%) remaining outpatients were evaluated by a diagnostic interview (K-SADS-PL) [ 26 ] and those 218 (33.0%) with a current depressive mood disorder were included in the study. The control sample was drawn from the enrollment lists in four schools in the corresponding geographical area. It was a random sample of age- and sex-matched students equating the distribution of the educational level of the outpatients. Instruments 1) The Scale for Suicide Ideation (SSI) is a clinician-rating scale and is presented in a semi-structured interview format [ 11 ]. It consists of 19 items that evaluate three dimensions of suicide ideation: active suicidal desire, specific plans for suicide, and passive suicidal desire. Each item is rated on a 3-point scale from 0 to 2. The higher the total score, the greater the severity of suicide ideation. In some previous studies on adult suicidality a score of 6 or more has been used as a cutoff threshold for clinically significant suicidal ideation [e.g. [ 20 ]]. The psychometric properties of the SSI have been evaluated for adult psychiatric patient population; the internal consistency of the scale was found to be good (α = 0.89), and factor analysis yielded the three above-mentioned dimensions [ 11 ]. Among inpatient children rated by trained raters the factors could not be replicated; only two factors ("active suicidal desire" and a mixture of "active and passive desire") existed with miscellaneous items left over [ 13 ]. Nine trained raters did the SSI rating in our study. 2) The Schedule for Affective Disorders and Schizophrenia for School-Aged Children-Present and Lifetime (K-SADS-PL) [ 26 ] is a widely used semi-structured diagnostic interview. Suicidal behavior was determined using four questions from the screening-section of the K-SADS-PL diagnostic interview: item-1 suicidal thoughts ("1" = none, "2" = occasional, "3" = frequent), item-2 suicide attempts and their seriousness ("1" = none, "2" = ambivalent, "3" = serious) and item-3 suicide attempts and their lethality ("1" = none, "2" = not life-threatening, "3" = life-threatening). Self-harming behavior was asked using item-4, the question on deliberate self-harm without intent to die ("1" = none, "2" = occasional, "3" = frequent) in the screening section of the K-SADS-PL. The K-SADS-PL is considered internationally reliable and valid diagnostic instrument for adolescent population [ 27 ]. It has been translated (and back translated) into Finnish and used widely in studies concerning suicidality [e.g. [ 9 , 28 ]]. Nine trained raters did the rating. Inter-rater reliability, assessed using 15 randomly selected videotaped interviews, was good for mood disorder diagnoses [weighted kappa [ 29 ] for MDD, other mood disorder, no mood disorder 0.87 (95 % CI 0.81, 0.93)]. 3) Clinical suicidality assessment (CSA): A three-point mutually exclusive grouping of suicidality (1-non-suicidal, 2-suicide ideation, 3-suicide attempts) is a simplified version of the 5-item "Spectrum of Suicidal Behavior Scale" [ 30 ]. It has been used in both research and clinical purposes [e.g. [ 10 ]]. The grouping is done by a clinician, and is based on two simple questions "Have you thought of killing yourself?" and "Have you attempted suicide?" and on patient records when appropriate. There is some evidence supporting the predictive validity of this grouping [ 10 ] but it has not been validated by comparing it with more structured measures like the K-SADS-PL. In this study, after a brief training the treating clinicians of the outpatient clinic did the CSA. They were instructed to include in class-3 also self-mutilation and other self-harming behavior with no explicit suicide intent. Procedure After a description of the study, a written informed consent was obtained from the subjects. For subjects less than 18 years consent was also asked from the parents or other legal guardians. For the community sample the K-SADS-PL and the SSI were performed at the same day by an expert clinician. For the outpatient sample the K-SADS-PL was performed within variable time from the SSI rating. The CSA was performed for the patient sample by clinicians during the beginning of the treatment. Statistical analysis Central tendencies of some data were reported using medians and quartiles because of non-normal distribution. Mann-Whitney U test was used to assess the significance of differences between the two samples. Internal consistency of the SSI was evaluated by calculation of Cronbach's α for the whole scale. Concurrent validity of the instrument was examined by comparing it with the K-SADS-PL with the CSA classifications. SSI total scores were first assessed in 5 classes of increasing suicidality derived from the K-SADS-PL responses in the following way: 1-no suicidal ideation or acts, 2-mild suicidal ideation (score 2 on item-1), 3-severe suicidal ideation (score 3 on item-1), 4-mild suicidal acts (score 2 on any of items 2–4 regardless of ideation), 5-severe suicidal acts (score of 3 on any of items 2–4 regardless of ideation). Then the SSI total scores were measured in 3 classes of increasing suicidal ideation severity, regardless of possible suicidal acts, derived from the K-SADS-PL responses on item 1: 1-no ideation, 2-mild ideation, and 3-severe ideation. Severe ideation (score 3) in this item was considered as "clinically significant suicidal ideation". Finally the SSI total score was assessed in the three classes of the CSA: 1-no suicidality, 2-suicidal ideation, 3-suicidal or self-harming acts. The statistical significance of the between-class differences was evaluated by Kruskal-Wallis test. For the analyses of concurrent validity only SSI-measurements in a range of 30 days from the K-SADS-PL and the CSA were included. Construct validity was measured by performing a principal component analysis (PCA) with varimax rotation in the outpatient sample. The internal consistencies (Cronbach's α) of the extracted components as well as the originally reported factors [ 11 ] were calculated. ROC-analysis was performed to evaluate the screening properties of the SSI, and the cutoff threshold for the instrument was defined by optimal trade-off between sensitivity and specificity (Youden's index [ 31 ]). The K-SADS-PL (score 3 in item-1) was used as the standard to define cases with clinically significant suicidal ideation. SPSS 11.0 (Chicago, Illinois 60606, SPSS Inc) was used for the statistical analysis. Results Eighteen percent (n = 40) of the outpatient sample were boys and 82% (n = 178) girls, in the community sample the percentages were 18.6% (n = 37) and 81.4% (n = 162), respectively. The subjects' mean age was 16.4 (SD 1.6) in the outpatient sample and 16.5 (SD 1.6) in the community sample. The median SSI total score for the patient sample was 0 (Q1–3 = 0–6) and for the community sample 0 (Q1–3 = 0-0) (z = -9.6, p = 0.000). The median SSI total score for subjects aged 13–15 was 0 (Q1–3 = 0–1) and those aged 16–19 0 (Q1–3 = 0–1) (z = -0.685, p = 0.493). The median time distance between SSI and K-SADS-PL was 21.5 days (Q1–3 = 9–36) for the patient sample and 0 days (Q1–3 = 0-0) for the control sample (z = -18.0, p = 0.000). The median time distance between SSI and the CSA was 6 days (range 0–35). Forty-seven (21.6%) outpatients and one (0,5%) control subject had current clinically significant suicidal ideation (p = 0.000) according to the K-SADS-PL. Reliability Cronbach's α was 0.95 for the whole sample, 0.81 for the community sample and 0.95 for the outpatient sample. Concurrent validity 146 (67%) of the outpatients and 199 (99.5%) of the controls were included in the analyses for concurrent validity, as their measurements were within the required range of 30 days. The median SSI sum scores in the five suicidality classes derived from the K-SADS-PL were class-1 = 0 (Q1–3 = 0-0); class-2 = 5.5 (Q1–3 = 0–8); class-3 = 13 (Q1–3 = 0–18.5); class-4 = 4 (Q1–3 = 0–17.3); class-5 = 8 (Q1–3 = 0–13). The differences were significant (χ 2 = 111.6, df 4, p = 0.000). The median SSI sum scores in the three classes of suicidal ideation derived from the K-SADS-PL were class-1 = 0 (Q1–3 = 0-0); class-2 = 4 (Q1–3 = 0–8); class-3 = 13 (Q1–3 = 4–18). The differences were significant (χ 2 = 132.6, df 2, p = 0.000). The median SSI sum scores in the three clinical suicidality evaluation classes (only the outpatient sample) were class-1 = 0 (Q1–3 = 0–1); class-2 = 10 (Q1–3 = 5–18); and class.3 = 15 (Q1–3 = 13.3–16.6). The differences were significant (χ 2 = 57.9, df 2, p = 0.000). Construct validity Principal Component analysis could be performed only for the outpatient sample due to a small variance of responses in the community sample. The analysis yielded a strong first unrotated factor, which explained 53% of the variance, and two more factors with eigen value > 1. The three factors and their internal consistencies after varimax rotation are presented in Table 1 . The internal consistencies (Cronbach's α) of the originally reported [ 11 ] three dimensions were "active suicidal desire" α = 0.92, "preparation" α = 0.69, "passive suicidal desire" α = 0.79. Table 1 Factor loadings and internal consistencies of the varimax rotated Principal Component Analysis (PCA) of the SSI in an outpatient sample of 218 adolescent outpatients with mood disorder. (* = Items that loaded identically to Beck's [11] original study) In the original study items 8, 10, 11 loaded on "active suicidal desire"-factor; items 13 and 15 on "preparation"-factor; and items 14 and 18 on "passive suicidal desire"-factor; item 17 did not load adequately on any of the factors. Item Loadings Factor 1: Factor 2: Factor 3: (active suicidal desire) (passive suicidal desire) (preparation) 7. time dimension: frequency 0.824 * 0.208 0.180 6. time dimension: duration 0.764 * 0.321 0.225 4. desire to make active suicide attempt 0.753 * 0.389 0.142 9. control over suicidal action 0.746 * 0.140 0.140 1. wish to live 0.716 * 0.170 0.063 12. method: specificity/planning 0.714 * 0.383 0.303 2. wish to die 0.702 * 0.360 0.137 3. reasons for living/dying 0.689 * 0.364 0.219 14. sence of "capability" 0.657 0.405 0.322 13. method: availability/opportunity 0.649 0.409 0.285 α = 0.94 5. passive suicidal desire 0.256 0.720 * 0.016 19. deception/concealment of suicide 0.196 0.711 * 0.201 8. attitude toward ideation/wish 0.508 0.650 0.177 10. deterrents to active attempt 0.242 0.633 0.389 11. reason for contemplated attempt 0.527 0.619 0.058 15. expectancy/anticipation of event 0.445 0.603 0.226 α = 0.85 18. final acts 0.098 0.151 0.802 17. suicide note 0.358 0.001 0.787 16. actual preparation 0.133 0.319 0.646 * α = 0.65 Validity as a screening instrument ROC analysis (Fig. 1 ) of the SSI total score against the K-SADS-PL-confirmed suicidal ideation yielded an area-under-curve (AUC) of 0.84 for the whole sample (n = 418) and an AUC of 0.80 for the patient sample (n = 218). The optimal trade-off between sensitivity and specificity (Youden's index) was achieved at a cutoff threshold score of four or more in the whole sample as well as the patient sample. In the whole sample the sensitivity and the specificity at this threshold were 75% and 88.9%, respectively (Table 2 ) with 53 subjects classified incorrectly. In the patient sample the sensitivity and the specificity at this optimal threshold were 76.6% and 77.2%, respectively (Table 3 ) with 50 subjects classified incorrectly. Figure 1 Detection of suicidal ideation by the Scale for suicidal ideation (SSI) against the K-SADS-PL as a standard, at a sample of 146 depressed adolescent outpatients and 199 age- and sex-matched controls. ROC-curve with a reference line. Table 2 Validity coefficients of different SSI cutoffs against K-SADS-PL diagnosed significant suicidal ideation at a mixed adolescent sample of 146 outpatients and 199 community controls SSI cutoff 0–1 1–2 2–3 3–4 4–5 5–6 6–7 7–8 8–9 9–10 10–11 Sensitivity 77.1% 75% 75% 75% 66.7% 64.6% 58.3% 58.3% 58.3% 56.3% 50.0% Specificity 83.0% 86.5% 87.6% 88.9% 90.0% 91.6% 93.0% 94.6% 95.9% 96.2% 96.8% Youden 0.60 0.62 0.63 0.64 0.57 0.57 0.51 0.53 0.54 0.53 0.47 Table 3 Validity coefficients of different SSI cutoffs against K-SADS-PL diagnosed significant suicidal ideation at an adolescent sample of 146 outpatients SSI cutoff 0–1 1–2 2–3 3–4 4–5 5–6 6–7 7–8 8–9 9–10 10–11 Sensitivity 78.7% 76.6% 76.6% 76.6% 68.1% 66.0% 59.6% 59.6% 59.6% 57.4% 51.1% Specificity 66.7% 73.1% 74.3% 77.2% 78.9% 82.5% 85.4% 88.9% 91.2% 91.8% 93.0% Youden 0.46 0.50 0.51 0.54 0.47 0.49 0.45 0.49 0.51 0.49 0.44 Discussion This study was the first to evaluate the psychometric properties of the SSI in an adolescent population. It was a part of the ongoing Adolescent Depression Study (ADS) and the sample of patients was large compared to earlier similar studies in adult populations, and probably representative of adolescent psychiatric outpatients with depressive disorders. The main finding was that the SSI appeared to be a reliable and valid instrument for evaluation of suicidal ideation in a depressed adolescent population. Its internal consistency and different aspects of validity were good and similar to what has been reported among adults. The construct validity of the SSI was checked by Principal Component Analysis, which yielded 3 theoretically meaningful and coherent factors, only slightly different from the original ones, with good internal consistencies. This suggests good construct validity. The first factor ("active suicidal desire") was nearly identical to Beck's original one [ 11 ]. The second factor ("passive suicidal desire") included theoretically coherent items, two of which were identical to Beck's original factor of similar content. The third factor was also theoretically meaningful, included three items concerning final preparations, and had one item in common with Beck's original "preparations" factor. The SSI converged theoretically meaningfully with both the three-class K-SADS-PL suicidal ideation-item and the clinical suicidality assessment (CSA); growing SSI scores were found within categories with increasing severity of suicidality. As to the convergence with the 5-class K-SADS-PL suicidality instrument, the results were more complex. The Kruskal-Wallis test yielded significant differences between the SSI scores in the different categories as expected, but the SSI-scores in the K-SADS-PL classes 4 and 5, with the supposedly most severe suicidality were not higher than in class 3. Classes 4 and 5 inquire about suicidal acts, and may represent a partly separate domain from suicidal ideation, which may be related to the presence of comorbid personality traits or conduct disorders. The SSI was designed to tap suicidal ideation and it may not satisfactorily tap features related with suicidal acts. In accordance with the theory of multidimensional nature of suicidality [ 8 ], severe suicidal ideation may not always be a prerequisite for suicidal acts in adolescents. The authors are not aware of previous empirical estimations of clinically relevant cutoff for the SSI in adolescents. In this study ROC analysis of the whole sample yielded a reasonable result, but the validity coefficients for the different cutoffs of the SSI were somewhat difficult to interpret. In the community sample, there was only one subject with K-SADS-PL-diagnosed clinically significant suicidal ideation, which may have biased the analyses made with the whole sample. The results for both the whole sample and the patient sample suggest that a cutoff threshold score of four or more might be optimal for adolescents. Depending on the purpose the SSI is used, however, the emphasis between sensitivity and specificity may change, and a different threshold may be useful. For example, if the purpose is to detect the maximum number of potential suicides the cutoff threshold should be lowered to minimize the number of false negatives. Limitations Several methodological limitations should be noted, some suggesting caution in interpreting the findings. Inter-rater and test-retest reliabilities, which would have given a complete picture of the reliability of the SSI, could not be evaluated in our setting; they would have required repeated SSI measurements for each subject. However, the alpha-coefficients are a marker of internal consistency, which is one indicator of reliability. Although large and representative, the sample was a pure outpatient sample with age- and sex-matched controls, and females were over-represented. The absence of inpatients may have caused us to see the spectrum of suicidal ideation narrower than in real clinical situations. The sample was limited to an urban area in southern Finland, the generalizability of our findings to rural areas, or to other countries, is not known. The use of K-SADS-PL as a standard for clinically significant suicidal ideation and behavior may be criticized, as the authors are not aware of a data on its validity. It is used, however, as one of the best available standards in adolescent mood disorder diagnostics, and taps suicidality with 4 relevant items. The same rater rated the K-SADS-PL and the SSI, which is a weaker test of concurrent validity than correlating measures rated by separate raters. Clinical implications The SSI can safely be used to evaluate suicidal ideation in adolescents where it seems to perform as well as in adults, where it is considered to be well established. When screening clinically significant suicidality in adolescents, a total score threshold of 3/4 may be useful. Suicidal acts may occur among adolescents with only "mild" suicidal ideation. Thus, prevention of suicidal acts cannot rely solely on the SSI, which does not seem to tap them accurately. Furthermore, questionnaires should be only an adjunct to the clinical evaluation of suicidality. Conclusions SSI appears to be a reliable and a valid measure of suicidal ideation at depressed adolescents, with a cutoff threshold value of four or more of total SSI score being an appropriate for detecting significant suicidal ideation. Competing interests The author(s) declare that they have no competing interests. Authors' contributions MMH analyzed the data and wrote the paper. MP interviewed patients, participated in planning the study and analyses, and writing the paper. LK, TR, HH and VT participated in planning the study, interviewed patients, and commented on the manuscript. OK participated in planning the study and the analyses, and commented on the manuscript. MM supervised the study, interviewed patients and participated in planning of the study and analyses, and writing the paper. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549544.xml |
544935 | Secondary neurons are arrested in an immature state by formation of epithelial vesicles during neurogenesis of the spider Cupiennius salei | Background In the spider Cupiennius salei about 30 groups of neural precursors are generated per hemi-segment during early neurogenesis. Analysis of the ventral neuromeres after invagination of the primary neural precursor groups revealed that secondary neural precursors arise during late embryogenesis that partially do not differentiate until larval stages. Results In contrast to the primary groups, the secondary invaginating cells do not detach from each other after invagination but maintain their epithelial character and form so-called epithelial vesicles. As revealed by dye labeling, secondary neural precursors within epithelial vesicles do not show any morphological features of differentiation indicating that the formation of epithelial vesicles after invagination leads to a delay in the differentiation of the corresponding neural precursors. About half of the secondary neural precursor groups do not dissociate from each other during embryogenesis indicating that they provide neural precursors for larval and adult stages. Conclusions Secondary neural precursors are arrested in an immature state by formation of epithelial vesicles. This mechanism facilitates the production of larval neural precursors during embryogenesis. I discuss the evolutionary changes that have occured during neural precursor formation in the arthropod group and present a model for the basal mode of neurogenesis. | Background The arthropods form a diverse group with a correspondingly high variation of neural structures adapted to the specialized behaviour and lifestyles of individual species. This raises the question of how developmental processes have been modified during evolution to generate the wide diversity of nervous systems seen in adult arthropods. Evolutionary modifications that lead to variations in neural structures can occur during different processes of neurogenesis. The establishment of neural networks can be influenced by changes in the generation of neural precursors, modifications of cell fates or elimination of individual neurons as well as changes in axonal guidance. A comparative analysis of neurogenesis in chelicerates and myriapods has revealed that although the developmental program is genetically conserved, there is a major difference in the recruitment of neural precursors as compared to insects and crustaceans [ 1 - 5 ]. Groups of neural precursors invaginate from the ventral neuroectoderm in a regular, strikingly similar pattern in spiders (chelicerates) and myriapods, while in insects and crustaceans single neural precursors are selected. This modification may be the basis for variations in the functions of spider and myriapod neurons, since a comparison of early segmentally repeated neurons that pioneer the major axon tracts in crustaceans and insects has not revealed any similarities in cell body positions or axonal outgrowths to myriapod neurons [ 6 , 7 ]. In the spider 30 to 32 groups of neural precursors are generated per hemi-segment during neurogenesis. As in Drosophila melanogaster , the neural precursors arise at stereotyped positions that are prefigured by a proneural gene ( CsASH 1 ), while the neurogenic genes Delta and Notch restrict the proportion of cells that adopt the neural fate at each wave of neural precursor formation [ 1 , 2 ]. In Drosophila melanogaster , the Delta/Notch signalling pathway is used for a decision between two cell fates in the ventral neuroectoderm: delaminating cells become neural precursors, while cells that remain apical give rise to epidermis. This decision does not take place in the central neurogenic regions of the spider [ 2 ]. The epidermal cells are derived from lateral regions that overgrow the neuromeres after invagination of the neural precursors. Since each invagination group consists of five to nine neural precursors, it can be estimated that an embryonic hemineuromere consists of about 220 neurons on average, similar to Drosophila . However, in the adult spider Cupiennius salei the subesophageal ganglion consists of 49,000 neurons [ 8 ] indicating that over 40,000 neurons must be generated during late embryonic and larval stages. In Drosophila melanogaster , 'embryonic' neuroblasts proliferate again and give rise to larval and adult lineages after a phase of cell cycle arrest from late embryogenesis to first larval instar [ 9 - 11 ]. An analysis of the mitotic pattern during neurogenesis has revealed that neuroblasts are missing in the spider [ 1 ]. In addition, most of the neural precursors do not divide after invagination. This raises the question of how additional neurons are generated that contribute to the larval and adult CNS of the spider. Results In the spider Cupiennius salei the germband develops from aggregations of cells that form the cephalic lobe and the caudal lobe [ 12 ]. One to three prosomal segments are generated by a subdivision of the cephalic lobe, while the remaining segments arise sequentially from the caudal lobe, the so-called posterior growth zone [ 12 , 13 ]. At the beginning of neurogenesis (about 130 hours of development; stages after Seitz [ 12 ]) a longitudinal furrow forms that divides the germband into left and right parts that remain connected only at the cephalic lobe and the posterior growth zone. The two halves of the embryo move laterally until they finally meet at the dorsal midline (ca. 300 hours of developement). This process is called inversion [ 12 ]. The formation of neural precursors and the invagination of these cells occurs during inversion [ 1 ]. Secondary invagination sites form after invagination of the primary neural precursors Although the invagination sites in the ventral neuroectoderm of the spider are generated in four subsequent waves over a time period of three days, the neural precursors detach form the apical surface at about the same time between 200 and 230 hours of development. Fig. 1A shows the final arrangement of the invagination sites shortly before the neural precursors loose contact to the apical surface. After invagination, the neural precursors differentiate and a neuropil develops at the basal side of the neuromeres (Fig. 1B , arrow). Since the epidermis arises lateral and medial to the ventral hemi-neuromeres (Fig. 1C , arrow), all cells of the central neurogenic region are eventually incoporated into the ganglion, i.e., the ventral neuroectoderm does not give rise to epidermoblasts and neuroblasts as in Drosophila melanogaster (see above). A detailed analysis of the morphology of the ventral neuromeres after invagination of the neural precursor groups revealed that the cells that remain apical form secondary invagination sites (Fig. 1B , asterisks). Figure 1 (A-E): Morphology of the secondary invagination sites. Confocal micrograph (A, inverted) of a flat preparation of an embryo stained with phalloidin-rhodamine and light micrographs (B-D) and electron micrograph (E) of transverse sections through prosomal hemi-neuromeres. The midline is to the right. (A) Final pattern of the primary invagination sites in the opisthosomal segments 1 and 2. The invagination sites are arranged in 7 rows. The black dots correspond to the constricted cell processes of the individual precursor groups that are attached to the apical surface (arrow). (B) Morphology of the secondary invagination sites. At 250 hours the secondary invaginating cell groups (asterisks) are still attached to the apical surface. The individual groups are isolated by brighter sheath cells (arrowhead). The primary precursor groups have dissociated (arrow) and form basal cell layers. The longitudinal connective (lc) is already visible at the basal side. (C) The secondary invagination sites (asterisks) loose contact to the apical surface, when the epidermis (arrow) overgrows the ventral neuromeres. (D) After invagination the secondary neural precursors (asterisks) remain attached to each other forming epithelial vesicles. The cell processes run parallel to each other and extend to a lumen (arrow). (E) The cell processes (o) of the invaginating cells of a group are opposed to each other and the lumen between the cell processes is filled with microvilli (arrow). Cell junctions connect the individual processes (arrowheads). lc , longitudinal connective; o2 to o3 , opisthosomal hemi-segments 2 to 3. The secondary invagination sites can be distinguished from the primary invagination groups by several morphological features. (1) Each secondary invagination group contains up to 40 cells as compared to 5 to 8 cells that form the primary invagination sites (Stollewerk et al., 2001; Fig. 1B,1D ; see also Fig. 4F ). (2) The cell processes of the secondary neural precursors do not extend straight to the apical surface as the primary invaginating cells, but face each other (Fig. 1D,1E ). Microvilli extend into the lumen between the opposite cell processes (Fig. 1E , arrow). (3) While the primary neural precursors detach from each other after invagination, the secondary invaginating cell groups remain attached to each other and maintain their epithelial character. (Fig. 1D,1E ) The individual cell processes are connected by cell junctions (Fig. 1E , arrow heads). (4) In contrast to the primary precursors, the secondary invaginating cell groups are surrounded by sheath cells. These cells are visible in the light and electron microscope as brighter cells that separate the individual invagination sites (Fig. 1B , arrowhead; Fig. 2 , asterisks; Fig. 3B ). They extend long cell processes that ensheath each cell group (Fig. 2A , arrowhead). Interestingly, the sheath cells that are located in the apical cell layer form bizarre cytoplasmic shapes that extend into the cell-free space at the ventral side of the embryo (Fig. 2C , arrow). Double-stainings with a marker for cell nuclei and a dye that stains the actin cytoskeleton show that the nuclei of the secondary neural precursors are shifted basally similar to the primary invagination sites (Fig. 3A,A' ; see also Fig. 1B,1D ). The stained nuclei that surround the individual secondary invagination sites in the apical cell layer correspond to the sheath cells (Fig. 3A,A' ). Figure 2 (A-C): Secondary invagination sites are surrounded by sheath cells. Electron micrographs of transverse sections through prosomal hemi-neuromeres. (A,B) Invagination sites (arrows) are surrounded by sheath cells (asterisks) that appear translucent in the electron microscope. The sheath cells extend processes (arrowhead) that enwrap the individual invagination sites. (C) Sheath cells that are located in the apical cell layer form bizarre shapes that extend into the cell free space at the ventral side of the embryo (arrowhead). The sheath cells are labeled with asterisks, the arrow points to an invagination site. Figure 3 (A-B): The nuclei of cells within the secondary invagination sites are located basally. Confocal micrographs of flat preparations of embryos double-stained with phalloidin-rhodamine (red) and YOYO (green) (A,A') and single stained with phalloidin-rhodamine (B). (A,A') The apical optical section at 250 hours of development shows that the secondary invagination sites (arrow) are still attached to the apical surface. The nuclei of the secondary precursors are located basally, as revealed by the absence of nuclei staining in the apical cell layer. The asterisks in A' indicate the positions of the cell processes of the secondary invagination sites (compare to A). (B) The basal optical section shows the distinct morphology of the sheath cells (arrows) that subdivide the individual invagination sites. Secondary invagination sites persist as epithelial vesicles In contrast to the primary invaginating cell groups that are generated in four waves (see above), the secondary invagination sites appear almost at the same time (Fig. 4A ). A detailed analysis of the ventral neuromeres of embryos stained with phalloidin-rhodamine, a dye that stains the actin cytoskeleton and accumulates in the constricted cell processes of the invaginating cells, revealed, that about 25 invagination sites are generated per hemi-segment. There is no clear dividing line between the formation of secondary invagination sites and invagination of the primary neural precurors. At 220 hours of development all secondary invagination sites are visible (compare Fig. 4A and 4B ), while some of the primary precursor groups are still attached to the apical surface (Fig. 4A , arrowhead). However, at 240 hours all primary precursors have detached from the apical surface and dissociated (Fig. 4B ; see also Fig. 1B ). At about 250 hours, epidermal cells arise lateral and medial to the ventral neuromeres and overgrow the ventral nerve cord within 50 hours [ 2 ]. In Fig. 4C (280 hours of development) the border of the overgrowing epidermis is visible as a circle in the medial region of each hemi-neuromere. Although the secondary invaginating cell groups detach from the apical surface at this time, the individual cells of a group remain attached to each other and persist as epithelial vesicles (Fig. 4D ; see also Fig. 1C ). Due to morphogenetic movements at about 300 hours of development, the anterior-posterior extension of the individual hemi-segments is reduced leading to a rearrangement of the position of the epithelial vesicles (Fig. 4E,4F,4G,4H ). After 350 hours 8 of the 25 invaginated cell groups are no longer visible indicating that the cells have detached from each other (Fig. 4F,4H ). However, 10 cell groups are still visible at hatching (Fig. 4G,4I ). Similar to the ventral neuromeres, groups of cells invaginate from the cephalic lobe neuroectoderm and persist as epithelial vesicles until larval stages (Fig. 4J ). DiI labelling of cells within epithelial vesicles revealed that the cells of a group are attached to each other (Fig. 5A,5B,5C ) and their short, thin cell processes run parallel to each other (Fig. 5B , arrow). They do not show any morphological features of differentiation, i.e. they do not grow long thin dendritic or axonal processes. These data show that 10 groups of neural precursors per hemi-segment do not differentiate during embryogenesis but give rise to neural cells that will be incorporated into the larval ganglia. Figure 4 (A-j): Invagination of secondary neural precursors and formation of epithelial vesicles. (A-F) Confocal micrographs of flat preparations of embryos stained with phalloidin-rhodamine. (B-G) Flat preparations of the fourth prosomal hemi-segments. (A) At 220 hours about 25 secondary invagination sites form (arrow). There is no clear dividing line between the formation of secondary invagination sites (arrow) and invagination of primary neural precursors. Some primary invagination sites are still visible (arrowhead) The bars indicate the segment borders. (B) Apical optical section of the pattern of secondary invagination sites (arrow) at 240 hours of development. (C) Epidermal cells overgrow the ventral neuromeres between 250 and 300 hours (arrowheads) The arrow points to a secondary invagination group. (D) After invagination the individual cells of a groups remain attached to each other forming epithelial vesicles (arrow). (E) At 300 hours the anterior-posterior extension of the individual hemi-segments has been reduced leading to a rearrangement in the positions of the invaginated cell groups (arrow). (F) After 320 hours 8 of the 25 invaginated cell groups are no longer visible indicating that the cells have detached from each other. The arrow points to an invaginated cell group. (G) 10 cell groups are still visible at hatching (arrow). (H) Overview of the arrangement of epithelial vesicles (arrow) of the four prosomal hemi-segments corresponding to the four walking legs. The anterior-posterior reduction in size is clearly visible (compare to A). The bars indicate the segment borders. (I) Flat preparation of the prosoma at hatching. Epithelial vesicles are still visible (arrow). The bars indicate the segment borders, the arrowhead points to the midline. (J) Flat preparation of the brain at 350 hours. The arrow points to epithelial vesicles. ch , chelicera; l1 to l4 , prosomal neuromeres corresponding to walking leg 1 to 4; leg 1 , walking leg 1. p , pedipalp; ped , pedipalpal neuromere. Figure 5 (A-C): DiI-labeling of cells within epithelial vesicles. Flat preparation of the fourth prosomal hemi-neuromere of an embryo labeled with DiI (red) and stained with phalloidin-rhodamine (green). (A-C) Invaginated cells in 40 segments of 10 embryos were labelled with DiI (red) and stained with phalloidin-FITC (green). The cells of a group (A, asterisks) are attached to each other (B,C large arrow head) and their short, thin cell processes run parallel to each other (B,C arrow). They do not show any morphological features of differentiation, i.e. they do not grow long thin dendritic or axonal processes. The small arrows (B,C) point to a cell of an adjacent invagination group. Dissociation of epithelial vesicles is not associated with cell divisions Analysis of the mitotic pattern during neurogenesis has revealed that the formation of the primary invagination sites is not connected with cell divisions [ 1 ]. In addition, mitotic activity seems to be restricted to the apical layer of the ventral neuroectoderm with the exception of a few cells indiciating that most of the invaginated primary neural precursors differentiate without further divisions. However, there are two waves of mitosis during the course of neurogenesis [ 1 ]. After formation of most of the primary invagination sites many cell divisions can be observed in groups of cells and single cells in the apical cell layer. The second wave arises when the primary precursors detach from the apical surface and the secondary invagination sites begin to form. Cell divisions are also restricted to the apical cell layer with the exception of a few cells [ 1 ]. These data indicate that the number of neuroectodermal cells is increased by cell proliferation prior to the recruitment of secondary neural precursors. A further analysis of the mitotic pattern during late embryogenesis with the mitotic marker anti-Phospho-Histon 3 and phalloidin-rhodamine revealed that only scattered mitotic cells are present in the ventral neuromeres. The pattern of cell divisions in the cephalic lobe and the prosomal segments (310 hours of development) shown in Fig. 6A is representative for the late embryonic stages. Since only a few mitotic cells are associated with dissociating epithelial vesicles (Fig. 6B,6C ), it can be assumed that the secondary precursors differentiate without further divisions, similar to the primary neural precursors. Figure 6 (A-C): Mitotic pattern in the ventral neuromeres after formation of the secondary invagination sites. Flat preparations of embryos stained with phalloidin-rhodamine (red) and anti-Phospho-Histon 3 (green). (A) Only scattered mitotic cells (arrowhead) are present in the ventral neuromeres after invagination of the secondary neural precursors (arrow). The pattern of cell divisions in the cephalic lobe and the prosomal segments at 310 hours of development is representative for the late embryonic stages. (B) Optical section through apical cell layers of the fourth prosomal hemi-neuromere. Only a few mitotic cells (arrowhead) are associated with epithelial vesicles. (C) A similar pattern is visible in basal cell layers of the same neuromere. The arrowhead points to a dividing cell, the arrow points to a dissociating epithelial vesicle. ch , cheliceral neuromere; cl , cephalic lobe; l1 to l2 , prosomal hemi-neuromeres corresponding to walking legs 1 to 2; ped , pedipalpal hemineuromere. achaete-scute homologues and neurogenic genes are re-expressed during formation of the secondary precursors Two achaete-scute homologues have been identified in the spider [ 1 ]. CsASH1 is expressed like a proneural gene in the neurogenic regions prior to formation of the primary invagination sites and is necessary for the generation of neural precursors. CsASH2 , in contrast, shows a pan-neural mode of expression: it is exclusively expressed in all invaginating neural precursors. Simlar to Drosophila melanogaster , the neurogenic genes Notch and Delta restrict the proportion of cells that adopt a neural fate at each wave of neural precursor formation [ 2 ]. During formation of the secondary invagination sites, the spider achaete-scute homologues and neurogenic genes [ 1 , 2 ] are re-expressed in the ventral neuromeres (Fig. 7 ). After invagination of the primary neural precursors, the expression of the achaete-scute homologues CsASH1 and CsASH2 and the neurogenic genes CsDelta1 and CsDelta2 is down-regulated, while CsNotch remains expressed at low levels in the ventral neuroectoderm (Fig. 7A,7B,7C,7D,7E ). There is no clear dividing line between the invagination of the primary neural precursors and the formation of the secondary invagination sites (see above), which is also obvious by CsDelta1 staining: while CsDelta1 is down-regulated in the primary neural precursors (Fig. 7C , arrow), transcripts accumulate at high levels in the secondary invagination sites (Fig. 7C , arrowheads). Similarily, a transient stronger expression of CsNotch is visible in the secondary invaginating cell groups (Fig. 7E , arrow). Interestingly, CsASH1 is only expressed after formation of the secondary invagination sites, in single cells and groups of cells (Fig. 7F , arrow) while the gene shows a proneural mode of expression during formation of the primary precursors [ 1 ]. Like CsASH1 , CsASH2 shows a pan-neural expression in the invaginating secondary precursors (Fig. 7G , arrow). CsDelta1 transcripts accumulate only in a subset of the secondary invaginating cell groups while CsDelta2 seems to be expressed in all of them (Fig. 7H,7I ). CsNotch shows a ubiquituous expression in the ventral neuromeres (Fig. 7J ). Figure 7 (A-J): Proneural and neurogenic genes are re-expressed during formation of the secondary neural precursors. Flat preparations of the fourth and fifth prosomal hemi-segments after in situ hybridisation of whole embroys. (A-E) 220 hours of development, (F-J) 250 hours of development. Anterior is at the top, the midline to the left. (A) At 220 hours, CsASH1 expression has been down-regulated in all primary neural precursors (arrow) with the exception of one group (arrowhead). (B) At this time the pan-neural gene CsASH2 is still weakly expressed in the primary neural precursors (arrow). (C) CsDelta transcripts accumulate in the secondary invagination sites (arrow heads), while transcripts are down-regulated in the primary precursor groups. (D) A similar expression, although weaker, is visible after CsDelta2 in situ hybridisation. The arrow points to a region where CsDelta2 has been down-regulated, the arrowhead indicates expression in the secondary neural precursors. (E) CsNotch remains expressed at low levels in the ventral neuroectoderm. An up-regulation of CsNotch transcripts is visible in the secondary invagination groups (arrow). (F) At 250 hours CsASH1 expression can be detected in the secondary invagination sites (arrow), although it is not expressed in all of them. (G) CsASH2 seems to be expressed weakly in all secondary invaginating cells groups (arrow). (H) A high accumulation of CsDelta1 transcripts is visible in about 10 of the invagination sites (arrow), (I) while CsDelta2 seems to be xpressed in all invagination groups (arrow). (J) Cs Notc h transcripts can be detected in all neuroectodermal cells at this time. l2 to l3 , walking leg 2 to 3. Discussion Formation of epithelial vesicles – a conserved character in arthropod neurogenesis? Analysis of the ventral neuromeres of spider embryos after invagination of the primary neural precursor groups revealed that secondary neural precursors arise during late embryogenesis that partially do not differentiate until larval stages. In contrast to the primary groups, the secondary invaginating cells do not detach from each other after invagination but maintain their epithelial character. In common with epithelial cells, they show a pronounced apico-basal polarity. The apical surface is covered with microvilli, while the lateral surfaces adhere to those of neighbouring cells of a group via specialized cell junctions, i.e. zonulae adhaerentes. Although the formation of epithelial cell groups has not been observed in the ventral neuromeres of other arthropods, epithelial vesicles have been described during development of the stomatogastric nervous system and the brain in Drosophila melanogaster . After invagination of the individual neuroblasts that pioneer the frontal connective and recurrent nerve [ 14 ], three groups of cells invaginate from the stomatogastric nervous system primordium [ 15 ]. They loose contact with the surrounding stomodeal epithelium and form elongated, hollow epithelial vesicles, similar to the secondary neural precursors of the spider. Finally, they dissociate into apolar cells and are incorporated into different stomatogastric ganglia [ 15 , 16 ]. In a similar way, the vesicle forming the optic lobe invaginates from the posterior head region of Drosophila melanogaster embryos. In contrast to the stomatogastric vesicles, this cell group remains epithelial throughout embryogenesis and larval life [ 17 ]. It has been shown in Drosophila melanogaster that the Delta-Notch signaling pathway is involved in maintaining the epithelial character of the optic lobe and stomatogastric nervous system (SNS) precursors [ 16 ]. In Notch mutant Drosophila melanogaster embryos, cells with the identity of SNS and optic lobe precursors develop at approximately normal numbers, but they do not form epithelial vesicles. Instead, these cells appear as solid, irregular clusters of apolar cells [ 15 - 17 ]. In the spider, the function of CsNotch during development of the secondary neural precursors could not be analysed, because injection of ds CsNotch RNA leads to a premature differentiation of neural precursors due to an ealier function of CsNotch in lateral inhibition [ 2 ]. However, the up-regulation of CsNotch in the secondary invagination sites suggests a role in formation of the epithelial vesicles (see Fig. 7E ). Similar to Notch , the proneural genes achaete , scute and lethal of scute are continuously expressed in the SNS of Drosophila melanogaster [ 18 ]. Loss of proneural gene function leads to the absence of a subpopulation of SNS precursors and subsequently to an irregular invagination of the SNS placode. Furthermore, proneural genes seem to promote the dissociation of SNS precursors from the epithelial vesicles, since loss of proneural gene function results in a delay of this process. Similar to Drosophila melanogaster , both achaete-scute homologues of the spider are expressed in the epithelial vesicles that are formed by the secondary neural precursors. However, in contrast to its function in the recruitment of the primary neural precurors, the expression pattern of the spider proneural gene CsASH1 does not suggest a role in the establishment of the secondary neural fate. CsASH1 transcripts can only be detected in subsets of neural precursors after generation of the secondary invagination sites. A similar expression pattern can be observed for CsASH2 , although the transcripts in the primary neural precursors are down-regulated later than the CsASH1 transcripts. The function of these two genes during generation of the secondary invagination sites and the formation of the epithelial vesicles could not be analysed. Due to their ealier role in the recruitment and differentiation of the primary precurors, injection of ds RNA of either gene leads to severe morphological defects in the ventral neuroectoderm [ 1 ]. The formation of epithelial vesicles leads to a delay in neural differentiation As revealed by DiI-labeling, secondary neural precursors within epithelial vesicles do not show any morphological features of differentiation. Obviously, the formation of epithelial vesicles after invagination leads to a delay in the differentiation of the corresponding neural precursors. Although the epithelial vesicles are formed at about the same time, they dissociate from each other subsequently. About half of them are still visible at the end of embryogenesis indicating that they provide neural precursors for larval stages. In insects a distinct mechanism has evolved for generating larval neural precursors during embryonic life. After a phase of cell cycle arrest from late embryogenesis to first larval instar, 'embryonic' neuroblasts proliferate again. [ 9 , 10 ]. Both in Drosophila melanogaster and in Manduca sexta , the larval progeny of these neuroblasts accumulate in groups of cells that are separated by glial cell processes and do not finish their differentiation until the onset of metamorphosis [ 10 , 19 ]. It has been shown that the secreted glycoprotein anachronism (ana) regulates release of central brain neuroblasts from cell cycle arrest [ 20 ]. Ana is expressed in glial cells that ensheath central brain and optic lobe neuroblasts. In ana mutant larvae, neuroblasts proliferate earlier than in normal development which in turn leads to a premature differentation of neurons in certain brain regions. This heterochronic defect has an impact on the axonal pattern: the ana mutant phenotype ranges from subtle missrouting of fiber tracts to massive disorganization that affects the entire optic lobe [ 20 ]. These data show that factors regulating the differentiation state of neural precursors can have an important influence on the organization of neural networks. The distinct morphology of the sheath cells in the spider neuromeres, i.e. their translucent cytoplasm, the absence of microvilli and the extension of cell processes that enwrap the neural precursors suggests that these are glial cells. Further analysis will show if these cells express genes that can influence the epithelial organization, i.e. the differentiation state of the secondary neural precursors, comparable to the glial cells of Drosophila melanogaster . Formation of epithelial vesicles – a basal mode of neurogenesis? A recent study on neurogenesis in the onychophoran Euperipatoides kanangrensis shows that, rather than forming individual invagination groups, the whole medial regions of the hemi-segments invaginate into the embryo [ 21 ]. The invaginated cells remain attached to each other forming transitory epithelial vesicles. Although the phylogenetic position of Onychophora is still being debated, they are generally placed basally in the arthropodan clade [ 22 - 27 ]. Since onychophorans have retained many pleisiomorphic features, it can be assumed that they reflect a basal mode of CNS development [ 28 - 30 ]. This leads to the following model of changes in neural precursor formation during arthropod evolution: the basal mode of neurogenesis is the invagination of one large cluster of neural precursors from the central region of each hemi-neuromere. These clusters form transitory epithelial vesicles in the ventral neuromeres [ 31 ]. An advanced mode of neurogenesis is seen in chelicerates and myriapods: groups of cells that arise in several waves at stereotyped positions invaginate form the ventral neuroectoderm [ 1 , 3 , 4 ]. Interestingly, both chelicerate and myripod neurogenesis reflects some ancestral features. In the spider, epithelial vesicles are formed by secondary invaginating cell groups, while in myriapods the whole central regions of the hemi-neuromeres sink into the embryo after invagination of individual groups of neural precursors [ 3 , 32 ]. An even complexer mode of neurogenesis is seen in insects and crustaceans: individual neuroblasts are singled out from the ventral neuroectoderm that divide in sterotyped patterns to give rise to ganglion mother cells and finally neurons [ 33 - 42 ]. Conclusions To summarize, the model suggests that the invagination of large groups of neuroepithelial cells that form transient epithelial vesicles represents the basal mode of neurogenesis. Subsequently, more parameters have been introduced to the process of neurogenesis during arthropod evolution, i.e. sequential invagination/delamination of neural precursors and connection between neural precursor formation and cell proliferation. It can be assumed that these additional parameters have contributed to the diversity of neural precursor populations. This diversity might have been used as an evolutionary tool to develop neural networks that are adapted to the specialized behaviour and morphologies of the individual arthropod groups. Materials and Methods Cupiennius salei stocks Fertilized females of the Central American wandering spider Cupiennius salei Keyserling (Chelicerata, Arachnida, Araneae, Ctenidae) were obtained from Ernst-August Seyfarth, Frankfurt, Germany. Embryos were collected as described before [ 1 ]. Histology and stainings Whole-mount in situ hybridisations were performed as described [ 1 ]. Phalloidin-rhodamine staining of spider embryos was performed as has been described for flies [ 43 ]. Anti-Phospho-Histone 3 immunocytochemestry has been performed as described [ 1 ]. DiI-labeling After chemically removing the chorion, embryos were fixed in 4 % formaldehyde in PBS and 1 vol heptane. The vitelline membrane was removed with needles and the embryos stained with phalloidin-FITC. Flat preparations of these embryos were attached to a coverslip with a double-sticky tape and covered with PBS. 1,1'-dioctadecyl 3,3,3',3'-tetramethyl indocarbocyanine perchlorate (DiI) was dissolved in ethanol and applied with glass needles. A small droplet of DiI was injected into single or several cells of an invagination group using a 63 × water-immersion lens and a FITC filter on a Zeiss fixed stage microscope and a micromanipulator. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544935.xml |
521175 | X Chromosome Sites Autonomously Recruit the Dosage Compensation Complex in Drosophila Males | It has been proposed that dosage compensation in Drosophila males occurs by binding of two core proteins, MSL-1 and MSL-2, to a set of 35–40 X chromosome “entry sites” that serve to nucleate mature complexes, termed compensasomes, which then spread to neighboring sequences to double expression of most X-linked genes. Here we show that any piece of the X chromosome with which compensasomes are associated in wild-type displays a normal pattern of compensasome binding when inserted into an autosome, independently of the presence of an entry site. Furthermore, in chromosomal rearrangements in which a piece of X chromosome is inserted into an autosome, or a piece of autosome is translocated to the X chromosome, we do not observe spreading of compensasomes to regions of autosomes that have been juxtaposed to X chromosomal material. Taken together these results suggest that spreading is not involved in dosage compensation and that nothing distinguishes an entry site from the other X chromosome sites occupied by compensasomes beyond their relative affinities for compensasomes. We propose a new model in which the distribution of compensasomes along the X chromosome is achieved according to the hierarchical affinities of individual binding sites. | Introduction Most X chromosomal genes are essential or relevant to both sexes. To cope with the difference in the number of copies of these genes in females (XX) and males (XY), organisms have evolved a variety of mechanisms, collectively termed dosage compensation, to equalize the levels of X-linked gene products in the two sexes. In Drosophila males the expression of most of the genes on the single X chromosome is doubled. At least six protein-coding genes, collectively referred to as male specific lethal s ( msl s), are required for dosage compensation ( Baker et al. 1994 ; Marin et al. 2000 ; Meller 2000 ): msl-1, msl-2, and msl-3, whose functions remain unknown; maleless (mle), encoding an RNA helicase; males absent on the first (mof), encoding a histone acetyltransferase; and jil-1, encoding a histone kinase. The products of these genes, together with noncoding RNAs encoded by the RNA on the X genes (roX1 and roX2) ( Amrein and Axel 1997 ; Meller et al. 1997 ; Franke and Baker 1999 ), are all reproducibly associated with hundreds of locations along the length of the polytenized salivary gland X chromosome in males. MOF has been shown both in vivo and in vitro to acetylate H4Lys16, a specific histone modification also found at sites where compensasomes are associated with the male X ( Hilfiker et al. 1997 ; Smith et al. 2000 ; Akhtar and Becker 2001 ). Recently, JIL-1, which phosphorylates H3Ser10, was shown to be enriched at the MSL binding sites in males ( Wang et al. 2001 ). Thus, MSL proteins and roX RNAs are thought to function in a ribonucleoprotein complex (compensasome) to mediate dosage compensation by altering chromatin structure of the male X chromosome ( Stuckenholz et al. 1999 ; Franke and Baker 2000 ). In females translational repression of msl-2 mRNA by the Sex-lethal protein (SXL) prevents formation of compensasomes and hence dosage compensation ( Bashaw and Baker 1997 ; Kelley et al. 1997 ). The processes and constraints that generate the observed distribution of compensasomes along the male X chromosome are unknown. Although the hundreds of places where compensasomes are found along the X chromosome are referred to as “sites,” they are in fact not points, but rather bands (small segments of chromosome) that roughly span the size range of salivary chromosome bands seen with DNA stains (i.e., a few tens to several hundreds of kilobases in length). Thus, both the locations and the extents of these sites are somehow specified. Furthermore, the compensasome bands do not correspond to the bands where DNA is condensed ( Baker et al. 1994 ; Kelley et al. 1999 ; Demakova et al. 2003 ). In addition, non-dosage-compensated X-linked genes (e.g., LSP1-α ) are scattered throughout the X chromosome and can reside next to dosage-compensated genes ( Baker et al. 1994 ). Since there is no known DNA-binding component in the compensasome, and consensus DNA sequences required for binding have not yet been identified, an understanding of the distribution of compensasomes along the X chromosome needs to encompass not only how complexes are targeted to these several hundred sites, but also how the ends of each band are delimited. A proposal for how the distribution of compensasome bands along the X chromosome is generated ( Kelley et al. 1999 ) has come from the following findings. MSL-1 and MSL-2 represent core components of the complex: The presence of both is required for either to bind, and none of the other MSL proteins binds to the X chromosome in an msl-1 or msl-2 mutant male ( Lyman et al. 1997 ). Furthermore, in males mutant for mle, msl-3, or mof, binding of MSL-1 and MSL-2 is only maintained at a limited number of sites (35–40) on the X chromosome, which include the roX1 and roX2 genes ( Lyman et al. 1997 ; Kelley et al. 1999 ). Finally, roX transgenes inserted into an autosome retain binding of compensasomes, and in addition show compensasome binding in the autosomal region flanking the insertion site, a phenomenon termed spreading ( Kelley et al. 1999 ). Based on these observations, a reasonable model ( Kelley et al. 1999 ) emerged suggesting that the 35–40 sites of MSL-1 and MSL-2 binding on the X seen in mle, msl-3, or mof mutants represent nucleation sites or entry sites for the complex. From these sites, newly assembled compensasomes would spread in cis along the X to form the hundreds of final sites observed in a wild-type male. In this spreading model, roX RNAs would also be required for compensasome assembly ( Park et al. 2003 ). However, there is to date no direct evidence that entry sites and spreading play any role in the processes that generate the normal pattern of compensasome binding along the X chromosome. We thus directly tested this model by analyzing various pieces of the X chromosome transposed or translocated to autosomal locations for their ability to bind compensasomes and initiate spreading. Results The spreading model implies that a piece of the X chromosome translocated to an autosome must contain at least one of the 35–40 “entry” sites if that piece of the X is to recruit compensasomes and become dosage compensated. We looked at MSL binding in various chromosome rearrangements that inserted small pieces of X chromosome into autosomal locations. Table 1 summarizes the translocations, transpositions, and duplications examined. The insertions in the first set (lines I to XI) range in size from about 1% to 15% of the length of the X, and the corresponding stretch of X chromosome for each contains 1–19 distinguishable MSL bands. These insertions were examined in heterozygous condition so we could readily identify the junctions between X chromosomal and autosomal material. When large enough, they appear as a loop of unpaired chromosome protruding from the paired autosomes. We found that transpositions containing one (lines VI to VIII) or several (lines I to V) previously described entry sites ( Lyman et al. 1997 ) showed consistent MSL binding along the inserted piece ( Table 1 ; Figure 1 A, 1 B, and 1 D). Surprisingly, transposed pieces of X chromosome lacking any entry site also showed MSL binding when inserted into an autosome ( Table 1 , lines IX to XI; Figure 1 C, 1 E, and 1 F). For all of these 11 transpositions the binding pattern observed and the intensity of MSL bands reproducibly matched the expected pattern of that piece of the X chromosome in a wild-type male. Even the smallest piece we looked at (line X, approximately 200 kb) showed one to two MSL bands ( Figure 1 C). Thus, we found that any piece of the X chromosome moved to an autosomal location is able to bind compensasomes, whether or not the transposed piece of X chromosome contains an entry site. This finding suggests that each of the hundreds of MSL bands observed on the X in males carries the information necessary and sufficient to attract compensasomes, and does not require adjacent entry sites. Figure 1 MSL Binding to Pieces of X Chromosome Inserted into Autosomes Salivary glands from males heterozygous for each transposition were fixed (47% acetic acid in phosphate-buffered saline, then lactic acid/water/acetic acid [1:2:3]), squashed on slides, treated with anti-MSL-1 antibodies and a secondary Cy3 anti-rabbit immunoglobulin G antibody, then counterstained with DAPI and viewed using a Zeiss Axiophot microscope. Both duplications and transpositions were able to attract compensasomes, whether or not they contained predicted entry sites. (A) Line II. (B) Line I, which contains the roX1 gene. (C) Line X shows one to two bands on the smallest transposition we studied; the intensity of the second band was variable even on the X chromosome. (D) Line IV. (E) Line IX. (F) Line XI. Breakpoints (described in Table 1 ) were verified by cytology when possible and/or with specific probes by in situ hybridization. Gray value images were pseudo-colored and merged. Table 1 Summary of the Transpositions Studied: Transpositions, Duplications, and Reciprocal Translocations Variations in both the number of bands observed in the transpositions and their intensity are due to variable accessibility of the piece examined on the squash and the orientation of the chromosomes when flattened for observation. Similar variations were observed on the intact X. No additional MSL binding was observed into autosomal regions flanking translocated X material or onto autosomal material inserted onto the X chromosome. We found a breakpoint in line VI to be at 3F1 instead of 3E7–8, and 5A instead of 4A in line XIII. Line III contains a piece of the X inserted into a pericentric inversion of the second chromosome, while line IV carries an inversion of 77D5–81 See Materials and Methods for precise genotypes a ES, number of entry/high-affinity sites present in each transposition according to our observations and previous studies ( Lyman et al. 1997 ) b Number of MSL bands observed in a wild-type background on each piece of the X inserted onto an autosome (lines I to XI), or number of MSL bands observed in autosomal regions inserted into the X (lines XII to XV) c N, percentage of nuclei showing additional bands in autosomal regions flanking the site of insertion of a piece of the X chromosome; the number of nuclei scored is presented in parentheses Tp, transposition; Dp, duplication; T, reciprocal translocation Interestingly, duplications showed binding both along the autosomal insertion and on the X chromosome (lines II and XI), indicating that the supply of compensasomes is not limiting in these circumstances. We also tested homozygous transpositions and duplications for MSL binding in males and found that we could recover MSL binding on each homozygous transposed piece (unpublished data) as well as on the X. Thus, even three copies of the same segment of the X chromosome (two of the duplication plus the original piece on the X) were able to maintain MSL binding. This result extends previous data showing that, by using specific msl-2 transgenes escaping SXL repression, ectopic expression of MSL-2 in females induced binding to both X chromosomes, in a pattern identical to the single X of a wild-type male ( Bashaw and Baker 1997 ). Therefore, binding occurs regardless of the location and number of copies of the X-linked targeted sequences. The determinations listed in Table 1 of how many entry sites each of the transpositions contains were made by comparing the reported breakpoints of each rearrangement to the described locations of entry sites ( Lyman et al. 1997 ). As cytological determinations can vary, we directly confirmed the presence or absence of entry sites by examining MSL binding in an msl-3 or mle mutant background for a subset of these transpositions ( Figure 2 ). Each line used in these experiments contained the transposed region from the X inserted into an autosome and a wild-type X chromosome. For line XI we found that, in mle mutant individuals, MSL binding was undetectable in either the transposed region (3A5–E8) inserted at 87E17 ( Figure 2 A– 2 E) or in this region in the wild-type X. As expected, the same is true when only a subset of this region is duplicated: Line X did not show binding in mle mutants to region 3C2–3C6 on the X or to the transposition of that region inserted at 61D ( Figure 2 F– 2 K). These findings confirm that lines X and XI do not contain entry sites. Similarly, we confirmed that transpositions inferred to contain entry sites in two lines (IV and VI) did in fact contain such sites. Thus, for line IV in an mle mutant background we observed MSL binding to one to three sites on both the transposition and the corresponding region of the X ( Figure 2 N and 2 P), while for line VI in an msl-3 mutant background we observed one site of MSL binding on both the transposition and the corresponding region of the X ( Figure 2 S). These findings are consistent with those of Lyman et al. (1997) , who reported two entry sites in the region encompassed by the transposition in line VI, and one entry site in the region encompassed by the transposition in line IV. Our findings firmly establish that isolated subregions of the X chromosome display normal patterns of compensasome binding irrespective of whether they contain entry sites, and thus suggest that entry sites do not play a distinct role in the establishment of compensasome binding along the X as postulated by the spreading hypothesis. Hereafter we will refer to entry sites as high-affinity sites, their original name ( Lyman et al. 1997 ). During the course of this study, Oh et al. (2004) have reported similar results for binding of compensasomes to transpositions from lines I, VIII, and IX. However, the scale of the analysis and the limited number of rearrangements did not yield the same conclusions. Figure 2 MSL Binding to Autosomal Duplications of X Chromosome Pieces in mle or msl-3 Mutant Larvae Salivary glands from w ; pr mle 12.17 / cn bw mle ; Dp (1;3)/msl2Δ10 or w; Dp (1;2)/msl2Δ21 ; msl3 p /msl3 p females were squashed and stained as described in Figure 1 , followed by in situ hybridization with a biotinylated probe specific for regions carried by each duplication ( Lavrov et al. 2004 ) and incubation with Oregon green-coupled streptavidin. Conditions throughout the procedure were adjusted to maximize MSL staining. Specific biotinylated probes (green bars) appear in green in merges (A, F, I, L, O, Q, and R) and as bright bands in (B, D, G, J, and M). MSL bands are shown in red in merges and in (P) and as bright bands in (C, E, H, K, N, and S). DAPI stain is blue. MSL binding is absent from duplications or the matching region on the X in line XI (3A5–3E8) (A–E) and line X (3C2–3C6) (F–K) in mle mutants, confirming that they lack any entry sites. Probe maps region 3D–E in (A–E) and 3C in (F–K). (L–P) Illustrated are the one to three bands detected in mle mutant nuclei on the duplicated region from line IV (2C1–3C5) (O and P) and on the same segment on the X (M and N). (O) and (P) are from another nucleus. (Q–S) A single band is detected at the 3F1 breakpoint of the duplication (3C2–3F1, line VI) in msl-3 mutant nuclei (S), corresponding to the weakest band of the doublet at 3F on the X. Note the weak signal on duplications compared to the same region on the X chromosome. Probe maps region 2D5–3A2 in (L–P) and 3D–E in (Q–S). The two high-affinity sites identified to date correspond to the roX1 and roX2 genes ( Kageyama et al. 2001 ; Park et al. 2003 ), and it was the fact that roX transgenes inserted into autosomal locations are able to induce spreading—binding of the MSLs to some autosomal sequences surrounding a roX transgene insertion site—that led to the hypothesis that spreading gives rise to the wild-type distributions of compensasome bands along the male X chromosome. We therefore examined whether autosomal transpositions of a piece of the X were able to induce spreading. In cells heterozygous for each of the transpositions listed above we never observed additional MSL binding to the autosomal regions either cis or trans to the insertion site ( Table 1 ; see Figure 1 ). We also did not observe additional MSL binding in males homozygous for the transpositions described above. This was true irrespective of the number of high-affinity sites contained in the transpositions. Interestingly, lines I and V, which each contain several high-affinity sites, including the roX1 or roX2 gene, respectively, showed no spreading in males wild-type for the MSLs (see Figure 1 B). The dichotomy between our results and those obtained with roX transgenes suggests that spreading may be a phenomenon restricted to some roX transgenes (see below) and not an aspect of dosage compensation. To further assess if spreading in cis occurs on the X chromosome, we next asked if the complex could spread from the X onto an autosomal piece attached to the X by a reciprocal translocation. We tested two reciprocal translocations that interchanged large portions of the X and 3R or 2L (see Table 1 , lines XII and XIII, respectively). Both translocations separate roX1 (3F) and roX2 (10C) genes from one another and thus both pieces of each translocation contain a roX locus. Anti-MSL-1 staining revealed the absence of any bands on either of the 3R or 2L pieces of these translocations ( Figure 3 ), while the pattern observed on the two transposed pieces of the X was normal. These results strengthen the idea that spreading may be a phenomenon restricted to roX transgenes, since the breakpoints in line XII (10A) and line XIII (5A) are relatively close to the roX2 (10C) and roX1 (3F) loci, respectively. Figure 3 Compensasomes Do Not Spread from the X Chromosome onto Autosomal Regions Inserted on the X (A) Females expressing MSL-2 from an msl2Δ3–21 transgene and bearing a reciprocal translocation between the X and second chromosome (line XIII) do not show additional bands in the regions of the 2L arm juxtaposed to X chromosome material. (B) MSL binding pattern on the X chromosome of a wild-type male. (C and D) The autosomal region 81F–82F10–11 does not show MSL binding when inserted at 3D in the single X of a male (line XV) (C) or in MSL-2-expressing females heterozygous for the same transposition (D). Note that the MSL binding pattern on the X chromosome is not altered by the insertion. The light band (arrow) maintained on the wild-type unpaired region of the X of a female heterozygous for the transposition is also present next to the same insertion at 3D on the unique X chromosome of a male (compare C and D). We also tested two small transpositions of autosomal regions into the X ( Table 1 , lines XIV and XV; Figure 3 C): Neither of them showed MSL binding, even weak, to any part of the inserted autosomal sequences. Furthermore, females either heterozygous or homozygous for these transpositions and expressing ectopic MSL-2 did not show any MSL bands in either of these insertions of autosomal material into the X, although they displayed normal MSL binding both to the unpaired X region (in heterozygotes) and along the paired portions of the two X chromosomes ( Figure 3 D). Thus, insertion of a piece of an autosome into the X does not disrupt MSL binding to either the unpaired X homologue at the insertion site or the regions of the X immediately flanking the site of insertion of autosomal material. Moreover, these results are inconsistent with the model derived from the roX transgene studies where MSL binding is observed both in the autosomal regions adjacent to the insertion site and on the wild-type autosomal homologue. Discussion In summary, we have used chromosome rearrangements to test two central aspects of the proposed spreading model of dosage compensation in Drosophila. It is worth noting that our experiments were a priori neutral: They could have provided compelling evidence for or against the spreading model. In both cases our results are inconsistent with the clear predictions of that model. First, we show that pieces of the X chromosome inserted into an autosome bind compensasomes in precisely the pattern characteristic of that piece of the X at its endogenous location on the X, and this property is independent of the presence of sites previously described as entry sites. Second, compensasomes do not spread from the X into autosomal pieces inserted into, or translocated onto, the X. Moreover, there is not spreading of compensasomes from autosomal insertions of pieces of the X chromosome into the autosomal regions flanking the insertion, even when such pieces contain a roX gene close to the breakpoint. These results suggest that spreading in cis is not part of the process of dosage compensation in flies. We thus propose that all of the hundreds of sites along the X chromosome where compensasomes are found in wild-type males are competent to independently recruit compensasomes. Our findings raise several questions regarding previous data. Are the 35–40 sites that attract partial complexes in mle or msl-3 mutants qualitatively different from the other sites at which MSL bands are found in wild-type, and if so, how? Why do roX transgenes induce additional binding to adjacent autosomal sequences? With respect to the potential heterogeneity of compensasome binding sites, while most of the relevant data are indirect (only the roX1 and roX2 genes are identified binding sites), the data are consistent with the simple view that the binding sites are homogeneous in terms of their function, but have varying affinities for compensasomes. Our finding that pieces of X chromosome transposed to autosomal locations display normal patterns of compensasome binding, irrespective of whether or not they contain high-affinity sites, removes the one functional distinction between binding sites that had been proposed. That there are not two classes of binding sites in terms of affinity for compensasomes, but rather a continuum of affinities, is strongly suggested by the recent report of Demakova et al. (2003) , who carefully characterized the number and locations of compensasome bands in mutant females expressing various limiting amounts of MSL-2. They found only four bands in the most limiting case, and progressively higher numbers of bands as more MSL-2 protein was expressed. Interestingly, the intermediate 40 sites at which complete complexes are assembled in these conditions exactly matched with the 35–40 high-affinity sites bound by partial complexes in mle or msl-3 mutants. Their data are consistent with a model in which compensasomes continue to bind site specifically to additional sites after all high-affinity sites are occupied, as opposed to spreading from high-affinity sites as previously proposed. Given these findings, a reasonable scenario as to how dosage compensation is achieved would be the following. As MSL expression begins, the high-affinity sites progressively sequester nascent partial or full complexes in the early stages of dosage compensation. When the amount of available complexes or its components increases, sites of higher affinity would accumulate more complexes, while low-affinity sites would remain undetectable, until the former have preferentially assembled sufficient amounts of complexes to make components available for sites with lower affinities. Thus, the compensasomes would progressively bind to different sites along the X according to the different affinities of these sites. Consistent with our model, we found that in mle or msl-3 mutants, duplications maintain binding of partial complexes at the high-affinity sites ( Figure 2 N, 2 P, and 2 S), though with a lower affinity than the same site on the X. The latter observation suggests that, in conditions where components of the complex are limiting, binding might also be dependent on the location of these sequences in the cell (see discussion on spreading below). That compensasome binding sites would have a range of affinities is also consistent with what is known about DNA-binding proteins, which recognize with varying affinities a range of binding sites whose sequences are related to a common consensus. Variations from the consensus can allow temporal and quantitative modulation of individual genes, or subsets of genes. That compensasome binding sites are also likely to vary in sequence, and hence affinities, comes from what is known about sex chromosome evolution in Drosophila species ( Marin et al. 1996 , 2000 ). During the course of sex chromosome evolution in this genus there are a number of cases in which new X chromosomes have evolved, and in all cases examined to date, this has been accompanied by the new X chromosome gradually acquiring compensasome binding sites as the new Y chromosome, its former homologue, degenerates. The selective advantage of dosage compensation for each gene is determined both by the state of degeneration of the allele on the new Y chromosome and by the degree to which a gene in males requires its function, and thus its expression, to match the output of both wild-type female X chromosomes ( Marin et al. 2000 ). Hence, one would expect individually evolved binding sites to exhibit a range of affinities for compensasomes. Finally, we note that each of the final compensasome bands on the X chromosome displays a reproducible but specific intensity, likely to reflect not only different affinities for compensasomes, but also the length of X chromosome encompassed in each band. The last issue we wish to address is spreading. The fact that, in chromosome rearrangements that juxtapose pieces of X and autosome, we never observed spreading, even when entry sites or roX genes were near the breakpoints, suggests that spreading does not exist naturally on the X chromosome, and is not required to establish the final pattern of binding in Drosophila males. Yet spreading from roX transgenes is very well documented in a variety of situations. We therefore suggest that spreading is a phenomenon specific to the roX transgenes, and a consequence of the key function of roX RNAs in dosage compensation. In particular, we propose that the roX genes are the sites of assembly of compensasomes using newly synthesized roX RNAs, just as the ribosomal RNA genes are the sites where ribosomes are assembled. Thus, roX transgenes would generate a high local concentration of compensasomes in their vicinity, competing with other chromatin-binding factors that normally bind to nearby autosomal sequences. In some cases, compensasomes would displace these other factors, resulting in a new compensasome band in the autosomal region flanking the transgene (spreading). Several features of spreading are consistent with this proposal. First, additional bands corresponding to spreading from roX transgenes contain roX RNA and the H4Lys16 modification, suggesting that they correspond to mature complexes ( Kelley et al. 1999 ). Second, transcription from a roX transgene is required to observe spreading of the complex onto neighboring regions ( Park et al. 2002 , 2003 ). Third, roX transgenes show variable and often no additional bands in a wild-type background, suggesting that spreading is largely dependent on the insertion site and its environment on the autosomes. One possibility would be that these roX transgenes lacking spreading are inserted next to sites bound by factors normally counteracting the effect of compensasomes on the autosomes. Such a view is supported by recent data showing that association of compensasomes at some roX1 transgenes can overcome the effect of methylation-mediated silencers ( Kelley and Kuroda 2003 ). Finally, MSL-1 and MSL-2 co-overexpression leads to mislocalization of partial MSL complexes to the autosomes and the centromere, as well as a dramatic decompaction of the X ( Oh et al. 2003 ), a male-specific phenotype also observed in both iswi or nurf mutants, two chromatin regulators ( Deuring et al. 2000 ; Badenhorst et al. 2002 ; Corona et al. 2002 ). Thus, increasing locally the amount of available complexes can induce new binding of MSL complexes to usually non-dosage-compensated regions. Molecular studies of dosage compensation in flies, worms, and mammals have revealed some striking similarities between these systems. In all three systems dosage compensation is achieved by a widespread modification of the structure of X chromosome chromatin, and in mammals and flies this involves specific modifications of histones. Dosage compensation in mammals and flies is also similar in that noncoding RNAs are essential components of the dosage compensation machinery. With respect to the other components of the dosage compensation machinery the situation is less clear. While compensasome-related complexes might be present in mammals (orthologs of msl-1, -2, -3, mle, and mof genes exist in mammalian genomes), some of them have identified functions not related to dosage compensation, and orthologs of msl-1, -2, and -3 were not found in Caenorhabditis elegans ( Marin 2003 ). Up until now it had also been thought that spreading was involved in dosage compensation in all three systems ( Park et al. 2002 ; Oh et al. 2003 ; Csankoversuszki et al. 2004 ; Okamoto et al. 2004 ). However, our findings indicate that in flies each of the bands on the X chromosome at which compensasomes are found in males is able to independently attract those complexes. Thus, at the interband level spreading does not appear to be part of the dosage compensation process in flies. However, it should be noted that our results do not address either how compensasomes are distributed across the tens of kilobases of DNA that likely comprise individual compensasome bands in salivary gland chromosomes, or how that distribution is achieved; it is possible that, at the level of single bands, spreading may be part of the process of dosage compensation. Materials and Methods Fly strains and genetic crosses Flies were raised on standard cornmeal-yeast-agar medium. Fly stocks containing transpositions were obtained from the Bloomington Drosophila Stock Center. Their genotypes are: Tp(1;2)rb + 71 g, ct 6 v 1 /C(1)DX, y 1 w 1 f 1 (line I); Df(1)ct-J4, In(1)dl-49, f 1 /C(1)DX, y 1 w 1 f 1 ; Dp(1;3)sn 13a1 /+ (line II); Tp(1;2)sn + 72d, f 1 car 1 /C(1)DX, y 1 f 1 ; Dp(?;2)bw D , bw D (line III); Tp(1;3)w vco , v 1 f 1 : in w vco /ClB, B 36d (line IV); Tp(1;3)v + 74c/FM7a (line V); Tp(1;2)w-ec, ec 64d cm 1 ct 6 sn 3 /C(1)DX, y 1 w 1 f 1 (line VI); Tp(1;3)f + 71b/FM6 (line VII); Tp(1;3)JC153, v 1 /FM7a (line VIII); Tp(1;3)sta, sta 1 : ss sta /FM3 (line IX); Tp(1;3)w zh , sc 1 z 1 w zh (line X); Df(1)w258–45, y 2 sn 3 /C(1)DX, y 1 w 1 f 1 ; Dp(1;3)w + 67k/+ (line XI); T(1;3)v, v A /FM6 (line XII); Tp(2;1)odd 1.10 , b 1 pr 1 cn 1 sca 1 /CyO (line XIII); Df(2 l)sc19–7/In(2 l)Cy L t R In(2R)Cy, Cy 1 amos Roi-1 cn 2 sp 2 or Dp(2;1)B19, y 1 ed 1 dp o2 cl 1 (line XIV); Dp(3;1)2–2, w 1118 ; Df(3R)2–2/TM3, Sb 1 (line XV). Breakpoints and insertion site are referred in Table 1 . Some lines contain additional rearrangements referenced in Lindsey and Zimm (1992) . Depending on their genotype, each line was crossed to Canton-S males or females for studies of MSL binding in their male progeny. For homozygous transpositions studies, stocks were balanced to give w; Tp(1;2)/Cyo-GFP or w; Tp(1;3)/TM3-GFP stocks. Non-GFP third instar male larvae were dissected for analysis. For autosome-to-X transpositions, females from lines XIV and XV were mated with w; msl2Δ3–21/CyoGFP or Dp(A;1)/Y; msl2Δ3–21/CyoGFP males. Non-GFP female larvae were dissected. For mle and msl-3 mutant analysis, stocks were balanced to give w; Tp(1;2)/CyoGFP; msl3 p /TM3-GFP or w; prmle 12.17 /CyoGFP; Tp(1;3)/TM3-GFP stocks. Females were crossed to w; msl3 p /CyoGFP; msl2Δ3–10/TM3-GFP or mle 1 cnbw/CyoGFP; msl2Δ3–21/TM3-GFP males, respectively. Non-GFP third instar female larvae were dissected for salivary glands polytene chromosomes analysis. Lines expressing MSL-2 from transgenes msl2Δ3–21 and msl2Δ3–10 are described in Bashaw and Baker (1995) . Mle and msl-3 mutants are described in Fukunaga et al. (1975) , Kuroda et al. (1991) , and Gorman et al. (1995) . All crosses to generate larvae for immunostaining were carried out at 18 °C. Polytene chromosome immunostaining Glands from male third instar larvae were dissected in PBS/0.7% NaCl, prefixed in 45% acetic acid for 10 s, and then fixed for 2–3 min in lactic acid/water/acetic acid (1:2:3) solution on siliconized coverslips. Glands were squashed and coverslips flipped off after freezing the slides in liquid nitrogen. Slides were then incubated in PBS for 15 min followed by incubation with affinity-purified anti-MSL-1 antibodies (dilution 1:100) as described previously ( Gorman et al. 1995 ). Chromosomes were viewed under epifluorescence optics on a Zeiss Axiophot microscope or a confocal microscope; pictures were taken using Spot software and colored. Immunofluorescent in situ hybridization of polytene chromosomes Clones RP-98 17.E.2, RP-98 03.D.13, and RP-98 48.O.22 from the Drosophila melanogaster BAC library (BACPAC Resources, Oakland, California, United States) were used to map regions 3D–E, 3C, and 2D5–3A2, respectively. Specific probes were obtained from BAC clone DNA preparations using the Bionick Labelling System (Invitrogen, Carlsbad, California, United States) according to the manufacturer's instructions. Squashes were prepared as described above. Immunostaining with affinity-purified anti-MSL-1 antibodies was followed by incubation with the appropriate biotinylated probe according to the method of Lavrov et al. (2004) . Supporting Information Accession Numbers The LocusLink ( http://www.ncbi.nlm.nih.gov/LocusLink/ ) accession numbers for the genes and gene products discussed in this paper are jil-1 (LocusLink 39241), mle (LocusLink 35523), mof (LocusLink 31518), msl-1 (LocusLink 35121), msl-2 (LocusLink 33565), msl-3 (LocusLink 38779), roX1 (LocusLink 43963), roX2 (LocusLink 44673), and SXL (LocusLink 44872). | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC521175.xml |
535896 | Atypical presentation of hepatocellular carcinoma: a mass on the left thoracic wall | Background Hepatocellular carcinoma is a common malignancy for which chronic hepatitis B infection has been defined as the most common etiologic factor. The most frequent metastatic sites are the lung, bone, lymphatics, and brain, respectively. Metastases to the chest wall have been reported only rarely. Case presentation We report a patient with hepatocellular carcinoma who presented with an isolated metastatic mass on the left anterolateral chest wall in the axillary region. Conclusions Metastasis of HCC should be included in the differential diagnosis of rapidly growing lesions in unusual localizations, particularly in patients with chronic liver disease even if a primary tumor can not be radiologically identified. | Background Hepatocellular carcinoma (HCC) is the most frequent primary malignant tumor of the liver [ 1 ]. It is usually seen in the 6 th and 7 th decades of life, and the most common etiologic factor is chronic viral hepatitis, particularly in the presence of cirrhosis. Chronic hepatitis B has been described as the most frequent cause [ 2 - 4 ]. Hematogeneous extrahepatic metastases are common, with lungs, regional lymph nodes, kidneys, bone marrow and adrenals being the most frequent sites [ 3 , 4 ]. Rarely HCC may present as a mass without a primary liver tumor being identified [ 5 - 7 ]. This report presents an unusual metastasis of HCC which presented as a mass on the left chest wall in the axillary region. Case presentation A 71-year-old male was admitted to our clinic with complaints of left shoulder pain, swelling in the left anterolateral chest wall, jaundice, weight loss, dyspnea and weakness. At three months prior to admission, he had noticed a less movable mass approximately 2 cm in diameter and then the tumor had grown rapidly, becoming increasingly painful during exertion. There was no history of expectoration, fever, or night sweats. The patient's physical examination revealed yellow discoloration of the sclerae, pitting edema up to the knees and a 5 cm × 6 cm fixed mass in the left axillary region. There was mild splenomegaly but no hepatomegaly, liver masses or ascites. Laboratory findings were as follows: sedimentation rate: 90 mm/hr; Hb: 11.5 mg/dL; WBC: 3900/mm 3 ; Plt: 126 000/mm 3 ; PT: 15.6 sec; INR:1.3; total protein: 6.5 g/dL (N: 6.4–8.5 g/dL); albumin: 2.2 g/dL (N: 3.4–4.8 g/dL); AST: 124 iu/L (N: 15–40); ALT: 52 iu/L (N: 10–40); ALP: 173 iu/L (N: 37–147); GGT: 213 iu/L (N:0–40); total bilirubin: 3.8 mg/dL (N: 0.1–1.2), and direct bilirubin: 1.57 mg/dL (N: 0–0.3). Hepatitis B virus surface antigen, IgG antibody to the core antigen, anti-HBe and HBV DNA with polymerase chain reaction were positive. HBe antigen, anti-Delta and serological markers of hepatitis C were negative. Abdominal ultrasonography showed ascites, splenomegaly and diffusely nodular and heterogeneous echogenic patterns in the liver. There was no history of chronic liver disease. Upper gastrointestinal endoscopy was normal except for esophageal varices. Computerized tomography of the thorax revealed a mass on the left anterolateral chest wall (Figure 1 ). Cytological examination of a fine needle aspirate taken from the mass was consistent with metastatic hepatocellular carcinoma (Figure 2 ). Abdominal computerized tomography detected thrombosis in the right portal vein. The liver parenchyma was diffusely heterogeneous with irregular borders and without a clear mass or infiltrating lesion. There was no lymphadenopathy on thoracic, abdominal or pelvic computerized tomography. The patient had an elevated serum alpha-feto protein (AFP) level of 60 000 ng/mL (0 – 13.6 ng/ml). Cytological examination of the liver confirmed the diagnosis of hepatocellular carcinoma. The patient was discharged on palliative treatment, and he died 21 days later. Discussion Although extrahepatic metastasis of HCC was reported in 18% of untreated patients in a retrospective study, metastatic lesions were found at a higher incidence in an autopsy study of deaths related to primary liver cancer [ 8 , 9 ]. The most common sites of extrahepatic involvement are the lungs, lymph nodes, adrenal glands, and bones [ 3 , 8 , 10 - 12 ]. Metastasis of HCC occurs frequently by way of intrahepatic blood vessels, lymphatic permeation, or direct infiltration. Hematogenous spread occurs with the involvement of hepatic or portal veins or the vena cava. Metastases have also been found in collaterals and varices, and this appears to have been the route of metastasis in the patient reported here. His right portal vein was thrombosed, most likely due to tumoral infiltration. Tumor cells might have passed through the left thoracic wall via portosystemic collaterals, the azygous system and finally intercostal veins. Another possible route is through subcutaneous collaterals communicating to thoracoepigastric veins and draining into the axillary vein. Although bone metastases are seen in 10% of HCC patients [ 13 ], it infrequently appears as the first manifestation of HCC. The most common sites are the vertebra and pelvis [ 14 ]. Isolated metastases of HCC to the ribs have been rarely reported [ 5 , 6 , 15 - 17 ]. The establishment of the diagnosis of metastatic HCC can occasionally be problematic, particularly when the primary tumor has not been identified. To the best of our knowledge there are only two reports in the literature describing cases of solitary metastasis to the chest wall from an unknown primary HCC. One patient had an HCC metastasis involving the sternum, and the other patient had a metastasis on the right 4 th rib [ 5 , 6 ]. In the present patient, the etiology of HCC was due to chronic hepatitis B, and the HCC was diffuse in nature. Yuki et al. reported an association between HBs antigen positivity and diffuse-type HCC. Intrahepatic, hematogenous and lymphogenous metastases have been frequently observed in diffuse-type HCC [ 18 ]. Nevertheless, diffuse-type HCC is often difficult to detect on imaging studies because of its permeative appearance and heterogeneity of background chronic liver disease [ 19 ]. In our patient, although HCC was of the diffuse type, metastasis involvement was isolated. There was also no regional lymph node infiltration. AFP, when elevated, usually correlates with tumor size. AFP doubling time is also closely related to tumor doubling time. A rapidly growing axillary mass, diffuse tumoral infiltration of the liver and a very high level of AFP in our patient are consistent with these observations. However, nearly one-fourth of patients with HCC may have normal AFP values [ 19 ]. The need for liver biopsy might be questionable in the our patient. The presence of chronic liver disease with a very high AFP level was highly suggestive for HCC. However, there are several reports in the literature describing ectopic HCC without a liver origin. These patients can be treated with surgical resection and have a good prognosis [ 7 , 20 ]. Furthermore, several reports have claimed that ectopic livers are particularly predisposed to developing HCC [ 20 - 22 ]. In conclusion, metastasis of HCC should be included in the differential diagnosis of rapidly growing lesions in unusual localizations, particularly in patients with chronic liver disease even if a primary tumor cannot be radiologically identified. Competing interest The author(s) declare that they have no competing interests. Authors contributions SC, involved in the patient active management and preparation of the figures. OY, carried out the literature search. SK, revised it for scientific content. KC, performed the cytological examination. MB, involved in the patient active management. AD, edited the manuscript and coordinated the submission. All authors contributed to the preparation of the manuscript. All authors read and approved the final version of the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535896.xml |
517821 | Genomic Insights into Methanotrophy: The Complete Genome Sequence of Methylococcus capsulatus (Bath) | Methanotrophs are ubiquitous bacteria that can use the greenhouse gas methane as a sole carbon and energy source for growth, thus playing major roles in global carbon cycles, and in particular, substantially reducing emissions of biologically generated methane to the atmosphere. Despite their importance, and in contrast to organisms that play roles in other major parts of the carbon cycle such as photosynthesis, no genome-level studies have been published on the biology of methanotrophs. We report the first complete genome sequence to our knowledge from an obligate methanotroph, Methylococcus capsulatus (Bath), obtained by the shotgun sequencing approach. Analysis revealed a 3.3-Mb genome highly specialized for a methanotrophic lifestyle, including redundant pathways predicted to be involved in methanotrophy and duplicated genes for essential enzymes such as the methane monooxygenases. We used phylogenomic analysis, gene order information, and comparative analysis with the partially sequenced methylotroph Methylobacterium extorquens to detect genes of unknown function likely to be involved in methanotrophy and methylotrophy. Genome analysis suggests the ability of M. capsulatus to scavenge copper (including a previously unreported nonribosomal peptide synthetase) and to use copper in regulation of methanotrophy, but the exact regulatory mechanisms remain unclear. One of the most surprising outcomes of the project is evidence suggesting the existence of previously unsuspected metabolic flexibility in M. capsulatus, including an ability to grow on sugars, oxidize chemolithotrophic hydrogen and sulfur, and live under reduced oxygen tension, all of which have implications for methanotroph ecology. The availability of the complete genome of M. capsulatus (Bath) deepens our understanding of methanotroph biology and its relationship to global carbon cycles. We have gained evidence for greater metabolic flexibility than was previously known, and for genetic components that may have biotechnological potential. | Introduction Methanotrophic bacteria such as Methylococcus capsulatus are responsible for the oxidation of biologically generated methane ( Soehngen 1906 ), and they are therefore of great environmental importance in reducing the amount of this greenhouse gas released to the Earth's atmosphere. Atmospheric methane levels have been increasing over the last 300 years, and it is thought that this is mostly due to human activity. Methane is a very effective greenhouse gas; it has been estimated that methane contribution to climate change is about 26 times that of carbon dioxide (mole for mole) ( Ehalt 1974 ; Ehalt and Schmidt 1978 ; Lelieveld et al. 1993 ). The effect is further amplified by the reduction of hydroxyl radical concentrations due to increasing atmospheric methane levels; these radicals oxidize methane photochemically, thus their loss from the atmosphere increases the persistence of methane ( Lelieveld et al. 1993 ). Biological methane oxidation is known to occur aerobically in both terrestrial and aquatic habitats, and anaerobically in sediments and anoxic salt water. It acts on methane biologically generated in situ and on methane scavenged from the atmosphere ( Figure 1 ). Deep-sea environments such as cold gas seeps and hydrothermal vents exhibit a photosynthesis-independent food chain based on methanotrophs and chemolithotrophs, some of which form symbiotic partnerships with invertebrates (e.g., Cavanaugh et al. 1987 ). Methanotrophs are also able to metabolize or co-metabolize xenobiotic compounds, including chlorinated solvents such as trichloroethylene, and hence have potential as bioremediation tools ( Large and Bamforth 1988 ). Figure 1 Global Methane Cycle Methane is oxidized either photochemically in the atmosphere or biologically in terrestrial and aquatic systems. The ocean, grasslands, and desert form major methane sinks, whereas wetlands, agricultural and grazing lands, and other anthropogenic sources such as landfills, are major sources. The cow depicted in the figure represents diverse ruminants. Anthropogenic inputs of nitrogen in the form of ammonia compete for MMOs, reducing methane oxidation and leading to the formation of nitrous oxide, another greenhouse gas. Distribution of methanotrophy within the Bacteria is currently thought to be relatively limited, being found so far in only 11 genera of the Proteobacteria. These methanotrophs are classified into two types, based primarily on their phylogenetic relationships but also on differences in their physiology and internal membrane structure. Type I methanotrophs (including Methylococcus ), which are all members of the Gammaproteobacteria, utilize ribulose monophosphate (RuMP) as the primary pathway for formaldehyde assimilation, whereas those of type II, which are all Alphaproteobacteria, use the serine pathway ( Hanson and Hanson 1996 ). M. capsulatus is classified as an obligate methanotroph ( Whittenbury et al. 1970 ); methane is oxidized via methanol to formaldehyde, which is then assimilated into cellular biomass or further oxidized to formate and CO 2 for energy production. The conversion of methane to biomass by M. capsulatus has been exploited for large-scale commercial production of microbial proteins by fermentation ( Skrede et al. 1998 ). The powerful tool of whole-genome sequencing has been applied to microorganisms that carry out other important components of the carbon cycle, such as photosynthesis ( e.g., Eisen et al. 2002 ; Dufresne et al. 2003 ) and methanogenesis ( Bult et al. 1996 ; Smith et al. 1997 ; Slesarev et al. 2002 ), but there is a paucity of genomic information on the methanotrophs, which are equally important contributors to global carbon cycles. Many insights into methylotrophy have been gained from the recently available partial genome sequence of Methylobacterium extorquens (AM1) ( Chistoserdova et al. 2003 , 2004 ), but this organism, like other nonmethanotrophic methylotrophs, is limited to the oxidation of C1 compounds other than methane. We undertook the whole-genome sequencing of M. capsulatus (Bath) to obtain a better understanding of the genomic basis of methanotrophy, a globally important microbial process. Results/Discussion Genome Properties The genome of M. capsulatus (Bath) comprises a single circular molecule of 3,304,697 bp. General features of the genome and its 3,120 predicted coding sequences (CDSs) are summarized in Table 1 . GC skew ( Lobry 1996 ) and oligoskew ( Salzberg et al. 1998b ) analyses were used to identify a putative origin of replication, and basepair 1 was assigned upstream of the glucose-inhibited division protein A (gidA) gene(MCA0001), adjacent to the chromosome-partitioning proteins encoded by gidB, parA, and parB and the operon that encodes F 1 F 0 ATP synthase. Table 1 General Features of the M. capsulatus (Bath) Genome EUS, enzymes of unknown specificity The M. capsulatus (Bath) genome contains 51 identifiable insertion sequence elements from various families ( Chandler and Mahillon 2002 ). As is found in other sequenced bacterial genomes, many of these elements share higher intra- than intergenome similarity. This suggests several possible mechanisms: expansion of these elements since their introduction into the M. capsulatus (Bath) genome, repeated cycles of duplication and subsequent deletion, or gene conversion. Twenty elements belonging to the IS4 family of insertion sequences ( Chandler and Mahillon 2002 ) encode a 366-amino-acid transposase with 100% amino acid sequence conservation between copies. One copy (MCA1197) is found within the soluble methane monooxygenase (sMMO) operon, although not in all sequenced clones, suggesting that this element is highly mobile. Other examples of insertion of this element include disrupted genes encoding tRNA pseudouridine synthase (split into two putative CDS—MCA1311 and MCA1313) and an exopolysaccharide export protein (split into MCA1176 and MCA1178). Two putative prophages (one of approximately 58.5 Kbp, spanning from MCA2632 to MCA2689 and the other, a Mu-phage-like element of approximately 45 Kbp spanning from MCA2900 to MCA2959) were identified in the genome. The Mu-like prophage is unusual in encoding an intein within F (Mu gp30), a cofactor in head assembly. This intein region is most similar to inteins present in several archaeal translation initiation factor IF-2 sequences from the genera Pyroccocus and Methanococcus, sharing 41% sequence identity and 62% sequence similarity with the Pyrococcus horikoshii intein. The intein lacks a recognizable endonuclease sequence and appears degenerate compared to the archaeal IF-2 intein, casting doubt on its ability to be mobile. If functional, the presence of the intein in this protein suggests that head morphogenesis could be regulated by conditions that influence the rate of intein excision. Another intein sharing sequence similarity with archaeal inteins was identified in the gene encoding ribonucleotide reductase (MCA2543). Inteins are rare, but when present are often found in genes associated with nucleotide metabolism, such as ribonucleotide reductases. Bacteriophages have been important tools for genetic manipulation of bacterial genomes, and such tools are currently lacking for M. capsulatus (Bath). The M. capsulatus (Bath) Mu-like prophage could be engineered to resemble the Mu derivatives, which have been excellent tools for random mutagenesis in other species ( Casadaban and Cohen 1979 ). Conditional protein splicing via inteins is used as a tool for protein engineering, drug therapy, and vaccine development ( Humphries et al. 2002 ; Mootz et al. 2003 ; Nyanguile et al. 2003 ). The putative inteins could be designed as a tool either for generating protein material for vaccination of salmon (or other animals feeding on M. capsulatus proteins) or for manipulating M. capsulatus live vaccine vectors. Metabolism and Transport: Genomic Basis of the Methanotrophic Lifestyle We have attempted to predict central metabolic pathways in M. capsulatus (Bath), including the methane oxidation pathway, mechanisms for carbon fixation, glycolytic and gluconeogenic conversions, and the tricarboxylic acid (TCA) cycle, from analysis of the genome data. These pathways, together with those known from previous studies, are depicted in Figure 2 , along with the locus numbers for predicted enzymes. Some of these pathways have not been experimentally verified, so we present Figure 2 as a hypothesis of metabolic activity in M. capsulatus (Bath) that is based on available genome data. Figure 2 Predicted Central Metabolic Pathways of M. capsulatus Genomic information was used to predict the flow of carbon from methanotrophy pathways into carbon fixation pathways, and thence into glycolysis/gluconeogenesis and the TCA cycle. Locus names are indicated next to key steps. Some intermediates are omitted. Methane oxidation Methanotrophs are unique in their possession of methane monooxygenases (MMOs), which catalyze the first step of methane oxidation ( Figure 2 ). M. capsulatus is known to possess both a particulate membrane-bound form, pMMO (detected by centrifugation studies and encoded by pmo ), and a soluble form, sMMO (encoded by mmo ), and these enzymes have been extensively studied ( Murrell 1994 ; Nguyen et al. 1998 ; Stolyar et al. 1999 ; Coufal et al. 2000 ; Murrell et al. 2000a , 2000b ; Whittington and Lippard 2001 ). The pMMO was previously known to consist of three subunits encoded by pmoCAB ( Zahn and DiSpirito 1996 ); two complete copies of pmoCAB and a third copy of pmoC (pmoC3) were previously identified ( Stolyar et al. 1999 ). Our genomic analysis suggests the pMMO genes have been recently duplicated ( Table 2 ). The pmoC3 gene is located in a putative operon with three additional genes of unknown function, and we can speculate that these three are also related to methane oxidation. Table 2 Selected Putative Lineage-Specific Gene Duplications in M. capsulatus (Bath) These genes were identified as those encoding proteins with better BLASTP matches to other proteins in M. capsulatus than to all other complete genomes Only one chromosomal locus for the components of sMMO was identified ( mmoXYB– transposase –mmoZDC– hypothetical protein –mmoGQSR [MCA1194–1205]). The transposase (MCA1197) is oriented in the same direction as the mmo genes and, thus, may be transcribed under sMMO-promoting growth conditions. The functions of these sMMO components have been previously determined ( Stainthorpe et al. 1990 ; Nielsen et al. 1997 ; Coufal et al. 2000 ; Merkx and Lippard 2002 ; Csaki et al. 2003 ). Methanol oxidation Methanol is available to M. capsulatus from the oxidation of methane and presumably also from exogenous sources (e.g., pectin and lignin degradation) ( Hanson and Hanson 1996 ), and its oxidation is catalyzed by methanol dehydrogenases ( Anthony and Williams 2003 ). We have detected three sets of genes encoding homologs of the structural components of methanol dehydrogenase (homologs of MxaF and MxaI) and the proteins required for its catalytic function ( homologs of MxaJGRACKLD ) ( Figure 2 ). There is one large cluster of genes (MCA0779–0790) including homologs of mxaFJGIRACKLD, which probably encodes a heterodimeric methanol dehydrogenase, as is found in other methylotrophs ( Amaratunga et al. 1997a , 1997b ). Also like other methylotrophs, M. capsulatus (Bath) contains a second methanol dehydrogenase–like cluster, mxaFJ, (MCA0299–0300) lacking a homolog of mxaI that normally encodes the small subunit of methanol dehydrogenase. The function of mxaJ is unknown, and it is not clear whether this mxaFJ cluster encodes components active in methanol oxidation. There is a third cluster of genes required for methanol dehydrogenase synthesis, mxaACKL (MCA1527–1530). Formaldehyde and formate oxidation Formaldehyde is the substrate for carbon fixation through the RuMP pathway in M. capsulatus ( Attwood and Quayle 1984 ) and thus is an important intermediate in both catabolism and anabolism ( Figure 2 ). However, formaldehyde is also highly toxic, and the cell needs to tightly control its production ( Attwood and Quayle 1984 ). We were able to compare the results of genomic analysis with previously reported pathways for formaldehyde oxidation in M. capsulatus . A membrane-bound pyrroloquinoline quinine protein is known to be the major formaldehyde dehydrogenase under high-copper growth conditions, while a soluble NAD(P) + -linked formaldehyde dehydrogenase is active when copper concentrations are low ( Zahn et al. 2001 ). Other previously characterized formaldehyde oxidation pathways in M. capsulatus include the cyclic pathway that uses enzymes of the RuMP pathway ( Strom et al. 1974 ), and the tetrahydromethanopterin (THMPT)-linked pathway ( Vorholt 2002 ). Genome analysis showed the previously determined N-terminal sequence of the pyrroloquinoline quinine–containing formaldehyde dehydrogenase ( Zahn et al. 2001 ) to match MCA2155, a protein resembling a sulfide-quinone reductase (SQR), on the basis of both its motifs and its phylogenetic relationship to other SQRs. Zahn et al. (2001) reported homology of their N-terminal sequenced enzyme with SQRs from other bacteria, but were unable to obtain experimental evidence for quinone reductase activity. In the absence of this evidence, we have annotated the gene as a formaldehyde dehydrogenase. We found that the previously sequenced 63-kDa subunit of NAD(P) + -linked formaldehyde dehydrogenase ( Tate and Dalton 1999 ) best matches the N-terminal of the large subunit of methanol dehydrogenase (MCA0779), although the sequences differ slightly. The sequence of modifin, the 8.6-kDa subunit thought to confer substrate specificity to the enzyme ( Stirling and Dalton 1978 ; Tate and Dalton 1999 ), could not be clearly identified in the genome. A recent paper ( Adeosun et al. 2004 ) helps resolve this conflict between the genome and these previous results; apparently the NAD(P) + -linked activity is due to an artefactual mixture of methanol dehydrogenase and methylene dehydrogenase. We also identified components of the RuMP (Entner-Douderoff pathway)-linked and THMPT-linked pathways of formaldehyde oxidation ( Figure 2 ). In addition to the formaldehyde oxidation systems described above, genome analysis suggests an additional complete tetrahydrofolate (THF)-linked pathway ( Figure 2 ) previously undescribed in M. capsulatus, as was recently found alongside the THMPT pathway in M. extorquens ( Vorholt 2002 ; Chistoserdova et al. 2003 ). In M. extorquens , the THF pathway is thought to play a role in assimilation of carbon from both formaldehyde and formate, while THMPT is involved in catabolic oxidation of formaldehyde to formate using the same enzymes used for methanogenesis; free formaldehyde is thought to be the substrate for hexulose-6-phosphate synthase in the RuMP pathway ( Strom et al. 1974 ; Vorholt 2002 ). M. capsulatus possesses homologs to genes encoding proteins in all of these pathways; therefore, it may have the capability to assimilate/detoxify formaldehyde in the same way. We have identified three previously undescribed homologs of formate dehydrogenases (MCA2576–2577, MCA1208–1210, and MCA1391–1393). Multiple formate dehydrogenases occur in other bacteria ( Sawers 1994 ; Chistoserdova et al. 2004 ); in M. extorquens, all three are expendable, indicating that this last step in methane oxidation plays a minor role and that formate can be dissimilated in other ways ( Chistoserdova et al. 2004 ). The importance of the formate dehydrogenases in M. capsulatus remains to be determined. The genomic redundancy in the set of methane oxidation pathways suggests that M. capsulatus exploits different systems under variable environmental conditions (e.g., copper levels). It is plausible that M. capsulatus balances its requirement for formaldehyde-derived carbon and reducing power with the toxicity of formaldehyde by taking advantage of three enzymes for formate oxidation and multiple pathways for formaldehyde oxidation under different environmental conditions. This redundancy has implications for future attempts to manipulate the genes of this pathway; simple knockouts may not be achievable. Carbon fixation M. capsulatus is known to assimilate formaldehyde through the RuMP pathway ( Strom et al. 1974 ) ( Figure 2 ). Genomic analysis suggests that RuMP components have experienced a lineage-specific duplication ( Table 2 ); there is a 5,267-bp identical direct repeat centered around the transaldolase gene that contains the genes for hexulose-6-phosphate isomerase , hexulose-6-phosphate synthase , fructose-1,6-phosphate aldolase , and transketolase . There is evidence that the RuMP pathway is also used for gluconeogenesis (see below); this dual function may have been facilitated by the redundancy resulting from this tandem duplication. The M. capsulatus (Bath) genome appears to contain some parts of the alternative serine pathway of formaldehyde assimilation ( Figure 2 ), including a candidate for the key serine cycle enzyme malyl-CoA lyase (MCA1739). Activities associated with this pathway have been reported as “sometimes” present ( Hanson and Hanson 1996 ). However, the majority of enzymes with a putative role in the serine cycle also have putative roles in other metabolic pathways (e.g., there are candidate genes encoding proteins that may be able convert malate to malyl-CoA—MCA1740–1741 are similar to the two subunits of malate Co-A ligase from M. extorquens [ Chistoserdova and Lidstrom 1994 ] but are also similar to the two subunits of succinyl Co-A synthases). In addition, the genome apparently lacks any good candidates for other steps in the serine cycle such as the conversion of phosphoenolpyruvate to oxaloacetate (i.e., phosphoenolpyruvate carboxylase). The latter enzyme may be circumvented by a likely oxaloacetate decarboxylase (MCA2479–2481), which converts pyruvate to oxaloacetate ( Figure 2 ). It is possible that M. capsulatus fixes formaldehyde through the serine cycle as far as oxaloacetate ( Figure 2 ). It appears that the Calvin cycle operates with transketolase (MCA3040 and MCA3046), reversibly converting glyceraldehyde-3-phosphate to xylulose-5-phosphate, bypassing the typical ribose-5-phosphate to fructose-6-phosphate segment ( Figure 2 ); cell suspensions of M. capsulatus grown on methane do not exhibit seduheptulose-1,7-bisphosphatase activity ( Strom et al. 1974 ), and a gene encoding this enzyme was not identified in the genome sequence. Redundancy in serine and glycogen biosynthesis pathways Serine is an important intermediate in M. capsulatus metabolism, and genomic evidence suggests three potential pathways for serine synthesis not previously described in M. capsulatus: a phosphorylated pathway from glycerate-3-phosphate, a nonphosphorylated pathway from glycerate, and a nonphosphorylated pathway from glycolate-2-phosphate ( Ho and Saito 2001 ) ( Figure 2 ). Homologs of genes encoding enzymes predicted to catalyze the three steps in the phosphorylated pathway (3-phosphoglycerate dehydrogenase, phosphoserine aminotransferase, and phosphoserine phosphatase) are present, but the latter two may have alternate functions in vitamin B6 biosynthesis ( Lam and Winkler 1990 ) or as homoserine kinases. Genes normally encoding homoserine kinases (thrB and thrH) were not identified in the M. capsulatus (Bath) genome, and phosphoserine phosphatase may perform this function as described in Pseudomonas aeruginosa ( Patte et al. 1999 ). The nonphosphorylated pathway is not well characterized at the molecular level, but it is known to be initialized by the dephosphorylation of phosphoglycerate to glycerate ( Ho and Saito 2001 ); subsequently, glycerate is oxidized to hydroxypyruvate and hydroxypyruvate is transaminated to serine ( Figure 2 ). The genome encodes a homolog of glycerate kinase, a 2-hydroxyacid dehydrogenase that may function as a hydroxypyruvate reductase, and a serine-glyoxylate aminotransferase, which may also have serine-pyruvate transaminase activity. Genes encoding these three enzymes appear to be organized in an operon (MCA1406–1408), supporting their proposed roles in serine formation from phosphoglycerate. The M. capsulatus (Bath) genome also encodes a putative phosphoglycerate mutase (MCA0753), to interconvert 3- and 2-phosphoglycerate ( Figure 2 ), allowing the organism to carry out either the phosphorylated or nonphosphorylated pathway. In the third pathway, M. capsulatus may derive glycolate-2-phosphate from the oxygenation reaction of ribulose bisphosphate carboxylase (MCA2743–2744, previously identified by Baxter et al. [2002] ), convert it to glycine, which is split into carbon dioxide, ammonia, and methylene-tetrahydrofolate ( Figure 2 ). A second glycine molecule and methylene-tetrahydrofolate ligate to form serine. There is experimental evidence ( Taylor et al. 1981 ) for this pathway of glycolate-2-phosphate assimilation, which resembles that of plants. Evidence for novel gluconeogenesis pathways and a complete TCA cycle A key enzyme in gluconeogenesis is fructose-1,6-bisphosphatase, which catalyzes the irreversible dephosphorylation of fructose-1,6-phosphate to fructose-6-phosphate; genes encoding this enzyme are absent. However, there are three potential alternative pathways for gluconeogenesis, previously unknown in this organism ( Figure 2 ). First, there is a transaldolase homolog (MCA3045) that may convert glyceraldehyde-3-phosphate directly to fructose-6-phosphate. Second, there is a putative phosphoketolase (MCA1587), which can condense pyruvate and glyceraldehyde-3-phosphate into xylulose-5-phosphate, which in turn is fed into the ribulose-5-phosphate pool for eventual formation of glucose-6-phosphate through the pentose phosphate pathway. Third, hydrolysis of fructose-1,6-bisphosphate to fructose-6-phosphate by a pyrophosphate-dependent 6-phosphofructokinase and a pyrophosphatase may occur, as was recently proposed in Nitrosomonas ( Chain et al. 2003 ). An incomplete TCA cycle lacking 2-oxoglutarate dehydrogenase activity has been found in nearly all type I methanotrophs, including M. capsulatus ( Hanson and Hanson 1996 ). However, genes encoding homologs of 2-oxoglutarate dehydrogenase are present (MCA1952 and MCA1953). Thus, a complete TCA-cycle might operate in M. capsulatus , not under methane oxidation, but under other conditions. Lack of experimental evidence precludes speculation as to the nature of these conditions; however, catabolite repression ( Wood et al. 2004 ) may play a role here. Another type I methylotroph, Methylomonas sp. (761), uses a complete TCA cycle to grow on glucose as its sole carbon and energy source ( Zhao and Hanson 1984 ); M. capsulatus may utilize the same mechanism as carbon is stored as glycogen. Consistent with its autotrophic lifestyle, M. capsulatus possesses only a limited array of membrane transporters for organic carbon compounds. However, although M. capsulatus is not known to utilize any sugars (although in the Texas strain they have been reported to support growth), one complete (MCA1941–1944) and one partial (MCA1924) ATP-binding casette (ABC) family transporter with predicted specificity for sugar uptake were identified. Additionally, components of transporters for peptides (MCA1264 and MCA1268), carboxylates (MCA1872), and a variety of amino acids (e.g., MCA0840) are present. Diversity of nitrogen metabolism M. capsulatus (Bath) is able to fix atmospheric nitrogen ( Murrell and Dalton 1983 ), conferring an advantage in environments where fixed nitrogen is limiting, and the structural genes for nitrogenase (nifH, nifD, and nifK) were previously shown to be contiguous ( Oakley and Murrell 1991 ), as they are in other nitrogen fixers. Genome analysis extends this contiguous region to include the genes nifE, nifN, and nifX, which are involved in synthesis of the nitrogenase iron-molybdenum cofactor (MCA0229–0239); this organization has been found in Chlorobium tepidum and some nitrogen-fixing methanogenic Archaea. Two 2Fe-2S ferredoxins (MCA0232 and MCA0238) and two genes identified as conserved hypotheticals (MCA0236–0237) are interspersed with the nif genes in the same orientation. The conserved hypothetical genes share the highest sequence similarity with genes from other organisms capable of nitrogen fixation, suggesting that they also have a role in this process. M. capsulatus exhibits considerable versatility in its combined nitrogen conversions, including nitrification and denitrification. Ammonia is oxidized to nitrite by both pMMO and sMMO because of their lack of substrate specificity ( Colby et al. 1977 ; Dalton 1977 ); the absence of a separate ammonia monooxygenase, and the redundancy of MMOs, suggests that the MMOs are the sole nitrification enzymes active in M. capsulatus . Four predicted ammonium transporters were identified (MCA0268, MCA0490, MCA1581, and MCA2136), suggesting that ammonium is an important nitrogen source for M. capsulatus . Methane oxidation is inhibited by the presence of ammonia and ammonia oxidation is inhibited by methane ( Whittenbury et al. 1970 ), and input of ammonia to wetland systems (e.g., through fertilizer runoff) may have significant effects on the consumption of biogenic methane by methanotrophs in these systems. It is also interesting to note that, in general, ammonia oxidation produces small amounts of nitrous oxide, which is also a greenhouse gas. Electron transport complement suggests unexpected metabolic flexibility The M. capsulatus (Bath) genome has a relatively large complement of putative c-type cytochromes; 57 proteins containing a heme-binding motif were identified, and 23 of these contain two or more heme-binding motifs. Analysis of the genome reveals electron transport components previously known to be associated with the methane oxidation pathway, such as cytochrome C L (MCA0781), a specific electron acceptor for methanol dehydrogenase. Other novel electron transport components are encoded in several physical locations on the chromosome; there is genomic evidence for chemolithotrophy and the ability to live at a variety of oxygen tensions. The genome encodes three predicted hydrogenases: (a) a multisubunit formate hydrogenlyase (MCA1137–1142), most likely involved in the conversion of formate to dihydrogen and carbon dioxide; (b) a soluble cytoplasmic NAD-reducing hydrogenase (MCA2724–2726), which transfers electrons to NAD + ; and (c) a membrane-bound Ni-Fe hydrogenase (MCA0163–0165). Activity of two hydrogenases (one soluble and one membrane-bound), and the role of molecular hydrogen in driving MMOs, was previously reported ( Hanczar et al. 2002 ). The membrane-bound Ni-Fe hydrogenase was previously sequenced ( Csaki et al. 2001 ). The presence of these two hydrogenases suggests that M. capsulatus is able to capture and oxidize hydrogen that is generated either exogenously or as a by-product of the ATP-dependent reaction of nitrogenase, and recycle it into the electron transport chain. Microorganisms that undergo fermentative metabolism are likely to be encountered in the habitat of M. capsulatus (e.g., soils) and could supply exogenous hydrogen for chemolithotrophic oxidation. The removal of this hydrogen by M. capsulatus metabolism may aid in driving these reactions forward and hence constitute a syntrophic partnership. Nitrogen fixation requires reducing power, which in aerobes can be supplied by reduced flavodoxin or ferredoxins. There is one candidate flavodoxin present (MCA1697) in the M. capsulatus (Bath) genome; however, two ferredoxins (MCA0238 and MCA0232) physically located within a cluster of genes encoding proteins involved in nitrogen fixation (see Nitrogen Metabolism section above) more likely serve in this capacity. In aerobes, these carriers are usually reduced by NADH/NADPH, although reverse electron transport could be involved in this organism. The M. capsulatus (Bath) genome encodes homologs of a two-subunit high oxygen-affinity cytochrome d (MCA1105 and MCA1106), which suggests the ability to live under microaerophilic conditions. This evidence for life at low oxygen tensions is supported by the presence of enzymes indicative of fermentative activity ( Table 3 ). Further support for anaerobiosis is provided by a putative large c-type cytochrome (MCA2189) that contains 17 heme groups and is located adjacent to several hypothetical proteins, including an oxidoreductase and an alkaline phosphatase important to the central metabolism of phosphorous compounds. This cytochrome has significant matches only to high molecular-weight cytochromes in the metal-ion reducers Shewanella oneidensis, Desulfovibrio vulgaris, and Geobacter sulfurreducens, suggesting that M. capsulatus may have the ability to undergo metabolism at a lower redox potential than previously known. This large protein is most likely localized in the periplasm, as indicated by its signal peptide. The ability to oxidize methane under reduced oxygen tensions would provide an advantage to M. capsulatus in allowing it to be physically closer to environments in which methane is biologically generated. Table 3 Putative Enzymes Associated with Fermentation The M. capsulatus (Bath) genome includes a region of approximately 25 kb (MCA0421–0443) encoding novel proteins related to energy metabolism (c-type cytochromes, a c-type cytochrome biogenesis protein, a novel gene possessing a flavodoxin domain, two proteins that may be involved in heme transport, and an undescribed Fe-S binding protein) and hypothetical proteins. The same 25-kb region contains six multiheme c-type cytochromes that are members of the cytochrome c 553o family. This previously described family is unique to M. capsulatus (Bath); genome analysis has widened our knowledge of this family from three ( Bergmann et al. 1999 ) to six members (MCA0338, MCA0421, MCA0423, MCA0424, MCA2259, and MCA2160). The redundancy in cytochrome c 553o proteins suggests a more complex and plastic electron transport capability than previously known. A novel monoheme c-type cytochrome (MCA1187) has eight transmembrane-spanning regions and is located near another c-type cytochrome and a proton-translocating pyrophosphatase (a transmembrane-spanning protein pump important in establishing electrochemical gradients). This configuration suggests that the monoheme protein has a role in ATP production via chemiosmosis. Two other monoheme CDSs (MCA2188 and MCA2196) are members of a paralogous family similar to a cytochrome found in G. sulfurreducens; one has a signal for twin arginine transport (for export from the cytoplasm) and the second a signature for fumarate lyase. The genome also encodes homologs to the putidaredoxin family of ferredoxins. Genomic evidence for anaerobic synthesis of unique fatty acids The membrane phospholipids of methanotrophs are unique, with mono-unsaturated fatty acids consisting of a series of even- and odd-numbered positional isomers of both the cis and trans configuration ( Makula 1978 ). Our analysis supports the existence of an anaerobic mechanism for unsaturated fatty acid synthesis in M. capsulatus, as proposed by Jahnke and Diggs (1989) on the basis of biochemical data . A predicted enzyme 3-hydroxydecanoyl-ACP-dehydratase (MCA2878) appears to catalyze an alternative dehydratase reaction at the C 10 level of fatty acid synthesis, followed by a synthase reaction carried out by 3-oxoacyl-acyl carrier protein (ACP) synthase (MCA2879), resulting in cis -vaccenate (18:1, cis -Δ11). Trans -unsaturated acids are obtained by isomerization of preformed cis -unsaturated fatty acids, to control membrane fluidity; putative fatty acid cis / trans -isomerases were identified in the genome (MCA1585 and MCA1806) . Sterol and hopanoid biosynthesis: evidence for the mevalonate-independent pathway M. capsulatus is one of only a few prokaryotes known to synthesize sterols de novo ( Bird et al. 1971 ), and it is thought that they have a role in maintaining membrane fluidity in response to changes in environmental temperature ( Jahnke 1992 ). The main cholesterols in M. capsulatus (Bath) are methylated; genome analysis indicated four putative proteins involved in the conversion of squalene to 4,4-dimethylcholest-8(14)-en-3β-ol (MCA2872, MCA2873, MCA2711, and MCA1404). Squalene is also a precursor for hopanoid synthesis ( Figure 2 ); homologs of squalene hopene cyclase, squalene synthase, and other enzymes leading to hopanoid synthesis were identified (MCA0812, MCA0813, and MCA2873). Acetyl-CoA is usually the starting point for synthesis of hopanoids and sterols. However, with the exception of the final step catalyzed by geranyl- trans -transferase, the mechanism for converting acetyl-CoA to squalene (the first major intermediate) is not apparent from genome analysis. Instead, M. capsulatus (Bath) contains genes for the alternative mevalonate-independent pathway from glyceraldehyde-3-phosphate and pyruvate (MCA0817, MCA0573, MCA1055, and MCA2518), so it is possible that the organism employs the same squalene synthesis pathway found in plants and many Gram-negative bacteria ( Rohdich et al. 2003 ). Environmental Sensing, Response, and Survival Copper homeostasis, scavenging, and transport. Copper is known to be important in the regulation of MMO activity; high copper concentrations are essential for the formation of extensive intracytoplasmic membranes and pMMO activity, and copper is thought to play an active role in both the catalytic site and the electron transport chain ( Nguyen et al. 1994 ; Semrau et al. 1995 ; Basu et al. 2003 ). In contrast, sMMO activity is inhibited by copper; synthesis of sMMO may allow methanotrophs to survive in copper-limited environments where pMMO cannot be active. Two distinct copper transporting systems have been identified in other bacteria, a P-type ATPase (the Cop system) and a resistance/nodulation/cell division (RND)-type copper ion efflux complex (the Cus complex) ( Petersen and Moller 2000 ; Rensing et al. 2000 ; Franke et al. 2003 ). Analysis of the M. capsulatus (Bath) genome reveals several elements that may relate to processing of copper. The genome encodes a putative nonribosomal peptide synthetase (NRPS) that may scavenge copper; another methylotrophic bacterium (Methylosinus) is known to excrete copper-binding compounds ( DiSpirito et al. 1998 ; Tellez et al. 1998 ). The NRPS comprises a starting module (MCA2107) containing an adenylation domain (probably recognizing a 5-hydroxy ornithine residue or another derivative of ornithine), a thiolation domain, and an unusual acetyltransferase domain. The starting module may interact with a second module (MCA1883) that contains a condensation domain and a terminal thioesterase needed for peptide release, leading to synthesis of a heavily charged peptide that could be involved in binding/scavenging of copper or other metals. A single polyketide synthase gene (PKS) (MCA1238) is found adjacent to a gene encoding a sensor protein with diguanylate cyclase and diguanylate phosphodiesterase activities (MCA1237), often found in environmental sensing proteins and H + /heavy metal cation antiporters. The two-module and six-domain organization of this PKS is atypical; it contains domains with unknown functions, and its role is difficult to predict. A cation membrane transport system (MCA1900, MCA1907, MCA1911, and MCA1915) is located near the NRPS, and the 4′-phosphopantetheinyl transferase needed for activation of both PKS and NRPS is also present (MCA1522), indicating that these multimodular enzymes may be active. M. capsulatus (Bath) has a large repertoire of 12 P-type cation ATPases, including multiple predicted copper ion pumps, which correlates with the role of copper in regulation of methane oxidation in this organism. There are also 18 resistance/nodulation/cell division–type metal ion and drug efflux pumps; the large number of these pumps, together with a variety of other metal cation uptake and efflux systems, highlight the significance of metal ion homeostasis in M. capsulatus . The genome encodes three homologs (MCA0705, MCA0805, and MCA2072) of P-type ATPases with the characteristic copper-binding P-type ATPase motif ( Solioz and Stoyanov 2003 ), which makes them likely to function like CopAs from other species. In Escherichia coli , CopA is regulated together with CueO, a multicopper oxidase, by CueR, a member of the MerR-family transcription regulators. No evidence for a CueR homolog was found, indicating a different mechanism of regulation of the copA and cueO genes in M. capsulatus . The genome encodes one potential cusCBA gene cluster (MCA2262–2264), with the cusA candidate (encoding the central transport protein CusA) having the same copper-binding and transport motif found in the E. coli gene. No indication of homologs of the CusF periplasmic copper chaperone was found. However, it is interesting to note that the cusB candidate (which encodes the CusB outer membrane protein) carries the metal-binding motif typically found in CusF, suggesting that the putative CusB may have a dual function as CusF. No evidence for the CusRS two-component response system regulating the E. coli cusCFBA operon was found in M. capsulatus (Bath). In summary, there are elements of previously studied copper transport and regulation systems in the genome of M. capsulatus (Bath); the lack of the same full complement of genes and identifiable regulators raises questions about the exact operation of copper regulation and suggests future experiments to resolve this central mechanism. Possible pathways for capsule biosynthesis. As evidenced by its specific epithet, M. capsulatus possesses an insoluble polysaccharide capsule ( Whittenbury et al. 1970 ), the composition of which has not previously been determined. Genome analysis reveals several possible pathways for the synthesis of capsular material, including colanic acid and alginate. The former includes putative colanic acid biosynthesis glycosyl transferases (MCA2124 and MCA1168), the DnaJ-like protein DjlA (MCA0020), which interacts with DnaK to stimulate colanic acid capsule synthesis ( Genevaux et al. 2001 ), and guanine diphosphate-mannose 4,6-dehydratase (MCA1146). Also present are rfb genes involved in the synthesis of O antigen; O antigen can serve as capsular material, but given its additional role in lipopolysaccharide synthesis, this cannot be determined with certainty. Colanic acid is generally not produced at temperatures higher than 30 °C in E. coli ( Whitfield and Roberts 1999 ), so alginate and O antigen may constitute capsular material at the higher growth temperatures favored by M. capsulatus. M. capsulatus (Bath) is somewhat desiccation resistant ( Whittenbury et al. 1970 ), and there is evidence that desert soil methanotrophs can survive long periods of water deprivation ( Striegel et al. 1992 ); capsule biosynthesis may aid in this. Primitive pathway for asparaginyl- and glutaminyl-tRNA synthesis. In common with the genomes of Archaea and some Bacteria, the M. capsulatus (Bath) genome lacks genes for asparaginyl-tRNA synthetase and glutaminyl-tRNA synthetase. However, a heterotrimeric glutamyl-tRNA amidotransferase (MCA0097–0099) is present, suggesting that a single amidotransferase forms asparaginyl-tRNA and glutaminyl-tRNA by transamidation of mischarged aspartyl-tRNA or glutamyl-tRNA, as found previously in many Gram-positive and some Gram-negative bacteria, archaea, and eukaryal organelles ( Ibba et al. 1997 ; Curnow et al. 1998 ; Becker et al. 2000 ; Raczniak et al. 2001 ; Salazar et al. 2001 ). This indirect transamidation pathway has been proposed as the more ancient route to Gln-tRNA Gln formation ( Curnow et al. 1997 ), because glutamine is thought to be among the last amino acids to be added to the current repertoire of 20 amino acids. It has also been suggested that when this indirect transamidation pathway is the primary source of Gln-tRNA Gln within the cell, it acts as a regulatory mechanism for glutamine metabolism ( Curnow et al. 1997 ). Evidence for Evolution of Genomic Novelty Genomic redundancy. The genome of M. capsulatus (Bath) exhibits redundancy in many pathways, as described in more detail in the relevant sections above. The redundant genes fall into two categories. The first category comprises those that appear to be lineage-specific duplications (see Table 2 ), identified as genes encoding proteins with better BLASTP matches to other proteins in M. capsulatus (Bath) than to all other complete genomes. In some cases, these genes are found adjacent to each other in the genome, implying that they may have been generated by a tandem duplication, and that they may be transient (tandem arrays are prone to deletion). Other lineage-specific duplications, including those of many genes encoding hypothetical and conserved hypothetical proteins, may not simply be transient mutations and may instead have been maintained because they confer an evolutionary advantage on the organism. The second category contains redundant genes that do not appear to be recently duplicated, and are evolutionarily divergent, suggesting ancient duplications or exogenous acquisition. The divergent phylogeny of these genes is inconsistent with ancient duplication and subsequent vertical transmission, but we cannot determine the origin, direction, and exact path of a possible lateral transfer. Past exchange of genetic material between a methanogen and a methanotroph ancestral to M. capsulatus, whether direct or indirect, is certainly plausible, given their biochemical dependency. The presence of both categories of redundancy for a given enzyme or pathway makes it more plausible that the enzyme or pathway is functionally important and that its redundancy is advantageous to the organism. M. capsulatus, like some other bacteria ( Karunakaran et al. 2003 ), contains multiple copies of the chaperonins GroES and GroEL (see Table 2 ) that appear to be recent duplications. The two GroEL genes found in an operon structure with GroES share more sequence similarity with each other than either does to the third distal GroEL. MCA1202 is part of the mmo operon and was previously identified as a GroEL ( Csaki et al. 2003 ). The genome also contains two sets of ATP synthase genes, as has been found in four other completed genomes ( Chlorobium tepidum, Pirellula [1], and two Listeria spp.), one of which is located at the putative origin of replication. Only one of these genes, that encoding the ATP synthase F 1 epsilon subunit, appears to be recently duplicated (see Table 2 ), and has not been reported to be present in more than one copy in other genomes. This subunit is thought to regulate the H + /ATP ratio ( Jones et al. 1998 ); it is possible that M. capsulatus alters the H + /ATP ratio by two different ATP synthases depending on its growth substrate (methane or glycogen/sugars). Phylogenetic analysis suggests that the other ATP synthase genes are not recently duplicated. Genes of the operon located at the origin of replication (MCA0006–0013) are of a type found only in Gammaproteobacteria or Betaproteobacteria, whereas the nonorigin genes (MCA2699–2708 and MCA1556) are divergent and related to the methanogenic Archaea and C. tepidum . The cooccurrence of the divergent genes in another organism able to fix nitrogen (C. tepidum) suggests that they may be involved in generation of extra ATP required to fix nitrogen. Other redundant genes with divergent phylogenies include those that encode the MetK S-adenosylmethionine synthetase (MCA0450 and MCA0139), which is involved in methionine and selenoamino acid metabolism and has a role in activation of formate dehydrogenase; the GlpG glycogen phosphorylase (MCA0067 and MCA2540), which has a role in starch and sucrose metabolism; and the cell division protein FtsH (MCA0851 and MCA1848), which is a proteolytic regulator of cell division under stress. One of each duplicated pair is most closely related to genes from other Gammaproteobacteria, whereas its partner is either most closely related to genes from cyanobacteria, or occupies a deep-branching position. The genome encodes formylmethanofuran dehydrogenase (MCA2860), an enzyme central to methanogenesis in Archaea. The fact that methanogenesis has not been previously reported in M. capsulatus suggests that this enzyme (along with methenyl-THMPT cyclohydrolase and formylmethanofuran THMPT formyltransferase) is instead functioning in reverse, in THMPT-linked formaldehyde oxidation ( Pomper and Vorholt 2001 ) ( Figure 2 ), as seen in some methylotrophs. Genes encoding subunits A, B, and C are found in an operon structure (MCA2857–2860), and there is a second distal subunit A gene (MCA2319) upstream of pmoCAB, together with ftr, which encodes the previous step in the TMPT pathway ( Figure 2 ), which appears to be a recent duplication (see Table 2 ). Subunits A and C were previously known in M. capsulatus ( Vorholt et al. 1999 ). Other genes similar to those of Archaea include those containing archaeal inteins (described above), His A/His F (involved in histidine biosynthesis in Archaea) (MCA2867), a putative arsenite transporter (MCA0791), and four conserved hypothetical proteins (MCA0196, MCA0197, MCA2834, and MCA2732). Non-homology-based functional prediction. Phylogenetic profiling ( Pellegrini et al. 1999 ; Eisen et al. 2002 ) and comparative analysis of M. capsulatus (Bath) with the incomplete genome data from the methylotroph M. extorquens were used to identify additional novel genes. Phylogenetic profiling identified four genes not previously known to have a role in methane oxidation pathways in M. capsulatus. Two of them (MCA0180 and MCA3022) clustered with the gene that ecodes methylene THF dehydrogenase (transfer of C1 compounds) together with a gene from Pirellula, and the others (MCA0346 and MCA2963) grouped with pmoC3 (oxidation of methane to methanol) and a gene from Nitrosomonas . Specific comparisons with M. extorquens, which possesses a much larger genome (7.6 Mb) than that of M. capsulatus (Bath) ( Chistoserdova et al. 2003 ), revealed shared genetic elements for methylotrophy. Determination of putative orthologous genes shared between M. capsulatus (Bath) and M. extorquens (best hits) yielded a total of 572 genes in 88 role categories. The majority of these shared genes are of unknown function. Putative orthologs detected included 24 conserved hypothetical genes. Phylogenetic profiling showed that ten of the 24 occur in a species distribution similar to proteins of the methane oxidation pathway, suggesting that they may also have a role in methane oxidation. Three of the ten (MCA1278, MCA1279, and MCA2862) were found within methanotrophy gene “islands” (see below), and three had the highest levels of similarity to M. extorquens (MCA1497, MCA1647, and MCA2862) supporting a putative role in the methane oxidation pathway. Of the 89 genes putatively involved in methylotrophy in M. extorquens, we found orthologs of 69, mostly in the categories of energy and carbon metabolism; the remaining 20 genes found in M. extorquens but not M. capsulatus (Bath) are involved in the metabolism of other C1 compounds not used by M. capsulatus. Many (41 of 69) of these shared methylotrophy genes are clustered on the chromosome into 13 groups, seven of which contain more than three genes, and the largest of which contains nine. Three hypothetical proteins were identified in these clusters, suggesting a role in C1 metabolism. Conclusions Our analysis of the M. capsulatus (Bath) genome has illuminated the genomic basis for the highly specialized methanotrophic lifestyle, including redundant pathways involved in methanotrophy and duplicated genes for essential enzymes such as the MMOs. We used phylogenomic analysis, gene order information, and comparative analysis with a partially sequenced methylotroph to detect genes of unknown function likely to be involved in methanotrophy and methylotrophy. Many methylotrophy genes were found to be clustered in gene islands in both organisms. We found genomic evidence for the organism's ability to acquire copper (including a previously unknown NRPS) and to use copper in regulation of methanotrophy, but the exact regulatory mechanisms remain unclear. The genome sequence suggests previously unexpected metabolic flexibility, including the ability to oxidize chemolithotrophic hydrogen and sulfur and to live under reduced oxygen tension, both of which have implications for methanotroph ecology. There is a clear need for experimental validation of these genome-based hypotheses. The availability of the complete genome of M. capsulatus (Bath) deepens our understanding of methanotroph biology, its relationship to global carbon cycles, and its potential for biotechnological applications, and it provides a set of hypotheses of gene function that can now be experimentally tested. In addition, the annotated genome provides a source of gene probes for detection and differentiation of methanotrophs in environmental samples. Materials and Methods Genome sequencing M. capsulatus (Bath) was purchased from National Collection of Industrial and Marine Bacteria (Aberdeen, United Kingdom) as strain NCIMB 11132, and its DNA was isolated as previously described ( Johnson 1994 ). The complete genome sequence was determined using the whole-genome shotgun method ( Venter et al. 1996 ). Clone libraries with insert sizes of 1.8–2.8 kb (small) and 6.5–11 kb (medium) were used for the random shotgun-sequencing phase. Physical and sequencing gaps were closed using a combination of primer walking, generation and sequencing of transposon-tagged libraries of large-insert clones, and multiplex PCR ( Tettelin et al. 1999 ). Sequence assembly was performed using The Institute for Genome Research (TIGR) Assembler ( Sutton et al. 1995 ). Repeats were identified using RepeatFinder ( Volfovsky et al. 2001 ), and sequence and assembly of the repeats were confirmed using medium-insert clones that spanned the repeat. Sequence annotation Identification of putative protein-encoding genes and annotation of the genome were performed as previously described ( Eisen et al. 2002 ). An initial set of open reading frames predicted to encode proteins (also termed CDSs here) was initially identified using GLIMMER ( Salzberg et al. 1998a ). Open reading frames consisting of fewer than 30 codons and those containing overlaps were eliminated. Frame shifts and point mutations were corrected or designated “authentic.” Functional assignment, identification of membrane-spanning domains, determination of paralogous gene families, and identification of regions of unusual nucleotide composition were performed as previously described ( Tettelin et al. 2001 ). Phylogenomic analysis ( Eisen 1998a , 1998b ; Eisen and Fraser 2003 ) was used to assist with functional predictions. Initially, all putative M. capsulatus (Bath) proteins were analyzed using the Automated Phylogenetic Inference System (J. H. Badger, personal communication, 2003). This system automates the process of sequence similarity, alignment, and phylogenetic inference for each protein in a genome. Sequence alignments and phylogenetic trees were refined using the methods described previously ( Salzberg et al. 2001 ; Wu et al. 2004 ). Comparative genomics Proteins were searched by BLASTP ( Altschul et al. 1990 ) against the predicted proteomes of published complete organismal genomes and a set of complete plastid, mitochondrial, plasmid, and viral genomes. The results of these searches were used (a) for phylogenetic profile analysis ( Pellegrini et al. 1999 ; Eisen and Wu 2002 ), (b) to identify putative lineage-specific duplications (proteins showing the highest E-value scores in pairwise comparison to another protein from M. capsulatus [Bath]), and (c) to determine the presence of homologs in different species. Orthologs between the M. capsulatus (Bath) genome and that of M. extorquens were identified by requiring mutual best-hit relationships (E-values less than 10 –15 ) among all possible pairwise BLASTP comparisons, with some manual corrections. A total of 89 genes involved in methylotrophy in M. extorquens ( Chistoserdova et al. 2003 ) were obtained from GenBank and used in a BLASTP search against M. capsulatus (Bath) and M. extorquens . Comparative genome analyses were also performed using the Comprehensive Microbial Resource ( Peterson et al. 2001 ). Identification of prophage regions Putative prophage regions were defined as containing genes that encode proteins bearing sequence similarity to known phage or prophage proteins. We are using “prophage” to refer to sequences with similarity to lysogenic bacteriophages that have not been experimentally demonstrated to form infectious particles. Additional supporting information included the presence of direct repeats representing the att core of the putative prophage (identified using MUMmer [ Kurtz et al. 2004 ]), the conserved late-gene operon responsible for packaging and head morphogenesis of tailed dsDNA bacteriophages ( Duda et al. 1995 ), and, in the case of bacteriophage Mu-like phages, conserved gene order (putative phage repressor, transposase A and B subunits, and a Mu-like mom DNA methyltransferase) demarking the 5′ and 3′ boundaries of the region ( Morgan et al. 2002 ). Best matches were determined by searching a custom database containing 14,585 total amino acid sequences from 185 published completed bacteriophage genomes, one TIGR unpublished completed bacteriophage genome, five published incomplete bacteriophage genomes, 54 published prophage genomes, and 18 TIGR unpublished putative prophage genomes, for a total of 258 unique phage or prophage entries. WU-BLASTP version 2.0 ( Altschul et al. 1990 ) was implemented through an in-house modification of the Condor parallel search tool ( Litzkow et al. 1988 ), reporting only those hits having E-values less than or equal to 10 –6 . In-house Perl and Linux shell scripts were used to identify the best hit (lowest E-value) per protein sequence query. Supporting Information Accession Number The GenBank ( http://www.ncbi.nlm.nih.gov/Genbank/ ) accession number for the Methylococcus capsulatus genome discussed in this paper is AE017282. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC517821.xml |
515373 | Damage Response Protein Buys Time for Bacterial DNA Repair | null | It is often said that after a nuclear catastrophe, cockroaches will inherit the earth, because they are so resistant to the harmful effects of ionizing radiation. But should the unthinkable come to pass, the bacterium Deinococcus radiodurans is likely to outlast even the cockroach. Its ability to endure radiation is truly impressive: it can withstand a dose a thousand times that which will kill a human. How it accomplishes this phenomenal feat of survival is the subject of a study in this issue by John Battista and colleagues at Louisiana State University in Baton Rouge and at the University of Wisconsin at Madison. D. radiodurans R1 While radiation damages many cellular components, it is the fracturing of the cell's DNA that is the most harmful. DNA breaks can be repaired, but in doing so, the cell is racing against time. The exposed free ends of the DNA invite digestion by the cell's own enzymes, called exonucleases. If the DNA is not stitched back together quickly enough, the exonucleases will degrade it past the point of repair, and the cell will ultimately succumb. Large doses of radiation can fracture a chromosome in thousands of places, far in excess of the repair ability of most cells. D. radiodurans , however, largely prevents exonuclease digestion, an ability which has previously been shown to be linked to the activity of a gene with the rather uninformative name of DR0423. But how, exactly, does this gene accomplish this life-saving feat? To answer this question, Battista and colleagues first showed that, following radiation exposure, DR0423 was upregulated 20- to 30-fold, and that deletion of the gene renders D. radiodurans susceptible to ionizing radiation. Together, these results clearly indicate that the DR0423 gene product is critical for protecting the bacterium. Based on this, they dubbed the gene ddrA , for “DNA damage response.” They also found that DdrA, the protein encoded by ddrA , binds to single-stranded fragments of DNA, exactly like those found at the broken ends of the DNA double helix when damaged by radiation. Finally, they showed that when DdrA bound to these broken ends, they were protected from digestion by exonucleases. An important question about this system is what it is actually good for. Since the level of radiation tolerated by the bacterium is found nowhere on earth, of what use is such an efficient DNA protection system? The answer might be that it also protects D. radiodurans from the effects of desiccation, a condition much more common in the life of a bacterium, and one which also induces widespread DNA damage. While ddrA cannot prevent the damage, it can preserve the DNA from degradation until conditions once again allow the bacterium to function, and repair its DNA. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC515373.xml |
529273 | Improving the accuracy of malaria-related laboratory tests in Ghana | Background Inaccurate malaria results can lead to patient mismanagement, misperceptions about malaria resistance patterns and public health misinformation. All laboratories need to be able to demonstrate that their results are accurate. Establishing and maintaining a system for monitoring test accuracy is a complex, expensive and technically demanding process, which very few poor countries have been able to implement. This study described the process and assessed the feasibility of establishing a nation-wide system for improving the accuracy of malaria-related tests in peripheral laboratories in Ghana. Programme implementation A baseline survey of all 693 laboratory staff in 205 sub-regional government and mission health laboratories in Ghana was conducted by a national network of laboratory supervisors. Survey results guided a training programme to improve test accuracy. Outcomes included changes in the quality of laboratory tests and the system was considered to be feasible if >50% of laboratory staff in each region received training and if test accuracy could be documented. Programme indicators 74% (mean) of the 693 laboratory staff were assistants with no professional qualifications. There were marked differences between regions in the availability of essential resources for malaria diagnosis (e.g. microscopes). 93% of laboratory staff received training; in six months there were increases of 11% and 7% respectively in the number of laboratories producing haemoglobin and malaria microscopy results of acceptable quality. Conclusions It is possible to establish a system for improving and monitoring test accuracy in peripheral laboratories on a country-wide basis in a developing country using a model that could be adapted for use in other countries and for other components of health care provision. | Background Clinical laboratory services are a critical component of health systems. They are essential for patient management and for providing accurate public health data including early detection of malaria resistance. In many poorer countries, laboratory services have been neglected due to chronic under-investment. The emergence of drug-resistant malaria and the extra burden of supporting diagnosis and treatment of HIV/AIDS has increased the strain on laboratories. The widespread resistance of malaria to cheap, antimalarial drugs such as chloroquine and the increasing use of relatively expensive combination therapy, means that presumptive treatment of all fevers as malaria may no longer be a sustainable option for some countries [ 1 ]. It is, therefore, imperative that malaria diagnosis at peripheral level health facilities, where the greatest burden of malaria health care provision is focused, is accurate. Establishing and maintaining an accurate and reliable laboratory service is a complex, expensive and technically demanding process, which very few poor countries have been able to implement. It depends on good laboratory management to oversee processes such as documentation, audit cycles, quality assurance and external validation, safety practices, and supervisory and accountability structures [ 2 ] and should be combined with improvements in clinical practice. Sub-standard laboratory services waste public and individuals' resources and result in clinical mismanagement and inaccurate health information. They also generate a culture of mistrust and communication breakdown between laboratory and clinical staff which contributes to low morale within the technical profession. Poor quality laboratory services have the greatest impact on the poorest people who use the service because they have the largest burden of ill-health [ 3 ]. To our knowledge there is no comprehensive nationwide system for monitoring the accuracy of malaria-related laboratory tests (i.e. malaria microscopy, haemoglobin estimation, tests associated with blood transfusion) currently operational in any sub-Saharan country. In January 2000, the Ministry of Health in Ghana commenced a two-year programme to determine the feasibility of establishing a nationwide quality assurance system for common tests performed at peripheral (district and sub-district) laboratories. This programme was specifically designed to complement the Ministry's 5-year Programme of Work [ 4 ]. In Ghana, laboratory services exist at most levels of health facility, except smaller health centres. Ghana's decentralization policy means that regional and district health managers are responsible for delivering and evaluating health care provision, including laboratory services. Prior to this programme there were no local or national quality assurance systems and no functional supervisory network for laboratories in Ghana. This feasibility programme was built on established laboratory management structures and locally available resources and was implemented by a national network of senior laboratory technicians. The aim of the programme was to determine the feasibility of establishing a nationwide system for improving the accuracy of malaria and other common laboratory tests. All staff performing laboratory tests in all public sector peripheral laboratories in Ghana were included in the programme irrespective of their grade. Programme implementation Establishing a national network of laboratory 'supervisors' Two senior technicians from each of Ghana's ten administrative regions were chosen by the Ministry of Health to constitute the national network of laboratory supervisors who would implement the programme. They were selected on the basis of their geographic location, seniority and commitment to improving laboratory services. All the supervisors had technical qualifications (2–3 years certificate or diploma course), but none had any higher technical or educational qualifications. Programme objective and baseline survey A workshop for the supervisors was held in Ghana at the beginning of the programme to define the objectives and to develop a workplan, timetable and monitoring processes. The agreed programme objective was to establish a system in all peripheral government laboratories in Ghana, for monitoring the accuracy of results of malaria microscopy, haemoglobin estimation and other commonly performed tests. The ability to train over 50% of laboratory staff in each of the ten regions and to monitor changes in the accuracy of results, were used as indicators of programme feasibility. As very little information was available about the state of Ghana's laboratories, supervisors initially carried out a nationwide baseline survey of all Ghana's peripheral laboratory facilities. They personally visited every government and mission laboratory in Ghana and collected first hand information about staff, equipment and tests offered. For each laboratory they also completed a safety checklist based on the World Health Organization's recommendations for laboratory safety [ 5 ]. Programme design and methods Programme planning and training Ideally training should be targeted towards the tests which are performed most poorly. However, as there were no monitoring systems in place in Ghana for laboratory tests and none of Ghana's laboratory staff had had any experience of these systems, it was not possible initially to identify the worst performed tests. The supervisors therefore started by providing training on seven of the most commonly performed tests. They used Ghana's Standard Laboratory Operating Procedures [ 6 ], which describe nationally standardized methods for individual tests, as their core teaching manual. Their choice of teaching methods, predominantly combinations of workshops and workplace training, varied between regions depending on local resources, needs and geography. Monitoring test accuracy Quality assurance systems used in industrialised countries and published in the literature are not appropriate for rural laboratories in poor countries such as Ghana. They are too complex for a workforce that is primarily made up of laboratory assistants and they are based on assumptions that the methods are generally automated and communication and transport networks are reliable. The supervisors therefore had to devise workable methods for externally monitoring test results from the peripheral laboratories. They distributed samples with known values to peripheral laboratories who processed them under normal working conditions. Once the methods for distributing, preserving and measuring test accuracy had been optimized for local use they were piloted in a small number of laboratories, before being introduced throughout each region. The ways in which the test accuracy was determined varied for each test. For example, for haemoglobin estimations a whole blood sample with known value (determined by repeated measurements in a regional laboratory) was distributed to peripheral laboratories and supervisors decided that laboratories with results outside the target value ±10% would receive priority for training. As the quality of results from peripheral laboratories improved, the supervisors reduced the acceptable range of results to ±5% of the target value. For malaria microscopy, the target malaria result of a whole blood sample or malaria smear was determined by consensus of several technicians from the regional hospital. Results from peripheral laboratories were considered 'accurate' if laboratories reported the presence (or absence) of malaria parasites and quantified them to within one grade (on a grading system of 0,+,++,+++) of the target result. For sickle cell tests a blood sample of know sickle status, determined by the regional laboratory, was distributed. Accurate results were those that correctly identified the sample as containing sickle haemoglobin or not. If the majority of results from peripheral laboratories for a single test varied from the target result, the supervisors re-evaluated their test target values. Whole blood samples were distributed in order to check the complete process of testing (e.g. pipetting accuracy, malaria smear preparation) rather than just an individual component. Supervisors used results of the quality monitoring to provide constructive feedback to the laboratory staff and to target their training towards tests and laboratories that performed particularly poorly. Programme funding Initially the programme was funded directly from Ministry of Health headquarters but subsequently, through collaboration with regional Ministry of Health administrators, several supervisors were able to access their own regional funds. Programme indicators Baseline survey There were 205 laboratories in Ghana located in regional and district government hospitals (97), mission hospitals (53) or health centres (54). The total staff complement in these laboratories was 693. The percentage of staff in each of the ten regions with no professional qualifications (assistants or bench-trained with 6 weeks to 1 year training) varied from 60–83% (mean 74%). Even after correcting for differences in regional populations and excluding regions with teaching hospitals, there were wide variations between regions in the number of laboratories and the availability of essential laboratory equipment such as microscopes, (Table 1 ). The safety survey showed that over 50% of 62 laboratories surveyed lacked essential items such as automatic pipettes (necessitating mouth-pipetting) (74%), protocols for waste disposal and equipment maintenance (82% and 100% respectively), first aid kits (100%) and fire safety equipment (94%). Other unsafe practices included allowing patients inside the laboratory (89%) and eating and drinking within the laboratory (55%). Table 1 Regional variations in laboratory resources corrected for population (year 2000) (excluding two regions with teaching hospitals) Region 2 3 4 6 7 8 9 10 Mean Range Population (millions) 2.10 1.55 2.65 2.10 1.25 0.70 1.80 1.85 Laboratories Total 16 15 40 7 26 9 57 32 per 100,000 pop. 0.76 0.97 1.51 0.33 2.08 1.29 3.17 1.73 1.48 0.33–3.17 Trained staff Total 13 9 21 11 11 7 19 24 per 100,000 pop. 0.62 0.58 0.79 0.52 0.88 1.00 1.06 1.30 0.85 0.52–1.30 Microscopes Total 30 20 51 NI 16 6 76 20 per 100,000 pop. 1.40 1.29 1.92 NI 1.28 0.86 4.22 1.08 1.72 0.86–4.22 Colorimeters Total 17 8 46 6 6 5 14 14 per 100,000 pop. 0.81 0.52 1.74 0.29 0.48 0.71 0.78 0.76 1.22 0.52–1.74 Centrifuges Total 20 15 46 12 6 6 54 21 per 100,000 pop. 0.95 7.42 1.74 0.57 0.48 0.86 3.00 1.14 2.02 0.48–7.42 Haematocrit centrifuge Total 9 4 39 7 8 2 5 10 per 100,000 pop. 0.43 0.23 1.47 0.33 0.64 0.29 0.28 0.54 0.53 0.23–1.47 Spectrophotometer Total 5 3 13 2 5 2 3 2 per 100,000 pop. 0.24 0.19 0.49 0.95 0.40 0.29 0.17 0.11 0.39 0.11–0.95 Blood mixer Total 5 3 20 0 0 1 10 0 per 100,000 pop. 0.24 0.19 0.75 0 0 0.14 0.56 0 0.24 0–0.75 NI = no information Extent of training and impact on test accuracy After 18 months, a mean of 93% (regional variation 58%–100%) of all Ghana's laboratory staff had been trained in malaria-related and other common tests. During the final six months of the programme the supervisors monitored the quality of haemoglobin estimations, sickle cell test and malaria microscopy results in 48% of all Ghana's 205 laboratories. In 4 regions the quality of up to 11 tests had been monitored. Tests that gave quantitative results were consistently the most poorly performed tests in all regions with only 78%, 78% and 84% of laboratories producing acceptable results for haemoglobin measurements, white blood counts and malaria microscopy respectively. Tests for HIV, hepatitis B, sickle-cell screen and ZN stain were consistently performed well with over 95% of laboratories meeting agreed target results. After a further six months training there were improvements in the accuracy of several tests, particularly in haemoglobin estimation and malaria microscopy with 89%, 83% and 91% of laboratories producing acceptable results for these tests respectively, (Table 2 ). Table 2 Changes in test accuracy in six months Mean number of laboratories with accurate results/total surveyed (%) Test Month 1 Month 6 Haemoglobin 45/58 (78) 49/58 (89) Malaria microscopy 49/58 (84) 49/54 (91) Sickle screen 53/53 (100) 51/51 (100) White blood count 31/40 (78) 40/48 (83) Blood group 23/24 (96) 57/58 (98) Cross match 7/7 (100) 7/7 (100) HIV test 27/28 (96) 28/28 (100) Hepatitis B antigen 22/22 (100) 40/40 (100) ZN stain 38/39 (97) 39/39 (100) Discussion There is very little information available about the state and quality of laboratory services in peripheral health facilities in poorer countries. The baseline survey showed that in Ghana three quarters of public sector laboratory staff do not have any technical qualifications, and even the supervisory cadre only have Diplomas. Through this study we have shown that the lack of professional qualifications amongst the laboratory workforce and the inequitable distribution of laboratory equipment, are not obstacles to establishing a simple and far-reaching quality assurance process. This study has shown that the worst performed tests at sub-regional level are those that generate quantitative or subjective results (such as haemoglobin estimation and malaria microscopy) rather than simple 'positive' or 'negative' results (such as HIV and hepatitis B screening tests). Further research is needed to examine the cost-effectiveness of potentially more accurate or less complex tests (e.g. malaria rapid diagnostic tests, HemoCue). Although these tests may be more expensive than those in use in most districts, their simplicity and accuracy may save downstream costs. One of the major limitations of this study is the lack of quality assurance at regional and supra-regional level for common laboratory tests. These were the laboratories responsible for determining the target values of tests distributed to peripheral laboratories. To overcome this problem, regional laboratories tested the samples several times before determining the target values and they also exchanged samples with each other as an inter-regional check on results. Plans are in place to expand this into an inclusive national quality monitoring system for higher level laboratories. We have confirmed previous work showing that at peripheral health facilities haemoglobin using manual methods is the most inaccurate test [ 7 ]. Safety issues are often not a priority for laboratories in poorer countries but overcrowding and poor laboratory organisation are recognized to be associated with a significant risk of accidents and consequent infection of health workers. Many of the safety issues identified through this project could be easily and cheaply rectified but because there was no supervisory system in place they had been ignored. To our knowledge this is the first external laboratory quality assurance programme in sub-Saharan Africa to demonstrate that it is feasible to achieve widespread coverage of peripheral health laboratories and staff. Through this programme an effective and practical model has been developed for improving the accuracy of common laboratory tests in Ghana. The model is based on a national network of senior technicians who implemented and supervised a continuous cycle of test monitoring and targeted training for all laboratory staff in their regions. The programme's success was dependent on the quality and commitment of the national network of laboratory supervisors. Participation in the programme was itself a strong motivational force for the supervisors. Because of the supportive attitude of the supervisors, the laboratory staff viewed the monitoring visits as educational rather than punitive and this ensured cooperation when remedial action was required. Health planners can use the lessons learnt from this programme to introduce measures to improve morale and job satisfaction of key health workers especially those in the neglected laboratory services. Components of this model have already been adopted by other countries for their laboratory programmes and could be adapted for other health disciplines in poorer countries. Several steps are necessary before this model could be sustainably implemented. The network of supervisors needs to have pro-active long-term support from all stakeholders, combined with career and promotion packages. Health managers need to provide secure funding for laboratory quality assurance programmes at all levels. This includes establishing internal quality control measures and external validation systems that are linked into national and international external quality monitoring schemes. To be successful, this model will require prioritization of the laboratory service by policy makers at national and supra-national level and adequate representation of the laboratories in decision-making processes. Implementing a quality assurance system for laboratories in poorer countries is expensive and logistically complicated. Managers will need to balance cost against quality, taking into account that it is those who are most vulnerable to ill health who can least afford to bear the brunt of the consequences of inaccurate laboratory results. Authors' contributions IB and VB designed the outline framework for the programme. IB documented and collated results. VB was the local programme coordinator. AAA provided an overview and ensured the programme was compatible with Ministry of Health policies. All authors read and approved the final manuscript. Funding sources This study was supported by the Ministry of Health in Ghana and the Malaria Knowledge Programme (funded by the Department for International Development, UK) at the Liverpool School of Tropical Medicine. The Department for International Development had no role in study design, the collection, analysis and interpretation of data, in the writing of the report or in the decision to submit the paper for publication and accepts no responsibility for the information or views expressed. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529273.xml |
555757 | Laparoscopic cholecystectomy in situs inversus totalis: a case report | Background Laparoscopic cholecystectomy is one of the commonest surgical procedures carried out in the world today. Occasionally patients present with undiagnosed situs inversus and acute cholecystitis. We discuss one such case and outline how the diagnosis was made and the pitfalls encountered during surgery and how they were overcome. Case presentation A 32 year old female presented to our department with epigastric pain radiating through to the back. A diagnosis of acute cholecystitis in a patient with situs inversus totalis was made following clinical examination and radiological investigation. Laparoscopic cholecystectomy was subsequently performed and the patient made an uneventful recovery. Conclusion Situs inversus presenting with acute cholecystitis is very rare. The surgeon must appreciate that care should be taken to set up the operating theatre in the mirror image of the normal set-up for cholecystectomy, and that right handed surgeons must modify their technique to adapt to the mirror image anatomy. | Background In 1600 the first known case of situs inversus in humans was reported by Fabricius [ 1 ]. The incidence is thought to be in the region of 1:5000 to 1:20000 [ 2 ]. The condition may affect the thoracic organs, abdominal organs or both. It is associated with a number of other conditions such as Kartagener's (bronchiectasis, sinusitis, situs inversus) and cardiac anomalies. There is no current evidence that situs inversus predisposes to cholelithiasis [ 3 ]. Since Mouret first performed it in 1987, laparoscopic cholecystectomy has become the standard operative procedure for gallbladder disease. It is associated with reduced hospital stay, fewer respiratory complications, less pain and a faster return to work. Case presentation A thirty two year old female was admitted with a three hour history of epigastric pain radiating into her back in keeping with biliary colic. She had vomited a number of times. In the previous week she had two episodes of a similar nature. On examination there was no jaundice or pyrexia. The apex beat was in the right fifth intercostal space, midclavicular line. She had epigastric tenderness but was not tender in the right or left upper quadrants. Her white cell count and amylase level was normal but her C-reactive protein level (CRP) was elevated at 290 mg/L. An electrocardiograph showed right axis deviation and right ventricular hypertrophy, in keeping with dextrocardia. An ultrasound scan of the upper abdomen identified the gallbladder, which contained stones, in the left upper quadrant. The spleen was visualised in the right upper quadrant. There was no evidence of common bile duct or intrahepatic duct dilatation. Chest X-Ray confirmed the clinical and electrocardiograph diagnosis of dextrocardia. The diagnosis of acute cholecystitis and situs inversus was made. The patient settled clinically over two to three days and was discharged home to be admitted electively for laparoscopic cholecystectomy. In order to conduct the laparoscopic cholecystectomy all theatre equipment including diathermy, monitors and CO 2 insufflator were positioned in the mirror image of their normal position. The surgical team also changed sides with the primary surgeon and first assistant on the patients right and the second assistant on the left. The ports were inserted in the usual way but on the left side. At laparoscopy the entirety of the abdominal contents were indeed reversed. The main difficulty encountered was that the primary surgeon, who was right handed, would have had to cross hands to retract on Hartmann's pouch while dissecting Calot's triangle. We overcame this difficulty by allowing the first assistant to retract on Hartmann's pouch, while the primary surgeon dissected Calot's triangle using his right hand via the epigastric port without hindrance. The common bile duct and cystic duct were identified, as was the cystic artery, which lay anterior to the cystic duct. The surgery proceeded without incident and the patient recovered and was discharged the next day. Conclusion In this case the patient presented with epigastric pain only and had no definite left upper quadrant pain. It has been noted in 30% of previous reported cases of acute cholecystitis in patients with situs inversus that the pain was felt in the epigastrium alone and in 10% the pain was localised to the right upper quadrant [ 4 ]. The proposed explanation for this is that the central nervous system may not share in the general transposition. Previous reports have confirmed that situs inversus is not a contraindication for laparoscopic cholecystectomy [ 1 - 3 , 5 , 6 ]. The procedure is, however, more difficult and care and time must be taken to re-arrange the equipment set-up in theatre, and to recognise the mirror-image anatomy which can cause difficulties with orientation. At least two thirds of surgeons are right handed. It is necessary for these surgeons, and their assistants, to modify their usual surgical technique to comfortably and safely carry out the procedure. Rather than the clumsy crossing of hands to retract on Hartmann's pouch for dissection of Calot's triangle, we suggest that retraction on Hartmann's pouch may be carried out by the assistant, thus allowing the surgeon to operate in a more ergodynamic fashion. Learning points 1. Detailed clinical examination is important in diagnosing previously unknown situs inversus. 2. Patients with gallabladder disease and situs inversus may have pain in the right upper quadrant, epigastrium or left upper quadrant. 3. Theatre equipment must be moved to the mirror image of their normal positions before surgery. 4. The surgeon must recognise the mirror image anatomy and modify his or her technique appropriately. Abbreviations CRP- C reactive protein. CO 2 - Carbon dioxide. Competing interests The author(s) declare that they have no competing interests. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555757.xml |
387264 | Mimotopes for Alloreactive and Conventional T Cells in a Peptide–MHC Display Library | The use of peptide libraries for the identification and characterization of T cell antigen peptide epitopes and mimotopes has been hampered by the need to form complexes between the peptides and an appropriate MHC molecule in order to construct a complete T cell ligand. We have developed a baculovirus-based peptide library method in which the sequence encoding the peptide is embedded within the genes for the MHC molecule in the viral DNA, such that insect cells infected with virus encoding a library of different peptides each displays a unique peptide–MHC complex on its surface. We have fished in such a library with two different fluorescent soluble T cell receptors (TCRs), one highly peptide specific and the other broadly allo-MHC specific and hypothesized to be much less focused on the peptide portion of the ligand. A single peptide sequence was selected by the former αβTCR that, not unexpectedly, was highly related to the immunizing peptide. As hypothesized, the other αβTCR selected a large family of peptides, related only by a similarity to the immunizing peptide at the p5 position. These findings have implications for the relative importance of peptide and MHC in TCR ligand recognition. This display method has broad applications in T cell epitope identification and manipulation and should be useful in general in studying interactions between complex proteins. | Introduction The identification of peptide epitopes associated with particular αβ T cell receptors (αβTCRs) is often still a bottleneck in studying T cells and their antigenic targets in, for example, autoimmunity, hypersensitivity, and cancer. A direct genetic or biochemical attack on this problem can be successful, especially with class I major histocompatibility complex (MHCI)-presented peptides. For example, tumor ( Van Der Bruggen et al. 2002 ) and transplantation ( Scott et al. 2000 ; Simpson et al. 2001 ; Shastri et al. 2002 ; Sahara and Shastri 2003 ) peptide epitopes have been found this way. Identification of the antigenic peptide in a mix of peptides stripped from MHC molecules isolated from antigen-presenting cells (APCs) has sometimes been possible using a combination of a biological assay, peptide fractionation, and peptide sequencing ( Guimezanes et al. 2001 ). However, this method is extremely labor intensive and depends on relatively high peptide frequency in the mix and a very sensitive bioassay. These conditions are not always achievable, especially with peptides presented by MHCII, in which peptide loading of surface MHC may require peptide concentrations orders of magnitude higher than those required for MHCI loading. The reward for the labor involved in identifying peptide epitopes directly can often be the identification of the protein source of the peptide, especially as the sequencing of the genomes of many organisms approaches completion. However, in many situations, rather than identifying this precise peptide epitope, it is sufficient to identify a peptide “mimotope.” Mimotopes can be defined as peptides that are different in sequence from the actual peptide recognized in vivo, but that are nevertheless capable of binding to the appropriate MHC molecule to form a ligand that can be recognized by the αβTCR in question. These peptides can be very useful for studying the T cell in vitro, for altering the immunological state of the T cell in vivo ( Hogquist et al. 1994 ), for vaccine development ( Partidos 2000 ), and potentially in preparing multimeric fluorescent peptide–MHC complexes for tracking T cells in vivo ( You et al. 2003 ). Mimotopes can sometimes be identified in randomized peptide libraries that can be screened for presentation by a particular MHC molecule to the relevant T cell ( Gavin et al. 1994 ; Linnemann et al. 2001 ; Sung et al. 2002 ; reviewed in Hiemstra et al. 2000 ; Liu et al. 2003 ). Thus far, strategies for screening these types of libraries have involved testing individual pools of peptides from the library and then either deduction of the mimotope sequence from the pattern of responses or sequential reduction in the size of the pool until a single peptide emerges. There are several limitations to this type of approach. Again, a very sensitive T cell bioassay is needed in which the activity of the correct stimulating peptide is not masked by competition with the other peptides in the pool. Also, an APC that expresses the relevant MHC molecule, but not the relevant peptide, must be found or constructed. Finally, because the screen relies on T cell stimulation, only agonist mimotope peptides are identified. In other applications, another powerful library method has been sequential enrichment/expansion of a displayed library of protein–peptide variants by direct ligand–receptor binding, e.g., using bacterial phage or yeast (also reviewed in Liu et al. 2003 ). These methods have not yet been developed for the routine identification of T cell antigen mimotopes, because of the lack of a suitable system for the display of peptide–MHCs or for screening via αβTCR binding using these organisms. In this paper, we describe such a method using modifications of previously described systems for producing soluble peptide–MHC complexes ( Kozono et al. 1994 ; Crawford et al. 1998 ; Rees et al. 1999 ) and αβTCRs ( Kappler et al. 1994 ) from baculovirus-infected insect cells. We constructed a library of peptides displayed on the surface of baculovirus-infected cells bound to the mouse MHCII molecule, IA b . The peptides in the library varied in five peptide amino acids known to be surface exposed and predicted to lie within the footprint of αβTCR interaction. Using fluorescent αβTCRs as probes, we have identified in the library mimotopes for two types of T cells, both originally produced by immunization of mice with the same IA b –peptide combination. One of these T cells was predicted from previous data ( Liu et al. 2002 ) to be very dependent on all of the peptide surface exposed amino acids. Consistent with these predictions, a single peptide mimotope was identified in the library for this T cell. The sequence of this peptide was highly related to the immunizing peptide. In contrast, the other T cell was hypothesized to be very peptide promiscuous ( Marrack et al. 2001 ) based on its broad allo-MHC reactivity. Consistent with this hypothesis, its αβTCR selected a large family of peptide mimotopes from the library. Comparison of these peptides indicated that attention of this αβTCR was focused primarily on a single position in the peptide. Results Characteristics of a Broadly Alloreactive and Conventional T Cell For this study we selected two T cell hybridomas, both prepared from IA b mice immunized with the peptide p3K. This peptide binds well to IA b ( Rees et al. 1999 ), and its crystal structure bound to IA b has been determined ( Liu et al. 2002 ) ( Figure 1 A). The hybridoma B3K-06 was produced from wild-type C57BL/6 immunized conventionally with the peptide ( Rees et al. 1999 ). Like most T cells resulting from immunization with a foreign peptide, it responds to IA b -expressing APCs in the presence, but not the absence, of p3K ( Figure1 B). It does not respond to APCs expressing other alleles of the IA MHCII molecule (data not shown). Also, as is commonly seen with conventional T cells, the interaction of the αβTCR of B3K-06 with IA b -p3K is very sensitive to changes in any of the peptide amino acids exposed on the surface of the IA b -p3K complex. Mutation of Q2, K3, K5, N7, or K8 to alanine virtually eliminates recognition of p3K by B3K-06 ( Liu et al. 2002 ; see Figure 1 B). Figure 1 Structure of IA b -p3K and Properties of T Cell Hybridomas Reactive to It (A) Ribbon structure of the α1 and β1 domains of IA b with a wire-frame representation of the bound p3K peptide ( Liu et al. 2002 ). Amino acids labeled in red are the five central peptide amino acids available for αβTCR interaction. (B) The figure shows the response of 10 5 B3K-06 hybridoma cells to various peptides presented by 10 5 IA b -bearing APCs, LB-15.13. (C) The figure shows the response of the T cell hybridoma YAe-62 to various MHCII molecules. In each case, 10 5 hybridoma cells were incubated overnight with MHCII presented in various ways. For IA b -p3K, soluble IA b -p3K was immobilized in the culture well before the addition of the hybridoma cells. In other cases, 10 6 spleen cells were used directly as APCs without additional peptide antigen. For pEα, the spleen cells came from IA b -pEα/ΔIAβ/ΔIi mice ( Ignatowicz et al. 1996 ). For wild-type IA b and allo-MHCII, the spleen cells came from H-2 congenic mice on the C57BL/10 background. Finally, spleen cells from ΔIAβ/ΔIi C57BL/6 mice were used. The hybridoma YAe-62 was chosen as a representative of broadly allo-reactive T cells present in mice carrying transgenes and gene knockouts that lead to expression of MHCII almost completely occupied by a single peptide ( Ignatowicz et al. 1996 ). It was produced from IA b -p3K-immunized mice that express the IA b molecule covalently linked to pEα, a dominant IA b -binding peptide derived from the MHCII IEα chain. Its properties are shown in Figure 1 C. YAe-62 responds to APCs bearing IA b -p3K, but not to APCs lacking MHCII nor to IA b -pEα-bearing APCs from the mouse from which the hybridoma was derived. However, YAe-62 has additional reactivities common to many T cells isolated from these mice ( Ignatowicz et al. 1996 ). In the absence of any added peptide, it also responds to all APCs expressing wild-type IA b , including those from mice with a much reduced MHCII peptide repertoire due to lack of the invariant chain. YAe-62 also responds well to APCs from a variety of mice carrying other alleles of IA. We have postulated that these T cells are focused on structural features of the MHCII molecule and are minimally dependent on direct peptide interaction ( Marrack et al. 2001 ). Display of Functional Peptide–MHC on Baculovirus-Infected Insect Cells We previously established methods that used baculovirus-infected insect cells to produce soluble MHC molecules with covalently bound antigenic peptides ( Kozono et al. 1994 ; Crawford et al. 1998 ; Rees et al. 1999 ). These constructions were the starting point for developing insect cells displaying functional peptide–MHCIIs. Several modifications were made to constructs that encoded the mouse MHCII molecule, IA b , with various bound peptides. First, to increase the stability of the molecule, an acid–base leucine zipper ( O'Shea et al. 1993 ) was attached to the C-termini of the extracellular portions of the MHC α and β chains, replacing what would normally be the transmembrane regions of these proteins. The basic half of the zipper was attached to the α chain ( Figure 2 A) and the acidic half to the β chain ( Figure 2 B). In addition, sequence encoding the transmembrane and cytoplasmic tail of the baculovirus major coat glycoprotein, gp64, was attached to the end of the acid zipper ( Figure 2 B). Sf9 insect cells infected with virus encoding this construction produced the MHCII molecule at a high level anchored on the cell surface via the gp64 transmembrane ( Figure 3 A). Also, to make Sf9 cells better APCs ( Cai et al. 1996 ), we established a version transfected with the genes for mouse ICAM and B7.1 ( Figure 3 B). When we tested the ability of Sf9 cells displaying the IA b -p3K complex to present the antigen to B3K-06 or YAe-62, the presence of ICAM/B7.1 greatly improved IL-2 production ( Figure 3 C). These results showed that IA b -p3K could be displayed on the surface of insect cells in a form easily recognized by T cells. In all of the experiments described below, infected conventional Sf9 cells were used for flow cytometry and infected ICAM/B7.1-expressing Sf9 cells were used in IL-2 stimulation assays. Figure 2 Constructions Used in These Experiments (A and B) Previously described constructions ( Rees et al. 1999 ) for the coexpression in a single baculovirus of soluble version of the α (A) and β (B) chains of IA b were modified as described in the Materials and Methods to anchor the molecule on the surface of infected insect cells. (C) The construction was further modified as described in the Materials and Methods to disrupt the IA b β chain with sequence encoding enhanced GFP flanked by sites for the enzymes SbfI and CeuI. (D and E) A degenerate DNA fragment was produced by PCR (D) and cloned into the construct replacing the GFP-encoding sequence (E) as described in the Materials and Methods . Figure 3 Functional Display of IA b -p3K on the Surface of Insect Cells (A) Sf9 insect cells were infected with baculovirus encoding a membrane-bound form of IA b -p3K. After 3 d, the surface expression of IA b -p3K was detected with an anti-IA b mAb using flow cytometry. (B) The genes for mouse ICAM (CD54) and B7.1 (CD80) were cloned into an insect cell expression plasmid as described in the Materials and Methods . The plasmids were used to cotransfect Sf9 cells, and a stable transfectant (Sf9-ICAM/B7.1) was cloned expressing both proteins detected with mAbs using flow cytometry. (C) Either Sf9 (open bars) or Sf9-ICAM/B7.1 (closed bars) cells were infected with baculovirus expressing IA b -p3K. After 3 d, the infected insect cells were used as APCs to stimulate IL-2 production from B3K-06 and YAe-62. Uninfected cells were used as negative controls. Detection of Displayed Peptide–MHC with Multimeric αβTCR Next we prepared fluorescent, soluble αβTCR reagents for use in flow cytometry to detect insect cells displaying the appropriate peptide–MHCII combination. Fluorescent multivalent versions of the soluble αβTCRs of B3K-06 and YAe-62 bound to insect cells displaying the IA b -p3K, but not a control peptide–MHCII combination ( Figure 4 A). Figure 4 Detection of IA b -p3K-Expressing Insect Cells with Polyvalent, Fluorescent αβTCRs (A) Sf9 insect cells were infected with baculovirus encoding IA b bound either to p3K (filled histogram) or a control peptide (FEAPVAAALHAV) (unfilled histogram). After 3 d, the infected insect cells were incubated with polyvalent, fluorescent soluble αβTCRs from B3K-06 or YAe-62. The binding of each αβTCR was assessed by flow cytometry. (B) Cells, prepared as in (A), were simultaneously analyzed with fluorescent αβTCRs and a mAb specific for IA b (17–227) that does not interfere with αβTCR–IA b interaction. (C) The binding of the αβTCRs is shown only for those infected insect cells that bear a high level of surface IA b (dotted region in [B]). Insect cells displaying IA b -p3K bound the αβTCR reagents very heterogeneously ( Figure 4 A), probably owing to heterogeneous expression of IA b -p3K due to variations in the multiplicity of infection (MOI) and the lack of synchrony in viral infection and expression. To focus on cells bearing a particular level of IA b , we stained the cells simultaneously with the fluorescent αβTCR reagents and with an anti-IA b monoclonal antibody (mAb) that did not interfere with αβTCR binding. In this case, there was a direct correlation between the amount of surface IA b -p3K expressed by an individual insect cell and the amount of αβTCR bound ( Figure 4 B) with cells bearing a particular level of IA b -p3K, binding the αβTCRs uniformly ( Figure 4 C). Therefore, comparing the two types of staining gave us a useful tool to evaluate the relation between peptide sequence and the strength of αβTCR binding (see below). Recovering Baculovirus Carrying a Particular Peptide–MHC Combination Our experiments showed that fluorescent αβTCRs could be used with flow cytometry to identify insect cells infected with a baculovirus encoding a specific peptide–MHC combination. We next tested whether this system could be used to enrich baculoviruses encoding a particular peptide–MHC. Insect cells were infected at an MOI of about 1 with a mixture of baculoviruses. Of these viruses, 1% encoded the IA b -p3K molecule and 99% encoded a control molecule (an αβTCR β chain). The infected cells were stained with fluorescent YAe-62 αβTCR and analyzed by flow cytometry. Although a distinct population of brightly fluorescent cells was not seen, the 1% of the cells with the brightest fluorescence were sorted, as were an equal number of cells that were very dully fluorescent ( Figure 5 A). The recovered infected cells were cultured with fresh insect cells to produce new viral stocks. These stocks were used to infect insect cells that were tested again with the fluorescent αβTCR reagent ( Figure 5 B). The cells infected with virus from the few fluorescent positive cells in the original population were now nearly all brightly fluorescent, and those infected with the virus from the fluorescently dull cells were nearly all negative for binding of the αβTCR. These results showed that flow cytometry could be used with a fluorescent multimerized αβTCR to find and greatly enrich insect cells infected with a virus encoding a specific peptide–MHC combination. Figure 5 Recovery of IA b -p3K Virus-Infected Cells with Fluorescent αβTCR (A) Sf9 cells were infected with a mixture of virus, 99% of which encoded a control protein (a TCR β chain linked to the gp64 transmembrane/cytoplasmic tail) and 1% of which encoded IA b -p3K. After 3 d, the infected cells were analyzed as in Figure 3 A for binding fluorescent αβTCR from YAe-62. The 1% of the infected cells with the brightest fluorescence was sorted (high sort, 15,700 cells). As a control, a similar number of cells that fluoresced as dully as the background fluorescence were also sorted (low sort). (B) The sorted cells were incubated with fresh Sf9 insect cells to allow propagation of the viruses and production of new stocks. The stocks were used to infect new Sf9 cells, and after 3 d the analysis of αβTCR binding was repeated. Construction of a Peptide Library Attached to IA b in Baculovirus The most widely used method for introducing gene constructions into baculovirus involves assembling the construct first in an Escherichia coli transfer plasmid, where it is flanked by sections of baculovirus DNA. The complete construct is then introduced into baculovirus by homologous recombination using any of the commercially available modified baculovirus DNAs that require homologous recombination with the plasmid in order to generate functional circular viral DNA ( Kitts and Possee 1993 ). Based on this procedure, we constructed an IA b –peptide library in two steps. In the original transfer plasmid that encoded the displayed IA b -p3K, we flanked the site encoding the peptide with unique restriction sites, one in the section encoding the β chain leader and the other in the section encoding the linker from the peptide to the N-terminus of the β chain. The DNA between these sites was replaced with DNA encoding enhanced green fluorescent protein (GFP) (Clontech, Palo Alto, California, United States) in-frame with the IA b signal peptide and with a 3′ termination codon (see Figure 2 C). Thus, cells infected with baculovirus carrying this construct produced GFP, but not an IA b molecule, because of disruption of the IA b β chain gene. We then designed a peptide library based on the structure of p3K bound to IA b (see Figure 1 A) We used oligonucleotides with random nucleotides in codons encoding five peptide amino acids (p2, p3, p5, p7, and p8) corresponding to the central surface-exposed amino acids of p3K bound to IA b . Other positions were kept identical to p3K, including alanines at the four standard anchor residues at p1, p4, p6, and p9. These oligonucleotides were used in a PCR to create a DNA fragment randomized in these five codons and with 5′- and 3′-end restriction enzyme sites compatible with those in the signal peptide and linker (see Figure 2 D). This fragment was ligated into the restricted plasmid, replacing the GFP sequence and restoring a functional IA b β chain gene (see Figure 2 E). The mixture of plasmids was then used to transform E. coli and a bulk plasmid preparation was made. The plasmids were cotransfected with BaculoGold baculovirus DNA into Sf9 insect cells to produce a mixed viral stock in which each virus carried the genes for IA b with a different peptide bound. Although it is difficult to calculate the efficiency with which recombination yields infectious baculovirus, we estimate the size of this library was between 3 × 10 4 and 1 × 10 5 independent viruses. Successive Enrichment of Baculovirus Carrying Peptide–MHC Combinations That Bind a Particular αβTCR A large number of Sf9 insect cells were infected at an MOI of about 1, with baculovirus carrying the IA b –peptide library. After 3–4 d, the cells were analyzed with fluorescent B3K-06- or YAe-62-soluble αβTCR, as described above. Fluorescent cells were sorted and cultured with fresh uninfected Sf9 cells to create new infected cells for analysis and an enriched viral stock. This process was repeated three to four times. In each case, when no clear fluorescent population was apparent, the brightest 1% of the infected cells was sorted. In later rounds the majority of the cells in a clearly distinguishable fluorescent population were sorted. Figure 6 summarizes the successive enrichment of viruses that produced IA b –peptide combinations that could be detected with each of the fluorescent αβTCRs. Infected cells binding the B3K-06 αβTCR were apparent only after two rounds of enrichment, but eventually yielded a population with uniform binding ( Figure 6 A). Infected cells that bound the YAe-62 αβTCR were detectable even with the initial library of viruses and enriched rapidly to yield a population with more heterogeneous levels of binding to the receptor ( Figure 6 B). Figure 6 Summary of Successive Screening of the IA b –Peptide Libraries with Fluorescent αβTCRs Sf9 insect cells (1 × 10 7 to 1.5 × 10 7) were infected at a MOI of approximately 1 with an aliquot of baculovirus encoding the IA b –peptide library. After 3 d, the infected cells were analyzed for binding the αβTCR of either B3K-06 or YAe-62. Either obviously fluorescent cells or the brightest 1% of the cells were sorted (2 × 10 4 to 8 × 10 4 cells) and added to 3 × 10 6 fresh Sf9 cells to propagate and reexpress the viruses contained in the sorted cells. These infected cells were then reanalyzed and sorted using the fluorescent αβTCRs. This process was repeated until no further enrichment of αβTCR binding was seen. In most cases, the reanalysis was done directly from the cells that were cocultured with the sorted cells. In a few cases, an intermediate viral stock was made and then used to infect additional Sf9 cells. The turn around time per cycle was 4–7 d. The figure shows the reanalysis in a single experiment of the initial viral stocks and all of the various intermediate enriched viral stocks. Sf9 cells were infected at an MOI of less than 1 with the viral stocks and analyzed as in Figure 4 for either B3K-06 (A) or YAe-62 (B) αβTCR binding. Comparison of αβTCR-Selected Peptides from the Library At the time of the final enrichment, single infected cells binding each of αβTCRs were sorted into individual wells of 96-well culture plates containing fresh Sf9 cells in order to prepare clonal viral stocks. These stocks were used to infect fresh Sf9 cells, which were reanalyzed for binding to the appropriate αβTCR as in Figure 4 . Viral DNA from the clones that showed homogeneous TCR binding at a particular level of IA b were used as template in a PCR using oligonucleotides that flanked the peptide site in the construct, and a third internal oligonucleotide was used to sequence the PCR fragment. The majority of PCR fragments yielded a single unambiguous peptide sequence. These viruses were used to infect Sf9 cells that expressed mouse ICAM and B7.1. The infected cells were used as APCs for either the B3K-06 or YAe-62 hybridoma, with IL-2 production being a measure of IA b –peptide recognition. Viruses expressing IA b –peptide combinations that produced high levels of surface IA b , but that neither bound to the αβTCR nor stimulated the T cell hybridomas, were used as negative controls, and virus producing IA b -p3K was used as the positive control. Results with a few representative virus clones are shown in Figure 7 A and 7 B, and a summary of all of the results is shown in Table 1 . Figure 7 Analysis of Baculovirus Clones from the αβTCR-Enriched IA b –Peptide Library (A) Sf9 cells were infected with stock from four baculovirus clones (B9, B13, B17, and B23) isolated from the virus pool enriched with the αβTCR of B3K-06. After 3 d, an aliquot of cells from each infection was analyzed as in Figure 4 to assure uniform binding of the fluorescent B3K-06 αβTCR (top). Viral DNAs prepared from other aliquots of the cells were used as templates in a PCR with oligonucleotides that flanked the DNA encoding the IA b -bound peptide. The fragment was sequenced directly with a third internal oligonucleotide (middle). The clone stock was then used to infect Sf9-ICAM/B7.1 cells. After 3 d, the infected cells were used as APCs for B3K-06 production of IL-2 (bottom). Virus encoding IA b -p3K was used as a positive control. Virus encoding pEα was used as the negative control. (B) Same as (A), but using YAe-62 and clones (Y2, Y14, Y28, Y52) derived from the IA b –peptide library using the YAe-62 αβTCR. Table 1 Summary of Peptides Selected by p3K-Reactive αβTCRs a Amino acids homologous to those in p3K are shown in red b Determined from mean fluorescence as in Figure 4 B and 4 C c Sorted by frequency and then by level of TCR binding Given our previous data indicating that the B3K-06 αβTCR interacted with all five of the p3K amino acids varied in this library ( Liu et al. 2002 ; see also Figure 1 B), we expected that mimotopes satisfying this receptor would be infrequent or perhaps even absent in a library of this size. Indeed, only one peptide was recovered from the library with the B3K-06 αβTCR, FEAQRARAARVD. It was found in all 42 clones analyzed with unambiguous αβTCR binding and peptide sequences. The sequence of this peptide was strikingly similar to that of p3K. Like p3K, it had a glutamine at p2. It had arginines at positions p3, p5, and p8, corresponding to the lysines found in these positions in p3K, most likely reflecting the importance of the positive charges at these positions. We do not know the relative importance of lysine versus arginine at these three positions, but given that there are two codons for lysine and six for arginine, there was of course a much higher probability of finding arginine than lysine. The most significant difference between this peptide and p3K was an alanine instead of asparagine found at p7. When bound to IA b on ICAM/ B7.1-expressing Sf9 APCs, FEAQRARAARVD was able to stimulate B3K-06 to produce IL-2, but not nearly as well as did p3K. This loss of stimulating activity was caused by one or more of the lysine-to-arginine substitutions and/or the asparagine-to-alanine substitution at p7. Interestingly, the substitution of alanine for asparagine in p3K eliminated the response of B3K-06 to soluble peptide presented by an IA b -bearing mouse APC (see Figure 1 B). Perhaps the very high density of IA b –peptide on the surface of the insect cells allows for responses to peptides that would normally not be stimulatory with peptides presented by conventional APCs, although another possibility is that somehow the arginine (particularly at p8) compensated for the absence of the asparagine sidechain. Consistent with the hypothesis that the αβTCR of YAe-62 would be more peptide promiscuous than that of B3K-06, we found 20 different peptide sequences among the analyzed clones that produced an IA b –peptide combination that bound the YAe-62 αβTCR. It is likely that many more would be identified if more clones were analyzed. Five sequences were found multiple times. Not unexpectedly, these were among those that bound the YAe-62 αβTCR most strongly. There was a 100-fold range in the intensity of αβTCR binding to the different IA b –peptide combinations, ranging from about 4-fold to 400-fold binding above that seen with a negative control peptide. One obvious property of these peptides stands out. There appeared to be a very strong selection for a basic amino acid at position 5. In 16 of 20 of the peptides, a lysine, arginine, or histidine was found at position 5, matching the lysine found in p3K. As a control, we sequenced random clones picked either from the original E. coli construction of the library (17 clones) or from the baculovirus library that expressed IA b –peptide well, but did not bind either αβTCR (11 clones). The frequencies of basic amino acids at p5 in these sequences were only 12% and 9%, respectively (data not shown). There was no strong selection for amino acids homologous to those of p3K at positions p2, p3, p7, or p8. The amino acids at positions p2 and p3 appear nearly random, suggesting little or no essential contact between this part of the peptide–MHC ligand and the receptor, although these positions may contribute to the wide range of apparent αβTCR affinities seen. While not homologous to the asparagine in p3K, leucine was found at p7 in six of 20 (30.0%) of the YAe-62 αβTCR-selected peptides and three of 11 (27.2%) of the IA b -binding peptides that were not bound by the YAe-62 αβTCR, but only two of 17 (11.8%) of the random E. coli plasmids. The amino acid in this position is only partially exposed on the surface and can contribute significantly to peptide–MHC interaction ( Liu et al. 2002 ). After asparagine, leucine is the most common amino acid found at this position in peptides found naturally bound to IA b ( Dongre et al. 2001 ; Liu et al. 2002 ). Therefore, although more data would be required to test its significance, there may have been some slight enrichment of leucine at p7 in the expressed library prior to αβTCR selection, reflecting the role of p7 in stable peptide binding to IA b . The amino acid at position p8 is predicted to be fully surface exposed. In the selected peptides, rather than an amino acid homologous to the lysine of p3K, there may be an overrepresentation of amino acids with small neutral sidechains (threonine, serine, alanine, glycine) at this position. Perhaps this indicates that, in general, larger sidechains can be inhibitory at this position, but again more data would be required to test this idea. The 12 IA b –peptide combinations that bound the YAe-62 αβTCR most strongly were also the ones that were able to induce IL-2 production from YAe-62. Among these, a number with the very highest apparent affinities stimulated YAe-62 better than did p3K. However, there was not a direct correlation between apparent affinity and the level of IL-2 production; i.e., several peptides that yielded complexes with IA b with about the same apparent affinity for the αβTCR nevertheless stimulated very different levels of IL-2 production from YAe-62. This may be related to the phenomenon of altered peptide ligands ( see Discussion ). Overall, our results supported our original prediction that for conventional T cells, such as B3K-06, most of the surface-exposed residues of the peptide would be important in peptide–MHC recognition, while for broadly allo-MHC-reactive T cells, such as YAe-62, peptide recognition would be much more promiscuous. Discussion The peptide degeneracy allowed for a given αβTCR–MHC combination has been a subject of study over many years. While minor changes in the exposed amino acids sidechains of the peptide can often destroy αβTCR recognition, usually at least some variation is tolerated within the predicted footprint of the αβTCR on the peptide–MHC ligand ( Evavold and Allen 1992 ; Reay et al. 1994 ). We can understand this flexibility to some extent from the X-ray structures of αβTCR–MHC–peptide complexes that show poor or even absent interactions between some peptide amino acid sidechains and the complementarity region (CDR) loops of the receptor (reviewed in Garcia et al. 1999 ). We have reported the properties of mice that have been genetically manipulated to express their MHCII molecules virtually completely occupied by a single peptide ( Ignatowicz et al. 1996 ; Marrack et al. 2001 ). One of the most unusual features of the repertoire of T cells that develop in these animals is that they show an unusually high frequency of broadly allo-MHC–self-MHC-reactive T cells. These T cells are lost when these animals are repopulated with MHCII wild-type bone marrow cells. We have concluded that cells of this type are commonly positively selected in normal animals, but to a large extent negatively selected by self-MHC occupied by a variety of self-peptides. Their survival in single peptide–MHC mice may reflect the need for many different peptides to expose all MHC amino acids and their various conformers during T cell-negative selection. We have proposed that the αβTCRs of these cells are focused on the common conserved features of peptide–MHC complexes rather than on the specific sidechains of the exposed amino acids of the peptide ( Marrack et al. 2001 ). A consequence of this hypothesis is the prediction that T cells of this sort should be much more peptide promiscuous than conventional T cells. The experiments reported here were designed to test this prediction by comparing the peptide promiscuity of one of these broadly allo-reactive T cells, YAe-62, typical of T cells from these mice, to that of a T cell with the same nominal specificity produced by immunization of conventional mice. The results support the conclusion that the broadly allo-reactive T cell has a much greater peptide promiscuity than does the conventional T cell. This question of T cell promiscuity is an important one in that it addresses the existence of a very large set of TCRs that apparently make it through positive selection, but never see the light of day in normal animals, because they are negatively selected on self-MHC with little input from the MHC-bound peptide. Thus, studying the peripheral fully negatively selected T cell repertoire gives a false impression of the interaction requirements necessary or sufficient for positive selection. These promiscuous T cells may also give us insight into possible evolutionary conserved αβTCR–MHC interactions that have been hard to sort out with conventional T cells. While perhaps much less frequent than in single peptide–MHC mice, peptide-promiscuous T cells have been described in normal individuals ( Brock et al. 1996 ). Consistent with the idea that this property may be linked to allo-MHC reactivity, in a parallel study we have shown that peptide-promiscuous T cells are enriched in normal mice in the population of T cells reactive to foreign MHC alleles and isotypes ( Huseby et al. 2003 ). In order to study the relationship between peptide sequence and αβTCR recognition, we developed a baculovirus-based display method for rapid identification of peptides that form complexes with MHC that bind a particular αβTCR. Display is one of the most powerful library techniques available. Its underlying principle is that the protein or peptide members of the library are expressed on the surface of organisms that harbor the DNA encoding them. A binding assay that isolates all members of the library with the appropriate properties copurifies the organism and the encoding DNA. The DNA is then amplified and reexpressed and the process repeated as many time as necessary to enrich fully the relevant molecules, whose sequence can be deduced from the copurified DNA. The great advantage of display libraries is that all members of the library that satisfy the screening conditions are enriched simultaneously without the need to identify them one by one. In order for peptides to be tested for αβTCR binding, they must be complexed with the relevant MHC molecule on a platform suitable for interaction with the T cell and/or its αβTCR. For display libraries, one aspect of this problem has been solved by the ability to express MHC molecules with sequence for a covalently attached antigenic peptide imbedded in the MHC genes ( Kozono et al. 1994 ; Mottez et al. 1995 ; Uger and Barber 1998 ; White et al. 1999 ). However, the most commonly used bacterial display systems do not yet allow for the assembly and display of complex multichain MHC molecules. There is a recent report of the successful display of a single-chain peptide–MHCI on yeast cells ( Brophy et al. 2003 ), but our own previous attempts with yeast had failed to yield displayed peptide–MHCII in a form capable of recognition by T cell hybridomas (data not shown). Our previous success with producing soluble MHC and αβTCR molecules using a baculovirus expression system and a report of peptide libraries displayed in baculovirus ( Ernst et al. 1998 ) led us to adapt these methods for surface display of peptide–MHCII on insect cells. We constructed a library of peptides attached to the displayed mouse class II molecule, IA b . Using fluorescently labeled multimeric soluble αβTCRs as bait and insect cells infected with the IA b –peptide library as fish, we were able to identify rapidly the members of the library that encoded peptide mimotopes for two αβTCRs. In these studies, the immunizing peptide (epitope) for the αβTCR was already known. However, this method should be useful as well in identifying mimotopes for αβTCRs whose peptide epitope is not known, provided that suitable peptide anchor residues for MHC binding are known. One limitation of this display method as presented here is the size of the peptide library. The bottlenecks caused by the preparation of the library in an E. coli plasmid and then its introduction into baculovirus by homologous recombination resulted in a library with only 3 × 10 4 to 1 × 10 5 members. This is far below the size required to have all 3.2 × 10 6 versions of the peptide present when randomizing five amino acids. Large libraries of this size require more efficient baculovirus-cloning methods, such as incorporation of DNA fragments directly into baculovirus DNA by ligation ( Ernst et al. 1994 ) or in vitro recombinase-mediated recombination ( Peakman et al. 1992 ). In preliminary experiments, we have constructed an IA b –peptide library with over 10 7 clones by directly ligating ( Ernst et al. 1994 ) a randomized PCR DNA fragment encoding the peptide into linearized baculovirus DNA using unique homing restriction enzyme (SceI–CeuI) sites introduced flanking the peptide-encoding region of the construct (data not shown). Since recircularized baculovirus DNA is directly infectious when introduced into insect cells by transfection, there is no theoretical reason why this method cannot be used to create libraries as large as those reported for yeast and phage. We have developed this method using IA b as the displayed MHCII molecule carrying the peptide library. However, using the same strategy, we have successfully displayed numerous other MHCII molecules, such as murine IE k and human DR4, DR52c, and DP2 (data not shown). While the leucine zippers that we included in this construct are not strictly required for expression of IA b , they have helped considerably in expression of some of these other MHCII molecules. Moreover, we ( White et al. 1999 ) and others ( Mottez et al. 1995 ; Uger and Barber 1998 ) have shown that peptides can be tethered to MHCI molecules via the N-terminus of either β2m or the heavy chain, making this approach feasible for searching for MHCI-bound peptide mimotopes as well. In preliminary experiments we have successfully displayed on the surface of Sf9 cells the mouse MHCI molecule, D d , with a β2m-tethered peptide from HIV gp120 (data not shown). Given that baculovirus has been such a successful expression system for many different types of complex eukaryotic proteins that express or assemble poorly in E. coli , this method may have broad applications to other receptor–ligand systems. As opposed to methods that use T cell activation as the peptide-screening method, an advantage of display methods that use flow cytometry for screening and enrichment is that the strength of binding of receptor and ligand can be estimated and manipulated. In the results reported here, by limiting the analysis to insect cells bearing a particular level of peptide–MHC, a uniform level of αβTCR binding was seen for an individual peptide sequence, but the strength of binding varied over two orders of magnitude for different peptides, presumably reflecting the relative affinity of the receptor for different IA b –peptide combinations. Thus, depending on whether one was interested in high- or low-affinity ligands for the αβTCR, one could enrich for peptides with these properties directly during the screening of the library. Such an approach has been used with antibody ( Boder and Wittrup 2000 ) and αβTCR ( Shusta et al. 2000 ) variants displayed on yeast to select directly for receptors with increased affinity. It is worth noting that there was not a direct correlation between the strength of αβTCR binding to a particular peptide–MHC combination and the subsequent level of IL-2 secretion seen from the T cell responding to this combination. While in general the best IL-2 secretion was obtained with complexes with the highest apparent affinities, some IA b –peptide combinations with apparent high affinity stimulated IL-2 production poorly. One interesting possibility is that this observation is related to the phenomenon of altered peptide ligands in which amino acid variants of fully immunogenic peptides only partially activate or even anergize the T cell ( Evavold et al. 1993 ), despite minor differences in affinity. In some cases, this phenomenon has been related to αβTCR binding kinetics, rather than just overall affinity ( Lyons et al. 1996 ). Future experiments using surface plasmon resonance or fluorescence peptide–MHC multimers might help to test this idea. In summary, the very properties that have made baculovirus a very successful expression system for complex eukaryotic proteins also make it suitable for library display methods, with potential application not only in T cell epitope/mimotope discovery, characterization, and manipulation, but also in studying a wide variety of other protein–protein interactions. Materials and Methods Synthetic peptides, oligonucleotides, and DNA sequencing The peptides pEα (FEAQGALANIAVD), p3K (FEAQKAKANKAVD), and various alanine-substituted variants of p3K were synthesized in the Molecular Resource Center of the National Jewish Medical and Research Center (Denver, Colorado, United States), as were all oligonucleotides used in PCR and DNA sequencing. Automated DNA sequencing was also performed in this facility. Cell lines and T cell hybridomas The insect cell lines Sf9 and High Five were obtained from Invitrogen (Carlsbad, California, United States). The IA b -p3K-reactive T cell hybridoma B3K-06 was produced from C57BL/6 mice as previously described ( Rees et al. 1999 ). The IA b -expressing B cell hybridoma LB-15.13 ( Kappler et al. 1982 ) was used to present soluble peptides to B3K-06. The T cell hybridoma YAe-62 ( Marrack et al. 2001 ) was produced from previously described ( Ignatowicz et al. 1996 ) C57BL/6 mice that lacked expression of the endogenous IA b β gene (ΔIAβ) and the invariant chain (ΔIi) and that carried a transgene for the IA b β gene that was modified to insert sequence encoding pEα and a flexible linker between the signal peptide and the N-terminus of the β chain. These mice were immunized intravenously with 3 × 10 6 dendritic cells from ΔIAβ/ΔIi C57BL/6 mice. These cells had been retrovirally transduced ( Mitchell et al. 2001 ; Schaefer et al. 2001 ) with the IA b β gene that was modified as above to express with a tethered p3K. T cells from these immunized mice were propagated in vitro and converted to T cell hybridomas, by standard techniques ( White et al. 2000 ). The hybridomas were initially screened for binding of multivalent, fluorescent IA b -p3K ( Crawford et al. 1998 ; Rees et al. 1999 ) and subsequently for IL-2 production in response to immobilized, soluble IA b -p3K, but not to spleen cells from the host ΔIAβ/ΔIi IA b -pEα transgenic mice. Further characterization of YAe-62 is described in the Results. Soluble αβTCRs cDNA, prepared from B3K-06 and YAe-62, was used as template in a PCR using oligonucleotides that flanked the Vα and Vβ regions and introduced restriction enzyme sites that allowed cloning of the PCR fragments into a previously described baculovirus expression vector for soluble αβTCRs ( Kappler et al. 1994 ). The cloned fragments were sequenced and incorporated into baculovirus and αβTCRs were purified from the supernatants of infected High Five cells. For B3K-06, the α chain was AV0401/AJ27 and the CDR3 sequence was CALVISNTNKVVFGTG. The β chain was BV0801/BJ0103 and the CD3 sequence was CASIDSSGNTLYFGEG. For YAe-62, the α chain was AV0412/AJ11 and the CD3 sequence was CAANSGTYQRFGTG. The β chain was BV0802/JD0204 and the CD3 sequence was CASGDFWGDTLYFGAG. Expression of ICAM and B7.1 in Sf9 cells DNA fragments encoding the baculovirus hr5 enhancer element, IE1 gene promoter, and IEI poly(A) addition region were synthesized by PCR using baculovirus DNA as template. The fragments were used to construct an insect cell expression vector (pTIE1) on a pTZ18R (Pharmacia, Uppsala, Sweden) backbone with the hr5 enhancer at the 5′-end, followed by the IE1 promoter, a large multiple cloning site (Esp3I, MunI, SalI, XhoI, BsrGI, HpaI, SpeI, BstXI, BamHI, BspEI, NotI, SacII, XbaI), and the IE1 poly(A) addition region. The complete sequence of the pTIE1 vector has been deposited in GenBank (see Supporting Information). DNA fragments encoding mouse ICAM and B7.1 were cloned between the XhoI and NotI sites of the multiple cloning site. Sf9 cells were transfected with a combination of the plasmids by the standard calcium phosphate method and cells expressing both molecules on their surfaces were cloned without selection at limiting dilution to establish the line Sf9-ICAM/B7.1. IL-2 assays T cell hybridoma cells (10 5 ) were added to microtiter wells containing either (1) saturating immobilized peptide–MHC, (2) 10 μg/ml peptide plus 10 5 LB-15.13 cells, (3) 5 × 10 4 Sf9-ICAM/B7.1 insect cells infected 3 d previously with baculovirus encoding a displayed peptide–MHC, (4) 10 6 spleen cells from IA b -pEα single peptide mice, or (5) 10 6 spleen cells from various knockout or MHC congenic mice. After overnight incubation the culture supernatants were assayed for IL-2 as previously described ( White et al. 2000 ). mAbs and flow cytometry The following mAbs were used in these studies: 17/227, a mouse IgG2a antibody, specific for IA b ( Lemke et al. 1979 ); ADO-304, an Armenian hamster antibody specific for an epitope on the αβTCR Cα region not accessible on the surface of T cells, but exposed on recombinant αβTCR and on CD3-dissociated, NP-40-solublized natural αβTCR ( Liu et al. 1998 ); 3E2 (PharMingen, San Diego, California, United States), specific for mouse ICAM (CD54); and 16–10A1 (PharMingen), specific for mouse B7.1 (CD80). For flow cytometry, an unlabeled version of 17/227 was used with phycoerythrin-coupled goat anti-mouse IgG2a (Fisher Biotech, Foster City, California, United States). To assemble multivalent fluorescent versions of the soluble αβTCRs, first a biotinylated version of ADO-304 was prepared. In brief, purified ADO-304 at 1–3 mg/ml in 0.1 M NaHCO 3 was labeled with Sulfo-NHS-LC-Biotin (Pierce Chemical Company, Rockford, Illinois, United States) at a molar ratio of 2.5:1 (biotin:antibody) for 4 h at room temperature. The reaction was quenched with 0.1 M lysine and the product dialyzed extensively against PBS. The resulting derivative contained about one biotin per molecule of mAb. The biotinylated mAb was complexed in excess with AlexaFlour647–streptavidin (Molecular Probes, Eugene, Oregon, United States). The complex was separated from the free biotin–antibody using Superdex-200 size exclusion chromatography (Pharmacia). In preliminary experiments, the amount of soluble αβTCR required to saturate an aliquot of a large single batch of this reagent was determined. To prepare the multivalent αβTCR, the appropriate amount of soluble αβTCR was mixed with an aliquot of the fluorescent anti-Cα reagent overnight. For staining for flow cytometry, this mix was used without further purification. Each 100 μl sample contained approximately 2 μg of the fluorescent reagent plus 10 5 Sf9 insect cells. This mixture was incubated at 27°C for 1–2 h. The cells were then washed for analysis. The advantages of this method for preparing fluorescent multimers over using direct enzymatic biotinylation ( Schatz 1993 ) of the αβTCR were that only one fluorescent reagent needed to be prepared for all αβTCRs, the mAb–streptavidin complex was very stable over a long period of time, and no special peptide-tagged version of the soluble αβTCR was required. Analytical flow cytometry was performed with a FacsCaliber flow cytometer (Becton-Dickinson, Palo Alto, California, United States). For sorting, a MoFlo instrument was used (Dako/Cytomation, Glostrup, Denmark). IA b and peptide library constructions For displaying IA b on the surface of baculovirus-infected insect cells, modifications were made as described in Figure 2 A and 2 B to a previously reported baculovirus construct for producing soluble IA b -p3K ( Rees et al. 1999 ). Other versions of this construction were prepared encoding other IA b -binding peptides. The constructions were incorporated into baculovirus by homologous recombination using the BaculoGold system (PharMingen). As described in Figure 2 C, this construction was altered in the E. coli transfer plasmid to replace the portion encoding p3K with sequence encoding enhanced GFP, flanked by sites for the restriction enzymes SbfI and CeuI. A PCR fragment was produced as described in Figure 2 D that encoded an IA b -binding peptide randomized at positions p2, p3, p5, p7, and p8, but identical to p3K at other positions. This sequence was flanked by sites for the restriction enzymes PstI and BstXI, such that the cohesive ends generated by these enzymes were compatible with those generated by SbfI and CeuI in the GFP-containing plasmid. Cloning the restricted fragment into this site regenerated a covalent peptide in-frame with the signal peptide and flexible linker of the IA b β chain (see Figure 2 E). After ligation of the fragment into this plasmid, a bulk transformation was performed using XL1-Blue E. coli (Stratagene, La Jolla, California, United States). An estimated 3 × 10 4 to 10 × 10 4 independent transformants were obtained that were used to make a mixed plasmid preparation. This mixture was incorporated into baculovirus by homologous recombination as above. In order to assure a high efficiency of conversion of plasmids to virus, 1.5 × 10 7 Sf9 cells were cotransfected with 6 μg of the plasmid mixture and 1.5 μg of BaculoGold DNA. Supporting Information Accession Numbers The GenBank ( http://www.ncbi.nlm.nih.gov/Genbank/) accession numbers for the sequences described in this paper are B7.1 (AJ278965), baculovirus DNA (L22858), ICAM (X52264), and pTIE1 vector (AY522575). | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC387264.xml |
524025 | A knowledgebase system to enhance scientific discovery: Telemakus | Background With the rapid expansion of scientific research, the ability to effectively find or integrate new domain knowledge in the sciences is proving increasingly difficult. Efforts to improve and speed up scientific discovery are being explored on a number of fronts. However, much of this work is based on traditional search and retrieval approaches and the bibliographic citation presentation format remains unchanged. Methods Case study. Results The Telemakus KnowledgeBase System provides flexible new tools for creating knowledgebases to facilitate retrieval and review of scientific research reports. In formalizing the representation of the research methods and results of scientific reports, Telemakus offers a potential strategy to enhance the scientific discovery process. While other research has demonstrated that aggregating and analyzing research findings across domains augments knowledge discovery, the Telemakus system is unique in combining document surrogates with interactive concept maps of linked relationships across groups of research reports. Conclusion Based on how scientists conduct research and read the literature, the Telemakus KnowledgeBase System brings together three innovations in analyzing, displaying and summarizing research reports across a domain: (1) research report schema , a document surrogate of extracted research methods and findings presented in a consistent and structured schema format which mimics the research process itself and provides a high-level surrogate to facilitate searching and rapid review of retrieved documents; (2) research findings , used to index the documents, allowing searchers to request, for example, research studies which have studied the relationship between neoplasms and vitamin E; and (3) visual exploration interface of linked relationships for interactive querying of research findings across the knowledgebase and graphical displays of what is known as well as, through gaps in the map, what is yet to be tested. The rationale and system architecture are described and plans for the future are discussed. | Background An unfortunate consequence of specialization in the sciences is poor communication across research domains – which can hamper the knowledge discovery process. Research findings in one area may be pertinent to another, researchers may be unaware of relevant work by others that could be integrated into their work and important findings just outside a researcher's focus can be overlooked. Compounding this problem is the difficulty of keeping current with new research findings that continue to grow at an exponential rate. Reliance on keywords and/or subject indexing to find relevant literature limits the researcher's ability to precisely search for and locate specific research findings. For example, a typical database query to locate all research articles reporting a statistically significant relationship between caloric restriction and cancer would retrieve articles reporting both concepts as represented by the indexing and keyword search – but not necessarily linked together as a research finding, with information regarding reported statistical significance of the finding, nor, perhaps most importantly, lacking representation of the linkages among the retrieved document sets. This lack of "interactivity" among retrieved citations is a critical limitation of traditional search and retrieval systems. As stated by Swanson (1986) in his examination of "mutually isolated literatures:" "Knowledge can be public, yet undiscovered, if independently created fragments are logically related but never retrieved, brought together, and interpreted [..] This essential incompleteness of search and retrieval therefore makes possible, and plausible, the existence of undiscovered public knowledge [ 1 ]." In addition to this limitation of search and retrieval, there are questions about representing a set of documents: What format or display of the retrieval set most enhances users' ability to identify which documents need to be examined in more detail? How can users navigate across document sets (i.e., to explore linkages) to enhance the discovery process? The bibliographic citation format is used by virtually all bibliographic databases today to report the results of database searches. However, it does not provide a way for the user to quickly review retrieved results for research methods and findings or to quickly view the relationships among the documents in the document set. Abstracts, even structured abstracts, simply do not provide a format conducive to rapid review of retrieved citations. In fact, the bibliographic citation format itself has changed little for the past two hundred years – even though it does not present an accurate representation of either the research methods or the research findings in a document [ 2 ]. In spite of great improvements in document retrieval over the past twenty years, most information systems developed to promote scientific discovery (e.g., [ 3 - 6 ]), are based on traditional search and retrieval approaches and the tools for locating and inter-relating research methods and findings are imprecise. This is the impoverished state Nobel laureate economist Herbert Simon described in his oft-cited remark on information as a commodity: "What information consumes is rather obvious: it consumes the attention of its recipients. Hence, a wealth of information creates a poverty of attention and a need to allocate that attention efficiently among the overabundance of information sources that might consume it" [ 7 ]. For researchers, such a poverty of attention can translate into delays in the progress of scientific discovery. A comprehensive approach to these challenges is the goal of the Telemakus research program. Telemakus was named for the son of Odysseus who went searching for his father, the legendary Greek hero of Troy. Similarly, the Telemakus research program is developing approaches and tools for searching, knowledge discovery and mapping domain knowledge. The overall vision is to enhance the knowledge discovery process through retrieval, visual and interactive interfaces and tools. In close collaboration with researchers in the biology of aging, a working knowledgebase system has been designed to present aggregated citation information and research methods and findings for display in a conceptual schema. The Telemakus KnowledgeBase System provides the user with both a macro- and micro-view of domain knowledge. The macro-view facilitates identification of patterns – both expected and unexpected occurrences of relationships among research concepts – and permits visualization and dynamic navigation of scientific domains. The resulting maps are analogous to citation mapping work done by Small [ 8 ] but, instead of citations, rely on reported research findings. The micro-view presents consistent and detailed document attributes including research methods and findings for each document in the database. This article describes the theories underlying the Telemakus KnowledgeBase System, provides an overview of its implementation, reports initial user feedback and explores future directions. Telemakus system builds on prior research in the areas of: (1) schema theory, (2) concept representation and (3) information visualization. Schema Theory Schemas are generalized mental models that provide a guide for structuring the process of production and comprehension of texts: "...at the simplest level, a schema is a description of a complex object, situation, process or structure. It is a collection of knowledge related to the concept [ 9 ]." According to schema theory, we understand the world in terms of prototypical patterns: people capture global coherence or structure their knowledge of the world based on scripts, schemas and narratives in which are embedded a vast array of relationships, concepts and vocabulary words. Individuals not only think and store knowledge in terms of scripts, frames and schemas but they also produce texts this way. Schema theory originated in linguistics and cognitive psychology as a model for holistically representing texts; it is primarily concerned with the "glue" which holds texts together, i.e., grammatical markers that allow texts to be cohesive as well as coherent [ 10 ]. "The crucial fact is that the cognitive constraints on information processing which require the formation of semantic macro-structures and which organize acts and speech acts in global units, at the same time have social implications: they determine how individuals wish, decide, intend and plan, execute and control, "see" and understand their own and others' speaking and acting in the social context. Without them the individual would be lost among a myriad of detailed incoherent pieces of visual, actional and prepositional information [ 10 ]." Research on the application of schema theory to scientific research includes the schematic representation of psychological reports [ 11 ], clinical trials [ 12 , 13 ] and, more recently, Dillon's work on the superstructure and predictability of text [ 14 ]. An elaboration and refinement of schemas are frames [ 15 ]. Like schemas, a frame is a data-structure for representing a stereotyped situation, a remembered framework to be adapted to fit reality by changing details as necessary. When one encounters a new situation (or makes a substantial change in one's view of a problem) one selects from a memory structure called a frame. "Attached to each frame are a number of kinds of information. Some of this information is about how to use the frame. Some of it is about what one can expect to happen next. Some is about what to do if expectations are not confirmed [ 15 ]." The example commonly used is entering a restaurant: If the tables have checkered tablecloths and paper napkin dispensers, we assume that the prices on the menu will be lower than at a restaurant with white linen tablecloths and napkins rolled around polished silverware. Understanding a written text is a process of fitting it into a larger schema known to reader as part of their previous knowledge about the world. It is reasonable to expect that presenting written texts in familiar formats can enhance and potentially speed up an individual's ability to review and analyze large document sets rapidly. Fuller [ 13 ] investigated the application of schema theory in the symbolic representation of full-text research reports to improve representation of research findings. She concluded that schema analysis offered promise for representing the document structure above the level of individual words and sentences and that such schemas, "...offer a means of writing precise computer programs in terms of the specific schema elements, based on the portions of the document where they are most likely to occur. [..] Schema theory appears to offer a paradigm or framework for indexing which provides a means of capturing both the intra-document (i.e., research design, methods and outcomes) and inter-document relationships. [..] It seems likely that the schema will prove effective as a means of improving the retrieval of documents both in terms of precision and relevance [ 13 ]." The predictability provided by schemas also applies to a document's metastructure. For example, Dillon [ 16 ] has proposed a model of navigation in electronic environments which assumes that experienced users of information form schematic representations of a document that in part represent its layout and structure. The Telemakus system utilizes the inherent and predictable research report layout and structure to create schematic representations or surrogates of research studies with extracted representations of research environment, methods and outcomes [ 17 , 18 ]. A second core component of the Telemakus system is based on concept representation. Concept Representation Concept representation is an important component in accurately representing facts from the document. Characterizing the location of concepts in a scientific document can greatly facilitate accurate document representation. Indexing a document – using a vocabulary or thesaurus of terms to represent the document – is a standard method employed to improve retrieval of relevant documents. Yet traditional approaches to indexing fall short of true document representation: reducing the words found in the abstract, title or full-text of the document may be suggestive of the content but are not truly representative of the methods and research findings. The indexing literature is replete with studies documenting interindexer inconsistency, even among experienced professionals using familiar well-documented systems [ 19 ]. Studies indicate that human indexers usually select the most frequently occurring words in a document, yet they will disagree on the terms used and the same indexer will use different terms to index the same document at different times [ 20 ]. Many current automated retrieval programs also rely on word frequency, thus equating frequency with importance for retrieval purposes, which may be a faulty assumption [ 21 ]. "The information retrieval (IR) problem can be described as a quest to find the set of relevant information objects (i.e., documents D) corresponding to a given information need, represented by a query Q. The assumption is that the query Q is a good description of the information need N. An often used premise in IR is the following: if a given document D is about the request Q, then there is a high likelihood that D will be relevant with respect to the associated information need. Thus the information retrieval problem is reduced to deciding the aboutness relation between documents and queries [ 22 ]." Another problem with current indexing practice lies in the way the unique structure of the information elements in the document is obscured. Scientific research reports have a highly predictable structure, with an introduction, methods, results and conclusion. Concepts mentioned in the introduction or conclusion section of a scientific article may not be the primary focus of the research described within the document. For example, the Introduction may include discussion of research among several animal models whereas the target of the research study itself is a specific breed of mouse. However, current indexing processes (whether human or automated) rarely discriminate between locations of concepts in the document for indexing purposes. In addition, index terms do not represent the connections between the various elements in the document; thus, a significant amount of critical information for the scientist is lost. For example, it is not possible to unambiguously retrieve citations from PubMed ® or any other bibliographic database today that will answer the question: "Has anyone ever published data that supports a connection between cancer and caloric restriction? If so, what was the intervention, what type of experiments were done and what were the findings?" A successful response to a query of this type is extremely difficult or impossible in traditional information retrieval systems because: " [..] conventional IR systems that employ isolated term assignments seem inadequate for queries which are specific and empirical in nature. If, on the other hand, retrieval systems provide a link to represent the relationships between the variables of interest as reported in the documents, queries [..] would be better answered. That is, precision might be enhanced for specific and empirical queries when the relationships between the index terms were specified in retrieval systems [ 23 ]." In other words, the researcher asking the questions above can retrieve a set of citations that contain both topics but still must go through the full-text of each document to determine if the research specifically answers the question. Several research studies have explored the utility of relationships captured from data tables and figures in scientific research studies. Fuller, et. al. [ 24 ] described the application of the relationship analysis process for quality filtering of the scientific literature and found it compared favorably with other measures of quality, including the Science Citation Index Impact Factor. Weiner, et. al. [ 25 , 26 ] applied relationship analysis and a mapping method to represent research findings from a database of cancer studies and found they could identify directions for new research studies. And Yamaguchi, et. al. [ 27 ] studied the relative importance of quantitative ideas as expressed in sentences in the text of the Results sections of research reports and in the data tables and concluded that the text ideas were more difficult to find and extract and were found to be less important when compared with ideas derived from the data tables. The importance of data tables for expert decision-making was underscored by Malogolowkin, et. al. who found that cancer researchers rely on ideas presented in numerical displays in published research studies for much of their design of new research protocols [ 28 ]. Malogolowkin, et. al. concluded that innovative aspects of the design can be traced and better understood by mapping the numeric relationships [ 28 ]. Oh [ 23 , 29 , 30 ] investigated the utility of empirical variables and their associated statistical relationships in document representation and retrieval and designed an empirical fact retrieval system (EMFRS). Results of the evaluation indicated that the EMFRS generally outperformed the traditional retrieval system in terms of precision, search effort and measures of user satisfaction. Identifying semantic relationships in text involves looking for certain linguistic patterns in the text that indicate the presence of a particular relationship (or research finding) using pattern-matching to identify the segments of the text or the parts of the sentence that match with each pattern: "If semantic relationships can be identified accurately in the text, retrieval results can be improved by eliminating documents containing the required concepts but not the desired relationships between the concepts [ 31 ]." The third component of the Telemakus system is based on visual mapping of reported research findings. Mapping Inter- and Intra-document Relationships As previously mentioned, indexing strategies rely on "isolated term assignments." This approach leads to the loss of two important sources of information: (1) intra-document information, i.e., the research relationships studied and tested and (2) cross-document information, which captures and links research relationships across groups of documents or domains. This loss is the result of breaking apart the context of clearly linked in research findings in the data tables and figures, concepts typically linked together (the x-y axes of the tables and graphs). Once the research relationships have been extracted, concept mapping, a means of spatially representing knowledge in a visual format, provides a potential solution to the challenge of maintaining the inter-relationships between documents and reported research findings. Spatial representations can assist in understanding conceptual relationships across a domain. They can also assist in identifying previously overlooked potential research connections. Numerous approaches to visualizing an information retrieval space have been explored (e.g., [ 8 , 32 , 33 ]), all seeking to capitalize on the natural strength people have for rapid visual pattern recognition. Most mapping work to date has focused on similarity between journal articles using citation analysis [ 8 ], co-occurrence or co-classification using keywords, topics, or classification schemes [ 34 - 36 ], or journal citation patterns [ 37 ]. Latent semantic analysis (LSA) has been used to map co-occurrence of words (or authors) in titles, abstracts, or full-text sources [ 33 ] and domain maps have been used to visualize author co-citation analysis [ 32 ]. While a review of information visualization strategies is outside the scope of this paper, there is a growing body of work related to mapping metaphors and visualizing large document sets and database search results to provide the user with the ability to visualize relationships among documents and their contents [ 38 , 39 ]. In addition, several tools have been developed that graphically present inter-document relationships, most commonly using some form of link-node diagram [ 40 , 41 ]. Concept mapping represents knowledge graphically through networks of ideas. Such networks consist of nodes (points) and links (arcs/edges). Nodes represent concepts and links represent connections between concepts. Concept mapping has been used for a variety of purposes, including to communicate complicated ideas and, as in the Telemakus system, to demonstrate connections among research findings. Methods How might one apply the theories previously described in developing a comprehensive "real world" information retrieval and knowledge discovery system? As reviewed in the previous section, the Telemakus system is built on and extends prior research in the areas of concept representation, schema theory and information visualization. Work on components of what has become the Telemakus system has been underway for many years with a particular emphasis on the importance and utility of relationships extracted from data tables and figures [ 24 , 42 - 44 ]. Fuller [ 12 , 13 ] identified key objective elements important in representing a clinical research report and developed a schematic representation. The clinical trials schema has been adapted for representing basic sciences research reports in the Telemakus system. Based on how scientists use and want to use the research literature, Telemakus brings together three innovations in analyzing, displaying and summarizing research reports across a domain: 1. Research Report Schema: Research methods and findings are extracted and presented in a consistent, coherent and structured schema format which mimics the research process itself and provides a high-level research report surrogate to facilitate searching as well as rapid review of retrieved documents. 2. Research Findings extracted from data tables and figures are used to index the documents, allowing searchers to request research studies which report a relationship between two concepts of interest. 3. Visual Exploration Interface provides a dynamic map of extracted research findings to graphically display what is known as well as, through gaps in the map, what is yet to be tested. Knowledgebase Creation & Components The Telemakus system consists of a database, research report schema and tools to create relationship maps among concepts across documents. The research report schema serves as a surrogate for the study, methods and research findings for each document as well as providing an interactive search interface. The schematic representations include standard bibliographic information (author, title, journal), information about the research design and methods (age, sex, number of subjects, pre-treatment and treatment regimen, organism and source of organism) and, most importantly, research findings derived from data tables and figures. The elements extracted by the Telemakus system from full-text documents are listed in Table 1 . There are 22 fields for each document, with 12 routinely obtained from PubMed. Of the remaining fields, entries to 4 are controlled by thesauri. Two fields, Authors and SourceOfOrganisms use customized thesauri developed specifically for the Telemakus system. Two additional fields, the ResearchFindings and Organism fields, use the Unified Medical Language System ® (UMLS ® ) Metathesaurus ® as the basis for creating a controlled vocabulary. Table 1 Research report schema database fields FieldName Source Comment Thesaurus? RecordID system provided unique identifier Author bibliographic database Y Year bibliographic database Title bibliographic database AuthorAddress bibliographic database author's email address Journal bibliographic database Volume bibliographic database Issue bibliographic database Pages bibliographic database Keywords bibliographic database subject headings Abstract bibliographic database entire abstract Tables Figures document extracted captions & URLs of tables/figures ResearchFindings document extracted pairs of related concepts Y (UMLS) Organism document extracted type of experimental subject Y (UMLS) Age document extracted Sex document extracted Pre-Treatment Characteristics document extracted NumberOfSubjects document extracted TreatmentRegimen document extracted SourceOfOrganisms document extracted Y AbstractURL bibliographic database FullTextURL bibliographic database URL of online article The UMLS Metathesaurus is a rich database of information on concepts that appear in one or more of a number of different controlled vocabularies and classifications used in the field of biomedicine. It provides a uniform, integrated distribution format of over 95 biomedical vocabularies and classifications and contains syntactic information. All Metathesaurus concepts are assigned to specific types or categories – e.g., "Disease or Syndrome," "Virus" – and the Semantic Network contains information about the permissible relationships among these types – e.g., "Virus" causes "Disease or Syndrome" [ 45 ]. The 2004 edition of the UMLS Metathesaurus includes over 1 million biomedical concepts and 2.8 million concept names in its source vocabularies [ 46 ]. The thesauri are reviewed (curated by expert indexers) in order to create a consistent controlled vocabulary structure. As indicated in Table 1 , research concepts and organism type thesauri are derived from the UMLS. As new concepts are identified from the document's data tables and figures, the UMLS is used to identify preferred terms that are added to the controlled vocabulary database. In addition to the preferred term, its synonyms, semantic type, broader and narrower terms and Unique Identifier are captured. The UMLS provides a very powerful approach to rapidly creating a robust scientific thesaurus in support of consistent and precise searching. Further, the semantic type descriptors for each concept and semantic network may offer some interesting opportunities for intelligent searching and mapping of research findings and their relationships in the future. At present, data extraction utilizes systems with both manual and automated processes. An evolving thesauri-building and revising approach are important components of the Telemakus system to ensure that vocabulary identification and management reflect the specialized needs of the knowledge domain as new research concepts are identified and reported. The knowledgebase construction process begins with an Internet search of a bibliographic database (e.g., PubMed, Web of Science ® , etc.). Database elements are extracted and verified against the relevant thesaurus. As new concepts are identified the UMLS is checked for the preferred term and it is added to the appropriate Telemakus thesaurus – along with synonyms, narrower and broader terms. One of the key innovations in the Telemakus system is the use of the data tables and figures for locating the concepts studied (and tested) by the researchers. Concentrating on the legends from data tables and figures focuses the extraction process and reduces the background noise of the full-text document, making the process tractable. In general, the information content of data tables and figures can be broken into two types: "facts" and "findings." Facts include reporting experimental design and comparative characteristics of animals in the study group (e.g., weight, age, pre-existing conditions, etc.). Findings are the results of the study (the research findings). Research findings are extracted from each of the "findings" data tables in a process described in Figure 1 . Figure 1 Process for deriving research relationships from data tables Table 2 provides a list of legends (the descriptions of content) from data tables and figures from a single research report and the end results of the extraction process. The legends are categorized into information content type (Fact or Research Finding), extracted concept relationships and concept relationships normalized (preferred terms) using the UMLS tools. In Table 2 , the first two legends report "facts" (the experimental design and the composition of the diets of the research animals) while "findings" are reported in the remaining legends. The third column displays the noun phrases extracted from the legends which are then mapped to their corresponding UMLS preferred terms, as seen in the fourth column. When mapped, the term "dietary intake" maps to "energy intake" and "mammary gland carcinomas" maps to "mammary neoplasms." This provides a "controlled vocabulary" which enhances the consistency of retrieval from the knowledgebase. Table 2 Information content type categorization and relationship concept candidates for a sample of table/figure legends Extracted from – Zhu Z, Haegele AD, Thompson HJ: Effect of caloric restriction on pre-malignant and malignant stages of mammary carcinogenesis. Carcinogenesis 1997, 18 (5) :1007–12 Table/Figure Legend Type Extracted Concept Relationships Concept Relationships (after Normalization using UMLS tools) Table I. Sequence of events that comprised the experimental design FACT None none Table II. Composition of diets FACT None none Fig 1. Effect of caloric restriction on dietary intake, body weight gain and the ratio of cumulative body weight/cumulative diet intake. FINDING • dietary intake – caloric restriction • body weight – caloric restriction • energy intake – caloric restriction • body weight – caloric restriction Table III. Effect of calorie restriction on the proportion of intraductal proliferations, ductal carcinoma in situ and carcinomas in mammary glands FINDING • intraductal proliferations – caloric restriction • ductal carcinoma in situ – caloric restriction • mammary gland carcinomas – caloric restriction • intraduct carcinoma of breast – caloric restriction • mammary neoplasms – caloric restriction Fig 2. Effect of calorie restriction on cumulative and final incidences of mammary carcinomas. FINDING • mammary carcinomas – caloric restriction • mammary neoplasms – caloric restriction Fig 3. Percentage distribution of lesions in a dietary group that were: intraductal proliferations, ductal carcinoma in situ and adenocarcinoma. FINDING • lesions – dietary group • dietary group – intraductal proliferations • dietary group – ductal carcinoma in situ • dietary group – adenocarcinoma • lesions – energy intake • intraduct carcinoma of breast – energy intake • adenocarcinoma – energy intake Fig 4. Effect of caloric restriction on urinary excretion of immunoreactive cortical steroid. FINDING • immunoreactive cortical steroid – caloric restriction • adrenal cortex hormones – caloric restriction A current focus is the application of natural language processing (NLP) techniques to assist in the automation of concept extraction process. MetaMap, a program developed by the National Library of Medicine ® , (NLM ® ) is being tested as a means of automatically parsing the legends from the data tables and figures to identify preferred UMLS concepts for addition to the Telemakus thesauri. MetaMap maps arbitrary text to concepts in the UMLS Metathesaurus; or, equivalently, it discovers Metathesaurus concepts in text. With this software, text is processed through a series of modules. First it is parsed into components including sentences, paragraphs, phrases, lexical elements and tokens. Variants are generated from the resulting phrases. Candidate concepts from the UMLS Metathesaurus are retrieved and evaluated against the phrases. The best of the candidates are organized into a final mapping in such a way as to best cover the text [ 47 ]. Telemakus KnowledgeBase System Architecture The Telemakus system architecture centers on: a relational database; a set of tools used to populate the knowledgebase with data extracted from bibliographic databases and full-text research reports; and several server side tools and programs responsible for delivering the content of the database to the public via the WWW. The entire system is built from open-source components, leveraging standard protocols and tools whenever possible. The document processing system is initiated by an analyst who runs, reviews and edits as necessary extractions from the document being processed. It currently consists of a number of discrete phases to download, extract and analyze each document. These services are built primarily in Java running behind Tomcat and Apache and accessed by the analyst through the browser. For the public Telemakus website interface, a number of open-source solutions have been selected and configured. An Apache web server intercepts all requests and delegates them to surrogate processes dedicated to each respective task. For requests to display the data from the database, the request is delegated to Zope, a content management service, for responding to the user's request. This typically includes running SQL queries against a PostgreSQL database and rendering the results in the conceptual schema that serves as a surrogate for each document. For tasks beyond simple queries and HTML requests, a Java Servlet™ is employed. As plain HTML is insufficient to effectively display and interact with the relationship map, a Java™ applet, TouchGraph, is used. TouchGraph is an open-source concept-mapping tool for creating and navigating links between information sources. The tool was chosen for Telemakus because of its flexibility, customizing capabilities, high quality source code and compatibility with most browsers and operating systems (OS). The TouchGraph visualization package serializes maps to and from XML. By using Java, HTTP and XML, TouchGraph makes it easy to dynamically feed content to generate interactive nodes-and-edges maps. Database Query and Navigation: How Does it Work? Figure 2 shows the initial search screen – the starting point for a search of a knowledgebase. The user can search using Boolean logic on a number of fields, including the abstract, keywords, full-text, title, research findings, etc. Each of the thesauri – Author, Research Findings, Organisms and Source of Organisms – are also available for browsing and are directly searchable. Sorting is supported by year, first author, journal title. Figure 2 Telemakus search screen Figures 3 and 4 show the results of the search. Clicking on the first listing (Chung) results in the retrieval of the complete record for that item in the research report schema format (Figure 5 ), a rapid summary of research methods and organism characteristics that provides quick links to a variety of types of information including the full-text of the research article. Clicking on any blue highlighted item under "Table/Figure" takes the searcher to the respective figure in the full-text article. Figure 3 Display of retrieval set for a search on "neoplasms" (part 1) Figure 4 Display of retrieval set for a search on "neoplasms" (part 2) Figure 5 Research report schema Schema for one of the retrieved scientific reports from a search on caloric restriction and neoplasms The research report schema also serves as a convenient interface for searching for related research concepts, offering a rapid way of following research connections through the database. For instance, clicking on "killer cells, natural – ad libitum" would retrieve additional articles that present data tables linking those two concepts. The "map it" function, at the bottom of the retrieval set (Figure 4 ) provides access to the visualized maps of research findings connections for the current retrieval set. Examples of the concept maps generated by clicking on "map it" from Figure 4 are presented in Figures 6 and 7 . Figure 6 presents a subset, more focused, map of research findings relating to the research concept of interest. Blue links highlight a reported (by the authors of the research report) statistically significant finding. The visualization tool permits moving from link to link and expanding the view to include a map of all research relationships reported in the retrieved set of documents (Figures 3 and 4 ). The user can also initiate a new search of a research term or link of interest (e.g., the relationship between survival rate and antioxidants) to retrieve all research papers which have reported this linkage. The iterative nature of the search process and ability to explore research connections from both the research schema as well as the research concept map supports the process of hypothesis exploration in a way that mimics the way many scientists work-by providing a means of exploring a variety of types of connections. Figure 6 Concept map of research findings linked to neoplasms Figure 7 Expanded concept map of research findings relationships Results The first completed Telemakus knowledgebase focuses on caloric restriction in aging and is freely available at . Caloric restriction was an ideal starting point for Telemakus because it is an important and rapidly expanding specialized area of the biology of aging that is also highly interdisciplinary. Telemakus is a component of the Science of Aging (SAGE) project funded by the Ellison Medical Foundation. Other SAGE partners include the American Association for the Advancement of Science and Highwire Press, Stanford University (SAGE Knowledge Environment web site: ). Formal usability testing of Telemakus is underway and will be the subject of a future article. Because a major goal of the Telemakus research program is to study scientists' approaches and preferences for accessing and using the scientific literature in order to create models and approaches for user-centered knowledgebases, researchers have been involved in the iterative design and testing of the system from its inception. The primary goals of this evaluation are to: 1. Determine scientists' preferences for working with the research literature. 2. Model preferred features based on those preferences. 3. Test the completeness of schema elements and structure as a document surrogate. 4. Experiment with and identify optimal visual representations to meet user needs. 5. Iteratively review/evaluate/test for improved performance in response to user feedback. 6. Identify domain(s) for future knowledgebase creation. In general, response to each successive iteration of Telemakus has been positive and included constructive feedback for system enhancements and expansions. User feedback affirms that retrieval based on research findings is a unique and highly desirable core function. Further, the Telemakus schematic document surrogate has been enthusiastically received as a major improvement over the traditional citation format with abstract. As one researcher stated (and several others have echoed), "The strengths of Telemakus are doing what PubMed does not do, which is to give an outline of the main points and to allow searching off the figure/table legends, organisms/sources and outcome fields." Additional feedback relates to the labeling of concept relationships as "statistically significant." Some researchers are interested in knowing the level of reported significance (i.e., p value) and asked for a detailed labeling to document this. In addition, there have been requests to consider labeling the relationships (i.e., directionality, type, etc.). Early testing of the mapping function resulted in the observation that color-blind individuals would not be able to see lines that were labeled with red or green, which led to a change in the mapping color scheme. There have been many additional suggestions for improving the visualization, including addition of three-dimensional representations and allowing more user control of the presentation itself. Some researchers have expressed interest in being able to build maps based on the date a particular research finding was reported. This functionality would create time sequence maps that show the progression of research over time and, perhaps, will demonstrate paths of research that have been discarded prematurely and may be worth re-visiting. A number of researchers have indicated the utility of this approach for teaching purposes – for a student to quickly get a sense of the research "facts" in a domain. There have also been requests for tools to support downloading subsets of the knowledgebases, as well as tools to allow individuals to manipulate maps and add their own research findings and ideas to the concept maps. Discussion While initial Telemakus development has focused on the research literature related to caloric restriction and the biology of aging, the goal is to expand into additional domains. For example, tables of genetic sequence information, which display reported relationships between gene sequences and diseases, are a natural area of expansion for Telemakus. There is great potential for building linkages between Telemakus knowledgebases and other factual databases, e.g., NCBI entrez resources. In addition, scientists from other domains beyond biomedicine (statistics, environmental research) have indicated that a customized schematic representation of research findings could be very useful in their domains. Speeding up document processing so Telemakus can easily and efficiently scale for comprehensive treatment of domains is a key priority. As discussed previously, the UMLS Metathesaurus resources (in particular, MetaMap) are proving extremely useful. In addition, the Semantic Network will be tested for enhancing searching and visualization of research findings. We will continue to utilize an iterative development method so that results of usability evaluation can immediately inform development of additional features. In particular, we want to test our hypothesis that the mapping feature will promote knowledge discovery by showing graphically what is known as well as, through lack of links, what research linkages have not yet been tested. Since basic sciences researchers tend to initially focus on the data found within a report's tables and figures (sometimes before or instead of actually reading the article), extracting the headings and providing linked research concepts mimics a researcher's traditional approach to reading the research literature [ 42 , 48 ]. When users understand regularities in information spaces (layout, structure, landmarks, etc.) as schemata they acquire navigational knowledge in the form of a cognitive map of the information space. By providing the conceptual schema based on the scientist's own view of scientific research as a document surrogate, Telemakus provides a roadmap for reading and rapidly browsing through numerous research reports and aids in acquisition of the navigational knowledge required for a user to successfully explore complex information spaces [ 49 ]. One of the long-term goals of the Telemakus system is not to build knowledgebases "ad infinitum" but rather to create flexible tools for users to quickly and efficiently locate and visualize aggregate research findings from any domain which reports research findings as data. As more and more full-text research reports are available on the Internet, we believe the tools we are developing will provide an important approach for focusing on research findings and providing visual cues for quick review and assimilation. Conclusions The Telemakus KnowledgeBase System builds on a good deal of prior research in a variety of domains. It provides a flexible new approach for creating knowledgebases to facilitate retrieval and review of scientific research reports. In formalizing the representation of the research methods and results of scientific reports, Telemakus offers a potential strategy to enhance the scientific discovery process. While other research has demonstrated that aggregating and analyzing research findings across domains augments knowledge discovery, the Telemakus system is unique in combining informative document representations with interactive concept maps of linked relationships across groups of research reports. Telemakus presents a novel approach to creating useful and precise document surrogates and may re-conceptualize the way we currently represent, retrieve and assimilate research findings from the published literature. Competing interests None declared. Authors' contributions SF conceived the study and contributed to its design, coordination and evaluation. DR and PB participated in the design of the study. DR led the overall coordination and drafted the manuscript. PB led the technical implementation. GMM contributed to the design, coordination and evaluation. All authors read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524025.xml |
524031 | Metabolic response of people with type 2 diabetes to a high protein diet | Background One of the major interests in our laboratory has been to develop a scientific framework for dietary advice for patients with diabetes. Knowledge regarding the metabolic consequences and potential effects on health of protein in people with type 2 diabetes has been a particular interest. Results We recently have completed a study in which dietary protein was increased from 15% to 30% of total food energy. The carbohydrate content was decreased from 55% to 40%, i.e. dietary protein replaced part of the carbohydrate. This resulted in a significant decrease in total glycohemoglobin, a decrease in postprandial glucose concentrations and a modest increase in insulin concentration. Renal function was unchanged. Currently we also are determining the metabolic response to a diet in which the carbohydrate content is further decreased to 20% of total food energy. The %tGHb decrease was even more dramatic than with the 40% carbohydrate diet. Conclusion From these data we conclude that increasing the protein content of the diet at the expense of carbohydrate can reduce the 24-hour integrated plasma glucose concentration, at least over a 5-week period of time. The reduction was similar to that of oral agents. Renal function was not affected significantly. Thus, increasing the protein content of the diet with a corresponding decrease in the carbohydrate content potentially is a patient empowering way of reducing the hyperglycemia present with type 2 diabetes mellitus, independent of the use of pharmaceutical agents. | Background Our research group has been and continues to be interested in the metabolic response of people with type 2 diabetes to macronutrients in the diet in general. More recently, we have been particularly focused on the metabolic response to a high protein diet. The reason for this is three fold: First, for several years, one of our major goals has been to develop a scientific framework for dietary advice based on sound metabolic principles. Second, we have data that suggest that an increase in dietary protein may be salutary for people with diabetes. And lastly, knowledge regarding the metabolic consequences and potential effects on health of dietary protein has lagged far behind that of dietary fats and carbohydrates. In this paper we will focus on the concept that an increase in dietary protein may be salutary for people with diabetes, and particularly for the control of blood glucose. Results The concept that an increase in dietary protein may be useful in controlling the blood glucose would appear to be counterintuitive, since amino acids derived from ingested or endogenous proteins are major net gluconeogenic substrates. The first step in the metabolism of amino acids is the removal of the amino group. This is condensed with CO 2 to form urea. The remaining deaminated product is largely converted into glucose through gluconeogenesis, although a small amount is converted into other products. (Figure 1 ). Figure 1 The α amino group from an amino acid is condensed with CO 2 to form urea. The remaining carbon skeleton can be used to synthesize glucose. Indeed, in 1915, Dr. Janney [ 1 ] reported that 3.5 g glucose can be obtained from 6.25 g of ingested meat or beef protein. Thus, theoretically for every 100 g of protein ingested, 56 g of glucose can be produced. For other proteins this varies between 50 and 84 grams. Thus when developing a dietary regimen for diabetic patients, dietitians were taught to count not only carbohydrate, but also to count 56% of the protein as carbohydrate. The rationale behind this recommendation was that carbohydrates raised blood glucose, proteins are converted to glucose, therefore, dietary proteins will raise blood glucose. However, in 1924, Dr. MacLean [ 2 ] reported that when a man with diabetes, and a fasting blood glucose of 280 mg/dl, ingested 250 g of meat, which is the equivalent of 50 grams of protein, and which should result in the production of ~25 g of glucose, there was no change in blood glucose over the 5 hours of the study. When the same subject ingested 25 g of glucose, there was a very large increase in blood glucose; indeed, it increased up to 600 mg/dl. This lack of increase in blood glucose concentration following the ingestion of protein was confirmed by Conn and Newburgh in 1936 [ 3 ]. These investigators fed a relatively enormous amount of beef, i.e. 1.3 pounds of beef, which is the equivalent of ~136 g of protein and which should yield 68 g of glucose, to a normal subject with a fasting blood glucose of 65 mg/dl and to a subject with diabetes whose fasting blood glucose concentration was 150 mg/dl. In neither case was there an increase in blood glucose concentration over the 8 hours of this study. However, when the same subjects were given 68 g of glucose, there clearly was an increase in glucose concentration in both cases. That ingested protein did not raise the blood glucose was largely ignored, in spite of this evidence in the scientific literature. Indeed, in his textbook in 1945 [ 4 ], Dr. Joslin, one of the most influential diabetologists at that time, was still counseling dietitians and patients to consider 56% of dietary protein as if it were carbohydrate. Single meal studies done in our laboratory With this background information, we decided to do a study expanding on these early observations. Seven subjects with type 2 diabetes [ 5 ], and 8 subjects without diabetes [ 6 ] ingested 50 g of protein in the form of very lean beef. In the non-diabetic subjects, there was no change in blood glucose concentration over the 4 hours of the study, as had been noted previously. However, in the subjects with type 2 diabetes, the glucose concentration actually decreased over the 5 hours of that study (Figure 2 ). Figure 2 Glucose (left panel) and insulin (right panel) response to ingestion of 50 g of protein in the form of lean beef. Data from 8 non-diabetic subjects (white lines, bottom) and 7 subjects with type 2 diabetes (yellow lines, top). (From [5,6]) We also determined the serum insulin response to the ingested protein and in confirmation of the studies of Berger [ 7 ], Fajans [ 8 ] and others, we observed a modest increase in the insulin concentration in the non-diabetic subjects [ 6 ]. However, there was a relatively large increase in insulin concentration in the subjects with type 2 diabetes [ 5 ]. Indeed, it was about four-fold greater than in the non-diabetic subjects (Figure 2 ). We also determined that the rise in insulin following the ingestion of 50 g of beef protein was just as potent in raising the insulin concentration as was the ingestion of 50 g of glucose [ 5 ]. That is, meat protein and glucose were equipotent in stimulating insulin secretion. In addition, we also demonstrated a linear dose-response relationship between the amount of beef ingested and the insulin response [ 5 ]. Since beef protein strongly stimulated insulin secretion, we next determined whether the simultaneous ingestion of protein with glucose would stimulate even more insulin and thus reduce the rise in glucose expected when glucose alone is ingested. We also were interested in determining if all common protein sources were equal in this regard. Therefore, we designed a study in which 9 – 15 males with untreated type 2 diabetes were given 50 g of glucose with or without 25 g of protein [ 9 ]. Seven protein sources were used: beef, turkey, gelatin, egg white, cottage cheese, fish and soy. The rationale behind giving 25 g of protein with 50 g of glucose, was that this ratio more closely resembles the ratio of protein to carbohydrate typically found in the diet. The plasma glucose and serum insulin concentrations were determined over a 5-hour period and the areas under the curves were calculated. The glucose area response clearly was decreased when glucose was ingested with 25 g of protein as beef, turkey, gelatin, cottage cheese, fish and soy. Only egg white did not result in a significant decrease in glucose area response when compared to the response to ingestion of glucose alone (Figure 3 ). Figure 3 Five hour integrated glucose area response to ingestion of 50 g glucose alone (pink bar) or 50 g glucose + 25 g protein in the form of beef, turkey, gelatin, egg white, cottage cheese, fish or soy (yellow bars, left to right) . (From [9]) When any of the proteins was added to the ingested glucose, the insulin area response was greatly increased (Figure 4 ). The smallest response was obtained with egg white, which was 190% or 1.9 fold over the response to glucose ingested alone. The greatest increase was with cottage cheese, which was 360% or 3.6 fold. As indicated previously, beef protein, on a weight basis, was just as potent as glucose in raising the insulin concentration. Since only 25 g of beef protein was ingested in the present study, the expected response would be 150% of that observed with just glucose ingestion [ 5 ]. With beef and every other protein source studied, the insulin response was greater than the theoretical expected response (Figure 4 ), strongly suggesting that there is a synergistic insulin response when protein is ingested with glucose [ 9 ]. In summary, in single meal studies in people with type 2 diabetes, dietary protein strongly stimulated insulin secretion and decreased the plasma glucose response to ingested glucose. Figure 4 Five hour integrated insulin area response to ingestion of 50 g glucose alone (pink bar) or 50 g glucose + 25 g protein in the form of beef, turkey, gelatin, egg white, cottage cheese, fish or soy (yellow bars, left to right). The horizontal line indicates the expected insulin area response. (From [9]) Insulin and Glucose Response to Mixed Meals Based on the above observations, we decided to determine whether an increase in dietary protein in association with a decrease in carbohydrate would decrease the 24 hour integrated plasma glucose concentration, increase the 24 hour integrated insulin concentration and decrease the % total glycohemoglobin in people with type 2 diabetes ingesting mixed meals over an extended period of time. We designed a study in which the protein content of the diet was increased from 15% of total food energy in a standard diet, to 30% protein in the experimental diet [ 10 ]. The carbohydrate content was decreased from 55% carbohydrate to 40% carbohydrate. However, it should be understood that since the additional protein can result in an increase in glucose production, the actual carbohydrate available theoretically would be about 48%, or a decrease in potential carbohydrate of only 7%. The fat content remained the same in both diets. Monounsaturated, polyunsaturated and saturated fat ratios were 10:10:10, respectively. Twelve people with untreated type 2 diabetes were studied using a randomized, crossover design. The subjects received each diet for 5 weeks with a washout period in between. The diets were isocaloric, and all food was provided. The subjects came to the Special Diagnostic and Treatment Unit 2–3 times each week to pick up the food, to be weighed, and to provide a urine specimen for creatinine and urea nitrogen determination. The major end-point of the study was to determine if there was a significant decrease in % total glycohemoglobin (%tGHb). The reason that 5 weeks was chosen for the study is because this is the time required for the % total glycohemoglobin to decrease 50% of its ultimate value after a rapid stable decrease in blood glucose concentration (Figure 5 ), [ 11 ], i.e. the results obtained should represent 50% of the ultimate % total glycohemoglobin response. Figure 5 Rate of change in % tGHb The subjects were weight stable on both diets. We considered this to be a very important aspect of the study because we wanted to attribute any metabolic changes to the diet per se, and not to be confounded by weight loss (or gain) [ 10 ]. Urine urea, normalized to the urine creatinine, was measured as an index of compliance. Since the protein content of the diet was doubled, one would expect that the urine urea:creatinine ratio also would approximately double if the subjects were compliant. The ratio on the standard diet was ~7 and was stable throughout the 5 weeks. When the same subjects were given the 30% protein diet, the urine urea:creatinine ratio was ~13–14, i.e. a value that one would expect with good compliance with the diet. The fasting glucose concentration did not change when the subjects received the 30% protein diet. However, the postprandial glucose concentrations were lower throughout the day [ 10 ]. Although the differences in postprandial glucose values were not very large, when integrated over the 24-hour period, there was a 38% decrease in postprandial glucose area response. If the 24-hour integrated area is considered to be 100% when the subjects ingested the 15% protein diet, when they ingested the 30% protein diet it was 62% (Figure 6 ). Figure 6 24-hr integrated plasma glucose area response in 12 subjects with type 2 diabetes after ingesting the 15% protein or the 30% protein diet for 5 weeks. (From [10]) Even though the postprandial glucose concentration was decreased on the 30% protein diet, the insulin area response was modestly increased (Figure 7 ). Figure 7 24-hr integrated serum insulin area response in 12 subjects with type 2 diabetes after ingesting the 15% or the 30% protein diet for 5 weeks. (From [10]) The % total glycohemoglobin decreased slightly from 8% to 7.7% during the 5 weeks of the study when the subjects were ingesting the 15% protein diet. When the subjects ingested the 30% protein diet, it decreased from 8.1 to 7.3%, i.e. the decrease was 0.8 (Figure 8 ). Figure 8 %tGHb response in 12 subjects with type 2 diabetes at weekly intervals while ingesting a 15% or a 30% protein diet. (From [10]) To put this decrease in % glycohemoglobin into perspective, the Physicians Desk Reference for 2003 [ 12 ] was consulted in regard to the decrease in %HbA1c or %glycohemoglobin when subjects with type 2 diabetes were given rosiglitazone or metformin, drugs commonly used to treat people with type 2 diabetes. In subjects receiving 4 mg rosiglitazone twice a day, which is a maximal dose, the mean decrease in HbA1c was 0.7% over a 16-week period of time (Table 1 ). For metformin, at a maximum dose of 2500 mg daily, the decrease was 1.4% over a 29-week period. Table 1 Comparison of treatment Agent Dose Duration of Treatment Decrease in %tGHb or %HbA1c Rosiglitizone 4 mg bid 16 weeks 0.7% Metformin 2500 mg 29 weeks 1.4% 30% Protein Diet 5 weeks 0.8% (1.6%) PDR 2003 [12] With the 30% protein diet, the decrease was 0.8% over the 5 weeks of our study. The ultimate decrease could be 1.6%, since at 5 weeks (35 days) the %tGHb would have decreased by only 50% of the expected final decrease (see Figure 5 ). Thus, the decrease would be similar to that obtained using either of the above two medications. Since there has been concern that a high protein diet may impair renal function, the creatinine clearance was determined at the end of the period of time the subjects ingested the 15% protein diet and at the end of the period of time that the subjects ingested the 30% protein diet. There was essentially no difference. The microalbumin excretion also did not change (Table 2 ). Table 2 Renal data 15% Protein Diet 30% Protein Diet Creatinine Clearance (ml/min) 122 ± 11 113 ± 27 Microalbumin (mg) 7.8 ± 1.7 7.0 ± 0.8 Also the differences in total cholesterol, HDL-cholesterol, LDL-cholesterol were not significant. The fasting triacylglycerol concentration decreased significantly when the subjects were on the 30% protein diet (Table 3 ). Table 3 Lipid data 15% Protein Diet 30% Protein Diet Total Cholesterol (mg/dl) 181 ± 15 171 ± 12 HDL-Cholesterol (mg/dl) 38 ± 3 39 ± 3 LDL-Cholesterol (mg/dl) 100 ± 12 101 ± 12 Triacylglycerol (mg/dl) 199 ± 20 161 ± 21* * P < 0.05 Discussion In summary, the integrated postprandial glucose area response was 38% less following ingestion of the 30% compared to the 15% protein diet. Total glycohemoglobin decreased significantly from 8.1 to 7.3% and potentially could result in a decrease to 6.5%. The integrated insulin concentration increased modestly. Renal function, LDL, HDL, and total cholesterol were unchanged. The triacylglycerol concentration decreased. Conclusions From these data we conclude that increasing the protein content of the diet at the expense of carbohydrate can reduce the 24-hour integrated plasma glucose concentration, at least over a 5-week period of time. The reduction was similar to that of oral agents and renal function was not affected significantly. Thus, increasing the protein content of the diet with a corresponding decrease in the carbohydrate content potentially is a patient empowering way of reducing the hyperglycemia present in people with type 2 diabetes mellitus, independent of the use of pharmaceutical agents. Results of a further modification in macronutrient content More recently we have completed study comparing an experimental diet to the standard diet, over a 5-week period of time. In the experimental diet, the protein was increased from 15% to 30% as in the above study. However, in this study the carbohydrate content was decreased from 55% to 20% of total food energy and the fat content was increased from 30% to 50%. The subjects studied were people with untreated type 2 diabetes. It was a weight maintenance diet, with a randomized crossover design. The %tGHb decrease was even more dramatic (9.8 to 7.6%) [ 13 ]. Competing interests None declared. Authors' contributions Both authors were equally responsible for designing the experiments, evaluating the statistics, interpreting the data, writing the manuscript, and organizing the figures and tables. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524031.xml |
549587 | Use of Procalcitonin and C-Reactive Protein to Evaluate Vaccine Efficacy against Pneumonia | Background Pneumonia remains the leading cause of death in young children. The poor specificity of chest radiographs (CXRs) to diagnose pneumococcal pneumonia may underestimate the efficacy of pneumococcal conjugate vaccine in preventing pneumococcal pneumonia. Methods and Findings The efficacy of nine-valent pneumococcal conjugate vaccine among children not infected with HIV (21%; 95% confidence interval, 1%–37%) increased when CXR-confirmed pneumonia was associated with serum C-reactive protein of 120 mg/l (12mg/dl) or more and procalcitonin of 5.0 ng/ml or more (64%; 95% confidence interval, 23%–83%). Similar results were observed in children infected with HIV. Conclusion C-reactive protein and procalcitonin improve the specificity of CXR to diagnose pneumococcal pneumonia and may be useful for the future evaluation of the effectiveness of pneumococcal conjugate vaccine in preventing pneumococcal pneumonia. | Introduction While pneumonia remains the leading cause of death in children, the absence of sensitive and specific tools to make an etiological diagnosis is a major limitation to our understanding of the efficacy of vaccines against pneumonia. Blood cultures lack sensitivity, and cultures from lung aspirates may be influenced by delayed presentation, antecedent antibiotic therapy, difficulties finding an accessible site to aspirate, lack of skills in performing the procedure, and the perception of both parents and clinicians that the procedure is too invasive [ 1 , 2 ]. Procalcitonin, at a low threshold (≥0.25 ng/ml), has been shown to be useful in directing the use of antibiotics in adults with pneumonia [ 3 ], and a recent meta-analysis concludes that procalcitonin may offer some advantages over C-reactive protein (CRP) for discriminating bacterial from nonbacterial infections [ 4 ]. Most studies support the observation that children with bacterial infections have higher levels of CRP and procalcitonin than those with viral infections [ 4 , 5 , 6 ]. Our recent observation that many children with viral-associated pneumonia have a bacterial super-infection [ 7 ], however, suggests that a high level of CRP and procalcitonin may be associated with unrecognised bacterial co-infection in a child with an established viral aetiology for pneumonia. This may also explain why some studies have found procalcitonin and CRP not to be useful in distinguishing between bacterial and “viral” pneumonia [ 4 , 8 ]. Based on the postulate that the most likely chest radiograph (CXR) manifestation of pneumococcal pneumonia is alveolar consolidation, we reported the efficacy of a nine-valent pneumococcal conjugate vaccine (PnCV) in reducing CXR-confirmed pneumonia based on definitions recommended by a World Health Organization working group [ 9 ]. The observed reduction in the incidence of CXR-confirmed pneumonia in the intent-to-treat analyses in PnCV recipients not infected with HIV (20%; 95% confidence interval [CI], 2% to 35%) and infected with HIV (13%; 95% CI, −7% to 29%) likely underestimated the reduction in the incidence of pneumococcal pneumonia [ 10 ]. The reason for this is that the outcome measure (i.e., CXR-confirmed pneumonia) is not highly specific for pneumococcal pneumonia. Consequently, it is likely that many of the CXR-confirmed pneumonia episodes were not pneumococcal in origin and therefore could not have been prevented by the vaccine under evaluation. We thus evaluated the usefulness of procalcitonin and CRP to improve the specificity of CXR-confirmed pneumonia as an endpoint in vaccine efficacy trials. This analysis was not a primary objective of the study, and is therefore an hypothesis-generating analysis, which should be tested as an a priori hypothesis in other study settings. Methods The methods of the randomised, double-blind, placebo-controlled trial, the objectives of which were to measure PnCV efficacy against invasive pneumococcal disease and CXR-confirmed pneumonia, have been published [ 10 ] ( Protocol S1 ). Briefly stated, the study included 39,836 children, including an estimated 6.47% infected with HIV [ 7 ], and was performed in Soweto, South Africa. Children were randomised to receive either a nine-valent PnCV conjugated to CRM 197 (Wyeth-Lederle Vaccines and Pediatrics, Pearl River, New York, United States) or a placebo at 6, 10, and 14 wk of age. The nine-valent PnCV included serotypes 1, 4, 5, 6B, 9V, 14, 18C, 19F, and 23F (i.e., vaccine serotypes) [ 10 ]. We now further measured procalcitonin and CRP levels in serum obtained within 12 h of hospitalisation among children with CXR-confirmed alveolar consolidation in the intent-to-treat analysis. CXRs were requested for any child hospitalised with a clinical diagnosis of lower respiratory tract infection according to the diagnosis of the attending study-physician. CRP values were determined at the time of admission by immunoturbidometry (717 Automated Analyzer, Boehringer Mannheim/Hitachi, Mannheim, Germany) and were reported in milligrams per liter with a lower threshold set at 3 mg/l or lower. In those children in whom CRP was not measured on admission to hospital, archived serum obtained within 12 h of admission and stored at −70 °C was used for analysis. The archived serum samples were also used for measuring quantitative procalcitonin levels using the LUMItest PCT assay (BRAHMS Diagnostica, Berlin, Germany). The CRP and procalcitonin tests were performed using commercially available assays according to the manufacturers' recommendations at the National Health and Laboratory Services, Johannesburg, South Africa. Bacterial Cultures Blood was cultured for bacterial growth on admission and processed using the BacT/Alert microbial detection system (Organon Teknika, Durham, North Carolina, United States). Isolates of Streptococcus pneumoniae were serotyped using the quellung method at the Respiratory and Meningeal Pathogens Research Unit and the results validated at the Statens Serum Institute in Copenhagen, Denmark [ 10 ]. Statistics Data were analysed using S
TATA version 8.0 (StataCorp, College Station, Texas, United States) and Epi Info version 6.04d (Centers for Disease Control and Prevention, Atlanta, Georgia, United States). Vaccine efficacy was calculated using Epi Info based on the formula: vaccine efficacy (percent) = [(ARU − ARV)/ARU] × 100, where ARU is attack rate in unvaccinated individuals and ARV is attack rate in vaccinated indivduals. The vaccine-attributable reduction in disease (VAR) was estimated by measuring the difference in incidence rate between vaccine and placebo recipients and expressed per 100,000 child years of observation. A p- value of 0.05 or lower was considered significant. All analyses were performed on an intent-to-treat basis, which included the first event of CXR-confirmed pneumonia in any child who had received at least a single dose of study vaccine. Vaccine efficacy calculations were also performed for bacteremic pneumococcal pneumonia that included all first episodes of pneumonia associated with growth of S. pneumoniae of any serotype from blood. Ethical Considerations The efficacy study and subsequent study evaluating the role of procalcitonin and CRP in children with CXR-confirmed pneumonia were approved by the Ethics Committee for Research on Human Subjects, University of the Witwatersrand, South Africa ( Protocol S2 ). Guardians gave informed consent for the collection of serum and its use for diagnostic assays that may improve the diagnosis of pneumonia ( Protocol S3 ). Results In children not infected with HIV with CXR-confirmed pneumonia, procalcitonin and CRP values were available for 132 (78.1%) of 169 vaccine recipients and 167 (78.8%) of 212 placebo recipients ( p = 0.88). The point estimate of vaccine efficacy in this cohort of children for whom serum was available was 21% ( p = 0.04; 95% CI 1%–37%), not significantly different from that determined in the original study (20%, p = 0.03; 95% CI 3%–35%). At CRP levels of 120 mg/l or more in conjunction with the primary CXR-confirmed outcome measure, the estimate of vaccine efficacy increased to 38% ( p = 0.05); at procalcitonin levels of 5 ng/ml or more, the estimated efficacy increased to 46% ( p = 0.04); and if both conditions were met, the estimated efficacy increased to 64% ( p = 0.006) ( Table 1 ). Table 1 CRPand Procalcitonin Improve the Specificity of CXR to Measure the Efficacy of PnCV in the Prevention of Pneumococcal Pneumonia CXR-confirmed alveolar consolidation based on World Health Organization guidelines for interpretation. Point estimates differ from those previously published because only children with CXR-confirmed pneumonia for whom serum for CRP and procalcitonin measurements were available are included a CXR-AC, CXR-confirmed alveolar consolidation; PCT, procalcitonin b Pneumonia associated with growth from blood of S. pneumoniae, of any serotype c Pneumonia associated with growth from blood of S. pneumoniae of vaccine-serotype-specific bacteremic pneumonia Among children infected with HIV with CXR-confirmed pneumonia, procalcitonin and CRP values were available for 139 (76.4%) of 182 vaccine recipients and 153 (73.2%) of 209 placebo recipients ( p = 0.47). Whereas vaccine efficacy was not significant for CXR-confirmed pneumonia (9%; p = 0.38), vaccine efficacy was significant when measured against CXR-confirmed pneumonia in conjunction with a CRP level of 120 mg/l or more (35%; p = 0.03) or a procalcitonin level of 5 ng/ml or more (33%; p = 0.04). As observed in children not infected with HIV, the point efficacy estimate in children infected with HIV with CXR-confirmed pneumonia was greatest if both CRP and procalcitonin were at or above 120 mg/l and 5 ng/ml, respectively (52%; p = 0.004; Table 1 ). The sensitivity of CXR-confirmed pneumonia plus CRP ≥120 mg/l plus procalcitonin ≥5 ng/ml in detecting vaccine efficacy against pneumococcal pneumonia was analysed by comparing the VAR per 100,000 child years to that observed using an outcome of bacteremic pneumococcal pneumonia. In children not infected with HIV, the combined outcome of CXR-confirmed pneumonia plus CRP ≥120 mg/l plus procalcitonin ≥5 ng/ml (VAR = 37) identified 5.3-fold more cases of pneumonia that were prevented by vaccination than was identified by bacteremic pneumococcal pneumonia (VAR = 7) and 4.1-fold more cases than was identified by vaccine-serotype-specific bacteremic pneumococcal pneumonia (VAR = 9). Similarly, CXR-confirmed pneumonia plus CRP ≥120 mg/l plus procalcitonin ≥5 ng/ml in children infected with HIV (VAR = 772) identified 1.6-fold more cases of pneumonia that were prevented by vaccination than was identified by pneumococcal bacteremic pneumonia (VAR = 483) and 2.2-fold more cases than was identified by vaccine-serotype-specific bacteremic pneumococcal pneumonia (VAR = 344). Among the children not infected with HIV with CXR-confirmed pneumonia, the likelihood of a difference between placebo and vaccine recipients was significantly increased by the presence of a CRP level of 120 mg/l or more and a procalcitonin level of 5 ng/ml or more (odds ratio = 1.37; 95% CI, 1.09–1.73; p = 0.027). This was true also among children infected with HIV (odds ratio = 1.41; 95% CI, 1.14–1.75; p = 0.005). Discussion Our results show that the true efficacy of the PnCV against pneumococcal pneumonia may be underestimated if one relies solely on a radiological diagnosis of pneumococcal pneumonia. The lack of specificity of CXRs for inferring bacterial versus non-bacterial etiology of pneumonia has been previously described [ 11 , 12 ]. The specificity of CXR-confirmed pneumonia as an outcome measure of efficacy of PnCV to prevent pneumococcal pneumonia was improved when it was analysed together with indirect evidence suggestive of bacterial infection, namely, elevated CRP and procalcitonin levels. This outcome measure of CXR-confirmed pneumonia plus CRP ≥120 mg/l plus procalcitonin ≥5 ng/ml, while more specific than CXR-confirmed pneumonia alone, will not be 100% accurate because pneumonia associated with other bacteria, including that due to pneumococcal serotypes against which the PnCV has no effect, may well present in a similar manner. An alternate explanation for our results, rather than that of improved specificity in diagnosing pneumococcal pneumonia, may be that the criteria of CXR-confirmed pneumonia coupled with a CRP level of 120 mg/l or more and a procalcitonin level of 5 ng/ml or more detected more severe disease against which the vaccine may have been more efficacious. Although we are unaware of data to support that CRP and procalcitonin are elevated in those with severe pneumonia, the absence of such data supporting this interpretation does not rule out this possibility. Importantly, the levels of CRP and procalcitonin used in this report are designed for specificity rather than sensitivity; it would be inappropriate to use the same threshold values of CRP or procalcitonin during the course of routine clinical practice when managing individuals suspected of having bacterial pneumonia. In that instance, it would be more appropriate to use thresholds that have a high sensitivity, to ensure that all children with possible bacterial infection are adequately treated [ 3 , 4 , 8 ]. Procalcitonin and CRP were equally useful in children infected and not infected with HIV to improve the specificity of the pneumococcal pneumonia efficacy endpoint in the vaccine trial among children with CXR-confirmed pneumonia. This is particularly important, as our initial observation of a non-significant reduction in CXR-confirmed pneumonia was attributed to the complexity of etiology of pneumonia, and thus the even poorer specificity of CXRs for identifying pneumococcal pneumonia in children infected with HIV compared to children not infected with HIV [ 13 , 14 ]. The current study thus indicates that the specificity of CXR-confirmed pneumonia for diagnosing pneumococcal pneumonia may also be improved in children with HIV through the concurrent use of procalcitonin and CRP. A distinction must be made between measures of vaccine efficacy for which the increased specificity of CXR-confirmed pneumonia plus CRP ≥120 mg/l plus procalcitonin ≥5 ng/ml over CXR-confirmed pneumonia alone is apparent, and studies of the burden of disease prevented (VAR), which are optimised when vaccine sensitivity is maximised. Our data show that the VAR is greater for CXR-confirmed pneumonia plus CRP ≥120 mg/l plus procalcitonin ≥5 ng/ml than that calculated using the very highly specific endpoints of serotype-specific or all-serotype pneumococcal bacteremic pneumonia. The best estimates of VAR will, however, be provided by endpoints that sacrifice some specificity for maximal sensitivity. Such studies include less specific clinical endpoints and are outside of the scope of the current analysis [ 7 ]. Our findings have implications for the planning of future studies aimed at evaluating the effectiveness of pneumococcal vaccines against bacterial pneumonia. Our data suggest that the outcome measure of CXR-confirmed pneumonia together with elevated CRP and procalcitonin levels may be more accurate as a surrogate of pneumococcal pneumonia than CXRs on their own. The increased specificity of this endpoint may allow smaller sample sizes for future vaccine efficacy/effectiveness studies. Based on the incidence of CXR-confirmed pneumonia in this report (167 [0.9%] of 18,626), a sample size of 80,058 children were required to detect the 20% reduction in CXR-confirmed pneumonia with 80% power and an alpha of 0.05. The actual power for this outcome in this study was thus only 46.8%. Using the more specific outcome measure of CXR-confirmed pneumonia with a CRP level of 120 mg/l or more and a procalcitonin level of 5 ng/ml or more, the sample size required to detect the observed 64% reduction in outcome was 44,734, i.e., 56% of the sample size required when measuring vaccine efficacy against CXR-confirmed pneumonia alone, and the power of the current study would have been increased from 46.8% to 71.5%. Our findings therefore suggest that CXR-confirmed pneumonia coupled with serological markers of CRP and procalcitonin is a more specific marker of pneumococcal pneumonia and may therefore provide a closer estimate of the efficacy of the PnCV against pneumococcal pneumonia. Supporting Information Protocol S1 Protocol of the Original Phase 3 Study (250 KB DOC). Click here for additional data file. Protocol S2 Ethics Committee Letter of Approval (378 KB DOC). Click here for additional data file. Protocol S3 Informed Consent Form for Original Phase 3 Study (24 KB DOC). Click here for additional data file. Patient Summary Background Pneumonia is the leading cause of death in children worldwide. Pneumonia can be caused by different bacteria and viruses, and there are no easy diagnostic tests to find out which bacterium or virus has caused the disease in a particular patient. This not only causes problems with prescribing the best treatment, but also makes it hard to evaluate vaccines that might protect against some causes of the disease but not others. Why Was This Study Done? The researchers involved in this study have evaluated vaccines against a particular bacterium (called Pneumococcus ) that is the leading cause of pneumonia in children. They have begun to test these vaccines in children, but were looking for more specific ways to distinguish cases of pneumonia caused by this particular bacterium from those caused by other bacteria or viruses. What Did the Researchers Do? They had previously done a trial for a vaccine that relied on chest X-rays to diagnose pneumonia. They had also collected blood from children who had participated in the trial and had become sick with pneumonia. They now checked those blood samples for two markers that indicate a bacterial infection and re-analyzed the study. What Did They Find? The vaccine was able to protect children—to some extent—against pneumonia. The vaccine appeared to offer greater protection against pneumonia when the pneumonia was diagnosed by a combination of X-rays and high levels of the two blood markers than when the illness was diagnosed just with a chest X-ray. What Does This Mean? These results raise the possibility that a combined test (chest X-ray plus two blood markers) is better at assessing whether the pneumonia vaccine works than just a chest X-ray alone. What Next? Because of the way this study was done—adding a specific analysis to a clinical trial after it was completed, rather than planning to test a hypothesis from the outset—it cannot be considered as proof of the idea tested. The results suggest that it is worth testing whether the combined diagnosis is more specific for pneumococcal pneumonia than chest X-rays alone, but new studies are needed to resolve the issue. More Information Online The Global Alliance for Vaccines and Immunization (GAVI): http://www.vaccinealliance.org/ GAVI Web page “Call for Intensified Research after Pneumococcus Trial Surprises”: http://www.vaccinealliance.org/Resources_Documents/Immunization_Focus/Download/update.php World Health Organization (WHO) Web site on vaccines: http://www.who.int/vaccines/ WHO Web page “Pneumococcal Vaccines”: http://www.who.int/vaccines/en/olddocs/pneumococcus.shtml ) | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549587.xml |
535909 | Host resistance does not explain variation in incidence of male-killing bacteria in Drosophila bifasciata | Background Selfish genetic elements that distort the sex ratio are found widely. Notwithstanding the number of records of sex ratio distorters, their incidence is poorly understood. Two factors can prevent a sex ratio distorter from invading: inability of the sex ratio distorter to function (failure of mechanism or transmission), and lack of drive if they do function (inappropriate ecology for invasion). There has been no test to date on factors causing variation in the incidence of sex ratio distorting cytoplasmic bacteria. We therefore examined whether absence of the male-killing Wolbachia infection in D. bifasciata in Hokkaido island of Japan, in contrast to the presence of infection on the proximal island of Honshu, was associated with failure of the infection to function properly on the Hokkaido genetic background. Results The male-killer both transmitted and functioned well following introgression to each of 24 independent isofemale inbred lines carrying Hokkaido genetic backgrounds. This was maintained even under stringent conditions of temperature. We therefore reject the hypothesis that absence of infection is due to its inability to kill males and transmit on the Hokkaido genetic background. Further trap data indicates that D. bifasciata may occur at different densities in Hokkaido and Honshu populations, giving some credence to the idea that ecological differentiation could be important. Conclusions The absence of the infection from the Hokkaido population is not caused by failure of the male-killer to function on the Hokkaido genetic background. | Background Selfish genetic elements that distort the sex ratio of their host are known widely in arthropods [ 1 ]. Despite over 70 years of research, we still do not fully understand the factors that dictate their presence or absence in different species in the field, nor the correlated question as to the factors causing variation in their frequency over geographical space within a species. A good approach to this problem is to examine the causes of variation within species, and in particular to identify the factors contributing to absence of elements from some populations in species known to bear the element in other areas. Factors causing variation in prevalence/incidence over space may be either ecological or associated with differences in host genetic constitution. For instance, in the case of X chromosome meiotic drive in Drosophila pseudoobscura , frequency declines at both high latitude and high altitude. This is not associated with resistance to the action of the driver (no resistance is known), and variation in rates of multiple mating is suspected as a cause [ 2 ]. In contrast, in the case of X chromosome drive in D. subobscura , X drive is present in North Africa and absent in Europe. Here, absence of X drive is associated with the less efficient function of X drive on the European genetic background [ 3 ]. This study pertains to the factors affecting the incidence of male-killing bacteria. These bacteria pass from a female to her progeny, and kill any males they enter. Male-killers are common in insects, but an appreciation of the factors underlying their incidence is lacking [ 4 ]. In the first place, host genetic factors may affect the ability of a male-killer to transmit or function. Within Drosophila prosaltans , for instance, there is intra-population host genetic variation in refractoriness to male-killer action/transmission [ 5 ]. Thus, it is logical to conjecture that a male-killer can be absent from a population because the host has evolved resistance to its action or transmission. In the second place, presence/absence of infection can be determined by ecological, environmental or genetic variation that influences the benefit of male-killing to the bacterium [ 6 ]. For instance, laboratory studies by Jaenike et al. [ 7 ] have demonstrated that the number of females ovipositing within a patch is a key determinant of invasion success. In this paper, we examine the factors that could cause incidence variation for the male-killing Wolbachia in Drosophila bifasciata in Japan. Drosophila bifasciata feeds on sap fluxes in deciduous forests, and sampling across 10 populations within Honshu island in Japan revealed a relatively constant frequency of infection, with between 5 and 7% of females infected with the male-killer [ 8 ]. The male-killer is a strain of Wolbachia still present on Honshu to this day [ 9 ]. In contrast to collections from Honshu, past surveys across Hokkaido, the North island of Japan, indicated flies in this area are not infected with the male-killer, despite the relative proximity of the sites to the infected populations in the Northern most sites in Honshu. In total, 559 flies from six locations within Hokkaido were tested, with no evidence of sex ratio distortion in any case [ 8 ]. We can be almost certain that the absence of infection is not due to the infection never arriving on this island. First, infection otherwise has a worldwide distribution, being found in Italian D. bifasciata populations [ 10 ]. Second, the 5 km wide Tsugara Straits between Honshu and Hokkaido may limit gene flow (and hence support differentiation), but are very unlikely to have been an absolute bar to the arrival of the infection. We tested whether the absence of infection in Hokkaido was associated with an effect of host genotype on the efficiency of male-killer transmission or strength of male-killing ability. Beyond this, we examined whether trap collection data were consistent with difference in D. bifasciata ecology between Hokkaido and Honshu islands. Our results indicated that the Hokkaido genetic background supported the transmission and action of the male-killer even under stringent conditions, ruling out genetic differentiation as a cause of the absence of the male-killer from Hokkaido. We did observe higher capture rates of D. bifasciata in the island of Hokkaido, and future work should therefore be focussed on the degree to which ecological differences affecting the drive of the infection dictates incidence in this species. Results Sex ratio of Hokkaido flies Twenty-eight female flies were collected from the field in Hokkaido. Of these, 4 failed to produce progeny. The remaining 24 all produced a normal sex ratio, consistent with continued absence of the male-killing trait in Hokkaido. Intensity of male-killing on the Hokkaido genetic background We tested whether the Hokkaido genetic background supported the male-killing Wolbachia from Honshu by introgression of the infection onto the Hokkaido genetic background. The progeny from the 24 Hokkaido females were maintained as isofemale inbred lines for three generations to capture genetic variation within them. The male-killing Wolbachia from Honshu was then placed onto each inbred background via crossing infected females from Honshu to males from each of the Hokkaido lines, with subsequent generations being maintained through further crossing to males from the appropriate Hokkaido line. The sex ratio produced following introgression of the infection to the Hokkaido genetic background was female biased and penetrance of the male-killing phenotype was perfect in the first generation. A few males appeared sporadically in six of the 24 lines in one or more subsequent generations, with highest frequency in lines 15 and 25 (Table 1 ). However, no males were produced in the F4 in any case, indicating no loss of infection or repeatable resistance to male-killer action. We particularly maintained observation over lines 15 and 25 over four subsequent generations, and adult males were not observed in the culture over this period (data not shown). These data are broadly comparable with data from Honshu control lines, where 4 of 27 lines showed sporadic male production. Table 1 The sex ratio produced during introgression of the male-killing infection to the Hokkaido genetic background. 'All-female' classification represent cases where both replicates produced all female broods. Where males were produced within a female-biased sex ratio, data is given separately for each replicate of the isofemale line. Generation Sex ratio No. of lines Line-replicate: male n progeny/total F1 All female 24 Female biased 0 F2 All female 21 Female biased 3 15-1: no males. 15-2: 2/74 21-1: 1/32. 21-2: no males 25-1 no males. 25-2: 1/10 F3 All female 19 Female biased 5 12-1 no males. 12-2: 1/18 15-1: 2/27 15-2: 2/52 17-1: 2/20 17-2: 1/46 20-1: 1/10 20-2: no males 25-1: 1/40 25-2: no males F4 All female 24 Female biased 0 Effect of stringent temperatures The lines above were moved to 23.5°C, the upper temperate before thermal induced loss of infection occurs in Honshu flies [ 11 ], and we maintained the lines at this temperature by backcrossing to males from the source uninfected Hokkaido line as before, for four further generations. No effect of elevated temperature on the penetrance or transmission of the male-killing trait was observed on the Hokkaido genetic background (Table 2 ). Sporadic males were observed in 4 of 20 lines over the four generations. However, males were never observed in both replicates within a line, nor were they ever observed in more than one generation within a line. Notably, none were produced in the final generation, indicating the infection was perfectly transmitted during the experiment. Control lines from Honshu maintained production of all female broods, in agreement with past observations. Table 2 The sex ratio in male-killer infected isofemale lines from Hokkaido following transfer of the introgressed infected lines to 23.5°. 'All-female' classification represents cases where both replicates produced all female broods. Where males were produced within a female-biased sex ratio, data is given separately for each replicate of the isofemale line. Generation Sex ratio No. of lines Line-replicate: male n progeny/total F1 All female 20 Female biased 0 F2 All female 20 Female biased 0 F3 All female 16 Female biased 4 11-1 no males 11-2: 1/21 19-1: 1/28 19-2: no males 23-1: 1/37 23-2: no males 25-1: 1/11 25-1: no males F4 All female 20 Female biased 0 Collection rates of D. bifasciata in traps We captured flies in the field and scored the samples for both absolute capture rate of D. bifasciata , and capture rate relative to other species of Drosophila . The capture rate of D. bifasciata was substantially higher in all four samples taken in Hokkaido province (Misumai, Koryukozan, Tomakomai, Matsumae) than in the two samples from the Northern Honshu populations (Mimmaya, Morioka) and the population from Mid Honshu (Kiyosumi). Increased capture rate in Hokkaido was also reflected in an increase in the proportion of all drosophilids sampled that were bifasciata (Table 3 ). This is consistent with the idea that the ecology of D. bifasciata varies between Honshu and Hokkaido, and that this may cause the presence of the infection in one island and absence in the other. Table 3 Catch rate of D. bifasciata in seven locations within Japan during early-mid October between 1973 and 1984. Catch rate is given as mean per trap per day, with number of traps and number of days trapped in parentheses. Proportion of catch that was bifasciata is given across all traps and days, with total Drosophila catch in parentheses. Island Location bifasciata caught per day/ per trap (traps, days) bifasciata as proportion of catch (n) Hokkaido Misumai 42°57' N 141°16' E 4.67 (5, 14) 13.6% (2407) Koryukozan 42°51' N 141°17' E 8.38 (5, 17) 15.94% (4458) Tomakomai 42°43' N 141°36' E 6.48 (6, 14) 2.01% (27033) Matsumae 41°26' N 140°08' E 2.54 (8, 7) 2.48% (5726) Honshu Mimmaya 41°10' N 140°24' E 0.21 (8, 7) 0.51% (2342) Morioka 39°15' N 141°10' E 0.79 (4, 7) 0.82% (2683) Kiyosumi 35°10' N 140°10' E 0.00 (4, 7) 0 (903) Discussion Previous study has shown that male-killing Wolbachia are absent from D. bifasciata in Hokkaido province of Japan, despite the infection being present in neighbouring Honshu. In this study, we have demonstrated that the absence of male-killing Wolbachia in the Hokkaido population is not caused by the inability of the male-killing Wolbachia to operate on the Hokkaido host genetic background. In contrast, the male-killer was perfectly transmitted in all 24 lines tested (after 4 generations of introgression, no males were produced in any of the lines). This high efficiency was maintained even at the threshold for complete male-killing in Honshu, 23.5 C (after a further 4 generations, all 20 lines still had all female broods). Thus, the male-killer is proficient at being transmitted and killing males on the Hokkaido background even under relatively stressful environmental conditions. Further, we continue to maintain the infection on the Hokkaido background (we are now at generation 15) without any loss of the infection and without appearance of males. Thus, it is not tenable to argue that the hosts themselves are not genetically suitable for the function of male-killer, at least on an ecological timescale. This situation contrasts with the case of meiotic drive in the related fly D. subobscura , where absence of drive in Europe was associated with refractoriness to the action of the sex ratio distorting element. In the absence of variation in the ability of the male-killer to function on the different genetic backgrounds, the question arises as to the features that cause infection to be absent from the Hokkaido population. Our study did reveal differences in trap collection rates between the populations of Hokkaido and those of Northern Honshu, with bifasciata captured at lower rate in the populations from Honshu than in Hokkaido. Thus, ecological heterogeneity is a possible source of the incidence pattern. There are three parameters in male-killer dynamics that may be environmentally influenced. First, the advantage to male-killing may not be as strong in Hokkaido populations. Second, the cost of infection to female flies may be higher in Hokkaido than in Honshu. Third, the transmission efficiency may alter, mediated via elevated temperature, or possibly by reduced overwinter temperature. In our view, the latter factor is unlikely to be driving the observed pattern. It is notable that the survey of Ikeda revealed the infection to be present in Northern Honshu, but not in Southern Hokkaido. Given these two areas are geographically and climatically very close, temperature differences have poor explanatory power. Explanations based on temperature are also weak because this species exhibits a degree of homeostasis in temperature, moving to elevated altitudes to avoid excess temperature. This leaves us with two hypotheses to explain absence of infection in Hokkaido. The first is that there is a weaker advantage to male-killing in Hokkaido than on Honshu, such that there is insufficient drive to maintain the bacterium or permit its spread. The advantage of male-killing to the bacterium depends upon the number of females ovipositing in a single patch [ 7 ]. If the high density of bifasciata observed in Hokkaido translates into many females laying eggs in a single sap flux, male death will not greatly increase the survival of infected females over uninfected, and infection will decline in frequency. The second factor that may cause infection to decline in frequency is if costs of infection are higher in Hokkaido than in Honshu. This factor can be ecologically contingent. Ikeda demonstrated that the relative fitness of infected flies compared to uninfected flies was lower under high densities in the laboratory [ 8 ]. Thus, if the high density of the adult fly we observed corresponds to a high density of larvae within a single sap flux, the direct cost of infection would be higher, and the infection would be expected to be less common or absent. Aside from these possibilities, which we consider most likely, other ecological discontinuities between Hokkaido and Honshu deserve investigation. Some symbionts, for instance, give the host protection against parasitoids [ 12 ], and thus differences in parasitoid infection rates could affect the frequency of a symbiont. Co-existing heterospecific competitors may also diminish the benefit of male-killing; if there are many species ovipositing within a patch, the advantage to male-killing may decline. Thus, the intensity of inter-specific competition also deserves investigation. Finally, the existence of other 'competing' inherited microorganisms should be excluded as a reason for the absence of the male-killing Wolbachia from Hokkaido. Conclusions It is not variation in the ability of Wolbachia to function on different host genetic backgrounds that drives presence or absence of infection in the D. bifasciata - male-killing Wolbachia system. We have demonstrated that despite being absent from Hokkaido, Wolbachia can both be maintained and express male-killing on the Hokkaido host genetic background. We observe that the two populations show differences in trap capture rates, and argue that either ecologically contingent benefits or ecologically contingent costs of infection may explain presence and absence of infection in this species, and that future research be focussed at this level. Methods Source of wild flies for introgression Twenty eight wild female D. bifasciata were collected on the campus of Hokkaido University, Sapporo (43°4'56"N, 141°20'21"E), Japan, in May 2003. These were then brought into the laboratory, where they were maintained individually in vials at 21°C on a modified cornmeal-agar diet (70 g sucrose, 60 g maize meal, 15 g yeast extract, 10 g agar, 2.5 g nipagin in a total volume of 1 liter). These female were checked for the presence of the male-killing trait through scoring of the sex ratio, and maintained by sib-sib inbreeding (2 males, 2 females) for three generations to make inbred isofemale lines. In total, 24 isofemale inbred lines were created (4 lines went extinct), which were maintained thenceforth by simple tossing every three weeks. Introgression of the infection onto the Hokkaido genetic background The genetic background of each of the 24 uninfected Hokkaido lines was then independently crossed onto the male-killer infected cytotype over four generations. To this end, 4 males were taken from each of the 24 Hokkaido lines, and mated to 4 Wolbachia infected virgin females extracted from a culture derived from Honshu island, Japan (established in [ 9 ]). This procedure was performed twice for each Hokkaido inbred line to give 24 introgression lines, each replicated with two replicates. Following this initial cross, introgression of the appropriate Hokkaido nuclear background continued for four generations, on each occasion four female offspring of each line being backcrossed to males from the respective Hokkaido uninfected inbred line. Flies were kept at 21°C throughout the experiment, and the sex of each line scored at each generation (n>10 individuals in every case). As a control against spontaneous loss of infection not associated with genetic differentiation, the male-killer was concurrently maintained on the Honshu genetic background, in lines maintained by backcrossing to individual isofemale lines (as established in [ 11 ]) with likewise monitoring of sex ratio. Temperature effect Temperature is known to affect the stability of the male-killing trait, and previous study demonstrated that an upper threshold of 23.5 C existed for stable maintenance of the infection on the Honshu genetic background [ 11 ]. We tested whether the infection remained stable at this stringent temperature on the Hokkaido genetic background. To this end, 20 of the above introgressed fly lines were transferred from 21° to the 23.5°, and the sex ratio of the offspring recorded for four further generations, with the lines maintained by backcrossing to males from the appropriater parental Hokkaido uninfected line as before. As a control, four Honshu isofemale lines were concurrently maintained at 23.5°. Collection rates of D. bifasciata in traps Evidence of differences in fly density can be derived from sampling the Drosophila communities. Drosophila communities were sampled in 4 deciduous forests in Southern Hokkaido, 2 sites in Northern Honshu, and one site in Mid Honshu, in early-mid October over a number of years between 1973 and 1984 (Figure 1 ). Collections were carried out using traps baited with fermented banana suspended from the canopy during early-mid October. The traps were especially designed for retaining trapped insects in a bottle of fixative solution and set vertically at different heights from the floor [ 13 , 14 ]. Since D. bifasciata is a typical forest-canopy dweller, sampling from the forest canopy is essential for estimating its population density. These traps were cleared 7 or 10 days after setting, and the capture rate of bifasciata per trap per day calculated at each locality to represent the density of this fly in this region. All drosophilid flies were identified, and the proportion of flies caught that were bifasciata recorded. Figure 1 Collection sites for D. bifasciata in Japan Authors' contributions ZV helped with the design of the crossing scheme, conducted the crosses involved, and performed the analysis of these crosses. MT designed and conducted the field sampling and scored trap collections. GH conceived the project, organised collection, helped with design of the crossing scheme, and wrote the paper. All authors read and commented on drafts of the manuscript, and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535909.xml |
549593 | Multiple Metabolic Hits Converge on CD36 as Novel Mediator of Tubular Epithelial Apoptosis in Diabetic Nephropathy | Background Diabetic nephropathy (DNP) is a common complication of type 1 and type 2 diabetes mellitus and the most common cause of kidney failure. While DNP manifests with albuminuria and diabetic glomerulopathy, its progression correlates best with tubular epithelial degeneration (TED) and interstitial fibrosis. However, mechanisms leading to TED in DNP remain poorly understood. Methods and Findings We found that expression of scavenger receptor CD36 coincided with proximal tubular epithelial cell (PTEC) apoptosis and TED specifically in human DNP. High glucose stimulated cell surface expression of CD36 in PTECs. CD36 expression was necessary and sufficient to mediate PTEC apoptosis induced by glycated albumins (AGE-BSA and CML-BSA) and free fatty acid palmitate through sequential activation of src kinase, and proapoptotic p38 MAPK and caspase 3. In contrast, paucity of expression of CD36 in PTECs in diabetic mice with diabetic glomerulopathy was associated with normal tubular epithelium and the absence of tubular apoptosis. Mouse PTECs lacked CD36 and were resistant to AGE-BSA-induced apoptosis. Recombinant expression of CD36 in mouse PTECs conferred susceptibility to AGE-BSA-induced apoptosis. Conclusion Our findings suggest a novel role for CD36 as an essential mediator of proximal tubular apoptosis in human DNP. Because CD36 expression was induced by glucose in PTECs, and because increased CD36 mediated AGE-BSA-, CML-BSA-, and palmitate-induced PTEC apoptosis, we propose a two-step metabolic hit model for TED, a hallmark of progression in DNP. | Introduction Diabetic nephropathy (DNP) is a serious and common complication of type 1 and type 2 diabetes mellitus, leading to end-stage renal failure in up to 30% of individuals with diabetes. Early abnormalities of DNP affect glomeruli and include an increase in glomerular filtration rate, microalbuminuria, glomerular hypertrophy, and thickening of the glomerular basement membrane, followed by expansion of mesangial extracellular matrix and glomerulosclerosis [ 1 , 2 ]. As with most chronic degenerative kidney diseases, the gradual decline of renal function at later stages of DNP is invariably associated with tubular epithelial degeneration (TED), also called tubular atrophy, and interstitial fibrosis (IF), hallmarks of degeneration to end-stage renal failure [ 3 ]. Pathomechanisms that may initiate and/or mediate TED in DNP remain poorly understood. While glomerular lesions consistent with human DNP have been described in various mouse models of diabetes, TED and IF have not been described in diabetic mice [ 4 ]. Combining detailed renal phenotype analysis with gene expression profiling of hyperglycemic mouse models of type 1 (streptozotocin [STZ]) and type 2 (db/db) diabetes, we recently reported that decreased mRNA levels of CD36 in kidneys were strongly correlated with albuminuria [ 5 ]. CD36 is a transmembrane protein of the class B scavenger receptor family and is involved in multiple biological processes [ 6 ]. CD36 is widely expressed and may interact with multiple extracellular ligands, including thrombospondin-1 (TSP-1), long-chain free fatty acids (FFAs), modified (oxidized) low-density lipoprotein (ox-LDL), advanced glycation end (AGE) products, and collagens I and IV [ 6 ]. CD36 mediates phagocytosis of apoptotic cells and malaria-parasitized erythrocytes [ 7 ]. Furthermore, CD36 mediates antiangiogenic activity associated with endothelial cell apoptosis induced by TSP-1 through p38 MAP kinase (MAPK) and caspase 3 [ 8 ]. Hyperglycemia-induced synthesis of CD36 protein in macrophages has been associated with increased uptake of ox-LDL by macrophages and foam cell formation in atherosclerotic lesions in people with diabetes [ 6 , 9 , 10 ]. While diabetic cardiovascular complications are closely linked epidemiologically with albuminuria and DNP, a role for CD36 in DNP and renal pathophysiology has not to our knowledge been described to date. Here we report a novel functional role for CD36 scavenger receptor and AGE and FFA palmitate (PA) in tubular epithelial apoptosis associated with TED and progression of DNP. Specifically, we show that glucose stimulates CD36 cell surface expression in proximal tubular epithelial cells (PTECs), and increased CD36 renders PTECs susceptible to both AGE- and PA-induced PTEC apoptosis by mediating sequential activation of src kinase, proapoptotic p38 MAPK, and caspase 3. Based on these findings, we propose a new two-step metabolic hit model for TED in the progression of DNP. Methods Animals Kidneys were obtained from 28-wk-old C57BLKS/J-lepr db/db , STZ-treated C57BL/6J, or STZ-treated 129SvJ mice and from age-matched control C57BLKS/J-lepr db/m , C57BL/6J, and 129SvJ mice as described [ 5 ]. Cell Culture Human proximal tubular cell line HK-2 and murine collecting duct cell line M1 were purchased from American Type Culture Collection (Manassas, Virginia, United States) and cultured according to the vendor's instructions. Mouse proximal tubular cell line MCT was provided by Fuad Ziyadeh (University of Pennsylvania, Philadelphia, Pennsylvania, United States). Transfections were performed with Fugene 6 (Roche Diagnostics, Indianapolis, Indiana, United States) according to manufacturer's protocol. CD36-containing plasmid was a kind gift of Nada Abumhrad (SUNY at Stony Brook, New York, United States). Cells were also co-transfected with EGFP (Clontech, Franklin Lakes, New Jersey, United States) to assess transfection efficiency. Cells were serum starved in 0.2% serum containing DMEM (1 g/l glucose) for at least 24 h prior to stimulation with AGE–bovine serum albumin (BSA), glucose, or FFA. Quantitative Real-Time PCR Quantitative real-time PCR analysis of mouse and human CD36, HPRT1, and beta actin was performed as described previously [ 5 ]. The following primers were used: mouse CD36 5′
TGCTGGAGCTGTTATTGGTG and 3′
CATGAGAATGCCTCCAAACA, mouse beta actin 5′
ACCGTGAAAAGATGATGACCCAG and 3′
AGCCTGGATGGCTACGTACA, mouse HPRT1 5′
TGTTGTTGGATATGCCCTTG and 3′
TTGCGCTCATCTTAGGCTTT, human CD36 5′
GCTCTGGGGCTACAAAGATG and 3′
TAGGGAGAGATATCGGGCCT, human beta actin 5′
GATGAGATTGGCATGGCTTT and 3′
CACCTTCACCGTTCCAGTTT, and human HPRT1 5′
AAAGGACCCCACGAAGTGTT and 3′
TCAAGGGCATATCCTACAACAA. Immunostaining and Immunoblotting Primary antibodies specific for the following proteins were used: monoclonal mouse anti-CD36 antibody, clone FA 6–152 (IgG) (Immunotech, Fullerton, California, United States), clone SMO (IgM) (Santa Cruz Biotechnology, Santa Cruz, California, United States), rabbit polyclonal anti-CD36 (Santa Cruz Biotechnology), rabbit polyclonal anti-aquaporin1, anti-aquaporin2, anti-Na/K/2Cl (Chemicon, Temecula, California, United States), rabbit polyclonal phospho38/MAPK and mouse monoclonal p38 (Cell Signaling Technology, Beverly, Massachusetts, United States), rabbit polyclonal p-src (Y418) (Biosource, Camarillo, California, United States), and mouse monoclonal anti-tubulin (Sigma, St. Louis, Missouri, United States). Immunostaining was performed on frozen sections with FITC- and Cy3-labeled secondary antibodies (Jackson Laboratories, USA), or on paraffin-embedded sections with immunoperoxidase, as described earlier [ 5 ]. Immunoblotting was performed with 30 μg of protein isolated from cultured cells. Protein samples were resolved on a 10% SDS-PAGE and immunoblotted with primary antibody and revealed with horse radish peroxidase (HRP)-conjugated anti-mouse IgM, or anti-rabbit IgG (Jackson Laboratory, Bar Harbor, Maine, United States). Immuncomplexes were detected by enhanced chemiluminescence (Pierce, Rockford, Illinois, United States). The proximal tubular immunostaining was evaluated semi-quantitatively by two independent pathologists who were unaware of the diagnosis; distribution and intensity of staining was scored on a ten-point scale. Fluorescence Flow Cytometric Analysis Cells were incubated in 0.5 mM EDTA in PBS at 37 °C for 20 min, scraped, and then washed with 1% fetal bovine serum. Cells were then exposed to monoclonal anti-CD36 IgG FA6 (5 μg/ml), or control mouse IgG1 (5 μg/ml) (Sigma), for 45 min on ice in the presence of 10% fetal bovine serum then washed with PBS. This was followed by an incubation with phycoerythrin-conjugated goat anti-murine secondary antibody (Southern Biotechnology, Birmingham, Alabama, United States) 1:50 for 45 min on ice. Cells (1 × 10 4 ) were analyzed by using a SCAN flow cytometer (BD, Franklin Lakes, New Jersey, United States), with appropriate gating. Flow cytometry data were analyzed using Cellquest (BD). Preparation of Glycated Albumin and Carboxymethyl-Lysine Albumin Briefly, to prepare AGE-BSA, essentially fatty-acid-free and endotoxin-free BSA (250 mg/ml) was incubated at 37 °C for 2, 5, and 10 wk with D-glucose (500 mM) in a 0.4-M phosphate buffer containing EDTA, ampicillin, Fungazone, polymixin B, and protease inhibitors. Control preparations were treated identically except that glucose was omitted. Carboxymethyl-lysine (CML)–BSA was prepared as described earlier [ 11 ]. Briefly, BSA with minimal CML content (CMLmin-BSA) was prepared by incubation of BSA (0.66 mM) with glyoxylic acid (2.15 mM) in the presence of sodium cyanoboronydrate (56 mM) in 200 mM sodium phosphate buffer (pH 7.8) at 37 °C under aseptic conditions. Finally, preparations were extensively dialyzed against phosphate buffer to remove free glucose. Preparations were then tested for the presence of LPS with a Quantitative Chromogenic LAL kit (Cambrex, East Rutherford, New Jersey, United States). The concentration of LPS was lower than 0.07 IU/mg protein in all preparations. Preparation of FFA Palmitic acid (P5585), oleic acid, and FFA-free low-endotoxin BSA (A8806) were purchased from Sigma. Palmitic acid was dissolved at 12 mM in PBS containing 11% fatty-acid-free BSA, sonicated for 5 min, shaken overnight at 37 °C, and sonicated for 5 min again [ 12 ]. For control experiments, BSA in the absence of fatty acids was prepared, as described above. The effective concentration of PA was determined using a commercially available kit (Wako Chemicals, Neuss, Germany). Apoptosis Detection In situ detection of DNA fragmentation was performed using the ApoTag TUNEL assay following the manufacturer's protocol (Intergen, Purchase, New York, United States) [ 13 ]. Apoptotic nuclei were detected using DAPI staining (1 μg/ml; 10 min) in cell cultures fixed with 4% paraformaldehyde, and analyzed via fluorescence microscopy to assess chromatin condensation and segregation. Caspase3 activity was detected by using the ApoAlert Caspase3 Fluorescent Detection system (BD) according to the manufacturer's protocol. Activity was normalized to total protein content. Z-DEVD-fmk, z-VAD-fmk, z-FA-fmk, and z-LEHD-fmk were purchased from BD. Human Kidney Biopsy Sample and Patient Characteristics Human kidney tissues (ten controls, ten with diabetic nephropathy, and ten with focal segmental glomerulosclerosis [FSGS]) were obtained from archived kidney biopsy samples or from discarded nephrectomy specimens. All diabetic samples were from patients with biopsy-proven advanced DNP with serum creatinine ranging from 1.7 to 5.6 mg/dl (151 to 444 μM/l), heavy proteinuria (3+ by dipstick or 3–6 gr/d), and hypertension. All patients with FSGS were from patients with creatinine levels of 1.7 to 4.9 mg/dl (151 to 435 μM/l), heavy proteinuria (3+ by dipstick), and hypertension. The diagnosis of FSGS was made on Periodic acid–Schiff staining in the absence of immunodeposits on electron microscopy. The diagnosis of diabetic nephropathy was based on the presence of diabetes, proteinuria, and the characteristic light microscopy findings. Institutional Review Board approval was obtained for procurement of kidney specimens at the Thomas Jefferson University Hospital. Statistical Methods Data are reported as mean and standard error of the mean (SEM) for continuous variables. All cell culture experiments were performed at least three times and summarized. Standard software packages (SPSS and Excel for Windows) were used to provide descriptive statistical plots (unpaired t -tests). The Bonferroni correction was used for multiple comparisons. Significance for the quantification of the CD36 staining in human biopsy samples was calculated via the Wilcoxon Rank Sum Test. Results Increased Expression of CD36 Specifically in Proximal Tubules of Human Diabetic Kidneys Is Associated with TED Using microarray-based gene expression profiling on whole kidney RNA together with supervised clustering methods, we previously identified and validated gene expression patterns for molecular classification of diabetic mice with albuminuria and mesangial expansion [ 5 ]. Reduced renal mRNA levels of the class B scavenger receptor CD36 were characteristic for diabetic mice with albuminuria [ 5 ]. Here we examined patterns of CD36 protein expression in kidneys of non-diabetic and diabetic mice and humans. CD36 protein was detectable in the thick ascending limb of loop of Henle and in the collecting duct, and absent in proximal tubules in both control and diabetic mouse kidneys ( Figure 1 A– 1 D). In contrast, CD36 was detectable only rarely in individual proximal tubular cells in sections from non-diabetic human kidneys (controls) ( Figure 1 E and 1 H), but was markedly increased specifically in PTECs in human diabetic kidneys ( Figure 1 F and 1 I). In addition, we did not observe increased proximal tubular CD36 expression in kidney biopsy samples from patients with FSGS ( Figure 1 J), that were matched with DNP samples for the severity of proteinuria (all in the nephrotic range) and renal insufficiency (all with elevated serum creatinine; 1.7–5.0 mg/dl). Semi-quantitative analysis of the distribution and intensity of CD36-positive PTECs (CD36 PTEC score), which was performed by two independent pathologists in a blinded manner, demonstrated that mean CD36 PTEC scores were not different between FSGS kidneys and normal human kidneys, but were significantly increased in DNP kidneys ( Figure 1 K). Figure 1 Differential Localization and Expression of CD36 Protein in Kidneys of Diabetic Mice with Glomerulopathy and of Humans with DNP (A and B) Indirect double-immunofluorescence labeling of kidney sections from non-diabetic control (A) and diabetic (B) mice with anti-CD36 (green) and proximal tubular marker anti-aquaporin1 (red). (C and D) Double labeling of non-diabetic control mice with anti-CD36 (green) and loop-of-Henle marker sodium potassium chloride cotransporter anti-NKCC (red) (C) and collecting duct marker aquaporin2 (red) (D) (arrow depicts colocalization of anti-CD36 and anti-aquaporin2 staining). (E and F) Double labeling of human kidney sections from control individuals (E) and individuals with diabetes with DNP (F) using anti-CD36 (green) and anti-aquaporin1 (red). (G) Higher-magnification image of (F) with arrows depicting colocalization of anti-CD36 and anti-aquaporin1. (Note that anti-CD36 labeling is heterogeneous: staining is isolated proximal tubular cells.) (H–J) Representative images of anti-CD36 immunoperoxidase staining of sections of normal human kidney (H), human kidney with DNP (I), and human kidney with FSGS (J). Arrow in (I) depicts proximal tubular epithelial staining. (K) CD36 PTEC expression score derived from blinded, semi-quantitative analysis of distribution and intensity of proximal tubular CD36 staining of human biopsy samples from ten normal control, ten DNP, and ten FSGS kidneys and the result shown on a dot plot. Significance was calculated by Wilcoxon Rank Sum Test, and PTEC scores for DNP kidneys were significantly different from those of FSGS kidneys and normal human kidneys. Periodic acid–Schiff–stained sections of kidneys from mice exposed to type 2 diabetes (db/db mice) for 20 wk ( Figure 2 A), or type 1 diabetes (STZ-treated C57BL/6J mice) for 20 wk (data not shown) demonstrated moderate to advanced mesangial expansion and glomerulosclerosis ( Figure 2 A). Tubular abnormalities were not detectable in either model ( Figure 2 A). In contrast, TED and IF were associated with moderate to advanced mesangial expansion and glomerulosclerosis on kidney sections of human DNP ( Figure 2 B). These findings indicate that in humans with DNP, diabetes-induced upregulation of CD36 expression in proximal tubules was associated with moderate to advanced stages of TED and IF. In contrast, in diabetic mice with albuminuria, mesangial expansion, and glomerulosclerosis, absence of CD36 expression was associated with normal appearance of the tubular epithelium and interstitial space. These findings suggest an association between diabetes-induced proximal tubular CD36 expression and TED. Figure 2 TED and IF Coincide with Proximal Tubular Apoptosis and CD36 Expression in Human DNP (A and B) Periodic Acid–Schiff staining of diabetic mouse kidney (28-wk-old C57BLKS/J-lepr db/db ) (A) and human DNP kidney (B). Arrowheads denote glomeruli with advanced mesangial expansion and glomerulosclerosis; arrows depict normal proximal tubule in diabetic mouse (A) and TED in human with DNP (B). (C) TUNEL assay (green) and anti-CD36 (red) double labeling of human DNP. Arrows indicate apoptotic, CD36-positive tubular epithelial cells. (D) TUNEL assay (green) and anti-aquaporin1 (red) double labeling of human DNP. Arrows depict TUNEL-positive and aquaporin1-positive PTECs. (E) Dot plot indicates the number of TUNEL-positive tubular cells per 100 total tubular cells in kidneys of control (CTL) and diabetic (DM) mice and humans, as indicated. Coincidence of Increased CD36 Expression and Increased Tubular Epithelial Cell Apoptosis in Human DNP CD36 has been shown to mediate apoptosis signaling induced by TSP-1 in endothelial cells [ 8 ] and by ox-LDL in macrophages [ 14 ]. We examined whether the strong upregulation of CD36 protein in PTECs, observed specifically in human DNP, was associated with tubular epithelial cell apoptosis in vivo. TUNEL-positive tubular epithelial cells also stained positive for CD36 protein ( Figure 2 D) and aquaporin1 ( Figure 2 C), indicating that apoptosis and CD36 expression coincided in PTECs in human DNP. In contrast, CD36 was not detectable in TUNEL-positive PTECs in non-diabetic FSGS kidneys and in normal human kidney (data not shown). Statistical analysis showed that the rate of TUNEL-positive tubular cells was significantly increased in kidneys of human DNP compared with normal control human kidney ( Figure 2 E). In addition, tubular epithelial apoptosis was increased, but highly variable, in FSGS kidneys (data not shown). In contrast, tubular epithelial apoptosis rates were comparable between non-diabetic control and all diabetic mouse kidneys ( Figure 2 E). The diabetic mouse group included 24-wk-old STZ-treated diabetic C57BL/6J or 129SvJ mice (0.23 ± 0.1 TUNEL-positive cells per 100 tubular cells) and 24-wk-old lepr db/db mice (0.2 ± 0.1 TUNEL-positive cells per 100 tubular cells). Together, these findings indicate that CD36 expression in PTECs is associated with apoptotic events of proximal tubular cells and TED specifically in human DNP, but not in FSGS with matched functional and clinical abnormalities. These in vivo findings demonstrate a strong association of diabetes-induced CD36 expression and apoptosis in PTECs in human DNP, suggesting that CD36 may play a critical role in TED by mediating PTEC apoptosis in progressive human DNP. High Ambient Glucose Induces CD36 Expression in Human PTECs High ambient glucose has been shown to induce CD36 protein synthesis in macrophages [ 9 ]. Because CD36 protein was markedly increased in proximal tubules in human DNP, we examined the effects of high ambient glucose on CD36 mRNA and protein expression in the human PTEC line HK-2 ( Figure 3 ). Exposure of cells to 30 mM D-glucose for 24 h, but not to control L-glucose, significantly increased levels of CD36 mRNA ( Figure 3 A), CD36 cell surface protein ( Figure 3 C), and CD36 protein expression in cell lysates ( Figure 3 D). In contrast, CD36 mRNA and protein were not detectable in the murine PTEC line MCT at either normal or high ambient glucose concentrations (data not shown). Interestingly, glucose stimulation decreased CD36 mRNA levels ( Figure 3 B) and CD36 cell surface protein ( Figure 3 C) in the murine collecting duct cell line M1, consistent with our previously reported findings in diabetic mouse kidney [ 5 ]. Exposure of human HK-2 and murine M1 cell lines to defined preparations of FFA PA or AGE-BSA had no effect on CD36 mRNA and protein expression levels (data not shown). These findings demonstrate that high ambient glucose causes upregulation of CD36 mRNA and protein specifically in human, but not in mouse, PTECs. Together with our in vivo observations, these results suggest that hyperglycemia may induce upregulation of CD36 mRNA and protein selectively in proximal tubules in kidneys of human DNP, but not diabetic mice with albuminuria. Figure 3 CD36 mRNA and Protein Synthesis Is Stimulated in Human, but Not in Murine, PTECs, and Is Suppressed in Murine Collecting Duct Cells by High Ambient Glucose (A) Relative CD36 mRNA abundance determined by quantitative real-time PCR in human PTEC line HK-2 treated with 30 mM D-glucose (open bars) or control L-glucose (black bars) for 4 and 24 h following maintenance of cells in 5 mM D-glucose medium. Bars represent mean ± SEM of three to five repeat experiments. Numbers on top of bars indicate significant p -values of experimental groups relative to 0 h. (B) Bar graphs show experiment as described under (A), using mouse collecting duct cell line M1 instead of human HK-2 PTECs. The relative expression of CD36 was normalized to internal control housekeeping genes HPRT and beta actin, and to baseline controls (untreated cells). (C) Relative cell surface expression of CD36 protein determined by FACS in M1 cells (open bars) and HK-2 cells (black bars) maintained in 5 mM D-glucose medium (CTL), or in medium containing 30 mM D-glucose (D-gluc) or L-glucose (L-gluc) for 72 h. (Original FACS histograms are provided in Figure S1 .) Bars represent mean ± SEM of three to five repeat experiments. Numbers indicate significant p -values of experimental groups relative to control. (D) Immunoblot showing CD36 protein levels in human HK-2 PTECs maintained in control 5 mM D-glucose (CTL), or after stimulation for 72 h with 30 mM L-glucose (L-gluc) or D-glucose (D-gluc), as indicated. Tubulin is shown for loading control. All data represent at least four independent repeat experiments. AGE-BSA, CML-BSA, and FFA PA Induce Apoptosis in Human PTECs via CD36 AGE albumin [ 15 ] and FFAs [ 16 ] have been implicated in the pathogenesis of diabetic complications, including tubular degeneration [ 17 ] and tubular epithelial-to-mesenchymal transition [ 18 ]. In addition, AGE albumin and FFA are known to interact with CD36 [ 19 , 20 ]. However, it is not known whether AGE and/or FFA can activate CD36 signaling and apoptosis in tubular epithelial cells. Treatment with AGE-BSA for 2, 5, or 10 wk or with CML-BSA induced a significant increase in the number of apoptotic nuclei in CD36-positive HK-2 cells compared with control BSA-treated or untreated HK-2 cells ( Figure 4 A). In contrast, AGE-BSA and CML-BSA had no effect on the rate of apoptotic nuclei in CD36-negative murine MCT PTECs (data not shown). Because AGE-BSA glycated for 5 wk (AGE-BSA5) induced robust apoptosis at concentrations between 20 and 40 μM ( Figure 4 A), we chose this preparation and concentration for further analysis in all subsequent experiments. AGE-BSA5-induced apoptosis was blocked when cells were preincubated with neutralizing anti-CD36 antibody, while preincubation with control IgG antibody had no effect ( Figure 4 A). These results were confirmed by DNA laddering assay (data not shown). Figure 4 AGE-BSA, CML-BSA, and FFA PA Induce Apoptosis in Human PTECs through CD36 Signaling Bar graphs show mean ± SEM of apoptotic nuclei, visualized by DAPI staining and normalized to 100 total cells, in human HK-2 PTECs. Data are derived from three independent repeat experiments. Numbers on top of bars indicate significant p -values of experimental groups relative to control, or as indicated by bracket. (A) Cells were treated for 48 h with control BSA (40 μM), TSP-1 (1 μg/ml), and AGE-BSA modified for 2, 5, or 10 weeks (AGE-BSA2, AGE-BSA5, and AGE-BSA10, respectively) in the absence or presence of control IgG (10 μg/ml) or anti-CD36 neutralizing antibody (10 μg/ml), as indicated. (B) Cells were treated with control BSA (40 μM), or CMLmin-BSA at 0.5, 1, 2, 5, and 10 μM, in the absence or presence of anti-CD36 neutralizing antibody, as indicated. (C) Cells were treated with monounsaturated FFA oleic acid (OA) or PA at increasing concentrations, in the absence or presence of control IgG (10 μg/ml) or anti-CD36 neutralizing antibody (10 μg/ml), as indicated. Among the most abundant glucose-modified proteins detectable in the plasma of diabetic individuals are CML proteins [ 21 ], which are typically present at 1.6 to 2.3 μM concentrations in the plasma and urine of diabetic individuals [ 22 , 23 ]. To use physiologically relevant CML proteins in our in vitro experiments, we prepared CMLmin-BSA, characterized by glycation of approximately 30% of lysine residues [ 21 ]. When applied to HK-2 PTECs at concentrations ranging from 0.5 to 10 μM, CMLmin-BSA increased apoptosis rates significantly ( Figure 4 B). The proapoptotic effect of CMLmin-BSA was blocked by CD36 neutralizing antibody, but not by control IgG ( Figure 4 B). CD36 has been shown to transport fatty acids in adipocytes [ 24 ] and in muscle cells [ 25 ]. Concentrations of FFAs may be substantially elevated, to levels of up to 700 μM, in individuals with type 2 diabetes or obesity [ 26 ]. Thus, we examined the effects of saturated FFA PA and monounsaturated FFA oleate on apoptosis of HK-2 PTECs in the absence or presence of anti-CD36 neutralizing antibody. PA significantly increased rates of apoptotic nuclei in a concentration-dependent manner in HK-2 PTECs ( Figure 4 C). Anti-CD36 neutralizing antibody, but not control IgG, blocked PA-induced apoptosis ( Figure 4 C). In contrast, oleate did not induce apoptosis, even at concentrations as high as 300 μM ( Figure 4 C), neither did it prevent PA-induced apoptosis (data not shown). Of note, these experiments were performed using a total fatty acid:BSA ratio of 6.6:1, in order to closely model pathophysiologic states in which unbound FFA concentration is high [ 27 ]. Taken together, our findings demonstrate that pathophysiologically relevant species of AGE-BSA and CML-BSA, as well as saturated FFA PA, induce apoptosis in human PTECs at concentrations previously observed in plasma and/or urine in humans with diabetes. AGE-BSA and PA Sequentially Activate src kinase, Proapoptotic p38 MAPK, and Caspase 3 through CD36 Receptor CD36 has previously been shown to trigger the activation of p59fyn, p38 MAPK, and caspase 3 (GeneID: 836) in response to thrombospondin in endothelial cells [ 8 ]. Therefore we examined phospho-src, phospho-p38 levels and caspase 3 activation in HK-2 PTECs treated with AGE-BSA and PA in the absence or presence of anti-CD36 neutralizing antibody. Both AGE-BSA5 and PA increased phospho-src levels rapidly (after as little as 5 min), and over a prolonged time interval (up to 3 h) ( Figure 5 A and 5 B). Phosphorylation of src kinase was blocked by anti-CD36 neutralizing antibody ( Figure 5 A and 5 B). This observation is consistent with previous findings demonstrating direct interaction between CD36 and p59fyn [ 8 ]; however, the involvement of other src kinases cannot be excluded. We also observed increased levels of phosphorylation of p38 MAPK beginning 1 to 2 h after treatment, and p38 activation was also completely blocked by anti-CD36 neutralizing antibody ( Figure 5 C and 5 D). These findings indicate that CD36 activates proapoptotic p38 MAPK possibly via src kinase activation in human PTECs when stimulated with AGE-BSA5 and PA. Chemical inhibition of p38 MAPK prevented the increase in the rate of apoptotic nuclei induced by both AGE-BSA5 and PA in HK-2 PTECs ( Figure 5 G), indicating that p38 MAPK function is required for apoptosis induced by AGE-BSA and PA through CD36 receptor. AGE-BSA and PA significantly increased activity of effector caspase 3 in human PTECs ( Figure 5 E and 5 F). Caspase 3 activation was blocked by anti-CD36 neutralizing antibody, but not by control IgG ( Figure 5 E and 5 F). Pan-caspase inhibitor z-VAD-fmk and the specific caspase 3 inhibitor z-DEVD-fmk prevented apoptosis induced by PA and AGE-BSA, while the specific caspase 9 inhibitor z-LEHD-fmk had no significant inhibitor effect ( Figure 5 G). Together these findings indicate that CD36 receptor mediates sequential phosphorylation of src kinases and p38 MAPK, leading to activation of caspase 3 and apoptosis in human PTECs exposed to AGE-BSA and PA ligands. Interestingly, we did not observe phosphorylation of Smad2 and p42/44 ERK MAPK under these conditions, as previously reported for AGE binding to the RAGE receptor [ 28 ]. Figure 5 Activation of Intracellular Pathways following AGE-BSA and PA Treatment of Human HK-2 PTECs (A and C) Immunoblots show levels of (A) phosphorylated (Y418) src kinase and tubulin or (C) phosphorylated p38 MAPK (pp38) and total p38 MAPK (p38) in HK-2 cells treated with AGE-BSA5 (40 μM) in the absence or presence of control IgG or anti-CD36 neutralizing antibody (10 μg/ml) for different time periods, as indicated. (B and D) As shown in (A) and (C), except HK-2 cells were treated with PA (150 μM) instead of AGE-BSA5. (E and F) Bar graphs demonstrate mean ± SEM of caspase 3 activity in three independent repeat experiments. Caspase 3 activity was measured by quantitative ELISA in HK-2 cells after 18 h of stimulation with AGE-BSA5 and PA, as per manufacturer's protocol. Numbers on top of bars indicate significant p -values of experimental groups relative to control, or as indicated by brackets. (G) Bar graphs demonstrate number of apoptotic nuclei of HK-2 cells, normalized to 100 total cells, treated with AGE-BSA5 (40 μM) or PA (150 μM) in the absence (black bars) or presence of pan-caspase inhibitor (z-VAD-fmk [100 μM]; open bars), caspase 3 inhibitor (z-DEVD-fmk [20 μM]; first striped bars), caspase 9 inhibitor (z-LEHD-fmk (20 μM); gray bars), or chemical inhibitors of p38 MAPK (SB203580 [10 μM]; second striped bars). Mean ± SEM of three independent repeat experiments is presented. Numbers on top of bars indicate the significant p -values for comparison relative to control (no inhibitor). CD36 Is Sufficient to Mediate Apoptosis Induced by AGE-BSA and FFA In contrast with CD36-positive human HK-2 PTECs, we found that treatment of CD36-negative mouse MCT PTECs with AGE-BSA had no effect on rates of apoptotic nuclei (data not shown). To test whether CD36 was sufficient to mediate AGE-BSA-induced apoptosis, we transfected CD36-negative mouse MCT PTECs with a plasmid expressing human CD36 or empty control vector, followed by treatment with control BSA or AGE-BSA5. AGE-BSA5 treatment had no significant effect on rates of apoptotic nuclei in MCT PTECs transfected with control vector ( Figure 6 ). In contrast, AGE-BSA significantly increased apoptotic nuclei compared with unglycated BSA in MCT PTECs transiently transfected with CD36 expression vector ( Figure 6 ). Nonglycated control albumin did not cause apoptosis. Thus, transgenic de novo expression of human CD36 in CD36-negative mouse PTECs was sufficient to mediate apoptosis induced by AGE-BSA. Figure 6 Expression of CD36 Transgene Confers Susceptibility to AGE-BSA-Induced Apoptosis (A–D) Representative images show DAPI (A and C) and FITC (B and D) labeling of CD36-negative MCT cells treated with 40 μM AGE-BSA5 for 24 h after co-transfection with green fluorescent protein plasmid pEGFP and pcDNA3.1 empty control vector (A and B), or pEGFP and CD36 expression plasmid pcDNA3.1/CD36 (C and D). (E) The dot plot shows results of four independent experiments where apoptotic nuclei per 100 total cells were quantitated in transfected cell cultures with or without treatments as indicated. Discussion Advanced diabetic nephropathies in humans with type 1 or type 2 diabetes are uniformly characterized by TED, or tubular atrophy, and IF leading to renal failure [ 29 , 30 ]. Although TED and IF are the strongest predictors for progression of DNP [ 31 ], mechanisms that underlie TED in DNP remain poorly understood. Based on our in vitro and in vivo findings we propose a two-step metabolic hit model for TED in DNP. High ambient glucose, but not AGE or FFA, cause stimulation of CD36 expression in PTECs specifically in diabetic kidneys. Increased CD36 expression mediates sequential activation of src kinase, proapoptotic p38 MAPK, and caspase 3 in PTECs in the presence of AGE and FFA PA, resulting in PTEC apoptosis. Proximal tubular epithelial apoptosis may be an initial mechanism for TED in DNP. Our conclusions are supported by several key observations. First, we identify a new functional role for CD36 as an essential mediator of proximal tubular epithelial apoptosis, inducible by AGE-BSA, CMLmin-BSA, and FFA PA. Previous reports demonstrated a role for CD36 in mediating apoptosis induced by TSP-1 in endothelial cells and ox-LDL in macrophages [ 8 , 14 ]. In the present study, we show for the first time, to our knowledge, that CD36 mediates apoptosis in differentiated epithelial cells that are exposed to AGE-BSA-, CMLmin-BSA-, and FFA-induced metabolic injury characteristic of the diabetic milieu. Interestingly, AGE albumins and CML are present in the urine of individuals with diabetes with albuminuria due to DNP and have been shown to bind proximal tubular epithelium [ 22 , 32 ]. While the presence or absence of FFAs in the urine of diabetics with DNP has not been determined to date, FFAs may cause tubular apoptosis [ 33 ]. It remains to be determined whether FFA interacts with CD36 to activate CD36 receptor signaling, or whether CD36 mediates FFA uptake to activate src kinase and p38 MAPK signaling. Irrespective of the upstream mechanism of FFA and CD36 interaction, our results demonstrate very rapid activation of a well-characterized intracellular kinase cascade of proapoptotic signaling. Our finding that AGE-BSA and PA induce apoptosis through a CD36-mediated and p38- and caspase-dependent mechanism in tubular epithelial cells, similar to TSP-1 and ox-LDL in endothelial cells and macrophages, respectively, suggests that multiple, context-dependent extracellular stimuli of apoptosis may converge on CD36 scavenger receptor to activate src kinase and proapoptotic p38 MAPK pathway. In the context of the diabetic milieu and diabetic complications, our findings provide new molecular insights into diabetes-induced AGE- and FFA-dependent injury of renal epithelial cells. Almost all TUNEL-positive apoptotic tubular epithelial cells showed increased expression of CD36, suggesting a strong correlation between upregulation of CD36 expression and increased apoptosis in PTECs specifically in human diabetic kidney in vivo. Importantly, biopsy samples from cases of FSGS that were matched for degree of proteinuria, renal function, and hypertension were characterized by TED, IF, and increased tubular epithelial apoptosis; however, proximal tubular CD36 expression was similar to that in normal human control kidney. Therefore, CD36 expression in PTECs is specifically associated with the diabetic condition and appears to be independent of degree of proteinuria and renal failure. Indeed, increased CD36 expression in PTECs in human DNP in vivo may be caused by hyperglycemia, as we show that high glucose concentration stimulates CD36 expression in vitro. It is intriguing that CD36 expression was not detected in PTECs in diabetic mice with longstanding hyperglycemia in vivo, although underlying mechanisms for the species-dependent differential regulation of CD36 in PTECs in vivo and in vitro between mouse and human remain unclear at this time. Comparisons of human CD36 and mouse Cd36 genes indicate a high degree of sequence and structural similarity in both coding and regulatory regions, suggesting that the mechanism or mechanisms that underlie our findings are likely determined by sequence-independent, epigenetically distinct response patterns to the diabetic milieu that differ between these species. It is also possible that dietary or metabolic factors account for the differences in CD36 expression, as mice were maintained on standard mouse chow characterized by significantly lower fat and cholesterol contents than typical western diets consumed by humans. However, dietary or other unknown environmental factors cannot explain the differential CD36 regulation by glucose in human and mouse PTECs. Thus, we are exploring whether biochemical or functional differences between mouse and human PTECs in glucose metabolism or glucose-induced signaling can be identified. However, current lack of understanding of the observed differential regulation between human and mouse does not diminish the translational research significance of our findings, with their clear therapeutic implications. Thus, the present study identifies a new CD36-dependent molecular signaling pathway that mediates tubular epithelial apoptosis, and may underlie TED and IF, hallmarks of disease progression, specifically in human diabetic nephropathy. Third, to our knowledge, our report provides the first controlled study demonstrating increased apoptosis specifically in PTECs in DNP with TED and IF. These findings are consistent with a recent uncontrolled case series of five patients with DNP [ 34 ], and with previous reports demonstrating tubular apoptosis in kidneys of STZ-treated diabetic rats [ 35 , 36 ]. Interestingly, our study shows that tubular epithelial apoptosis was associated with TED and IF in human DNP, while normal appearance of tubular epithelium and interstitium was associated with baseline apoptosis rates in diabetic mouse models. Together, published observations from experimental diabetes models in mouse and rat, and human DNP, and our own findings in diabetic mouse models and human DNP, suggest a striking association of TED and tubular epithelial apoptosis. However, whether tubular epithelial apoptosis causes TED in DNP will require further investigation. Interestingly, acute and chronic chemical inhibition of caspase activity in a nephrotoxic serum nephritis model of chronic progressive glomerulonephritis with TED and IF reduced tubular apoptosis and TED [ 37 ]. Decreased tubular apoptosis and TED were associated with significantly reduced IF and decreased collagen synthesis in this model. This finding suggests that tubular epithelial apoptosis may trigger TED and IF in this model of chronic glomerulonephritis in rat, and supports our conclusions that diabetes-induced tubular epithelial apoptosis may underlie TED associated with IF in human DNP. In conclusion, we report a new functional role for CD36 scavenger receptor in tubular epithelial apoptosis associated with tubular degeneration and progression of DNP. Specifically, we show for the first time that both AGE and FFA PA induce PTEC apoptosis through CD36-mediated activation of src kinase, p38 MAPK, and caspase 3. Because high glucose stimulates CD36 expression in human PTECs and because CD36 expression is increased in apoptotic tubular epithelial cells in human DNP, we propose a two-step metabolic hit model relevant for TED, a hallmark of progression of human DNP. Supporting Information Figure S1 Glucose Regulates CD36 Expression in Tubular Cells Flow cytometric analysis of (A) human (HK-2) and (B) murine (M1) tubular epithelial cells incubated with control IgG (green curve) or with anti-CD36 antibody (FA6) (black curve) in medium containing 5 mM glucose (empty curve) or in medium containing 30 mM glucose (red curve) for 3 d. (45 KB PPT). Click here for additional data file. Accession Numbers The LocusLink ( http://www.ncbi.nlm.nih.gov/LocusLink/ ) accession numbers for the gene products discussed in this paper are caspase 3 (GeneID: 836), CD36 (GeneID: 948), MAPK (GeneID: 1432), p42/44 ERK MAPK (GeneID: 50689), p59fyn (GeneID: 2534), and Smad2 (GeneID: 4087). Patient Summary Background The kidneys are often affected in people with diabetes. Around one in three people with type 1 (juvenile, or insulin-dependent) and one in ten people with type 2 (late onset, or non-insulin-dependent) diabetes will develop kidney disease (called diabetic nephropathy). Diabetic nephropathy is one of the leading complications of diabetes and is the leading cause of kidney failure worldwide. Some risk factors make it more likely that certain people with diabetes will develop kidney disease—for example, kidney disease occurs more often in patients from South Asian or African backgrounds, in men, in patients with poor control of their blood sugar levels, and in those with high blood pressure or who smoke. However, the details of how, exactly, diabetes damages the kidneys are not clear. What Did the Investigators Do? They studied samples taken from the kidneys of humans and mice with and without diabetes and looked at the effects of high glucose concentrations on the cells in the kidneys. They found that in one part of the human kidneys high glucose caused a change in the cell surface causing an increase in a protein called CD36. This change occurred in the samples from people with diabetes, but did not occur in the samples from mice with diabetes. The investigators also found that some substances that are often found in the blood of people with diabetes could join to CD36; in doing so, these substances triggered the death of these cells, which is one of the first steps that occurs in diabetic nephropathy. What Do These Findings Mean? This particular protein (CD36) could have a central role in triggering diabetic nephropathy. Although there are no immediate clinical implications of this research for the treatment of people with kidney problems, this research helps in understanding how high glucose damages the kidney. In particular, it highlights how important it is to keep blood glucose levels as normal as possible. Where Can I Get More Information? Medline Plus's article on diabetic nephropathy: http://www.nlm.nih.gov/medlineplus/ency/article/000494.htm Diabetes UK's online information centre: http://www.diabetes.org.uk/infocentre/index.html National Diabetes Information Clearinghouse: http://diabetes.niddk.nih.gov/ National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) Animal Models of Diabetic Complications Consortium (AMDCC): http://www.amdcc.org/ | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC549593.xml |
523825 | Obstetric Fistula in Ilorin, Nigeria | In this perspective, Andrew Browning of the Fistula Hospital in Addis Ababa discusses a study on obstetric fistula in Ilorin, Nigeria. The study was originally published in the West African Journal of Medicine [ 1 ]. With the journal's permission, we have made a PDF of the full-text article freely available on our website ( see Text S1 ). | The obstetric urogenital fistula has caused women misery ever since they first started delivering children. It was once common worldwide, but with the advent of safe obstetric care during the early part of the last century, the condition has become rare in rich countries. Urogenital fistulae do still occur in developed countries, but unlike in the developing world, they are usually a complication of a difficult pelvic surgery, cancer, or radiation [ 2 ]. Obstetric Fistula: A Disease of Poverty Obstetric fistula—a urogenital fistula from obstructed labour—is now only encountered in countries where health resources are scarce. The shame associated with incontinence drives affected women further into a life of poverty and begging. Many women with fistula either do not know that they can get medical help, or if they do, they are unable to pay. Furthermore, very little scientific research has been published about obstetric fistula and its management, partly because the people treating patients with this condition are working in remote areas, often with very limited resources for research. What has been written consists largely of personal case series and a few epidemiological studies [ 2 ]. To date there has been only one randomised trial in the developing world, involving 79 women operated on by a single surgeon in Benin, which found that intra-operative intravenous antibiotics did not reduce the risk of failed surgical repair or of objective incontinence [ 3 ]. There has been only one study in a developing country comparing different surgical techniques—a retrospective study of 46 patients operated on over a five-year period at a hospital in Mumbai, India [ 4 ]. This study suggested that a technique called the Martius procedure (which involves grafting of a labial pad of fat) may be better than simple anatomic repair. What we do know about the obstetric urogenital fistula is that the women who have these injuries are young, usually illiterate, and of a lower socioeconomic background. They are more often primiparous and short in stature, and they have an average length of labour of some 3.9 days. The labour is usually unattended, or if attended, it is by someone unskilled. The women inevitably deliver a stillborn child. About half of the women with fistula are divorced as a direct result of their incontinence [ 2 , 5 , 6 , 7 ]. The Injury and its Consequences The initial injury that leads to a fistula results from ischaemic necrosis of the soft tissues of the pelvis due to an impacted presenting part during the long labour. Incontinent women face a life of shame and isolation (Photograph: © 2004 Shaleece Haas. This is an open-access image distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.) The ischaemia then affects the bladder and vagina (and sometimes the rectum and vagina), resulting in a fistula. The process also affects other pelvic structures. These include the nerves of the sacral plexus, resulting in foot drop and hamstring compartment weakness (foot drop may also be a result of prolonged squatting in labour, injuring the common peroneal nerve as it traverses the head of the fibula). Bony abnormalities are common, separating or obliterating the symphysis pubis. Up to half of patients develop upper renal tract abnormalities: scarring of the ureter can cause obstructive uropathies [ 8 ]. Up to two thirds of women are rendered amenorrhoeic (their periods stop), either from disorders of the hypothalamic-pituitary axis or from Asherman syndrome (adhesions in the uterus due to scarring). The vagina may be completely destroyed, as may the cervix, causing an obstructive outflow tract resulting in cryptomenorrhoea (women menstruate, but the sloughed blood and tissue don't leave the body). The continual leakage of urine over the perineal skin can cause local and painful irritation, termed ‘urine dermatitis’. Bladder stones can occur, as women affected by fistula often drink less to try and pass less urine and the concentrated urine can form calculi. The obstetric injury has been termed a ‘field injury’, as the pathology is broad rather than isolated [ 9 ]. The resulting range of injuries can be daunting for health professionals who are working with limited resources. The Ilorin Experience A recently published retrospective case note review provides new data on obstetric urogenital fistula in northern Nigeria. Ijaiya and Aboyeji reviewed 34 cases of fistula managed at the University of Ilorin Teaching Hospital over a two-year period [ 1 ]. During this period, there were 32,188 deliveries—thus, the incidence of fistula was 1.1 per 1000 live births. The mean age of the women with fistula was 23.9 years, and 32 of the 34 women were illiterate. Half were primiparous. The most common cause of the fistula was obstructed, prolonged labour—the cause in 28 out of the 34 cases. The most common complications of the fistulae were divorce or separation (eight women) and amenorrhoea (seven women). What Is Obstetric Fistula? ‘[Obstetric fistula] usually occurs when a young, poor woman has an obstructed labour and cannot get a Caesarean section when needed. The obstruction can occur because the woman's pelvis is too small, the baby's head is too big, or the baby is badly positioned. The woman can be in labour for five days or more without medical help. The baby usually dies. If the mother survives, she is left with extensive tissue damage to her birth canal that renders her incontinent.’ Source: UNFPA Campaign to End Fistula: “What is Fistula?” ( www.unfpa.org/fistula/about.htm ). How does this study compare with other literature on obstetric fistula in Nigeria? First, the incidence reported in the study is lower than that of another hospital study of 22,774 deliveries in Zaria, also in northern Nigeria, which gave an incidence of 3.5 per 1000 deliveries [ 10 ]. However, both of these incidence figures are from hospital-based studies, and it is thought that most women do not get to a health facility to deliver their child. So the true incidence of obstetric fistula may well be much higher. In terms of prevalence, it has been estimated that there are up to 800,000 women in Nigeria who have a urogenital fistula from obstructed labour [ 11 ]. Second, although the figures given in Ijaiya and Aboyeji's study differ slightly from those of other publications, their study does reconfirm the trends in aetiology and epidemiology of obstetric fistula in the developing world. Surgery at the Fistula Hospital, Addis Ababa (Photograph: © 2004 Shaleece Haas. This is an open-access image distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.) Treatment and Prevention As in other resource-poor countries, many women with obstetric fistula in Nigeria do not get to a surgeon with expertise in fistula repair. There are, however, a few dedicated professionals in Nigeria helping women with fistula and operating on up to 1,600 women a year [ 11 ]. Resource-rich countries were able to eradicate the obstetric fistula almost 100 years ago, but the challenge to resource-poor countries is enormous. There are an estimated 2 million women with fistula in the world, with anywhere between 100,000 and 500,000 new cases developing each year [ 11 ]. At the world's current capacity for dealing with the problem, it would take up to 400 years to treat the backlog of patients. Clearly we need many more centres equipped to care for women with fistula. The United Nations Population Fund (UNFPA; www.unfpa.org ) and the International Federation of Gynecology and Obstetrics ( www.figo.com ) are endeavouring to help. UNFPA has already sponsored training workshops on fistula surgery for surgeons and fledgling fistula units in Bangladesh and some parts of Africa. The obstetric fistula is an entirely preventable condition. Several strategies have been proposed to eradicate this condition in developing countries ( Box 1 ), just as it has been eradicated in the developed world. However, to prevent any new cases of obstetric fistula from occurring, there would need to be 75,000 new emergency obstetric centres built in Africa alone [ 12 ]. This would require not only funds, but an appropriate number of trained doctors, nurses, midwives, and support personnel. Box 1. The UNFPA's Key Strategies to Address Fistula ‘Postpone marriage and pregnancy for young girls ‘Increase access to education and family planning services for women and men ‘Provide access to adequate medical care for all pregnant women and emergency obstetric care for all who develop complications ‘Repair physical damage through medical intervention and emotional damage through counselling' Source: UNFPA Campaign to End Fistula: “Fast Facts” ( www.unfpa.org/fistula/facts.htm ). Even if such centres are established, women will need to be convinced of the importance of seeking help without delay for a difficult labour. And then, to be able to receive that help, roads need to be built, transport systems need to be put in place, and communications need to be improved. The obstacles are clearly huge, and with currently very little money and very few professionals available, women with obstetric fistula will sadly be with us for many more years to come. Supporting Information Text S1 Full Text of Ijaiya and Aboyeji's Study [ 1 ] (234 KB PDF). Click here for additional data file. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC523825.xml |
529267 | Trends in suicide in Scotland 1981 – 1999: age, method and geography | Background Male suicide rates continued to increase in Scotland when rates in England and Wales declined. Female rates decreased, but at a slower rate than in England and Wales. Previous work has suggested higher than average rates in some rural areas of Scotland. This paper describes trends in suicide and undetermined death in Scotland by age, gender, geographical area and method for 1981 – 1999. Methods Deaths from suicide and undetermined cause in Scotland from 1981 – 1999 were identified using the records of the General Registrar Office. The deaths of people not resident in Scotland were excluded from the analysis. Death rates were calculated by area of residence, age group, gender, and method. Standardised Mortality Ratios (SMRs) and 95% confidence intervals were calculated for rates by geographical area. Results Male rates of death by suicide and undetermined death increased by 35% between 1981 – 1985 and 1996 – 1999. The largest increases were in the youngest age groups. All age female rates decreased by 7% in the same period, although there were increases in younger female age groups. The commonest methods of suicide in men were hanging, self-poisoning and car exhaust fumes. Hanging in males increased by 96.8% from 45 per million to 89 per million, compared to a 30.7% increase for self-poisoning deaths. In females, the commonest method of suicide was self-poisoning. Female hanging death rates increased in the time period. Male SMRs for 1981 – 1999 were significantly elevated in Western Isles (SMR 138, 95% CI 112 – 171), Highland (135, CI 125 – 147), and Greater Glasgow (120, CI 115 – 125). The female SMR was significantly high only in Greater Glasgow (120, CI 112 – 128). Conclusion All age suicide rates increased in men and decreased in women in Scotland in 1981 – 1999. Previous findings of higher than expected male rates in some rural areas were supported. Rates were also high in Greater Glasgow, one of the most deprived areas of Scotland. There were changes in the methods used, with an increase in hanging deaths in men, and a smaller increase in hanging in women. Altered choice of method may have contributed to the increased male deaths. | Background Compared to the adjacent countries of England and Wales, Scotland had a low suicide rate through most of the twentieth century [ 1 ]. This did not appear to be explained by differences in recording of suicide [ 2 ]. Suicide rates in Scottish men increased in the 1970s and 1980s [ 3 , 4 ]. Rates in younger men continued to increase in the late 1980s [ 5 ] and early 1990s [ 6 ], at a time when male rates in England declined [ 7 , 8 ]. By contrast, female rates decreased in Scotland, although not as rapidly as in England and Wales [ 3 ]. Several authors have noted the importance of suicide in Scotland as a public health problem [ 9 , 10 ]. There was an increase in hanging and motor vehicle exhaust fumes as methods of male suicide in 1970 – 1989 and Pounder [ 5 ] suggested that choice of method might contribute to the increase in Scottish male rates, as some methods are associated with higher case fatality rates. Crombie [ 11 ] found that some areas had higher rates of male suicide than the Scottish average, mainly in rural areas. Access to particular methods of suicide may contribute to this [ 12 ]. Gender, age, suicide method and geographical area therefore appear to be important considerations in the epidemiology of suicide in Scotland. No recent summary of suicide trends in Scotland has been available, and this paper describes trends in relation to these factors. Methods We used anonymised information on deaths by suicide and undetermined deaths provided by the General Register Office for Scotland (GROS). Deaths registered during 1981 – 1999 were included if the cause of death was recorded as suicide or as undetermined cause (ICD-9 E950-E959 and E980-E989 respectively). Undetermined deaths were included as suicide deaths may be misattributed [ 13 , 14 ]. Population figures were taken from the GROS annual reports for the mid-year of each period. For analyses by area, if a death was registered away from the person's home address, the death was allocated to their area of residence, rather than the area in which they died. Deaths of people resident outside Scotland were identified using country codes, and were excluded. As far as possible, therefore, results reflect the rates of suicide and undetermined deaths of people resident in each area of Scotland. Standardised Mortality Ratios were calculated for National Health Service administrative areas, with 95% confidence intervals. In time period descriptions, the periods 1981 – 1985, 1986 – 1990, 1991 – 1995, and 1996 – 1999 were used. Data were analysed using Excel and SPSS. Results There were 14502 deaths recorded as suicide or undetermined cause in the time period. Of these deaths, 28.5% occurred in females (n = 4137) and 71.5% in males (n = 10365). Gender and age group Table 1 shows changes by gender. The male suicide rate for suicide and undetermined deaths increased from 187 per million in 1981 – 1985 to 252 per million in 1996 – 1999, an increase of 35%. In the same period, the female rate per million decreased from 88 to 82 per million, a 7% decrease. The female decline occurred between 1981 – 1985 and 1986 – 1999. By contrast, male rates increased between all time periods. The female: male rate ratio in 1981 – 1985 was 2.1:1 By 1996 – 1999 this had increased to 3.1:1. Table 1 Suicide and Undetermined Deaths in Scotland 1981 – 1999 By Gender and Time Period Rate per Million Population 1981 to 1985 1986 to 1990 1991 to 1995 1996 to 1999 Gender No. of deaths Rate/million No. of deaths Rate/million No. of deaths Rate/million No. of deaths Rate/million % change from first to last time period Males 2324 187 2,587 210 2,948 238 2,506 252 35% Females 1180 88 1,030 78 1,058 80 869 82 -7% Total 3,504 136 3,617 142 4,006 156 3,375 165 21% Examining changes by age group in males (Table 2 ), there are increases in male rates in the under 15 years, 15– 24 year, 25 – 34 year and 35 – 44 year age groups, of 137%, 97%, 86% and 26% respectively. The increase in the youngest male age group, although based on very small numbers of deaths, occurred between 1986 – 90 and 1991 – 95. In the 15–24 and 25 – 34 year age groups, increases occurred in every time period. There were decreases in the 45 – 54 and 55 – 64 year age groups and increases, of 4% and 10%, in the 65 – 74 and 75 years and over age groups. Table 2 Suicide and Undetermined Deaths in Males in Scotland 1981 – 1999 By Age Group and Time Period 1981–1985 1986–1990 1991–1995 1996–1999 Age group No. of deaths Rate per million No. of deaths Rate per million No. of deaths Rate per million No. of deaths Rate per million % change from first to last period <15 years 11 4.1 11 4.5 25 10.1 19 9.7 137% 15–24 317 141.0 440 208.3 467 257.2 369 278.0 97% 25–34 410 226.1 536 276.1 736 355.9 677 421.4 86% 35–44 422 268.5 470 279.2 594 340.8 506 338.0 26% 45–54 437 311.0 420 301.6 473 313.6 377 290.6 -7% 55–64 382 290.7 339 263.3 306 241.3 229 226.4 -22% 65–74 228 244.6 223 238.5 212 216.0 200 254.3 4% 75+ 117 253.6 148 287.8 135 253.6 129 278.8 10% All Ages 2,324 187.0 2,587 209.7 2,948 237.8 2,506 252.1 35% In women, there were increases in the three youngest age groups, with a 76% increase in rates in the under 15 year old group, 150% in the 15 – 24 year group and 37% in 25 – 34 year olds (Table 3 ). There were decreases in every older age group from 35 – 44 years to 75 years and over. Table 3 Suicide and Undetermined Deaths in Females in Scotland 1981 – 1999 By Age Group and Time Period 1981–1985 1986–1990 1991–1995 1996–1999 Age group No. of deaths Rate per million No. of deaths Rate per million No. of deaths Rate per million No. of deaths Rate per million % change from first to last period <15 years 7 2.7 7 3.0 7 3.0 9 4.8 76% 15–24 70 32.4 89 44.1 100 57.6 103 81.0 150% 25–34 141 78.9 174 91.4 217 106.4 172 108.3 37% 35–44 179 112.3 158 93.3 192 109.1 157 103.9 -8% 45–54 231 154.9 151 103.3 179 115.0 153 115.1 -26% 55–64 264 176.3 187 129.6 138 98.4 96 86.7 -51% 65–74 174 136.5 153 122.5 107 84.8 101 102.6 -25% 75+ 114 114.3 111 103.0 118 108.3 78 87.0 -24% All Ages 1,180 88.4 1,030 78.1 1,058 80.1 869 82.4 -7% Method of suicide The commonest methods of suicide and undetermined deaths in men were hanging, strangulation and suffocation, poisoning with solid or liquid substances, drowning, use of gases and vapours and jumping from high places (Table 4 ). Hanging, strangulation and suffocation in males had a similar rate to poisoning with solid or liquid substances in 1981 – 1985, but by 1996 – 1999 it had increased by 96.8% from 45 per million to 89 per million, compared to a 30.7% increase for self-poisoning deaths, from 46 per million to 60 per million. Deaths from jumping and cutting also increased, by 44.2% and 18.8% respectively. 'Other gases and vapours', predominantly car exhaust deaths (data not presented), decreased slightly from the first to last periods, but this concealed a substantial increase between 1981 – 1985 and 1986 – 1990, followed by a decrease in 1996 – 1999. Unspecified means increased by 70.9% from 14 to 23 per million. Table 4 Methods of Suicide and Undetermined Death in Males in Scotland 1981 – 1999 By Time Period 1981 to 1985 1986 to 1990 1991 to 1995 1996 to 1999 Primary Cause No. of Deaths Rate/million No. of Deaths Rate/million No. of Deaths Rate/million No. of Deaths Rate/million % change from first to last time period Solid or liquid substances 572 46 594 48 768 62 598 60 30.7 Hanging, strangulation and suffocation 562 45 630 51 774 62 880 89 96.8 Submersion(drowning) 353 28 417 34 358 29 268 27 -5.1 Other gases and vapours 288 23 428 35 448 36 225 23 -2.3 Other, unspecified means 169 14 170 14 226 18 231 23 70.9 Firearms and explosives 166 13 119 10 114 9 66 7 -50.3 Jumping from high place 163 13 176 14 203 16 188 19 44.2 Cutting and piercing instruments 40 3 38 3 38 3 38 4 18.8 Gases in domestic use 8 1 7 1 7 1 8 1 33.4 Late effects of injury 3 . 4 8 1 12 1 4 1 99.9 Total 2,324 187 2,587 210 2,948 238 2,506 252 34.8 In females, the commonest methods of suicide and undetermined death were poisoning with solid or liquid substances, hanging, strangulation and suffocation, and drowning (Table 5 ). Self-poisoning decreased by 4.8% between first and last periods from 45 to 43 per million, while drowning, and jumping from high places decreased by 54.1% and 30.3% respectively. Gases and vapours showed an increase but, as in males, the rate was higher in the middle two time periods. The rate of hanging, strangulation and suffocation deaths increased by 53.5%. There was an increase in unspecified means of suicide, from 4 to 8 per million. Table 5 Methods of Suicide and Undetermined Death in Females in Scotland 1981 – 1999 By Time Period 1981 to 1985 1986 to 1990 1991 to 1995 1996 to 1999 Primary Cause No. of Deaths Rate/million No. of Deaths Rate/million No. of Deaths Rate/million No. of Deaths Rate/million % change from first to last time period Solid or liquid substances 601 45 577 44 570 43 452 43 -4.8 Submersion(drowning) 248 19 159 12 132 10 90 9 -54.1 Hanging, strangulation and suffocation 127 10 95 7 134 10 154 15 53.5 Jumping from high place 89 7 77 6 67 5 49 5 -30.3 Other, unspecified means 60 4 53 4 81 6 83 8 75.1 Other gases and vapours 31 2 51 4 50 4 30 3 22.5 Firearms and explosives 13 1 10 1 6 0.5 3 0.4 -68.9 Cutting and piercing instruments 9 1 8 1 9 1 5 1 -25.0 Late effects of injury 2 0.4 - - 7 1 3 0.4 1.4 Gases in domestic use - - - - 2 0.4 - - - Total 1,180 88 1,030 78 1,058 80 869 82 -6.8 Geographical areas In the nineteen-year period as a whole, there was substantial geographical variation (Figures 1 and 2 ). The highest male rates were in Western Isles, Highland, Orkney, Greater Glasgow and Tayside (Table 6 ). When considered as Standardised Mortality Ratios, Western Isles, Highland and Greater Glasgow were statistically significantly elevated (Table 6 ). Six areas, Fife, Ayrshire and Arran, Forth Valley, Lothian, Borders and Lanarkshire had significantly lower SMRs than the Scottish average. Figure 1 Male Standardised Mortality Ratios for Suicide and Undetermined Deaths in Scotland, 1981 – 1999 by Area Figure 2 Female Standardised Mortality Ratios for Suicide and Undetermined Deaths in Scotland, 1981 – 1999 by Area Table 6 Male Standardised Mortality Ratios by Health Service Area Suicide and Death by Undetermined Cause in Scotland 1981 – 1999 Health Board of Residence No. of Deaths Rate/million Standardised Mortality Ratio Ratio LCI UCI Argyll & Clyde 909 225 103 97 110 Ayrshire & Arran 676 197 91 84 98 Borders 176 186 84 72 97 Dumfries & Galloway 301 223 101 90 113 Fife 634 197 90 83 97 Forth Valley 496 197 89 82 98 Grampian 1,050 219 99 93 105 Greater Glasgow 2,252 264 120 115 125 Highland 555 294 135 125 147 Lanarkshire 907 174 81 75 86 Lothian 1,399 202 90 85 95 Orkney 50 274 124 92 164 Shetland 47 214 99 72 133 Tayside 828 231 105 98 112 Western Isles 85 300 138 112 171 Scotland 10,365 220 100 - - In women, the highest rates were in Glasgow, Tayside, Highland and Dumfries and Galloway (Table 7 ). Only the Greater Glasgow SMR was significantly elevated. One area, Lanarkshire, had a significantly low female SMR. Table 7 Female Standardised Mortality Ratios by Health Service Area Suicide and Death by Undetermined Cause in Scotland 1981 – 1999 Health Board of Residence No. of Deaths Rate/million Standardised Mortality Ratio Ratio LCI UCI Argyll & Clyde 330 77 93 84 104 Ayrshire & Arran 290 78 95 84 106 Borders 83 81 95 77 118 Dumfries & Galloway 122 85 101 84 120 Fife 277 82 100 89 112 Forth Valley 195 73 89 77 102 Grampian 383 77 95 86 105 Greater Glasgow 921 99 120 112 128 Highland 169 86 105 91 123 Lanarkshire 366 66 83 75 92 Lothian 598 81 97 90 105 Orkney 15 80 98 55 162 Shetland 16 74 95 54 154 Tayside 357 92 110 99 122 Western Isles 15 53 65 36 107 Scotland 4,137 82 100 - - There were changes within areas over the period studied. Male rates increased in all fifteen NHS Board areas (Table 8 ). The smallest percentage increases were in Orkney, Highland and Greater Glasgow, three of the areas with the highest male rates in the first time period. Shetland and Western Isles had large percentage increases, but this was based on small numbers of suicide and undetermined cause deaths. The largest increases in mainland Scottish Board areas were in Argyll and Clyde (62%), Borders (60%), Forth Valley 59%), Ayrshire and Arran (56%) and Grampian (49%). Table 8 Male Deaths from Suicide and Undetermined Cause in Scotland 1981 – 1999 Rates By Health Service Area and Time Period 1981 to 1985 1986 to 1990 1991 to 1995 1996 to 1999 Primary Cause No. of Deaths Rate/million No. of Deaths Rate/million No. of Deaths Rate/million No. of Deaths Rate/million % change from first to last time period Argyll & Clyde 190 175 212 199 272 259 235 283 62 Ayrshire & Arran 143 158 161 178 194 214 178 246 56 Borders 31 128 41 167 62 245 42 205 60 Dumfries & Galloway 71 201 74 209 79 220 77 269 34 Fife 148 177 155 183 169 198 162 239 35 Forth Valley 102 154 127 192 136 205 131 245 59 Grampian 219 181 258 208 291 224 282 270 49 Greater Glasgow 551 234 565 252 666 304 470 270 15 Highland 131 273 155 315 144 284 125 305 12 Lanarkshire 195 140 242 177 255 187 215 197 40 Lothian 307 172 350 195 408 223 334 222 29 Orkney 13 275 11 233 15 308 11 281 2 Shetland 12 202 10 179 7 121 18 388 92 Tayside 193 205 200 214 230 242 205 272 33 Western Isles 18 230 26 343 20 274 21 376 64 Scotland 2,324 187 2,587 210 2,948 238 2,506 252 35 In females, rates changed little or decreased in all mainland Boards other than Argyll and Clyde (16.9% increase) (Table 9 ). Rates increased in Orkney, Shetland and Western Isles, although these were based on very small numbers of deaths. The greatest declines in mainland Board areas were in Ayrshire and Arran (30% decrease) and Grampian (27.9% decrease). Table 9 Female Deaths from Suicide and Undetermined Cause in Scotland 1981 – 1999 Rates By Health Service Area and Time Period 1981 to 1985 1986 to 1990 1991 to 1995 1996 to 1999 Primary Cause No. of Deaths Rate/million No. of Deaths Rate/million No. of Deaths Rate/million No. of Deaths Rate/million % change from first to last time period Argyll & Clyde 96 82 73 64 76 68 85 96 17 Ayrshire & Arran 88 90 73 75 80 82 49 63 -30 Borders 26 98 20 75 18 66 19 86 -12 Dumfries & Galloway 34 91 29 77 32 84 27 89 -2 Fife 79 89 76 85 69 77 53 74 -18 Forth Valley 53 75 44 63 55 78 43 76 0.3 Grampian 133 105 85 66 84 63 81 76 -28 Greater Glasgow 270 104 238 97 235 98 178 94 -10 Highland 41 82 40 79 59 112 29 68 -17 Lanarkshire 98 67 93 64 93 64 82 71 7 Lothian 165 85 156 81 143 73 134 84 -0.3 Orkney 2 41 2 41 3 60 8 202 390 Shetland 1 17 4 72 7 125 4 88 412 Tayside 91 88 93 92 101 99 72 89 0.6 Western Isles 3 38 4 53 3 41 5 88 130 Scotland 1,180 88 1,030 78 1,058 80 869 82 -7 Discussion The epidemiology of suicide in Scotland has changed greatly between 1981 and 1999. Male suicide rates have increased in all age groups up to and including 35 – 44 years. The highest male suicide and undetermined death rates in 1996 – 1999 were in the 25 – 34 year age group. In women, rates dropped in age groups from 35 – 44 years up to and including 75 years and over. Rates increased in younger women. There is limited information on the factors underlying individual deaths from suicide in Scotland. Squires and Gorman [ 15 ] reviewed the deaths by suicide of a group of young men in Lothian, and reported that a third had experienced recent relationship difficulties with a partner. Half of the group studied had a previous history of attempted suicide. Cavanagh et al [ 16 ] reported largely similar findings in a case control study in south-east Scotland. The group who had died by suicide had an odds ratio of 9.0 (95% CI 1.3 – 399) for current family problems, and an odds ratio of 5.0 (95% CI 1.1 – 47) for physical health problems. There was felt to be limited scope to intervene in suicide and deliberate self-harm through family health services because of limited contact, and non-specific presentation of problems [ 15 , 17 ]. Some methods of self-harm have higher case fatality rates [ 18 ]. Firearms have the highest case fatality rates, followed by drowning and hanging [ 19 , 20 ]. The most striking changes in male suicide methods in Scotland were the marked increase in hanging deaths, and the increase and subsequent decrease in deaths from 'other gases and vapours', which are mainly car exhausts. It seems likely that the decrease in motor vehicle exhaust fume deaths was related to the introduction of catalytic converters. Not all countries have reported a decrease in suicide from motor vehicle exhausts after catalytic converter introduction [ 21 ], but deaths in England decreased [ 22 ]. The reduction in deaths from motor vehicle exhaust fumes in England and Wales was associated with an increase in hanging deaths [ 7 ]. In Scotland, our data suggest that hanging deaths were increasing in men before deaths from motor vehicle exhaust fumes began to decline. The increase in hanging also appears greater than the decrease in motor vehicle exhaust deaths. The relationship between vehicle exhaust fume and hanging deaths in Scotland does not appear to be identical to that reported in England and Wales, and deserves further investigation. The difference between areas was also of note. The lower rates of increase in the areas with the highest initial rates may reflect to regression to the mean. Method availability [ 23 ] may be important in rural/urban differences. Obafunwa and Busuttil [ 24 ] reported that, within the Lothian region of Scotland, hanging was commoner in younger deaths, while use of car exhaust fumes for suicide was particularly important in rural areas [ 24 , 25 ]. In Lothian, an area that includes the capital city of Scotland, suicide by firearms was uncommon. Previous work has suggested higher rates of male suicide in some rural parts of Scotland [ 11 ]. Stark et al [ 12 ] have suggested that this may be related to the use of methods of self-harm in rural areas, such as firearms, with a high case fatality rate. Gunnell and colleagues [ 26 ] have argued, in relation to England and Wales, that changes in method preference, and therefore in case fatality, should be considered before concluding that changes must relate to social trends. Availability of method would not explain the differences between apparently similar rural areas. Previous work has found that deprived areas of Scotland tend to have higher suicide rates [ 27 , 28 ]. Deprived areas in Scotland were reported to have had the greatest increase in young male suicide between 1981 – 3 and 1991 – 3 [ 29 ]. Greater Glasgow, the non-rural area with the highest rate over the time period, is an area with substantial deprivation. Rural deprivation is difficult to measure, and recent work suggests that rural areas of Scotland may suffer greater levels of deprivation than had been realised. It is possible that rural deprivation is underestimated, and deprivation may explain more of the elevation in some rural areas than has been assumed in the past. Using routine information allowed a large number of suicide and undetermined deaths to be included in this series. There are, however, limitations to the use of anonymised routine data. No qualitative information was available, and our exploration of the data was limited to trends with no examination of possible underlying causes. The increase in the rate of deaths recorded as suicide or undetermined cause of death, but where no detail on method was included, could conceal recent trends. The increase as a percentage of relevant registrations was small, however, increasing from 7.5% in the first period to 9.1% in the last period studied. The classification of deaths as suicide is often difficult, but the inclusion of undetermined deaths as well as deaths recorded as suicide should have helped to minimise bias from under identification [ 14 , 30 ]. Squires et al [ 31 ] reported that improved communication between pathologists and the Registrar General for Scotland from 1994 on was associated with a decrease in undetermined deaths and in increase in deaths coded as being caused by dependent or non-dependent use of drugs. It is possible, therefore, that the figures for the final two periods may under-represent deaths that would have been identified as 'unidentified' in the earlier periods. Using information on Scottish residents only allowed identification of the suicide rates of local populations. Deaths of non-residents can account for up to 10% of all suicide and undetermined cause deaths in some rural areas of Scotland [ 12 ]. Our findings indicate that, even when these deaths are excluded, rates remain increased in some rural areas. Conclusions A divergence between male rates in England and Wales and in Scotland, and in male and female rates within Scotland, had been identified for the first part of the time period described here. This work found that male rates of suicide and undetermined death continued to increase in Scotland, but also identifies increases in younger female age groups. Examination of changes in method by male and female age group will help to establish whether changes in case fatality because of altered method choice [ 26 ] may be part of the explanation for these findings. The shift to hanging seems to be a significant trend in men in Scotland. It will be important to understand the reasons for this to allow appropriate intervention strategies to be considered. Some rural areas of Scotland had significantly elevated male suicide rates. We have suggested that access to lethal means of suicide may be one contributing mechanism for this, and have also noted the higher than expected suicide numbers in some rural occupations [ 12 ]. Rural areas are subject to poverty of income and opportunity, so it is also possible that rural deprivation may play an important part. Occupational associations of suicide in Scotland deserve further exploration. Examination of the association between deprivation, rurality and suicide may assist in the identification of possible interventions in rural Scotland. Competing interests The authors declare that they have no competing interests. Authors' contributions CS had the idea for the study, wrote the grant application, contributed to the design and interpretation and drafted the paper. Diane Gibbs and Tracey Rapson analysed the data. Paddy Hopkins contributed to the design, undertook part of the analysis, and commented on the interpretation of the results. Alan Belbin and Alistair Hay contributed to the design and helped interpret the results. All authors read and approved the final draft of the paper. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529267.xml |
523831 | The Birth of Reproductive Health: A Difficult Delivery | In 1994, the landmark "Cairo Conference" on population and development promised reproductive health for all. Ten years later, what has been achieved? | About a decade ago, I went wandering around Cairo's City of the Dead. This might sound like a grim bit of tourism, but my connection to that vast necropolis runs deep—quite literally, as my family is buried there. After visiting their grave, I rambled through the city's dusty alleyways, past hundreds of years of history. Yet what I remember most about that day was not one of the many magnificent tombs, but a simple brick building with a sign, of all things, for a family planning clinic. I was certainly surprised by my discovery; in retrospect, I should not have been. That part of Cairo is home to hundreds of thousands of people for whom looking after the dead is a way of life. Their fertility invigorates the funereal air: the caretaker of my family's tomb, for example, had a blooming family of his own living near the grave. Where better to offer family planning than in a place so poor that reproduction seemed more a matter of fate than choice? The Cairo Conference That visit is a fitting metaphor for the field of reproductive health as a whole. Ten years ago, officials, experts, and activists from 179 countries also came to Cairo for the International Conference on Population and Development (ICPD). The conference produced a 20-year plan of action that focused on universal access to reproductive health services, including family planning and sexual health; reducing infant, child, and maternal mortality; better education, especially for girls; equality between men and women; and sustainable development. The ICPD's key achievement was to reorient thinking on reproduction away from narrowly defined, government-dictated population control to a broader appreciation of reproductive and sexual well-being within health care systems, a view driven by individual choice and rights, not official priorities. “The Cairo Conference was a peak moment,” says Sally Ethelston, vice president for communications at Population Action International, one member of a consortium of non-governmental organisations launching a report card to mark the anniversary of the Cairo Conference in early September. “There were times when people were excited that they had accomplished something, and you could see it on their faces.” Today, however, the mood is very different. While progress has been made on some of the plan's targets, effort has faltered on others. And the conference “camaraderie” that Ethelston describes has given way to conflict between faith and science, over abortion and condoms. Like signs of life in the City of the Dead, the Cairo Conference gave birth to great expectations, some of which have already expired. Baby Steps Towards Cairo's Goals So, how far has the developing world come towards meeting the ICPD goals? There has certainly been progress on institutional reform in some countries, according to a recent survey of national policies by the United Nations Population Fund (UNFPA) [ 1 ]. For example, more than a third of the 151 countries questioned have introduced legislation on reproductive rights, and almost half have expanded their primary health care services to include family planning. The birth of a baby on August 15, 2000, brought India's population to one billion (Photo: Raghu Rai, on behalf of the David and Lucile Packard Foundation) But translating policy into action has been difficult. Overall, the picture is one of patchy success, according to Susheela Singh, director of research at The Alan Guttmacher Institute, a nongovernmental research organisation. Official statistics, as limited as they are for many aspects of reproductive and sexual health, show mixed results. On a positive note, global population growth has slowed to roughly 77 million people a year [ 2 ]. But while fertility rates have fallen in some developing countries, such as Mexico, they remain stubbornly high in others, such as Ethiopia [ 3 ]. Over the past decade, contraceptive use has grown, but so has demand, and there are now an estimated 201 million women in developing countries whose need for modern birth control goes unmet, resulting in 60 million unintended pregnancies a year [ 4 ]. Progress on legalising abortion has been slow, and an estimated 19 million abortions a year still occur under unsafe conditions [ 5 ]. Despite growing awareness of sexually transmitted disease, the annual number of sexually transmitted infections remains worryingly high at 340 million worldwide [ 6 ]. While infant mortality rates have improved somewhat, maternal mortality figures have barely budged. An estimated 529,000 women still die every year from complications of pregnancy and childbirth. The highest rates are in sub-Saharan Africa, where, on average, 920 women die for every 100,000 live births, compared with 24 deaths per 100,000 live births in Europe [ 7 ]. This is all the more distressing, says Vivien Tsu, senior programme officer at the Program for Appropriate Technology in Health, because these women's lives could be saved through straightforward measures and basic technologies, such as access to skilled midwives, simple drugs like magnesium sulphate for eclampsia and oxytocin for post-partum bleeding, cellular phones to call for help, and transportation to emergency obstetric centres. Obstacles to Reproductive Health So why hasn't more been achieved? One problem is certainly money. The 1994 Cairo Conference estimated the cost of implementing programmes for family planning, maternal health, and prevention of sexually transmitted diseases, as well as data collection and analysis in developing countries, at $18.5 billion by 2005—or $24.3 billion in today's dollars. The goal was to mobilise one-third of the money from donor nations, and the rest from developing countries themselves [ 8 ]. Last year, global spending on reproductive health and services reached $14.7 billion, according to estimates from UNFPA, the Joint United Nations Programme on HIV/AIDS, and the Netherlands Interdisciplinary Demographic Institute [ 8 ]. Encouragingly, investment has increased since 2001, when the momentum of ICPD seemed to falter and international spending fell to $9 billion. But this is still wide of the mark. While developing countries have failed to meet their conference commitments, it is donor countries that are most remiss: rich country contributions reached an estimated $2.3 billion in 2003 [ 8 ], a far cry from the conference target of $6.1 billion (or $8.1 billion in today's dollars) by 2005. Reproductive health is not alone in waiting for donors to give generously. For all the rhetoric at international summits, few rich countries have lived up to their lofty pledges of debt relief and of dedicating 0.7% of their gross domestic product to overseas development assistance. But as Steve Sinding, head of the International Planned Parenthood Federation (IPPF), points out, there are other reasons too for the shortfall. In the past donor interest was largely stimulated by fears of a population crisis. When the Cairo Conference reframed issues in terms of women's health and reproductive rights, rather than an impending population explosion, Sinding argues, the “demographic rationale” was lost, taking funding with it. Moreover, there are other issues competing for international funding, most notably AIDS. At the time of the Cairo Conference, 20 million people were infected with HIV; today the number has grown to an estimated 38 million [ 9 ]. AIDS threatens to derail the Cairo Conference plan of action. Through maternal-to-child transmission, and wide-scale orphaning, HIV threatens to reverse small successes at reducing infant and child mortality. By killing off teachers and sapping household incomes, AIDS is sabotaging education. By killing off scarce medical workers and overwhelming fragile health care systems, the disease is compromising reproductive health services. Gender equity is undermined, as women and girls bear the brunt of the epidemic, as caregivers, breadwinners, or patients themselves. Roughly half of the money spent on reproductive health last year went towards HIV/AIDS. And billions more is on the way, from the likes of the Global Fund to Fight AIDS, Tuberculosis, and Malaria and the United States President's Emergency Plan for AIDS Relief, which promises $15 billion over five years to HIV/AIDS programmes [ 10 ]. But much of this money is going into AIDS-specific programmes that do not address reproductive health more broadly. Even as the world is gearing up to scale up AIDS prevention and treatment to millions worldwide, few of the agencies involved come from the world of reproductive and sexual health. This is a pity because it means that HIV/AIDS programmes are not making use of valuable infrastructure and expertise already on the ground in places where AIDS hits hardest. Given that 57% of HIV infections in sub-Saharan Africa are among women [ 9 ], and that, for many of them, family planning clinics are their sole contact with the formal health care system, it seems odd not to integrate such services into the wider battle against HIV. Such centres can offer not only HIV testing and counselling, as well as condoms (against the double whammy of unwanted pregnancy and HIV infection), but also a broad-based message of good sexual health that can help protect against HIV and other sexually transmitted diseases. Moreover, pre- and ante-natal care provide an opportunity to stop mother-to-child transmission of HIV in its tracks. Condom distribution in Soweto, South Africa (Photo: Arjen van de Merwe, Population Concern) Where once family planning was the darling of international donors, HIV is now the cause célèbre. “There's a lot of resentment about the spotlight moving on,” says Kevin O'Reilly, a former reproductive health specialist now at the department of HIV/AIDS at the World Health Organization. However, there are now attempts to bring the two together. Meetings earlier this year in Switzerland, New York, and Bangkok have led to calls to action to strengthen links between programmes addressing HIV/AIDS and sexual and reproductive health. While this should help in the battle against AIDS, the money which flows to AIDS should also benefit reproductive health. Ideological Battles Arguably the most formidable obstacle to that union, and indeed further progress in improving reproductive health, is ideology. Since the Cairo Conference, a fierce battle has emerged between religious conservatives who eschew abortion and condoms in favour of abstinence and fidelity, and more liberal voices who argue for a full armamentarium to tackle these problems. The clash is loudest in the field of HIV/AIDS, where the President's Emergency Plan for AIDS Relief allocates a third of its funding for disease prevention to programmes focusing on abstinence and fidelity; public health experts argue that such an approach is ineffective at best, and dangerous at worst, without an equal emphasis on the availability of condoms for all. But the clash resounds in the wider arena of reproductive health as well. Four years ago, the ICPD's central target—access to reproductive services for all by 2015—failed to make it into the Millennium Development Goals, largely because of political nervousness. But as Kofi Annan, United Nations secretary-general, has pointed out, progress on the other key targets, such as eradication of poverty and hunger, will not be achieved without a focus on women's rights, education, reproductive health, and family planning. The fight between conservatives and liberals is clearest in the case of the US, which is the world's leading bilateral donor on reproductive health, spending $429 million this year [ 11 ]. However, this money comes with strings attached, says Françoise Girard, a reproductive rights lawyer in New York. Some of these are subtle. For example, Girard points to American pressure on several Asian and Latin American governments—during recent regional meetings to mark the anniversary of the Cairo Conference—not to re-affirm their commitment to the ICPD plan of action, with its emphasis on a full suite of reproductive rights and services. Other strings are more obvious. In 2001, George W. Bush reinstated the Mexico City Policy, otherwise known as the “Global Gag Rule”, which denies US family planning assistance—including money and contraceptive supplies—to any non-American group unless it certifies that it neither performs nor endorses abortion. IPPF, Marie Stopes International, and their local affiliates have been hard hit by the Rule, scaling back services in Kenya, Ghana, and elsewhere that offered essential health care to thousands of women and children. Then there is the Kemp-Kasten Amendment, a piece of US legislation which prohibits US assistance to any organisation as deemed by the President that “supports or participates in the management of a program of coercive abortion or involuntary sterilization.” At the behest of conservative supporters, President Bush has used the amendment to withhold $34 million in annual congressional appropriations to the UNFPA for the past three years. The UNFPA says that the $34 million could have been used to prevent 2 million unintended pregnancies, 800,000 induced abortions, 4,700 maternal deaths, and 77,000 infant and child deaths. The White House accuses UNFPA of abetting coercive reproductive practices in China—a claim which the UNFPA strenuously denies. Moreover, a number of international delegations, including one from the US State Department in 2002, have investigated the UNFPA's activities in China and failed to find evidence to support such allegations. Fortunately, other donors are stepping in to fill the breach: earlier this year, for example, the United Kingdom announced it would raise its contribution to the UNFPA to £80 million over the next four years, as well as increase its support to IPPF by a third. But even if the shortfall is made up, the ill will such clashes have engendered cannot be so easily salved. A Call for Strong Leadership Getting it right on reproductive health cannot wait another decade. The largest generation of young people in history—a whopping 1.2 billion aged 10–19 years—is entering adulthood [ 1 ]. They are making their sexual debut at ever earlier ages, against a backdrop of rising sexually transmitted diseases and growing social conservatism, which makes clear information, frank discussion, and free choice on abortion, contraception, and sexual health extremely difficult. More than ever, reproductive health needs strong leaders in rich and poor countries alike to mobilise both money and political commitment. Reproduction is a sexy subject; it is time the world again paid it the attention it deserves. Useful Links The Cairo Conference: http://www.iisd.ca/cairo.html Population Action International: www.popact.org UNFPA: www.unfpa.org Program for Appropriate Technology in Health: www.path.org The Alan Guttmacher Institute: www.guttmacher.org The Joint United Nations Programme on AIDS: www.unaids.org Netherlands Interdisciplinary Demographic Institute: www.nidi.nl IPPF: www.ippf.org Global Fund to Fight AIDS, Tuberculosis, and Malaria: www.theglobalfund.org The World Health Organization HIV/AIDS Programme: www.who.int/hiv/en | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC523831.xml |
529449 | Overweight, obesity, and colorectal cancer screening: Disparity between men and women | Background To estimate the association between body-mass index (BMI: kg/m 2 ) and colorectal cancer (CRC) screening among US adults aged ≥ 50 years. Methods Population-based data from the 2001 Behavioral Risk Factor Surveillance Survey. Adults (N = 84,284) aged ≥ 50 years were classified by BMI as normal weight (18.5–<25), overweight (25–<30), obesity class I (30–<35), obesity class II (35–<40), and obesity class III (≥ 40). Interval since most recent screening fecal occult blood test (FOBT): (0 = >1 year since last screening vs. 1 = screened within the past year), and screening sigmoidoscopy (SIG): (0 = > 5 years since last screening vs. 1 = within the past 5 years) were the outcomes. Results Results differed between men and women. After adjusting for age, health insurance, race, and smoking, we found that, compared to normal weight men, men in the overweight (odds ratio [OR] 1.25, 95% CI = 1.05–1.51) and obesity class I (OR = 1.21, 95% CI = 1.03–1.75) categories were more likely to have obtained a screening SIG within the previous 5 years, while women in the obesity class I (OR = 0.86, 95%CI = 0.78–0.94) and II (OR = 0.88, 95%CI = 0.79–0.99) categories were less likely to have obtained a screening SIG compared to normal weight women. BMI was not associated with FOBT. Conclusion Weight may be a correlate of CRC screening behavior but in a different way between men and women. | Background Colorectal cancer (CRC) is the third most common cancer in the United States with approximately 150,000 cases annually leading to about 57,000 annual deaths [ 1 ]. A prospective study of over 900,000 US adults found that, compared to normal weight adults, death rates from CRC were 20% to 84% higher in overweight and severely obese men and 10% to 46% higher in overweight and severely obese women [ 2 ]. Although other factors (e.g., age, family history) also contribute to CRC risk, obesity is a significant risk factor [ 1 ]. Thus, overweight and obese adults should consider obtaining regular CRC screening because early detection and intervention might reduce mortality [ 1 , 3 ]. However, studies suggest that overweight and obese women are more likely to delay cervical and breast cancer screenings than normal weight women [ 4 , 5 ]. In contrast, data from the 2001 Behavioral Risk Factor Surveillance Survey (BRFSS) indicates that, among men, overweight and obesity associates with obtaining prostate-specific antigen testing (Fontaine, Heo & Allison, under review). Although these cancers are sex-specific, the disparity led us to evaluate whether the obesity CRC screening association differed between men and women. Methods The Center for Disease Control and Prevention's BRFSS collects state-based data on preventive health practices and risk behaviors in the non-institutionalized civilian population aged ≥ 18 years [ 6 ]. The analyses we report are derived from the 2001 BRFSS. Information on BRFSS design and sampling methods are reported elsewhere [ 7 , 8 ]. Study variables Body mass index (BMI; kg/m 2 ), calculated from self-reported weight and height, was the predictor. Outcomes were interval since the most recent use of fecal occult blood test (FOBT), and sigmoidoscopy (SIG) in adults aged ≥ 50 years who reported ever having had the respective screening examination. BRFSS codes FOBT responses as: 'within past year', 'within past 2 years', 'within past 5 years', '5 or more years ago', 'don't know/not sure', or 'refused'. SIG is coded as: 'within past year', 'within past 2 years', 'within past 5 years', 'within past 10 years', '10 or more years ago', 'don't know/not sure', or 'refused'. Consistent with screening recommendations [ 1 ], we dichotomized FOBT as 0 = > 1 year since last screening vs. 1 = screened within the past year. For SIG, the American Cancer Society recommends screening every 5 years for adults aged ≥ 50 1 . Thus, SIG was dichotomized as 0 = > 5 years since last screening vs. 1 = screened within the past 5 years. We included age, education, race, income, self-reported general health status, smoking, employment, and health insurance as covariates. Statistical analysis We grouped respondents into 5 BMI-defined categories (18.5–<25 "normal weight", 25–<30 "overweight", 30–<35 "obesity class I", 35–<40 "obesity class II", and ≥ 40 "obesity class III"). Respondents (n = 250; .3%) with BMI's <18.5 ("underweight") were omitted from the analyses. We used multivariate logistic regression to estimate BMI-screening associations by entering the BMI-defined categories and potential confounders into the model as either continuous (e.g., age [including polynomials up to the third order]) or dichotomous variables (e.g., health insurance). Using the guidelines proposed by Greenland [ 9 ], we retained covariates that were statistically significant at the two-sided 0.20 alpha level or caused a ≥ 10% change in any of the BMI-defined categories when deleted. As a result, education, income, self-reported general health, and employment were omitted. Responses coded as 'don't know/not sure', or 'refused' were treated as missing variables and excluded from analyses, as were respondents with missing data on any covariates. To ensure unbiased general population estimates, we used sample weights provided by the BRFSS. BMI categories were investigated as 4 contrasts with the normal weight category serving as the referent. To evaluate whether sex moderated the BMI-screening association, we ran adjusted logistic models that also included BMI × sex interaction terms. Finally, because we observed a significant BMI × sex interaction, we then analyzed the data for men and women separately. Analyses were performed with SPSS 11.5. Results The mean age of the respondents was 65 years (median = 63). The mean BMI was 30.2 (median = 31) and 93% reported having health insurance. Less than half reported ever having either a screening FOBT or SIG (Table 1 ). Among those who ever had a screening examination 54.1% of men and 52.7% of women (χ 2 (1) = 6.61, p = .010) reported obtaining a screening FOBT within the previous year, and 84.4% of men and 80.3% of women (χ 2 (1) = 98.4, p < 0.001) reported obtaining a screening SIG within the previous 5 years. Table 1 Selected characteristics of respondents aged ≥ 50 years Characteristic Value* N Age, yrs 64.6 ± 10.1 84,284 Body mass index (BMI), kg/m 2 30.2 ± 6.2 84,284 Sex, % Men 38.2 32,179 Women 61.8 52,106 Race, % White 82.3 68,639 Non-white 17.7 14,778 Health insurance, % Yes 93.0 78,260 No 7.0 5,904 Smoking, % Current smoker 18.2 15,265 Former smoker 35.4 29,709 Never smoker 46.4 38,959 Ever had fecal occult blood test (FOBT), % Yes 43.0 37,498 No 53.9 49,123 Ever had screening sigmoidoscopy (SIG), % Yes 45.3 39,574 No 51.1 46,584 Screening fecal occult blood test (FOBT), % within past year 53.2 18,449 greater that 1 year 46.8 16,238 Screening sigmoidoscopy (SIG), % within past 5 years 81.8 30,465 greater than 5 years 18.2 6,771 * Plus-minus values are means ± standard deviation BMI was not associated with obtaining a FOBT (OR's ranged from 0.90 to 0.98). Compared to normal weight adults, however, those in the overweight (OR = 1.15, 95%CI 1.02–1.31), obesity class I (1.21, 95%CI 1.09–1.35), II (1.17, 95%CI 1.04–1.44) and III (1.27, 95%CI 1.05–1.58) categories were more likely to have obtained a screening SIG within the previous 5 years (p's < 0.05). The interaction effect between sex and BMI categories on FOBT was not significant (χ 2 (4) = 8.64, p=.071). However, the interaction effect between sex and BMI categories on SIG screening was significant, (χ 2 (4) = 114.03, p < .0001). BMI was not associated with obtaining a FOBT for either sex (OR's ranged from 0.87 to 1.05). However, compared to normal weight men, men in the overweight (1.25, 95%CI 1.05–1.51) and obesity class I (1.21 95%CI 1.03–1.75) categories were significantly more likely to have obtained a screening SIG. In contrast, obesity class I (0.86 95%CI 0.78–0.94) and II (0.88 95%CI 0.79–0.99) women were less likely to have obtained a screening SIG compared to normal weight women (see Figure 1 ). Figure 1 Adjusted odds ratios (OR) for obtaining a screening sigmoidoscopy according to BMI-defined categories for men and women * Significantly different from normal weight reference group at p < 0.05. Discussion These data support an association between BMI and obtaining a screening SIG within the previous 5 years, after smoking, health insurance, race, and age are taken into account. Moreover, the BMI-SIG associations were different between women and men. Women in the obesity class I and II categories were less likely to obtain SIG screening as a function of BMI. This is consistent with associations between BMI and delayed cervical and breast cancer screening [ 4 , 5 ]. On the other hand, men in the overweight and obesity class I categories were more likely to obtain a screening SIG. The reasons for this disparity are unclear. Perhaps physicians encourage cancer screening more vigorously among their overweight and obese male patients. Differences between men and women on factors such as self-esteem and body image [ 10 ] may also contribute to explaining the differential BMI-screening associations. These speculations underscore the importance of identifying barriers that might deter overweight and obese women from obtaining screenings. This study has limitations including: the BRFSS, a telephone survey, is prone to measurement error; because the BRFSS is an observational study, the BMI-screening associations could be due to residual confounding or confounding from unmeasured variables; the cross-sectional design did not allow testing causal inferences; and people without telephones, approximately 3% of the US population [ 6 ], are not surveyed through BRFSS. Conclusions These data indicate that weight may be a correlate of CRC screening behavior but in a different way for men and women. Competing interests The author(s) declare that they have no competing interests. Authors' contributions MH drafted the paper and assisted with the statistical analysis and interpretation. DB assisted with the writing of the manuscript and in the interpretation of the results. KF obtained and analyzed the data and assisted with the preparation of the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529449.xml |
524186 | Identification of rehabilitation needs after a stroke: an exploratory study | Background Services to meet adequate rehabilitation needs of elderly stroke survivors are not always provided. Indeed, since 1995, in the wake of the Quebec shift to ambulatory care, home care services, mainly those related to rehabilitation of the elderly, are either unavailable or incomplete. The aim of this study was to examine the rehabilitation needs of this clientele from their hospitalization to their reintegration into the community. Methods The "Handicap Production Process" conceptual approach was chosen to help identify the rehabilitation needs of persons affected by physical or cognitive disabilities due to the interactions between personal and environmental factors, and (activities of daily living, social roles). This qualitative exploratory study was performed in 2003. Data were collected among four groups of experts: patients, caregivers, health care providers and administrators. Data triangulation was used to ensure a rigorous analysis and validity of the results. Results Unfulfilled needs could be found in the categories of pertaining to residence, community living, psychological and emotional needs. Indeed, it appears that a psychological follow-up to discuss acceptance and consequences of non-acceptance would facilitate mid-to long-term rehabilitation. Conclusion Improving accessibility to healthcare services, respecting priority parking spaces for the disabled as well as promoting public awareness would enable a better social reintegration and recovery of social roles, thus limiting the onset of handicap situations. | Background After a stroke, a good proportion of the elderly rapidly re-enter the community without having benefited from rehabilitation services to help reduce their impairments and disabilities. Indeed, in Canada, only about 10% of stroke sufferers have access to intensive rehabilitation services [ 1 ]. The others, being relatively independent in their activities of daily living (walking, dressing, eating), return home with or without support services. However, aside from physical problems, there often are less noticeable disorders of a perceptive-cognitive nature (hemineglect, attention or organizational problems, and impaired learning ability). These disorders could trigger handicap situations [ 2 ] and be a source of subsequent losses of autonomy, which in turn can increase use of health services, recurring hospitalizations and premature institutionalization, resulting in an expensive health system. Moreover, a person's disabilities almost always affect the lives of his/her/family. The World Health Organization [ 3 ] defines rehabilitation as the combination and coordination of medical, social, educational and vocational resources aimed at optimizing a person's functional independence. Rehabilitation methods are essentially intended to reduce a person's disabilities and prevent the onset of disabling situations in order to support an optimal quality of life. Local community service centers (CLSC) are deeply concerned that they are unable to assume their role of providing, for their elderly users, front-line services that should cover various areas such as prevention, screening, general services and social reintegration [ 4 ]. The shift to ambulatory care initiated in Quebec in 1995 did not result in the development of outpatient rehabilitation services, even though these services, regarded as essential, can noticeably reduce hospitalization and improve quality of life. The outpatient clientele is more often left to fend for itself and sometimes has to turn to private clinics. For those with insurance coverage, the costs of these services are partly reimbursed; but 1.5 million Quebecois (23%) do not own private health insurance. Moreover, there is no comparable information on the waiting delays of private clinics that would allow to evaluate accessibility to their rehabilitation services [ 5 ]. Because of that lack of information, the Ministry of Health and the Regional Health and Social Services Boards can hardly provide cost-effective human resources in rehabilitation and equitably distribute financial resources, all the while keeping in perspective intra-regional as well as inter-regional needs. In order to delimit these issues from different perspectives, this study examines the rehabilitation needs of stroke patients in relation to their being cared for in their own home and according to their capabilities and their. This research was based on the "Handicap Production Process" conceptual approach (HPP) [ 6 , 7 ]. Handicap Production Process conceptual approach (HPP) The HPP (handicap production process) ensues from the works of the International Network of the Handicap Production Process (Figure 1 ) – formerly known as the Quebec Committee for the International Classification of Impairment, Disability and Handicap (QC-ICIDH) [ 6 , 7 ] – which followed those of the World Health Organization [ 8 ]. This anthropologic model, used in research as well as in clinical situations, holds four components: risk factors (causes), personal factors (body systems and aptitudes), environmental factors (facilitator and obstacle) and (social participation and handicap situation). Figure 1 Handicap production process conceptual approach [8,9] Personal factors relate to the person's basic characteristics like age, sex, socio-cultural identity, body systems (integrity vs. impairment) and the aptitudes (capacities vs. disabilities). Body systems identify components of the whole body. Integrity of these systems is based on the human biological norms while an impairment relates to a degree of anatomical, or physiological damage. Aptitudes are associated with a person's capacity for physical or mental activity such as walking or understanding. The value of an aptitude is measured on a scale ranging from optimal capacity to total disability. In the HPP, ensure the survival and thriving of an individual within society, during his/her entire life. They are arranged in twelve categories, the first six referring to activities of daily living and the last six relating to social roles valued by the person or his/her socio-cultural context. These categories are: nutrition, body condition, personal care, speech, habitation, mobility, responsibilities, interpersonal relationships, community living, leisure activities as well as education and work. The environmental factors are the constituents of a person's surroundings than can affect the realization of a living habit. Interactions between personal and environmental factors create needs that impact a person in the execution of his/her activities of daily living, thus limiting his/her social participation. A positive interaction between personal and environmental factors supports social participation while a negative interaction produces the development of handicap situations. Definition of needs Needs are subjective, since they are felt by the person (Talbot L. Les besoins de santé des individus . Unpublished manuscript). It could be a need for a resource to provide adequate and ample services. Bradshaw's taxonomy [ 9 ], used by Pineault and Daveluy [ 10 ], distinguishes four types of needs: normative, perceived, expressed and comparative. Normative needs are those that agree with norms, as defined by health professionals. Perceived needs are those perceived by individuals, depending on health services available. They become expressed needs, once articulated. Generalization of evaluated needs in a population results in comparative needs. Finally, needs depend on factors related to the person and his/her environment, on organizational factors, on factors related to the service providers [ 11 ]. As observed by repeated measures done in the post-stroke period, needs evolve with time, as do their response, depending on which services were provided. Furthermore, existing definition of needs seem to justify resource constraints rather than to satisfy health care needs of the person. The gap between perceived needs and normative needs is an area of improvement in the quality of services [ 21 ]. According to the HPP, rehabilitation needs would result from a discrepancy between the capacities of the person and the different factors of his/her environment [ 12 ]. These rehabilitation needs are associated with the handicap situations of the person in doing his/her daily activities, as compared to the most desirable level. Needs can therefore be multilevel (personal or environmental factors). Using four groups of people (patients, caregivers, health care providers and administrators), the aim of this research was to investigate the expressed and normative rehabilitation needs of post-stroke elderly living in the community. Methods Instruments As proposed by Morgan and Krueger [ 13 ], an inductive qualitative research tool was selected. The method of focus group discussion [ 13 ] was used with four groups of key-informants: patients, caregivers, health care providers and administrators. Focus groups are an effective method of obtaining data in new or ill-defined research fields. The method is divided in four phases: 1) establishing the questions; 2) planning the focus groups (number and size of groups, time and place of meetings, selection and recruitment of participants, choice of moderator); 3) leading of the focus group; 4) analysis and report. Participants (n = 25) were explained the aim and procedures of the study and agreed by signing a consent form. This research was approved by an ethics committee. Recruitment procedure In order to achieve experimental diversification, participants were identified by purposive selection. The patients (n = 4) and the caregivers (n = 5) were recruited through social workers from the local community service centers, daycare centers and also chosen from a data bank of participants to previous studies. Health care providers (n = 9) were solicited through directors of professional services and coordinators of rehabilitation or home-based services. They worked in different fields of healthcare and services, in rural and urban areas. Administrators (n = 7) were recruited through hospital managers who would identify which one was more familiar with the study clientele and worked in various rural and urban areas. All participants were recruited because of their critical abilities and their experience with needs related to the stroke process from onset to reintegration into the community. The area were the research took place is known as rural and semi-rural with a population of 150,000. Data collecting procedure Each of the four meetings lasted approximately two hours and was recorded on audiotape. A interview guide (Appendix 1 [ additional file ]) was built according to the method suggested by Morgan and Krueger [ 13 ]. Each group included a moderator and her assistant, an observer and a co-investigator. To make sure the participants would convey their perceived and expressed needs, the first meeting was held with the patients' group. Then followed the meetings with the caregivers, the health care providers and finally the administrators. The summary of the first meeting (group of patients) was used as an introduction to discussion for the second group; the third group benefited from the summary of the two previous ones and so forth. Data analysis process In order to ensure the validity of results, data was triangulated as described hereafter. During the last minutes of the meetings, the assistant-moderator would ask the participants to clarify some statements she had jotted down during the encounter. She would then sum up the meeting to validate its content with the participants, at which time they could complement, rectify or add to the information already given. This information along with the notes of the observer would help summarize the focus after each of the meetings. Later, a member of the team listened to the audiotape, allowing further analysis of the data. In the following weeks, each participant was sent a brief report of his/her focus group meeting. One week later, each participant was contacted to corroborate the content of the report. During these interviews, the comments of the participants testified to their understanding and approval of each report and allowed to add some details. Verbal and written summarized reports on the four groups were presented to the research team. Normative and expressed needs were grouped according to the twelve categories of the HPP model. After analysis of the focus groups, two categories were added in order to include psychological-emotional needs as well as psychological-cognitive needs and the education and work categories were abandoned. The participants from the four groups identified their needs as: a) needs already fulfilled or not mentioned, b) needs partially fulfilled c) unfilled needs. Results Characteristics of the participants of the four focus groups are presented in Table 1 . Patients were aged between 71 and 85 years old and most of them had suffered a stroke less than three years before and were identified as having severe to moderate limitations requiring more than 5 hours of assistance per week. None received home-based care from public services and only one in four was assisted by a relative about two hours a week. Most of the caregivers were retired women – aged between 42 and 60 years old – who took care of their disabled husband. Notice that the participants from this group were not related to the participants from the patients' group. Generally, the number of weekly care-giving hours varied from 8 to more than 20 hours. Table 1 Description of characteristics of patients and caregivers Patients characteristics n = 4 Caregivers characteristics n = 5 Age : Age : 71–75 years old 3 41–59 years 3 81–85 years old 1 60–69 years 2 Time since stroke : Current occupation : 2–3 years 3 Work outside home 1 4–8 years 1 Retired 4 Homecare public services Relationship with the stroke victim : Yes 0 Wife 4 No 4 Daughter 1 Help from relatives: Length of caregiver's role: Yes 1 (2 hrs/wk.) 3–4 years 2 No 3 More that 4 years 3 Level of education : Hours of weekly help : Primary 2 8–12 hours 1 High-school 2 12–16 hours 1 16–20 hours 1 More that 20 hours 2 Type of help provided : A.D.L. 5 I.A.D.L. 5 Psychological support 3 Other (stimuli) 3 ADL: activities of daily living IADL: instrumental activities of daily living Eight out of nine health care providers had a fairly good knowledge of stroke and seven of them had more than nine years of experience with a stroke clientele. They were from different professional fields: special education (1), occupational therapy (1), social work (2), neuropsychology (1), speech therapy (1), physiotherapy or physical rehabilitation therapy (3). These professionals practiced in an active care hospital (2), a CLSC (4), a daycare center (2), a day hospital (1) and a community organization (1). The administrators came from the same kinds of facilities with the addition of an intensive functional rehabilitation unit and a rehabilitation center. Four of the administrators had a restricted knowledge of stroke and had other customers aside from stroke patients. Two out of seven took care only of persons over 65 years old or with neurological problems. Education fields of the administrators were either management or health related disciplines. The most important expressed needs of the patients were acceptance of their health problem (stroke) and accessibility to physiotherapy and occupational therapy services on an outpatient basis. Then subsequently followed the needs relating to adapted means of transportation, medical follow-up, home visit from healthcare personnel, and stimulus and motivation provided by a caregiver. Domestic help and encouragement from the healthcare personnel were also important needs that have been expressed. Patients were more communicative about their needs relating to community, their psychological and emotional needs, and house alteration requirements. Unfulfilled needs mainly included the occasional home visit from a health professional, domestic help, coaching by the CLSC and medical follow-up. The foremost rehabilitation need identified by the caregivers was that the patient be loved, surrounded, and that he felt secure. These were followed by the necessity to make home adaptations, the need to inform the stroke patient and to provide him with physical and mental stimuli. Finally, the needs relating to tactile sensation problems, supportive care and attention, and acceptance of the situation were also important. It is worth noting that participating caregivers looked after individuals much more severely impaired than our group of patients. These caregivers mentioned the same categories of needs but in a different order of priority. For them, the psychological needs were the top priority; then came the need to adapt the home followed by community living needs. Notice that the beneficiaries had not resumed many of their social roles in the community. According to the caregivers, the less fulfilled needs were mainly related to psychological needs and associated with community living. Specifically, they concerned medication and its side effects, adaptation to this new situation for the caregiver and acceptance by the beneficiary. Respite, emergency help, supportive care and attention, training of family members were also needs partially fulfilled in these categories. For the health care providers , the needs that were less fulfilled mainly concerned related to community living, psychological needs and speech impairment. More precisely, they included: leisure activities, awareness, long-term family support, bond between spouses, respect of the person's pace, delivery of timely and simplified medical information to the patient and his/her spouse, adaptation of the home and respite, because in rural areas, for instance, there is no specialized transportation. Re-education for basic activities like eating or getting dressed and psychological assistance for acceptance, self-esteem and dignity were the most important. Followed the capacity to communicate and the family's need to be supported in understanding the health problem. Finally, long-term follow-up, simplification of the information given and home alterations also constituted essential needs, again according to the health care providers. For the most part, the health care providers worked with severely disabled persons. In their view, the interventions having to do with personal care held a priority over those concerning psychological needs and communication needs. As also identified by the caregivers, needs related to community living and the home followed. In the administrators ' view, a psychological support intervention made by the case manager, for instance, was more necessary in rural areas. They mentioned more "partially fulfilled" needs in the areas of speech, mobility and community living. They underlined the necessity to become sensitive to this clientele with cognitive problems and their needs for spirituality, means of transportation, financial support, the need to maintain and improve speech re-education. Administrators emphasized their wish to ensure the complete management of the patient. They were interested in the list of needs identified by the patients, since the needs of the affected individuals cannot be dissociated from those of their family. Needs identified as a priority by the administrators were a stabilized health condition, recovery of biological, psychological and social abilities, compensation mechanisms, integration into the community and support in finding a new sense to one's life. The chronology of those needs was unanimous because it set a continuum in the rehabilitation process. Priority needs identified by the administrators seemed closer to those listed by the patients. Administrators seemed to believe the patients had made a fairly good functional recovery and insisted on needs related to community living, personal care and psychological well-being. Finally, they were the only ones to clearly talk about spiritual needs. On the whole, unfulfilled needs were identified in the four groups as from categories relating to housing, community living, psychological and emotional needs. Table 2 summarizes the response to rehabilitation needs according to the categories of from the HPP model. Table 2 Participants' perception of the response to rehabilitation needs Patients Caregivers Health care providers Administrators F PF U F PF U F PF U F PF U Nutrition x nm nm nm Body condition x x x x Personal care x nm x x Communication nm x x x Housing x x x x Mobility x nm x x Responsibilities nm nm nm nm Interpersonal relationships including sexuality x x x x Community living x x x x Leisure activities nm x x x Psychological x x x x Cognitive x x x nm F : Need fulfilled; PF : Need partially fulfilled; U : Need unfulfilled; nm: Need not mentioned by this group of participants Note: the «education» and «work» categories were removed. Although some of the categories were partially fulfilled for the health care providers and administrators, the gradient of the providers' opinion on each of the needs specifically mentioned, stood between "partially fulfilled" and "unfulfilled". Conversely, the gradient of the administrators' answers stood between "fulfilled" and "partially fulfilled". Interestingly, only the patients' group acknowledged nutrition needs. Discussion In Lewinter et al. [ 14 ], results from individual interviews show that the cognitive needs are not very well fulfilled, if at all, by the available rehabilitation services. Results from the current study tend to draw the same conclusions. Indeed, the patients from the groups mentioned they had to perform their own intellectual activities, in order to stimulate and maintain cognitive functions during rehabilitation and after reintegrating their home. Furthermore, the patients from the current study point out the lack of continuity between resources when discharged from rehabilitation services. This also corroborates the results of Zwygart-et al [ 15 ], who collected their data through a mail questionnaire, and those of Brandriet, et al. [ 16 ]. The perceived needs of the caregivers really stood out during focus groups and would be inseparable from those of the patients. Caregivers considered rehabilitation in terms of recovery and excluded compensation mechanisms. For instance, they were unable to imagine any possible means to compensate for poor vision, a need expressed by the patients. While patients referred to the psychological need of accepting the situation, caregivers mentioned the psychological need to face one's own limitations and the probability of complete rehabilitation. These results agreed with the study of Gauthier [ 17 ] who collected his data six to nine months post-stroke, and who noticed some unfulfilled psychological needs. From our results, it seems that supporting the caregiver at the time of the hospitalization would allow him to offer a better support to the stroke victim. That support would perhaps allow acceptance of the situation and an active implication in the rehabilitation. Long-term follow-up and respect of the person's pace were mentioned as being important. Caregivers felt useful and appreciated this phase of the rehabilitation where attendance, supportive care and attention as well as information were provided at the day hospital. All groups pointed out the need for a more aggressive early re-education and a continuing rehabilitation. None of the groups mentioned needs pertaining to responsibilities. Finally, needs associated with sexuality were acknowledged by all groups, but none offered any solution. Health care providers felt powerless, even frustrated, with so many needs to be fulfilled and so little human resources available. Health care providers proposed many possible avenues worth exploring in relation with community and some ways of addressing psychological problems. The health care providers identified most of the partially fulfilled needs, ranging the whole continuum of the rehabilitation process. They pointed out the necessity for health system administrators to acknowledge expressed and normative needs of stroke patients. Adding human resources would begin solving the problem. Patients who benefit from intensive rehabilitation (about 10% of the stroke clientele) are treated according to their impairments in order to diminish their disabilities and improve their functionality. Services to help the family and to modify to the physical environment are usually available. The only help rarely provided is that of fulfilling psychological or cognitive needs and supporting social integration. However for 90% of the stroke clientele who only benefits from rehabilitation services at the acute care hospital, little of their needs are fulfilled. This data concurs with that of Lewinter et al. [ 14 ] on the inadequate length of rehabilitation services, as reported by patients and caregivers. As inferred by these focus group discussion, it would appear more important, in the current health system, to assess capabilities and impairments of the stroke patients than to offer proper services to fulfill their expressed needs. And yet, the same health care providers considered the assessment more important to their own normative needs than to those of the patients. In general, the answers of the administrators were based on what the resources should offer instead of on the accessibility to services, which they were less aware of. It seems that most of the participating administrators were preoccupied by the needs of the patients throughout the whole process and not only when they were using their services. They proposed solutions focused on the needs of the patient and on the most adequate resource to fulfill those needs at this stage of the rehabilitation process. In fact, the need for specially trained personnel to help prevent contractures and dehydration, as well as spiritual needs, were reported by administrators from chronic care facilities. Spiritual needs had also been reported by McLean et al. [ 18 ] in their pilot study with individual interviews. Administrators were much more precise in their description of the patients' sexual needs than the health care providers. This need was also described in the Lewinter et al. [ 14 ] study, but solutions have yet to be proposed. Since they are often consulted on this topic, the administrators were surprised the health care providers had so little to say about it. Admitting that the providers may be uncomfortable about their patient's sexuality, administrators sometimes provide counselling themselves. Regarding prevention and treatment of secondary impairments, which is the second stage of the rehabilitation process described by Duncan, et al. [ 19 ], administrators pointed out the importance of intense early rehabilitative interventions, no matter how old the stroke victim was. They noticed gaps in the early rehabilitation process that bear long-term repercussions. They identified a lack or resources or of specific knowledge as possible reasons of these impairments in those services where the priority is to have versatile care providers to work with a diversified clientele. They also mentioned the importance of integration to promote long-term rehabilitation. They seemed to trust existing services and the competence of health care providers regarding functional rehabilitation and compensation needs. Even though the administrators said that all services were available, they acknowledged the necessity of a resource person to help access these services. In a system where, it seems, the responsibility of answering the needs of stroke patients is being discarded, case managers and administrators will share this difficult task of making services available and guiding the patients towards them. It is interesting to note the discrepancy between the perception of fulfilled needs identified by the patients and the caregivers and those identified by the health care providers and administrators. This inconsistency shows that the two categories of needs in Bradshaw's taxonomy [ 9 ], namely the perceived and expressed needs (patients and caregivers) are different from the normative needs (health care providers and administrators). Strengths and limits Many strengths and limitations have to be mentioned. Methodologically speaking, holding the first meeting with the patients' focus group ensured a valid identification of their needs, as they were perceived. As for the other participants, recruitment was done in a rigorous manner in order to obtain a fairly equal number of representatives from rural and urban areas, from various professions, and from different types of rehabilitation resources, including representatives of community organizations. It should be mentioned that recruitment of the patients and caregivers was difficult; participants from these two groups were probably not typical of persons being dismissed to their home after a stroke. However, the fact that the severity of the stroke ranged from severe to mild in those two groups increases the external validity of the results. Recall bias is significant in this study as patients and caregivers had to recall events that took place few years ago. The implication of the four groups of participants ensures a better validity of the data as was suggested by Liu and Mackenzie [ 20 ]. The rigorous triangulation analysis also ensures a good validity of the results. Data collection having been done more than two years after the stroke allowed for identification of needs in the continuing process of rehabilitation, until social integration. The obvious needs concern the actual day-to-day living of the elderly, because the study was intended for stroke patients who were over 65 years old when they had their stroke. During the focus group discussions, patients who had suffered a stroke at least two years before conveyed their perception of unfulfilled needs all through the rehabilitation process. However, the "unfulfilled" perception may have been attributable to memory loss. Nevertheless, since the results from this group of participants agree with those from the caregivers' and health care providers' groups, and sometimes even with those of the administrators' group, it seems that the cause is to be attributed to the inaccessibility to services, their unavailability or a poor knowledge of their existence. To help clarify this question, a multi-center longitudinal study is ongoing to follow-up on the evolution of the needs in rehabilitation for post-stroke victims after being released to their home. Conclusions After a stay in an active care hospital (ACH), before the patient is discharged, a meeting between family members and a stroke specialist (social worker, nurse and volunteer worker having had similar experience) could allow discussions about a possible mental depression of the stroke patient and about upcoming difficulties. It would also be necessary at this time to give the caregivers information on their loved one's health status and to inform them of their future needs. After hospital discharge, a specially trained healthcare worker could remain available by telephone to counsel the caregivers. Finally, a psychological follow-up of the caregiver would allow him to better support his/her relative in his/her rehabilitation process. After discharge from the hospital or from the rehabilitation center, a better communication between healthcare facilities and increased availability of services at the CLSC – like twice a year home visits by a health care provider – are suggested by the patients. As concluded by Lewinter et al [ 14 ], it seems that a professional psychological follow-up to discuss acceptance and consequences of non-acceptance would favour mid- to long-term rehabilitation. Recommendations to health services Improving accessibility to services, respecting priority parking spaces for the disabled and promoting public cooperation would allow for a better social integration and recovery of social roles. On the whole, a better distribution of financial resources between institutions having to deal with this clientele, would allow long-term support of the person and his/her family. Health care providers first suggest that full rehabilitation teams (from different professional fields) be created (for example stroke team). They also propose the appointment of a case administrator or pivotal health care provider, based on the total system concept in healthcare, which would simplify communication between partners (hospitals, local community service centers, rehabilitation centers, daycare centers...). Services would be more suited to the needs of the individual and available for the families. Intervention priorities would be centered on needs expressed by the person. Education programs on how to deal with different types of clienteles would be made available to the health care providers to help them adjust to the difficulties associated with all kinds of issues. For mid- to long-term needs, improving means of transportation, adding support groups in rural areas, improving long-term follow-up in urban areas, humanizing of care, making information accessible, educating the neighbours and demystifying stroke by informing the public, would help integration and recovery of social roles to counter handicap situations. Finally, it is imperative to consider the perceived and expressed needs of the patients and caregivers and to integrate those needs with the normative needs identified in planning rehabilitation programs. Supplementary Material Additional File 1 Appendix 1: Course of the group discussions for the patients, caregivers, health providers and administrators Click here for file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524186.xml |
555743 | Unbiased descriptor and parameter selection confirms the potential of proteochemometric modelling | Background Proteochemometrics is a new methodology that allows prediction of protein function directly from real interaction measurement data without the need of 3D structure information. Several reported proteochemometric models of ligand-receptor interactions have already yielded significant insights into various forms of bio-molecular interactions. The proteochemometric models are multivariate regression models that predict binding affinity for a particular combination of features of the ligand and protein. Although proteochemometric models have already offered interesting results in various studies, no detailed statistical evaluation of their average predictive power has been performed. In particular, variable subset selection performed to date has always relied on using all available examples, a situation also encountered in microarray gene expression data analysis. Results A methodology for an unbiased evaluation of the predictive power of proteochemometric models was implemented and results from applying it to two of the largest proteochemometric data sets yet reported are presented. A double cross-validation loop procedure is used to estimate the expected performance of a given design method. The unbiased performance estimates ( P 2 ) obtained for the data sets that we consider confirm that properly designed single proteochemometric models have useful predictive power, but that a standard design based on cross validation may yield models with quite limited performance. The results also show that different commercial software packages employed for the design of proteochemometric models may yield very different and therefore misleading performance estimates. In addition, the differences in the models obtained in the double CV loop indicate that detailed chemical interpretation of a single proteochemometric model is uncertain when data sets are small. Conclusion The double CV loop employed offer unbiased performance estimates about a given proteochemometric modelling procedure, making it possible to identify cases where the proteochemometric design does not result in useful predictive models. Chemical interpretations of single proteochemometric models are uncertain and should instead be based on all the models selected in the double CV loop employed here. | Background Current computational methods for prediction of protein function rely to a large extent on predictions based on the amino acid sequence similarity with proteins having known functions. The accuracy of such predictions depends on how much information about function is embedded in the sequence similarity and on how well the computational methods are able to extract that information. Other computational methods for prediction of protein function include structural similarity comparisons and molecular dynamics simulations (e.g. molecular docking). Although these latter methods are powerful and may in general offer important 3D mechanistic explanations of interaction and function, they require access to protein 3D structure. Computational determination of a 3D structure is well known to be resource demanding, error prone, and generally requires prior knowledge, such as the 3D structure of a homologous protein. This bottleneck makes it important to develop new methods for prediction of protein function when a 3D model is not available. Recently a new bioinformatic approach to prediction of protein function called proteochemometrics was introduced that has several useful features [ 1 - 4 ]. In proteochemometrics the physico-chemical properties of the interacting molecules are used to characterize protein interaction and classify the proteins into different categories using multivariate statistical techniques. One major strength of proteochemometrics is that the results are obtained directly from real interaction measurement data and do not require access to any 3D protein structure model to provide quite specific information about interaction. Proteochemometrics has its roots in chemometrics, the subfield of chemistry associated with statistical planning, modelling and analysis of chemical experiments [ 5 ]. In particular it is closely related to quantitative-structure activity relationship (QSAR) modelling, a branch of chemometrics used in computer based drug discovery. Modern computer based drug discovery is based on modelling interactions between small drug candidates (ligands) and proteins. The standard approach is to predict the affinity of a ligand by means of numerical calculations from first principles using molecular dynamics or quantum mechanics. QSAR modelling is an alternative approach where experimental observations are used to design a multivariate regression model. With x i denoting descriptor i among D different descriptors and y denoting the biological activity, (linear) QSAR modelling aims at a linear multivariate model y = w T x = w 0 + w 1 x 1 + w 2 x 2 + ... + w D x D (1) where w = [ w 0 , w 1 , w 2 ,..., w D ] T are the regression coefficients and x = [1, x 1 , x 2 ,..., x D ] T . The activity y , may be the binding affinity to a receptor but may also be any biological activity e.g., the growth inhibition of cancer cells. In comparison with numerical calculations from first principles and similar approaches, the main advantages of QSAR modelling are that it does not require access to the molecular details of the biological subsystem of interest and that information can be obtained directly from relatively cheap measurements. The joint perturbation of both the ligand and protein in proteochemometrics yields additional information about the different combinations of ligand and protein properties for an interaction than can be obtained in conventional QSAR modelling where only the ligand is perturbed. In recent years, various other bioinformatic modifications of conventional QSAR modelling have been reported. These include simultaneous modifications of the ligand and the chemical environment (buffer composition and/or temperature) in which the interaction take place [ 6 - 8 ], and three-dimensional QSAR modelling of protein-protein interactions that directly yields valuable stereo-chemical information [ 9 ]. Although proteochemometrics has already proven to be an useful methodology for improved understanding of bio-chemical interactions directly from measurement data, the quantitative proteochemometric models designed so far have not yet been subject to a detailed and unbiased statistical evaluation. A key issue in this evaluation is the problem of overfitting. Since the number of ligand and protein properties available is usually very large, to avoid overfitting, one has to constrain the fitting of the regression coefficients. For example, in ridge regression [ 10 ], a penalty parameter is tuned based on data to avoid overfitting, and in partial least squares (PLS) regression [ 11 - 13 ] the overfitting is controlled by tuning the number of latent variables employed. In proteochemometrics as well as in many QSAR studies reported, the performance estimates reported are obtained as follows: 1) Perform a K -fold CV for different regression parameters, 2) Select the parameter value that yields the largest estimated performance value, and 3) Report the most promising model found and the associated performance estimate. Although this procedure may seem intuitive and may yield predictive models (as we in fact demonstrate below) the performance estimates obtained in this way may be heavily biased. Interestingly, this problem was recently addressed in the context of conventional QSAR modelling [ 14 ], and has also been discussed in earlier work, see [ 15 , 16 ]. As an alternative or complement to constraining the regression coefficients, one may also reduce the variance by means of variable subset selection (VSS). In QSAR modelling, many algorithms for VSS have been proposed based on various methodologies, for example optimal experimental design [ 17 , 18 ], sequential refinements [ 19 ], and global optimization [ 20 ]. VSS is used to exclude variables that are not important for the response variable, in the process of model building. Variables that are not important receive low weights in both a PLS and a ridge regression model, however if the fraction of unimportant variables is very large [ 21 ] the overall predictive power of the model is reduced. In this case VSS can improve the predictivity. However, if the fraction of unimportant variables is rather small, the quality of the model will not be improved by using VSS, it might on the contrary be slightly reduced. However, the interpretability of the model will in both cases be improved. Although many of the advanced algorithms for VSS are powerful, they are all computationally demanding. Therefore, in order to keep the computing time down in our use of the double loop cross validation procedure employed here, conceptually and computationally simple algorithms for VSS were used instead of the more advanced ones presented, e.g. in [ 17 - 20 ]. Most likely, the more advanced algorithms would yield more reliable models with even higher predictive power than for the models designed here. However, the main issue of interest in this paper is to confirm the potential of proteochemometrics. In previous reported proteochemometrics modelling, all available examples were used in the VSS. These were split into K separate parts and a conventional K -fold cross validation (CV) was performed. However, since all available examples were used, there were no longer any completely independent test examples available for model evaluation. Interestingly, this problem of introducing an optimistic selection bias via VSS was recently also pointed out in the supervised classification of gene expression microarray data [ 22 ]. In this paper we employ a procedure that can be used to perform unbiased statistical evaluations of proteochemometric and other QSAR modelling approaches. An overview of this so-called double loop CV procedure is presented in Figure 1 , and may be regarded as a refinement of the current practice in proteochemometrics in the following respects: 1. K 1 different variable subset selections are performed, one for each step in the outer CV loop. This avoids optimistic selection bias. 2. The best performance estimates ( Q 2 ) found in the inner loop by means of K 2 -fold CV are computed, but not reported as the model's performance estimate. This avoids the second optimistic selection bias mentioned above. 3. An unbiased performance estimate, P 2 , is computed in the outer loop and is reported as the performance estimate of the modelling approach defined by the procedure in the inner loop (the methods of VSS, regression, and model selection employed). P 2 is the result of different models that are designed and selected in the inner loop. It reflects the performance that one should expect on average. 4. Repeated K 1 -fold CVs which yield information about the robustness in the results obtained (presented as confidence intervals). In addition to these refinements, this work also demonstrates the potential of fast and straight forward alternative methods for VSS and regression in the inner loop. Moreover, it indicates that the performance estimates reported by certain software packages for QSAR may be quite misleading. We reanalyzed two of the largest proteochemometric data sets yet reported. The first data set is presented in [ 2 ] and contains information about the interactions between 332 combinations of 23 different compounds with 21 different human and rat amine G-protein coupled receptors. In total, there are 23 × 21 = 483 possible interactions and the basic task is to fill in the 483-332 = 151 missing values. The second data is presented in [ 23 ] and contains information about the binding of 12 different compounds (4-piperidyl oxazole antagonists) to 18 human α 1 -adrenoreceptor variants (wild-types, chimeric, and point mutated). As for the first data set, there are not interaction data available for all the 12 × 18 = 216 possible interactions, but for 131, see [ 24 ] for more details about this data set. Below these two data sets are referred to as the amine data set and the alpha data set , respectively. Results Software Computer programs were written in MATLAB (Mathworks Inc., USA) to integrate the double loop procedure in Figure 1 with robust multivariate linear regression using partial least squares (PLS) regression and ridge regression. These programs also contained two simple and fast methods for variable ranking called corrfilter and PLSfilter. For details, see the Methods section. Parameters The joint variable selection and PLS tuning performed in the inner K 2 -fold CV loop was performed with K 2 = 5. The different values of N D (the number of molecular descriptors) evaluated were 10, 20, 50, 100, 200, ..., 1000, 1500, 2000, ..., 6000 for the amine dataset and 10, 20, 50, 100, 200, ..., 1000, 1500, 2000 for the alpha data set. The values of N L considered were either the number of latent variables 1, 2, ..., 8, for both the amine and alpha data set or the degree of RR penalty 0, 0.5, 1.0, ..., 3.0 for the amine data set and 10, 50, 100, 150, 200 for the alpha data set. In the outer K 1 -fold CV loop, the same number of splits ( K 1 = 5) was used as in the inner loop. On the global level, the complete experiments were performed 100 times using different random partitions of the complete data sets. Unbiased predictive power In Table 1 a summary of the results from 100 randomly selected partitions of the complete amine data set are presented in the form of the mean values and standard deviations obtained. The number of molecular descriptors and latent variables selected in the inner loop are summarized in Table 2 . The average values of the biased Q 2 obtained in the inner loops look quite promising for the PLSfilter method ( Q 2 = 0.90 for both PLS and RR) and is even higher than the value reported in earlier studies [ 2 ]. However, the corresponding unbiased performance estimate P 2 is much smaller ( P 2 = 0.52 or 0.51 for PLS and RR, respectively). The Q 2 values for the models obtained after variable selection using corrfilter are significantly lower than when using PLSfilter, but the P 2 values are almost on the same level for the two variable selection methods when no variable selection at all is used. corrfilter reduces the number of descriptors to about one third of the initial number, but corrfilter still selects more than twice as many descriptors than PLSfilter (see Table 2 ). Since the main reason for variable selection is improving the interpretation of the model by reducing the number of descriptors, this indicates that one should select PLSfilter instead of corrfilter. In Figure 2 , the external (unbiased) predictions used to compute P 2 for the PLS model using PLSfilter show that there is useful predictive power, but only for examples with mid-range pK i values. The model has poor predictability for both low and high pK i values, indicating that the standard design procedure used in the inner CV loop does not always yield reliable models. This confirms earlier findings [ 14 ], that maximization of the unbiased performance estimate Q 2 is not always reliable, and also indicates that unreliable designs can be detected by means of the outer CV loop employed in this work. The estimated performances of the models for the alpha data sets are presented in Table 3 . Here both the Q 2 and P 2 values are high and the difference between the two measures is smaller than for the amine data set. This indicates a lower level of overfitting. The number of descriptors selected in the variable selection is much lower for the alpha data set (see Table 4 ) than for the amine data set. Both the high P 2 values and the display of the external prediction in Figure 3 show that the models have high predictive power. Also, the predictive power is significantly higher after variable selection than without. This is an example when variable selection does not only improve the model interpretability, but also the the model predictivity. The above results indicate, for example, that a combination of PLSfilter, PLS regression and model selection by maximization of Q 2 produces individual models with predictive power. The relative standard deviation of the predictive power is less than 5% for the two data sets considered. However, the number of variables selected has a relative standard deviation of 455/1933 = 25% and 69/199 = 35%, respectively. Moreover, the standard deviation in the number of latent variables (an implicit constraint on the regression coefficients) is approximately one (1.4 and 0.8) or 15%. In conclusion, the individual models are quite different but essentially all of them yield useful predictions. Comparisons to other programs To verify that our computations using MATLAB are comparable to computations by other programs, such as SIMCA, GOLPE and UNSCRAMBLER, models without variable selection were performed with all four approaches. In the comparison we have compared Q 2 values for models based on all descriptors built with PLS using between one and ten latent variables for the amine data set. All the Q 2 values were computed using the leave out CV method with five random groups and are presented in Figure 4 . Remarkably, the Q 2 values obtained with SIMCA 7.0 are much higher than for the other methods. This is due to the fact that SIMCA does not use the standard formula (Eq. 3) to compute Q 2 (personal communication with Umetrics), for some general information see [ 25 ]. Robustness and interpretability To study the robustness and interpretability of the set of models obtained using the two data sets considered, two different levels of information were computed and presented. The first level of information consists of two histograms displaying, for each ligand block (L1–L6 for the amine data set, and L1–L3 for the alpha data set (for the alpha data set the three ligand blocks correspond to the three positions of modification in the ligand)), and for each transmembrane region (TM1–TM7), how often different kinds of descriptors are selected. The histograms are based on the 500 observations obtained in the 5-fold cross validation performed 100 times using different, randomly selected, partitions of the data set. The descriptors are divided into receptor descriptors and ligand descriptors that are further subdivided into original descriptors, cross term descriptors, and absolute valued cross terms. In Figure 5 , hit rates for receptor and ligand blocks in the 100 different 5-fold cross validations performed are presented, both for the original, the cross term, and the absolute valued cross term descriptors. In Figure 5 A and 5B , the results for the amine data set are presented. Figure 5 C and 5D displays the corresponding results for the alpha data set. The second level of information displays the average and the standard deviation of the contribution of the different TM regions in the receptors for creation of receptor-ligand affinity according to the 500 different models designed. The contributions were calculated exactly as described in [ 2 ] for a single model, and then the average and standard deviation were calculated. Therefore, the results presented in Figure 5 corresponds to Fig. 3 in [ 2 ] where the results for a single proteochemometric model were presented. As before, for each TM region, the contributions to the affinity by different ligands is displayed, this time the variance (uncertainty) information is added. The top of Figure 6 shows the detailed contributions of TM regions to affinity, for each possible combination of ligand and receptor, according to the 500 different proteochemometric models designed using the amine data set, PLS regression and the PLSfilter VSS algorithm. The bottom part displays the corresponding results for the alpha data set when employing ridge regression and the PLSfilter VSS algorithm. Discussion In summary, the results reported here confirm earlier reports on the potential of proteochemometrics modelling for prediction of biological activity. It is interesting to note that the VSS did increase the predictivity of the models for the alpha data set, but not for the amine data set. The VSS for the alpha data set did also reduce the number of variables to approximately 4% of the original variables, while for the amine data set 15–38% of the variables remained after VSS. This indicates that for models where many variables receive low weights (as for the alpha data set) the VSS can significantly improve the model, whereas for a data set like the amine data set, with less low weighted variables, the VSS does not improve the model even though it can improve the interpret ability of the model. The basic goal of proteochemometric modelling is to obtain a single quantitative model that can predict biological activities accurately and which can be easily interpreted biochemically. In this context, it is important to stress that the only role of the outer loop employed in this work is to obtain unbiased estimates of the average performance of the design procedure considered in the inner loop. The additional random splitting of the data sets is used on top of this to gain information about the stability in the performance estimates. Thus, for procedures in the inner loop that yield small variances around a high average of P 2 , there is statistical support that a single design will yield useful predictions. In order for a single model to be chemically interpretable as well, all the models selected in the inner loop should yield approximately the same number (same set) of variables and the constraints on the regression coefficients (e.g., the number of latent variable in PLS regression) in all models should be approximately equal. With this in mind, the results presented in this work indicate that it is possible to design single proteochemometric models with predictive power based on the two data sets considered but that there is a relatively large variance (from one design set to another) in the variables selected and the constraints put on the regression coefficients. This indicates that although a single proteochemometric model would be useful for predictions, a detailed chemical analysis of such a model would be uncertain. More reliable information should be gained from a careful joint analysis of all the models (and their variables) selected in the inner loops of the different evaluations performed. For example, as briefly discussed in [ 9 ], the variables selected with the highest frequency should be of great interest. Thus, systematic and simultaneous biochemical analyses of all the models selected in the inner loops of this kind are required. For illustrative purposes of the complexity and potential of such analyses, here we have presented frequency distributions indicating which variable blocks are selected frequently in the two modelling problems considered. Moreover, we have also presented estimates of the variability (uncertainty) in estimating the contributions to affinity, between various combinations of ligands and receptors, from different transmembrane regions. In Figure 5 (top), histograms display how often different kinds of descriptors were selected in the 500 models designed for the amine data set. One conclusions is that for corrfilter, the absolute valued cross terms are selected three times as often as ordinary cross terms. Another conclusion is that for PLSfilter, fewer variables are selected and there is no obvious preference for one of the two types of cross terms. For the alpha data set it is obvious from Figure 5 C that only TM2 and TM5 are important to the model. From Figure 6 C and 6D , it is also obvious that the cross terms (and also the absolute valued cross terms) are selected less often than the ordinary descriptors. Figure 6 A and 6B displays contributions to affinity decomposed separately for each TM region and each drug/receptor combination. One conclusion here is that there is substantial variance in the estimates of the contributions which now is revealed and should dampen the risk of over-interpretations. Another conclusion is that the different regression and variable selection methods employed give similar results. Therefore, only one result each for the amine and the alpha data sets are presented in Figure 6 . A third conclusion is that a more clear and more reliable pattern of contributions can be identified in the present study than from the estimated contributions in [ 2 ] which were based on a single model only. For example, a pattern of consistently negative average contribution is found from TM3 and the receptors 5HT1B to 5HT1F, but this pattern does not appear in Fig. 3 of [ 2 ]. A fourth conclusion is that for the alpha data set, there seem to be no significant contributions to affinity from TM1, TM3, TM4, TM6 and TM7. This result agree with previous results for this data set [ 2 ]. Although earlier findings have been confirmed, one should note that there are a number of differences between the present and earlier studies which makes detailed comparisons difficult: 1) In earlier work different variable subset selection methods were employed and in some attempts there were no subset selection at all. 2) The normalization and use of nonlinear cross terms differ between the present and earlier studies of the alpha data set. 3) The limited forms of external predictions attempted earlier e.g., in [ 2 ] are not directly comparable with the present results. 4) Different software packages have been employed for model selection and performance estimation. Conclusion This work employs a methodology for unbiased statistical evaluation of proteochemometric modelling and confirms that proteochemometric modelling is a new bioinformatic methodology of great potential. The statistical evaluation performed on two of the largest proteochemometric data sets yet reported indicates that detailed chemical analyses of single proteochemometric models may be unreliable and that a systematic analysis of the set of different proteochemometric models produced in the statistical evaluation should yield more reliable information. Finally, although this work has focused on confirming the potential of proteochemometrics, the kind of systematic unbiased performance estimation employed here is of course also relevant for closely related areas of bioinformatics like microarray gene expression analysis and protein classification. Methods Data sets In the amine data set, each of the 23 compounds was described by means of 236 different GRid INdependent Descriptors (GRIND) [ 26 ] computed for the lowest energy conformation found and organized into 6 different blocks associated with different kinds of physical interactions. In addition, each receptor was split into seven separate transmembrane regions by means of an alignment procedure and then each amino acid was described by means of five physico-chemical descriptors (z-scales). In total, 159 trans-membrane amino acids were translated into 795 physico-chemical descriptors organized into 7 different blocks (regions). In the alpha data set each of the 12 different compounds was described by means of 24 binary descriptors indicating the presence of different functional groups at three positions in the compound. Moreover, 52 amino acids in the trans-membrane regions of the receptors were identified to have varying properties between receptors and each of them were also coded into five or two physico-chemical properties each, yielding totally 96 descriptor values. Before the proteochemometric modelling step, the amine data set was subjected to preprocessing in order to reduce the dimensionality of the original descriptors. This step should be part of the design procedure, leaving external examples outside. However, this issue is not expected to be critical and was therefore ignored in this study. For the compounds in the amine data set, after mean centering (no normalization), principal component analysis (PCA) was employed separately to each of six different blocks of GRIND descriptors, each block representing a particular kind of physical interaction. Similarly, each of the seven trans-membrane receptor block descriptors was subjected to PCA. This resulted in 6 × 10 = 60 compound descriptors and 7 × 15 = 105 receptor descriptors. Finally, 12,600 additional "cross-term" descriptors were produced by combining the compound and receptor descriptors nonlinearly. The cross-terms were added to account for non-linearities and they are shown to significantly improve the model predictivity. For each pair of compound and receptor descriptor blocks (totally 6 × 7 = 42 pairs), the 150 possible products between a compound and receptor descriptor value were computed. In addition, the absolute value of the deviation of each product from the average of the product over the data set available was computed. This resulted in 300 descriptor values for each of the 42 block pairs i.e., 42 × 300 = 12,600 values. For the alpha data set, the cross terms formed were the 2 × 24 × 96 = 4,608 possible products between the descriptors of ligands and receptors. No block-wise PCA was employed to reduce the dimensionality. As a final step before entering the modelling phase, all descriptor values were mean centered and normalized to have unit variance. Robust PLS and ridge regression In PLS regression, first a latent variable model x = t 1 b 1 + t 2 b 2 + ... + t M b M (2) of the vector x of descriptor values is created where t m is latent variable and b m is the corresponding basis (loading) vector. As few uncorrelated latent variables as possible which have the largest covariances with the response variable y , are selected. Then, a linear model y = a 0 + a 1 t 1 + ... + a M t M is obtained from ordinary least squares fitting. Usually, this predictor is transformed back into the original variables yielding y = w T x as in (1). The robustness of PLS comes from the latent variable modelling which eliminates problems caused by strongly correlated variables and few examples. Ridge regression achieves its robustness by adding a penalty term (or, equivalently, a Bayesian prior) to the ordinary least squares criterion that reduces the variances in the regression coefficients. In the experiments considered below, the degree of penalty used in the RR and the number of latent variables used in the PLS regression were tuned in the inner CV loop to maximize their corresponding inner K 2 -fold cross validation performance estimates. Variable ranking algorithms In the PLS modelling, the subsets of molecular descriptors used were selected jointly with the latent variables. Before the joint selection was performed, the molecular descriptors were ranked using two simple and fast methods: A bottom-up algorithm, which we call corrfilter, and a top-down algorithm which we call PLSfilter, corrfilter ranks the molecular descriptors according to the Pearson correlation coefficient between the descriptor and the response variable (the affinity). PLSfilter first builds a PLS model using all available descriptors and between one and L latent variables, where L is the number of latent variables associated with the model in (2) that explain 99% of the observed variance in y . Then each descriptor is ranked according to the corresponding mean of the squared coefficients, w i , in the regression models (1) from the L different models. For the alpha data set below only PLSfilter is applicable. This is due to the discrete nature of the ligand descriptors. Inner loop: joint VSS and regression parameter selection After completing the variable ranking, the most promising combination of the number of top-ranked variables and the number of latent variables in the PLS regression modelling or the degree of penalty in the ridge regression modelling was selected as judged by a K 2 -fold CV performance estimate. The performance estimates for different combinations of values of N D , the number of top-ranked molecular descriptors, and values of N L , the number of latent variables (PLS) or degree of penalty (RR), were considered. Finally, the pair ( , ) of numbers yielding the highest estimated predictive power was selected. The predictive power of the models was measured by the commonly used dimensionless quantity Q 2 defined as where n is the number of examples, y i is the measured biological activity of example i , is the corresponding prediction, and is the arithmetic mean value of all the measured activities. Hence, Q 2 is a CV estimate of the fraction of the variance of the response variable explained by the model. In the case of ordinary least squares fitting, Q 2 is also a CV estimate of the squared Pearson correlation coefficient between the true ( y ) and the predicted ( ) response values. Thus, a value of Q 2 close to one is traditionally interpreted as a good (valid) model. Outer loop: external K 1 -fold CV As already mentioned, selection of a QSAR model that maximizes a K 2 -fold CV performance estimate is common in conventional chemometrics and is also applied in proteochemometrics. This method of tuning is more complicated and therefore slower than simpler alternatives (such as tuning to maximize a single conventional hold out performance estimate) but is expected to be less sensitive to overfitting. Although parameter tuning based on CV is attractive, overfitting may still occur and the performance estimate obtained may be too optimistic. Some aspects of this danger were recently pointed out [ 14 ] and has also been discussed in much earlier work [ 15 ]. In conclusion, it is important to employ a second external CV as in Figure 1 to estimate the true performance also of sophisticated design procedures that employ CV for parameter tuning. For each step in the external K 1 -fold CV loop, one of the K 1 subsets of the whole data set was kept for validation and the rest were used for design of a regression model. The predictions obtained in this outer CV loop were finally used in the formula for Q 2 in (3). However, since the predictions used for calculating Q 2 were kept outside the whole design procedure, as in earlier work [ 9 , 16 ], we denote the computed quantity by P 2 to indicate that this is an unbiased performance estimate based on external predictions. Repeated K 1 -fold CVs The results obtained from a single K 1 -fold CV are interesting but are sometimes heavily influenced by the particular data partitioning used. In the work reported here, we therefore performed repeated K 1 -fold CV in the outer loop. For each partitioning selected randomly, the corresponding value of P 2 was computed using the procedures described above. Thus, a set of different values of P 2 were obtained and used for determination of the variability in the results obtained. Computations The main body of programming and computations were performed using MATLAB on standard processors (900 MHz). For comparisons, we also employed the program packages SIMCA (Umetrics, Sweden), GOLPE [ 17 ] and UNSCRAMBLER (CAMO, Norway). Authors' contributions E.F. and M.G. devised and implemented the proposed double CV loop procedure, the feature selection algorithms, and a numerically efficient version of ridge regression required. P.P., M.L., J.E.S.W. provided the data sets studied together with experience and insights gained from their earlier work on proteochemometrics. V.M. and M.G. supervised the project. All authors read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555743.xml |
212690 | Out of the Way | By default, all published works are copyrighted. Creative Commons provides means for authors to share their work more freely | When did calling a lawyer become part of the scientific process? It hasn't officially, of course. But as a generation of researchers has grudgingly come to know, navigating the legal red tape of universities, corporations, and publishers is an inevitable part of the practice. Whether seeking access to information or sharing one's own findings with others, scientists increasingly find themselves having to ask an intermediary's permission. This “ask first” culture has developed at just the moment when technology has opened vast new possibilities for collaboration and information-sharing. The timing is not coincidental. Policymakers, under the influence of lobbies defending pre-digital business models, have reacted to new technology with ever more extreme intellectual property laws. The result is a legal regime tailored to a powerful minority but ill-suited to a number of other constituencies—scientists and scholars chief among them—that thrive on openness. Worries over broad notions of “piracy” and “asset management” have insinuated themselves into fields where those terms, until recently, held no meaning. The Public Library of Science (PLoS) is at the vanguard of a growing cross-disciplinary movement to counteract this trend by demonstrating that voluntary models of open publishing are not only viable, but crucial to scientific innovation. Yet PLoS’ goal of “immediate, unrestricted access to scientific ideas, methods, results, and conclusions” is not immediately compatible with the stringent rules of copyright, which apply fully and automatically to all published works, by default. The exercise of something less than full copyright requires, oddly, some legal tinkering—which is where Creative Commons, the organization I help manage, comes in. Creative Commons, a 501 (c) (3) nonprofit corporation based at Stanford University in California, is led by a board of expert legal and technical thinkers. (Its chairman, Lawrence Lessig, a law professor at Stanford and a recipient of the Scientific American 50 Award in 2003, recently joined the PLoS board of directors.) Creative Commons was founded on the idea that some people prefer to share their works on more generous terms than standard copyright provides. The organization offers such authors an easy and clear way to announce these preferences. The goal is to help endeavors like PLoS, as well as individual authors, expand access to quality content online while reducing the legal friction and uncertainty of copyright law. In other words, Creative Commons offers legal tools to help clear permissions, once and for all. We help get the law out of science's way. These tools are the Creative Commons licenses, a suite of form legal documents available for free on the Creative Commons website. Each license allows an author to retain his or her copyright while permitting certain uses of the work, on certain conditions: to declare “some rights reserved” as opposed to “all rights reserved.” From a simple menu, copyright holders mix and match their preferences: an attribution requirement; a prohibition on commercial reuse; a restriction on derivative works; or a “share-alike” provision that obligates licensees to offer any derivative works to the public on the terms they received. (PLoS has chosen the simplest and least restrictive of the licenses, permitting copying, as well as free commercial reuse and transformation, in exchange for simple attribution.) Since the licenses’ launch in December 2002, nearly 800,000 Web pages (well over 1,000,000 discrete works) have been made available under Creative Commons licenses. Because they're free and can apply to any kind of copyrighted work, the licenses have been popular with Webloggers, teachers, novelists, musicians, photographers, and hobbyists. Many institutional adopters, too, have used the licenses to facilitate innovative publishing techniques, particularly in the sciences. The Massachusetts Institute of Technology's Open Courseware project publishes materials from its university courses under a version of the licenses, inviting students and educators from around the world to reuse them royalty-free. Rice University's Connexions project, an interactive tool that helps instructors build courses and texts from a collective knowledge repository, requires authors to license their contributions for free reuse in return for authorial attribution. The American Museum of Natural History's Biodiversity Commons will soon use the licenses to facilitate search across a broad collection of conservation databases and websites. Like PLoS, all of these projects use Creative Commons licenses to simplify and streamline the process of rights clearance. But the licenses also serve another critical function: they formalize the collaborative ethos of the scientific and academic communities in a language that legal intermediaries cannot quarrel with. This standardization also helps otherwise disparate communities, whether across disciplines or geographic boundaries, to agree in advance on the rules for sharing. Creative Commons is now considering expanding into other fields where the law has begun to restrict open research: scientific data and patents, in particular. With a portion of a new US$1 million grant from the Hewlett Foundation (putting our total of funding received at over US$3 million), we hope to build the Science Commons, a branch of the organization dedicated to bringing a measure of reason, and restraint, to the legal thicket that has grown around scientific research. Like PLoS, Creative Commons’ goals and methods are designed to make the most of the opportunities created by new communications technology. But, also like PLoS, our inspiration reflects the wisdom and optimism of the Enlightenment as much as that of the Digital Age. We are trying to restore the sense of legal moderation that policymakers of a bygone era, heavily influenced by the philosophy of the first scientific revolution, understood would “promote the progress of science and the useful arts,” as the U.S. Constitution puts it. “Knowledge [is] not the personal property of its discoverer, but the common property of all,” wrote Benjamin Franklin, the great cosmopolitan, polymath, and patron saint of innovation. As we enjoy great advantages from the inventions of others, we should be glad of an opportunity to serve others by any invention of ours, and this we should do freely and generously.” Franklin, who knew a thing or two about both publishing and science, never practiced law. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC212690.xml |
535935 | Pregnancy weight gain and breast cancer risk | Background Elevated pregnancy estrogen levels are associated with increased risk of developing breast cancer in mothers. We studied whether pregnancy weight gain that has been linked to high circulating estrogen levels, affects a mother's breast cancer risk. Methods Our cohort consisted of women who were pregnant between 1954–1963 in Helsinki, Finland, 2,089 of which were eligible for the study. Pregnancy data were collected from patient records of maternity centers. 123 subsequent breast cancer cases were identified through a record linkage to the Finnish Cancer Registry, and the mean age at diagnosis was 56 years (range 35 – 74). A sample of 979 women (123 cases, 856 controls) from the cohort was linked to the Hospital Inpatient Registry to obtain information on the women's stay in hospitals. Results Mothers in the upper tertile of pregnancy weight gain (>15 kg) had a 1.62-fold (95% CI 1.03–2.53) higher breast cancer risk than mothers who gained the recommended amount (the middle tertile, mean: 12.9 kg, range 11–15 kg), after adjusting for mother's age at menarche, age at first birth, age at index pregnancy, parity at the index birth, and body mass index (BMI) before the index pregnancy. In a separate nested case-control study (n = 65 cases and 431 controls), adjustment for BMI at the time of breast cancer diagnosis did not modify the findings. Conclusions Our study suggests that high pregnancy weight gain increases later breast cancer risk, independently from body weight at the time of diagnosis. | Background Sensitivity of the breast to hormones and environmental exposures varies throughout a woman's life span [ 1 ]. During pregnancy, the breast undergoes extensive changes in preparation for lactation. High estrogenicity during pregnancy causes marked cellular proliferation, in both in the normal and tumor cells. Normal breast cells differentiate to milk-secreting alveoli, while tumor cells, if present, continue to multiple and eventually form a detectable tumor. These two events probably explain the dual effect of pregnancy on breast cancer risk: pregnancy before age 20 reduces, whereas first pregnancy after age 30 increases, breast cancer risk [ 2 ]. In young women, pregnancy may eliminate future targets for neoplastic changes by differentiating target cells [ 3 ]; the breast tissue of older first time mothers is more likely to have acquired malignant cells that are stimulated by high pregnancy hormonal environment. Women whose pregnancy estrogen levels are elevated are at an increased risk of breast cancer. For example, women who took the synthetic estrogen diethylstilbestrol (DES) during pregnancy are at an increased risk of developing breast cancer [ 4 ], as are women who suffered from severe pregnancy nausea [ 5 ] or who gave birth to heavy newborns [ 6 ]. Both nausea in pregnancy and high birth weight are linked to elevated pregnancy estrogen levels [ 7 , 8 ] Conversely, pregnant women having high alpha feto-albumin levels [ 9 , 10 ], or suffering from hypertension or pre-eclampsia [ 11 , 12 ], exhibit a reduced risk. Alpha feto-protein has direct antiestrogenic activity and binds estrogens, reducing their biological availability [ 13 , 14 ]. Hypertension during pregnancy is linked to reduced estrogen and increased testosterone levels [ 15 ]. A recent study in which estrogen levels were measured in stored blood samples of pregnant women later diagnosed with breast cancer, provides direct evidence in support of high estrogen and low progesterone levels in increasing maternal breast cancer risk [ 16 ]. However, some studies have failed to find an association between pregnancy estrogen levels, determined indirectly, and maternal breast cancer risk [ 11 , 17 ]. Adipose tissue aromatizes androgens to estrogens, and thus high body mass index (BMI) is linked to elevated estrogen levels in postmenopausal women [ 18 ]. Some studies suggest that high pregnancy weight gain may be associated with increased pregnancy estrogen levels [ 19 ], although this has not been confirmed in more recent studies [ 20 , 21 ]. The goal of this study was to determine whether high pregnancy weight gain affects breast cancer risk. Methods The cohort The study population was a historic cohort of women pregnant between 1954 and 1963 in Helsinki, Finland (n = 4,090). The cohort was a sample gathered for a study on hormone exposure, including 2,022 exposed, 2,062 controls and 6 women with unknown hormone exposure status. Information on the cohort was collected from the maternity cards of municipal maternity centers, which are used by most pregnant Finnish women. The hormone-exposed women had been prescribed estrogen or progestin drugs during pregnancy to prevent early abortion or preterm delivery. For each exposed woman, a woman next in the maternity center file who gave birth during the same year and had not been prescribed hormones during pregnancy, was chosen as a control. The cohort has been previously prescribed in detail [ 22 , 23 ]. There were no differences in breast or other estrogen-dependent cancers between hormone-exposed and control mothers [ 22 ]. Visits to a private doctor were used as an indicator of socio-economic status, since no information on education or occupation at the time of the index pregnancy was available. Cancer cases were identified through a record linkage to the national cancer registry until June 2001. Mortality and emigration data were obtained from the population registry until August 2001. The linkage between the cohort and the registries was based on a unique personal identification number. Inclusions and exclusions Inclusion criteria were the following: first and last visit at the maternity center between 4–45 th gestation weeks, the time between the body weight measurements 3–300 days, and delivery between 22–45 th gestation weeks. For each mother, the gestation week she gave birth was determined by using the date of estimated timing of delivery. Women who did not fulfill these criteria were excluded (Fig. 1 ). In addition, women with multiple births were excluded because their weight gain is not comparable to that of mothers of singletons. Mothers with pre-eclampsia or eclampsia were excluded because they accumulate weight as fluid retention during pregnancy, and have been reported to have a reduced breast cancer risk [ 11 , 12 ]. Figure 1 Study population and exclusions. Pregnancy weight gain was first calculated as the difference between the first and last visit to maternity center. However, this window varied considerably among included mothers (range 3–295 days). The time-period of calculated weight gain was therefore adjusted by extrapolating a line for each mother to reflect her weight increase during pregnancy. The calculations are described in detail in Additional File 1 . After the unstable period of early pregnancy, a mother's weight increases linearly [ 24 ]. Mothers usually begin to gain weight after the first trimester (e.g [ 25 ]). We extrapolated the line separately for 0–15 th (Line A) and 15–40 th gestation weeks (Line B) for each mother. Weight gain was extrapolated to continue until 40 th gestation week for all mothers, although 22.2% of mothers delivered at 39 th gestation week or before. For mothers for whom both Line A (n = 2,143) and Line B (n = 2,184) were available, total pregnancy weight gain was calculated by adding the extrapolated weight gains from both periods. Thus, total pregnancy weight gain could be extrapolated only for 2,089 women. Cases and controls (66.5% vs 65.0%) did not differ concerning the number of available weight measurements. For the rest of the women, either the first weight measurement was later than 24 th gestation week, the last weight measurement was before 30 th gestation week, or information on pre-pregnancy weight, weight at the first or the last visit, or timing of the visits was not available. As indicated above, these subjects were excluded from the analyses. We compared the characteristics of the mothers who were excluded (n = 2,001) to the characteristics of the mothers in the final study population for whom total pregnancy weight gain could be determined (n = 2,089). The two groups were similar in regard to breast cancer incidence, age at menarche, height and the frequency of visits to a private doctor. However, the excluded mothers were older (mean: 27.1 years vs. 26.5 years, p < 0.001), heavier (58.7 kg vs. 57.3 kg, p < 0.001; body mass index, BMI: 22.3 kg/m 2 vs. 21.8 kg/m 2 , p < 0.001), older at first birth (25.2 years vs. 24.7 years, p = 0.016), had more children during index pregnancy (1.92 vs. 1.81 at index birth, p < 0.001), and were more often exposed to estrogen or progestin drugs (50.3% vs. 48.6%, p = 0.021). Their children were shorter (mean 49.3 cm vs. 50.3 cm, p < 0.001) and weighed less (mean 3,310 g vs. 3,472 g, p < 0.001), suggesting that excluded mothers' pregnancy weight gain might have been lower. It is probable that exclusion of these women had no major effect on the findings. The case-control study A nested case-control study was performed to determine whether later weight development confounded the association between pregnancy weight gain and breast cancer risk. A sample of women was chosen from the final cohort (n = 2,089) that included all breast cancer cases with data on pregnancy weight gain (n = 123). For each case, we chose seven randomly selected controls (n = 856) who were born in the same year as the case. These 979 women were linked to the Hospital Inpatient Registry to obtain information on the women's stays in hospitals. 117 cases were identified with a hospital visit in average 0.4 months after breast cancer diagnosis (median 0.0, range from -16.3 to 17.2), and of these cases information on body weight and height was available for 65 (53% of 123 cases). Among the controls, 699 had been a patient in a hospital at a similar age than their respective cases (maximum difference +/-5 years), and 431 of them had weight and height available in the hospital archives (50% of 856 controls, 6.6 controls/case). The breast cancer cases with no information on later body weight did not differ from the cases used for the nested case-control analysis. The controls with no information on later body weight were approximately 1.5 years older at the time of the hospital visit than the controls included to the study (p = 0.027). Statistical analysis Statistical significance of possible differences in baseline characteristics of the study population, pregnancy weight gain and postpartum weight loss and weight retention by tertiles of pregnancy weight gain was tested by using analysis of variance for continuous variables and χ 2 -test for proportions. The incidence of breast cancer per 100,000 person years was counted by groups of 5 kg pregnancy weight gain, tertiles of pregnancy weight gain and tertiles of postpartum weight retention. Person years were calculated from the delivery to the diagnosis of breast cancer or other endpoint including death, emigration or end of the study. The association between pregnancy weight gain and breast cancer risk was further examined using a Cox regression model. Total pregnancy weight gain was included as a categorical covariate (tertiles) in the model. Age at menarche, age at first birth, age at index pregnancy, BMI before pregnancy, and parity (at index birth) were all used as continuous covariates in the model. Postpartum weight retention 51 days after delivery (mean) was later added to the model. The incidence of breast cancer was counted and the Cox regression model was carried out also separately for pre- and postmenopausal breast cancers. Information on the age at menopause was not available. Therefore all women were expected to have menopause at the age of 50 years. In the case-control study, weight and BMI change between pre-pregnancy and at the time of later hospital visit were compared between the tertiles of pregnancy weight gain (analysis of variance). A Cox regression model that included later BMI was also used to analyze the data. Results In the cohort, 123 (5.9%) women developed breast cancer during the mean follow-up of 38.9 years. The mean age at diagnosis was 56.0 years (range 35–74). Background characteristics and index pregnancies are described by tertiles of pregnancy weight gain in Table 1 . Low pregnancy weight gain (<11 kg) was associated to slightly higher ages during index pregnancy, during first pregnancy and at menarche, and to lower height, higher BMI before pregnancy, higher gestation weeks at delivery, lower weight of the placenta, smaller infant and lower proportion of users of estrogen drugs compared to women with higher pregnancy weight gain (11–15 kg or >15 kg). Table 1 Background characteristics of index pregnancy, by tertiles of estimated pregnancy weight gain. Means (and SD) or percentiles are shown. Pregnancy weight gain (kg) < 11 (n = 696) 11–15 (n = 697) >15 (n = 696) p-value Background Mother's age (year) 1 27.0 (5.3) 26.2 (4.9) 2 26.2 (5.0) 2 0.006 Mother's age at first birth (years) 25.2 (4.8) 24.4 (4.4) 2 24.5 (4.4) 2 0.003 Married (%) 1 97 98 98 0.720 Visits to a private doctor (%) 1 49 45 51 0.053 Mother's height (cm) 161.2 (5.5) 162.0 (5.2) 2 162.9 (5.2) 23 <0.001 Mother's body mass index before pregnancy (kg/m 2 ) 22.3 (2.9) 21.6 (2.3) 2 21.6 (2.5) 2 <0.001 Mother's age at menarche (year) 14.2 (1.6) 13.9 (1.6) 2 13.8 (1.6) 2 <0.001 Regular menstrual cycles (%) 94 93 95 0.296 Parity (at index birth) 1.8 (1.1) 1.8 (1.0) 1.8 (1.1) 0.896 Index pregnancy Gestation weeks at delivery (week) 40.6 (2.2) 40.4 (2.1) 40.2 (2.2) 2 <0.001 Exposed to estrogens (%) 45 49 52 0.040 Weight of the placenta (g) 4 603 (112) 634 (124) 2 661(205) 23 <0.001 Infant height (cm) 4 50.0 (2.3) 50.3 (2.1) 50.6 (2.5) 23 <0.001 Infant weight (g) 3,376 (514) 3,466 (504) 2 3,577(544) 23 <0.001 Low birth weight (%) 5 3 3 0.305 1 at the time of index pregnancy 2 a statistically significant difference compared to the lowest tertile (<11 kg) 3 a statistically significant difference compared to the middle tertile (11–15 kg) 4 total n = 2,055–2,089, except for placental weight (n = 1,217) and infant height (n = 2,006) Weight development during and after pregnancy is presented by tertiles of pregnancy weight gain in Table 2 . Higher weight gain during pregnancy was associated to higher weight loss after delivery, but also to higher weight retention and BMI at the postpartum check-up visit. Table 2 Weight gain during and after pregnancy by tertiles of estimated pregnancy weight gain. Means (and 95% confidence intervals) are shown. Pregnancy weight gain (kg) <11 (n = 696) 11–15 (n = 697) >15 (n = 696) p-value Mother's weight gain (kg) Total weight gain, weeks 0–40 8.6 (8.5–8.8) 12.9 (12.9–13.0) 18.2 (18.0–18.4) <0.001 Weight after delivery 1 Weight change from 40 th week (kg) -7.4 (-7.7 – -7.2) -8.8 (-9.0 – -8.6) -10.6 (-10.9 – -10.3) <0.001 Weight compared to pre- pregnancy weight (kg) +1.3 (1.0–1.6) +4.1 (3.8–4.3) +7.6 (7.3–8.0) <0.001 BMI (kg/m 2 ) 22.6 (22.4–22.9) 23.2 (22.9–23.4) 24.4 (21.1–24.6) <0.001 Weight at the hospital visit 2 (n = 167) (n = 170) (n = 159) Change from pre-pregnancy weight (kg) + 6.4 (5.1–7.8) +10.4 (9.0–11.9) +12.5 (10.8–14.3) <0.001 BMI (kg/m 2 ) 25.0 (24.3–25.6) 25.6 (25.0–26.2) 26.3 (25.6–27.0) 0.021 Change from pre-pregnancy BMI (kg/m 2 ) + 2.4 (1.9–3.0) + 4.0 (3.4–4.6) + 4.8 (4.1–5.4) <0.001 1 on postpartum day 51, range 40–78, total n = 1,314–1,713 2 29 years after pregnancy in average, range 9–47 Breast cancer incidence per 100,000 person years The mean BMI before pregnancy was 21.8 kg/m 2 and the mean total extrapolated weight gain during pregnancy was 13.3 kg (range -5.0–33.1 kg) in our cohort. Average pregnancy weight gain (and range) was 13.1 kg (-3.0–33.1) among primiparas, 13.5 kg (1.9–30.7) among women who gave birth to their second child and 13.2 kg (-5.0–32.4) among women who gave birth to at least their third child. The incidence of breast cancer by 5 kg categories of total pregnancy weight gain is shown in Table 3 . Higher pregnancy weight gain was associated with a higher incidence of breast cancer. However, the number of women in some of the weight gain categories was small, and therefore the statistical analyses were carried out in tertiles of total pregnancy weight gain (Table 4 ). The incidence of breast cancer was significantly higher in mothers in the highest tertile of pregnancy weight gain (15–33 kg), when compared to the middle tertile (11–15 kg) (p = 0.04). Breast cancer incidence was lowest in the middle tertile, but no differences in the risk were seen in the mothers of the lowest tertile of weight gain (less than 11 kg), when compared with the other two categories. Table 3 Breast cancer incidence (per 100,000 person years, py) by estimated total weight gain (weeks 0–40). Weight gain (kg) Breast cancer cases (n) Number of women Py 171 Incidence 0 <0 0 4 0–4.99 1 42 1,677 60 5–9.99 23 423 16,614 138 10–14.99 53 954 37,266 142 1 15–19.99 33 508 19,498 169 ≥20 13 158 6,086 213 total 123 2,089 81,312 151 1 Breast cancer incidence was exceptionally high (200 per 100,000) in mothers who gained 10–10.99 kg during pregnancy. Table 4 Incidence (per 100,000 person years) and unadjusted and adjusted rate ratios (RR) 1 and confidence intervals (CI) on the Cox model for breast cancer by tertiles of estimated total weight gain (kg) in pregnancy (weeks 0–40), and by tertiles of postpartum weight retention. Breast cancer cases Number of women Incidence p-value Unadjusted RR (95% CI) Adjusted RR (95% CI) 1 Pregnancy weight gain (kg) 0.09 <11 39 696 143 1.18 (0.74–1.88) 1.11 (0.68–1.83) 11–15 33 697 121 1.00 (ref.) 1.00 (ref.) > 15 51 696 190 1.59 (1.03–2.47) 1.62 (1.03–2.53) Postpartum weight retention (kg) 2 0.33 <3 26 558 121 1.00 (ref.) 1.00 (ref.) 3–5 35 539 170 1.29 (0.72–2.34) 1.36 (0.73–2.54) >5 35 545 170 1.54 (0.87–2.74) 1.56 (0.85–2.86) Pregnancy weight gain (kg), case-control study 3 <11 19 167 - - 1.01 (0.54–1.91) 0.95 (0.49–1.84) 11–15 19 170 - - 1.00 (ref.) 1.00 (ref.) > 15 27 159 - - 1.50 (0.83–2.69) 1.48 (0.81–2.69) 1 Adjusted for age at menarche, age at first birth, age at index pregnancy, parity (at index birth) and body mass index (BMI) before pregnancy. For the RR for postpartum weight retention, adjusted also for pregnancy weight gain (weeks 0–40). 2 Weight on postpartum day 51, range 40–78, compared to pre-pregnancy weight. 3 Adjusted for age at menarche, age at first birth, age at index pregnancy, parity (at index birth), BMI before pregnancy, and BMI during the hospital visit in average 29 years after pregnancy. All analyses were initially carried out separately for pre- and postmenopausal breast cancers. The results for postmenopausal women were similar to the results for the whole cohort (results not shown). The incidence of premenopausal breast cancer was too low for statistical analysis and pre- and postmenopausal breast cancers were not separated further in the analyses. When these analyses were restricted to mothers who delivered after 39 th gestation week, results were similar than in the whole cohort (results not shown). The incidence of breast cancer was calculated separately for early (0–15 th gestation weeks) and later pregnancy weight gain (15–40 th gestation weeks). Early pregnancy weight gain was not associated with breast cancer risk. The impact of later pregnancy weight gain was similar to the impact of total weight gain, but more modest (results not shown). Multivariate analysis Both unadjusted and multivariate adjusted rate ratios and confidence intervals for the risk of breast cancer are presented in Table 4 . In the Cox regression model, mothers in the highest tertile of pregnancy weight gain (>15 kg) had a 1.62-fold higher risk for breast cancer compared to mothers in the middle tertile (average weight gain 12.9 kg), when age at menarche, age at first birth, age at index pregnancy, BMI before pregnancy and parity at index birth were included in the model. To assess the sensitivity of these analyses, the lowest and highest weight gain groups were used as reference groups. When the lowest weight gain group was the reference group, no differences among the groups were seen. However, when the highest weight gain group was the reference group, women with average weight gain had significantly lower risk of breast cancer (multivariate adjusted RR 0.62, 95% CI 0.40–0.97). When the middle tertile of weight gain was again used as the reference group and the analysis was restricted to mothers who delivered after 39th week of gestation, the results were essentially similar although statistically not significant (data not shown). The results did not either change when adjusted additionally for the year of index birth. The increased breast cancer risk in the highest tertile of pregnancy weight gain was found only for postmenopausal breast cancer (relative risk, RR = 1.80, 95% confidence interval, CI 1.05–3.07, p = 0.03). The RR for premenopausal cancer was 1.00 (95% CI 0.40– 2.48, p = 0.99). However, the number of premenopausal breast cancer cases with the information on all variables in the model was too low (n = 25) to yield sufficient power. No statistically significant differences in breast cancer risk were observed between the tertiles of postpartum weight retention, determined approximately 51 days after delivery (Table 4 ). Other results Later age at menarche was marginally related to a decreased risk of breast cancer (adjusted RR = 0.99, 95% CI 0.97–1.00). Mother's age at the time of first pregnancy or at the index pregnancy, parity at index birth or BMI before pregnancy were not statistically significantly associated with the risk of breast cancer. The results were similar when height and weight were used as separate variables in the model, instead of BMI. Lower pre-pregnancy BMI was associated with higher weight gain during pregnancy (p < 0.001) and higher postpartum weight retention (p = 0.003), but not with postpartum weight loss. The differences in the incidence of breast cancer were not statistically significant between the pre-pregnancy BMI-categories. The case-control study Women who gained at least 15 kg weight during pregnancy had a higher BMI at the time of later hospital visit (29 years after pregnancy in average) than women who gained <11 kg weight during pregnancy (p = 0.021) (Table 2 .). Changes in body weight (p < 0.001) and BMI (p < 0.001) were also higher in women who gained 11–15 kg or >15 kg compared to women who gained <11 kg during pregnancy. These findings are in agreement with earlier reports showing a link between excessive pregnancy weight gain and becoming overweight/obese later on [ 26 , 27 ]. The time window between pregnancy and assessment of BMI during later hospital visit was similar among the tertiles of pregnancy weight gain (29.2 vs. 30.0 vs. 30.1 years, p = 0.397). In the Cox regression model, women's later BMI at the time of diagnoses was not associated with breast cancer risk (adjusted RR = 0.96, 95% CI 0.90–1.04). Further, results relating to pregnancy weight gain and breast cancer risk were not altered by adding data on later BMI to the model (Table 4 ). It is to be noted that since fewer women were included to this analysis, the effect of pregnancy weight gain did not reach statistical significance. Discussion The results obtained in our study indicate that higher than recommended pregnancy weight gain increased mothers' risk of developing breast cancer. Thus, women who gained more than 15 kg during pregnancy had a 62% increase in breast cancer risk, compared to those who gained between 11–15 kg. The Institute of Medicine (IOM) published their most recent recommendations for pregnancy weight gain in 1990 [ 28 ]. The recommended pregnancy weight gain in the USA is 11.5–16 kg for women with normal pre-pregnancy BMI; i.e., they are not obese or underweight. Pregnancy weight gain recommendations are lower (7–11.5 kg) for overweight women and higher (12.5–18 kg) for underweight women. As seen in Table 3 , the incidence of breast cancer in our study was highest among women who gained more than 20 kg during pregnancy, suggesting that the increase in risk may apply primarily to women at the most extreme range of pregnancy weight gain. An increase in breast cancer risk was seen mostly in women who were diagnosed with this disease after age 50 and thus were postmenopausal. However, the number of premenopausal breast cancers was low in the cohort, and we cannot exclude the possibility that pregnancy weight gain may also increase the risk of premenopausal breast cancer. Data generated in epidemiological studies rarely provide causal relationships. We propose four different mechanisms that may link high pregnancy weight gain to a later increase in breast cancer risk. First, weight retention in women who gained excessive amounts of weight during pregnancy may have persisted into their postmenopausal years. Women prone to postpartum weight retention might also be prone to long-lasting weight gain after pregnancy [ 29 ], and high BMI during postmenopausal years increases breast cancer risk [ 26 ]. To examine this possibility, information on body weight at the time of breast cancer diagnosis was obtained. If the association between pregnancy weight gain and breast cancer risk was affected by later weight development, breast cancer cases should have had higher BMI at the time of diagnosis. As this was not the case, we propose that high pregnancy weight gain increases breast cancer risk independently from body weight at the time of diagnosis. Another alternative is that women who gained an excessive amount of weight during pregnancy may have had higher pregnancy hormone and growth factor levels than women who gained within the recommended range, stimulating the growth of existing malignant cells in the breast, leading to development of a detectable tumor. Several studies have shown that markers of high pregnancy estrogen levels increase mother's breast cancer risk [ 4 - 6 , 9 - 12 , 16 ]. Estrogen levels may correlate with high pregnancy weight gain [ 19 ], but two recent studies have not confirmed this observation [ 20 , 21 ]. Other possible hormones that could be mediating the effect of pregnancy weight gain on breast cancer risk include leptin. Leptin levels correlate strongly with BMI [ 27 ], also during pregnancy [ 30 ], and leptin is suggested to increase breast cancer risk [ 31 ]. We did not have any biological samples available for hormone measurements. It is known that high hormone levels increase the proliferation of normal breast cells that then is accompanied by increased genomic instability and accumulation of DNA adducts [ 16 , 32 ]. Thus, the third explanation is that high pregnancy weight gain increased the likelihood of DNA damage and mutations in genes that initiate breast cancer. Since the window between index pregnancy and diagnosis of breast cancer was approximately 30 years, there was enough time for the initiation to have taken place during pregnancy. Finally, known and unknown causes of breast cancer may have confounded the results. For example, these causative factors might be more common in women who gain excessive amounts of weight during pregnancy or they caused women to gain excessive amounts of weight during pregnancy. A theoretical example is a gene mutation/polymorphism that could both make a woman more prone to gain weight during pregnancy and increases breast cancer risk. Methodological limitations have to be considered when interpreting the results, and they include high rate of exclusion and an exposure to estrogenic drugs during pregnancy. Of the 4,090 women available for the study, 48.9 % were excluded for reasons listed in Figure 1 (109 of which were diagnosed with breast cancer). Total pregnancy weight gain could be extrapolated only for 2,089 women, of which 123 had developed breast cancer. Other information on background and index pregnancies indicated that the excluded women might have gained less weight during pregnancy than the final study population (see chapter Inclusions and exclusions). However, we found no evidence that breast cancer incidence was different between the women excluded and included to the study. Some women in our cohort had been exposed to synthetic estrogens during pregnancy to avoid a threatening miscarriage, and this exposure might have affected the results. However, it was the initial reason for obtaining information from pregnant women, and we found no effect of the drug exposure on the incidence of breast cancer. We are not aware of any other cohort that could be used to assess the link between pregnancy weight gain and breast cancer risk, but if such a cohort becomes available, and it is not potentially compromised by high rate of exclusion of subjects or an exposure to drugs during pregnancy, the present results can be either confirmed or nullified. Follow-up of "old" cohorts similar to ours is rarely possible, making our study relatively unique. Another area of potential source for errors is the variability in time period between the weight measurements during pregnancy (range 3–295 days), requiring us to extrapolate the pregnancy weight gain for each woman. This step also has been successfully used in other studies [ 33 ]. A further weakness of the study was that no information on weight gain in previous and subsequent pregnancies was available. Therefore, we cannot exclude the possibility that a woman who did not develop breast cancer and during the index pregnancy gained less than 15 kg, might have had subsequent pregnancies that were characterized by excessive weight gain. Finally, in the case-control study that determined the impact of body weight at the time of diagnosis on breast cancer risk, information on this weight was obtained only for 53% of the cases and 50% of the controls. However, the direction of bias rising from exclusion may have diluted the effect, rather than caused it. In conclusion, our findings suggest that excessive pregnancy weight gain increased later risk of developing breast cancer. This association needs to be further confirmed in prospective studies. List of abbreviations diethylstilbestrol – DES; body mass index – BMI Competing interests The author(s) declare that they have no competing interests. Authors' contributions TK : PhD student who collected and put together all the material for the study, managed the data and did statistical analysis, and participated in writing the manuscript. MG : Participated in statistical analysis of the data and writing the manuscript. EH : Collected the original data base of pregnant women and was in charge of linking the data base to cancer registry, participated in statistical analysis planning and writing the manuscript. RL : Generated the idea of testing the hypothesis in the Hemminki data base, and participated in all stages of the study and in writing the manuscript. LH-C : Generated the hypothesis, obtained funding for the study, and was in charge of writing the manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional File 1 Calculation of line a and line b for each mother. Click here for file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC535935.xml |
544882 | Bioinformatic mapping of AlkB homology domains in viruses | Background AlkB-like proteins are members of the 2-oxoglutarate- and Fe(II)-dependent oxygenase superfamily. In Escherichia coli the protein protects RNA and DNA against damage from methylating agents. 1-methyladenine and 3-methylcytosine are repaired by oxidative demethylation and direct reversal of the methylated base back to its unmethylated form. Genes for AlkB homologues are widespread in nature, and Eukaryotes often have several genes coding for AlkB-like proteins. Similar domains have also been observed in certain plant viruses. The function of the viral domain is unknown, but it has been suggested that it may be involved in protecting the virus against the post-transcriptional gene silencing (PTGS) system found in plants. We wanted to do a phylogenomic mapping of viral AlkB-like domains as a basis for analysing functional aspects of these domains, because this could have some relevance for understanding possible alternative roles of AlkB homologues e.g. in Eukaryotes. Results Profile-based searches of protein sequence libraries showed that AlkB-like domains are found in at least 22 different single-stranded RNA positive-strand plant viruses, but mainly in a subgroup of the Flexiviridae family. Sequence analysis indicated that the AlkB domains probably are functionally conserved, and that they most likely have been integrated relatively recently into several viral genomes at geographically distinct locations. This pattern seems to be more consistent with increased environmental pressure, e.g. from methylating pesticides, than with interaction with the PTGS system. Conclusions The AlkB domain found in viral genomes is most likely a conventional DNA/RNA repair domain that protects the viral RNA genome against methylating compounds from the environment. | Background The purpose of this study has been to identify domains with homology to AlkB in viral genomes, in order to get a better understanding of distribution and possible function of such domains. The AlkB protein of E. coli , and probably most of its homologues, is involved in repair of alkylation damage in DNA and RNA. It repairs 1-methyladenine and 3-methylcytosine by oxidative demethylation and direct reversal of the methylated base back to its unmethylated form. Recently the protein was identified as a member of the 2-oxoglutarate (2OG)- and Fe(II)-dependent oxygenase superfamily [ 1 - 3 ]. The catalytic reaction requires molecular oxygen, Fe 2+ and 2-oxoglutarate, which is subsequently converted into succinate, CO 2 and formaldehyde [ 4 ]. The 2OG-FeII oxygenase superfamily is widespread in Eukaryotes and bacteria [ 1 ], and is currently the largest known family of oxidising enzymes without a heme group [ 5 ]. The 3D structure of several of these oxygenases is known, and they share a common fold with a structurally conserved jelly roll β-sheet core with flanking α-helices. Very few residues are totally conserved across these structures, basically just the residues involved in coordination of the Fe(II) ion and the 2-oxoglutarate. AlkB-like genes are widespread in most types of organisms except Archaea. However, whereas bacteria normally have just one or at most two AlkB homologues [ 6 ], multicellular Eukaryotes tend to have several homologues. In the human genome at least 8 different AlkB homologues (ABHs) have been identified [ 7 ]. These homologues seem to have slightly different properties with respect to substrate preference and subcellular localisation, and this may be a reason for the proliferation of ABHs e.g. in humans. However, a detailed functional mapping of all ABHs has not yet been carried out. A sequence alignment of known ABHs identifies very few residues as totally conserved, basically just a HxD motif, a H and a RxxxxxR motif. These residues are also conserved in the more general 2OG-FeII oxygenase superfamily as described above, except for the final R. The first three residues (HxD and H) are involved in Fe(II)-coordination, whereas the first R is involved in 2OG-coordination. The final R is most likely involved in AlkB-specific substrate binding. In addition to DNA repair, it has been shown that E. coli AlkB and the human AlkB homologue hABH3 may be involved in RNA repair. When expressed in E. coli both AlkB and hABH3 reactivate methylated RNA bacteriophage MS2 in vivo . This illustrates that direct repair may be an important mechanism for maintenance of RNA in living cells [ 4 ]. RNA repair proceeds by the same mechanism as DNA repair. Repair of damaged RNA was previously considered very unlikely, due to the natural redundancy of RNAs in a cell [ 8 ]. However, RNA is essential for cell function: unrepaired RNA can lead to miscoded or truncated proteins, and alkylated RNA could signal cell cycle checkpointing or apoptosis [ 9 ]. Consequently the occurrence of RNA repair does not come as a great surprise. The mechanism of direct reversal of methylation as used by AlkB homologues is particularly important for RNA repair, as it means that single-stranded regions may be repaired without introducing strand breaks. Repair of alkylation damage in DNA and RNA has recently been reviewed [ 10 ]. AlkB homologues have also been found in plant viruses. It has been suggested that methylation may be used in host-mediated inactivation of viral RNAs, and that AlkB homologues in some plant viruses may be used to counteract such defence mechanisms [ 1 ]. However, no detailed study of this has been published. The research project reported here has focused on a better understanding of the distribution and potential function of putative AlkB homology domains by using in silico mapping of viruses in which such domains have been found, as well as related viruses. Results The general mapping strategy of the project was to identify viral genomes with AlkB homology domains, identify common features of these genomes, and subsequently find additional genomes with similar features, but without AlkB homology domains. This data set could then be used to analyse the properties and distribution of AlkB-like domains in viruses, as a basis for generating hypotheses about the evolution and function of these domains. Identification of relevant viral protein sequences The PSI-Blast search for viruses in the NCBI nr protein sequence database was initiated with ALKB_ECOLI (NCBI gi113638), restricted to residues 110 to 210 and using the default inclusion threshold of 0.005 on E-values. The chosen residue range corresponds to the most conserved region in AlkB homologues [ 10 ]. The PSI-Blast search converged after 4 iterations, and included 43 hits below the 0.005 inclusion threshold, from 22 different ssRNA positive-strand viruses. The AlkB homologues were found in viruses belonging to Allexi , Ampelo , Carla , Fovea , Mandari , Potex , Tricho and Vitiviruses , all of which are known to infect plants [ 11 ]. In all of these viruses the AlkB domain is a part of the replicase polyprotein, which normally consists of a viral methyltransferase domain (MT), a viral helicase domain (HEL) and a RNA-dependent RNA polymerase domain (RdRp). Therefore separate PSI-Blast searches for the individual components of the replicase polyprotein were also initiated. All searches were done with PSI-Blast using the default inclusion threshold (E-value of 0.005). The searches for MT and HEL domains were initiated using residue ranges 449–841 and 1938–2178 respectively from Grapevine leafroll-associated virus 3 ( Ampelovirus , NCBI gi29650261). The search for RdRp was initiated with residue range 1361–1798 from Soil-borne cereal mosaic virus ( Furovirus , NCBI gi11546056). These sequences were chosen based on the output from the previous AlkB search. This gave a library of protein sequences with either AlkB, MT, HEL or RdRp domains, the general composition of which is illustrated in Figure 1 . From this library a subset was generated, consisting of all sequences containing MT, HEL and RdRp domains. This included processed (cleaved) polyprotein sequences where RdRp was found as a separate subsequence. However, whenever possible, the protein sequence corresponding to the genomic sequence was used. The final library, described in Table 1 and in Additional file 1 , consisted of 146 sequences from a large number of different viruses. Figure 1 PSI-Blast search results shown as a Venn diagram. Initial searches using methyltransferase, helicase and RdRp domains retrieved 163, 175 and 237 sequences, respectively. A total of 146 sequences contained all three domains, and 22 of these also contained an AlkB domain. Table 1 Summary of Pfam domains Classification Pfam domains b Host Family Genus n a AB OT PC A1 ot Plant Bromoviridae Alfamovirus 1 Bromovirus 4 4 Cucumovirus 3 Ilarvirus 11 Oleavirus 1 Unassigned 1 Closteroviridae Ampelovirus 4 2 Closterovirus 5 1 Crinivirus 4 Flexiviridae 1 Allexivirus 5 1 Mandarivirus 1 1 Potexvirus 17 3 Flexiviridae 2 Capillovirus 3 3 3 Carlavirus 6 5 5 6 Foveavirus 6 5 6 6 Trichovirus 2 2 2 Vitivirus 2 2 Unassigned 2 1 2 2 Tymoviridae Maculavirus 1 1 Marafivirus 3 3 3 Tymovirus 7 7 3 Unassigned Benyvirus 2 2 Furovirus 4 Hordeivirus 1 Idaeovirus 1 Pecluvirus 1 Pomovirus 4 Tobamovirus 18 Tobravirus 3 Unassigned 2 Invertebrate Tetraviridae Betatetravirus 1 Unassigned 1 Vertebrate Togaviridae Alphavirus 17 17 2 Unassigend Hepatitis E-like 2 2 a Number of sequences. b Number of sequences with each domain type, excluding the common MT, HEL and RdRp domains (AB – AlkB, OT – OTU, PC – Peptidase C, A1 – A1pp, ot – other). The library of protein sequences was screened for known domains in Pfam. This identified Pfam domains Viral_helicase1 and RNA_dep_RNApol2 in all sequences, corresponding to HEL and RdRp domains, respectively. In addition Vmethyltransf and 2OG-FeII_Oxy , corresponding to MT and 2OG-FeII oxygenase (AlkB) domains, were identified in several sequences. However, for sequences from Flexiviridae and Tymoviridae there was no clear identification of any MT domain by Pfam, although they had been retrieved by PSI-Blast in a search for MT domains. Therefore HMMER was used to build a Pfam type profile for these sequences. A PSI-Blast search was initiated using residues 1–500 of Potato virus M ( Carlavirus , NCBI gi9626090). Twelve representative sequences were selected from the search output, covering Carla , Fovea , Potex , Allexi , Capillo and Maculavirus . Subsequences representing the conserved region according to the PSI-Blast alignment, corresponding to residues 35–378 of the query sequence, were aligned using ClustalX, and a Pfam type profile was generated and calibrated using tools from the HMMER package. The resulting profile was able to identify putative methyltransferase domains in all Flexiviridae and Tymoviridae sequences in the data set. Other Pfam domains – Peptidase_C21 , C23 , C33 , C34 , C35 and C41 , A1pp and OTU – were also identified in subsets of sequences. A1pp is a member of the Appr-1-p processing enzyme family, and the domain is found in a number of otherwise unrelated proteins, including non-structural proteins of several types of ssRNA viruses. OTU is a member of a family of cysteine proteases that are homologous to the ovarian tumour ( otu ) gene in Drosophila . Members of this family are found in Eukaryotes, viruses and pathogenic bacteria. Phylogenetic analysis The MT, HEL and RdRp domains identified by Pfam as described above were extracted from the library sequences, aligned by ClustalX, and combined into a new alignment consisting of only these domain regions. This turned out to be necessary in order to get robust alignments. The intervening regions between the conserved domains are extremely variable in these sequences, and this tended to confuse alignment programs in the sense that conserved regions were not correctly aligned. The combined sequence alignment of domains from Closteroviridae , Flexiviridae and Tymoviridae was then used as input for building a phylogenetic tree with MEGA2. The final tree is shown in Figure 2 , with polyproteins containing AlkB-like domains indicated. Figure 2 Unrooted phylogenetic tree for Flexiviridae 1 and 2, Tymoviridae and Closteroviridae . Sequences are labelled with genus and NCBI gi accession number. Bootstrap values ≥ 80 are shown. Sequences with AlkB domains are indicated with black dots. A second alignment was generated from all sequences with AlkB-like domains, using only the regions corresponding to MT, AlkB, HEL and RdRp Pfam domains. The domains were aligned individually, and the combined alignment was used as input for MEGA2. However, this data set did not give a reliable phylogeny (data not shown), and the separate domains of this alignment were therefore analysed individually and compared. This analysis is summarised in Table 2 . For each domain a bootstrapped neighbour-joining (NJ) tree was generated with MEGA2. The average bootstrap support value over all branches was computed for each tree, and this value was clearly lower for the AlkB tree compared to the other trees. A maximum likelihood (ML) tree was generated for each domain with Tree-Puzzle. This showed the same trend, the likelihood values indicated that the AlkB tree was clearly inferior to the other trees. The individual trees were then compared using the quartet-based strict joint assertions (SJA) measure as implemented in the Component software package. Both the NJ and ML trees showed the same trend. The MT, HEL and RdRp domains gave similar tree structures, with SJA values between 0.053 and 0.161 for NJ trees and between 0.058 and 0.092 for ML trees when they were compared to each other. The AlkB domain gave a significantly different tree structure, with SJA values from 0.456 to 0.524 for NJ trees and from 0.258 to 0.317 for ML trees when compared to the MT, HEL and RdRp trees (the actual trees are given in Additional file 2 ). For comparison the SJA values for comparing the corresponding NJ and ML trees for MT, AlkB, HEL and RdRp were 0.054, 0.000, 0.040 and 0.003, respectively, showing that the NJ and ML procedures gave almost identical tree structures. Day has estimated expectation values and standard deviations for various distance measures (including SJA) for comparison of random trees [ 12 ]. The SJA values shown in Table 2 for comparisons between MT, HEL and RdRp NJ trees were 14.2 – 17.1 standard deviations from the expectation value of 0.665 for a tree with 22 nodes, whereas the corresponding values for the AlkB NJ tree were 4.4 – 5.4 standard deviations from the expectation value. Similar ranges were observed for the ML trees as well as for alternative distance measures, e.g. the Symmetric Difference (SD) measure (data not shown). Although this means that the SJA value for comparing AlkB trees to MT, HEL and RdRp trees were significantly better than for random trees, it also shows that the MT, HEL and RdRp trees were clearly more similar to each other than to the AlkB tree. Table 2 Strict joint assertions distances for NJ and ML trees ML\NJ a MT AlkB HEL RdRp log L b BS (%) c ID (%) d MT - 0.488 0.161 0.053 -14068 85 27 AlkB 0.263 - 0.524 0.456 -4016 35 38 HEL 0.058 0.317 - 0.117 -10425 87 28 RdRp 0.062 0.258 0.092 - -14543 91 37 a Strict joint assertions (SJA) values based on quartets as computed by Component for ML trees (lower left) and NJ trees (upper right). SJA is defined as resolved and different quartets divided by all resolved quartets. b The likelihood value from Tree-Puzzle. c Average bootstrap value for all branches in each NJ tree. d Average sequence identity for all pairs of sequences in each alignment. The alignment of the AlkB domain seemed to be of comparable quality to the other alignments. In fact the AlkB domain had the highest average pairwise sequence identity, as seen in Table 2 (see Figure 3 for the actual alignment). In other words, these AlkB domains were as similar to each other as the other three domains with respect to sequence identity, but they did not represent a consistent evolutionary history when compared to the other domains of this polyprotein. This may indicate that the AlkB domains have evolved separately from the other domains, and possibly as several independent instances. Figure 3 Multiple alignment of sequence regions corresponding to the AlkB domains. The alignment was generated with ClustalX. The residues involved in coordination of the essential Fe 2+ ion are completely conserved, except in one of the Vitivirus sequences. These residues are the HxD motif, a single H, and the first R in the RxxxxxR motif. The function of the remaining conserved residues is unclear, but at least some of them may be involved in coordination of the substrate [10]. The degree of co-evolution was analysed by computing pairwise distances between sequence regions in the alignment of MT, AlkB, HEL and RdRp domains described above. In Figure 4 selected results are shown as scatter plots, where the Blosum 50 score value between e.g. the MT domains in a pair of sequences is plotted against the score value for AlkB domains in the same pair of sequences. Plots for the MT, HEL and RdRp domains show that they are strongly correlated for MT vs. RdRp (r 2 = 0.95), MT vs. HEL (r 2 = 0.87) and HEL vs. RdRp (r 2 = 0.81). The plot of the AlkB domain vs. these three domains for the same set of sequences shows a very low degree of correlation for AlkB vs. RdRp (r 2 = 0.10), AlkB vs. MT (r 2 = 0.12) and AlkB vs. HEL (r 2 = 0.16). Figure 4 Pairwise distances between sequence regions corresponding to methyltransferase (MT), RdRp and AlkB domains. Each data point corresponds to e.g. RP-RP and MT-MT distances for the same pair of sequences, and sequences showing similar evolutionary distance in these two regions will fall on the diagonal. The pairwise distances were estimated from multiple alignments using the Blosum50 score matrix [47]. Trend lines were estimated with Excel. The trend line for AlkB vs. RdRp is heavily influenced by the point at (675, 670). It represents two Foveavirus sequences (NCBI gi3702789 and gi9630738), they are 98% identical over the full polyprotein sequence. As mentioned above the genome organisation of these replicase polyprotein sequences seems to be very flexible. In order to analyse domain organisation the location of identified Pfam domains were plotted for a number of sequences, as shown in Figure 5 . Figure 5 Location of Pfam domains in the variable region of Flexiviridae 2 sequences. The regions have been extracted directly from Pfam output, and sequences and regions are drawn to scale. The black bar at each end of a motif indicates that a full-length motif has been found, for partial motifs the bar at the truncated end would be missing. Similarity of viral AlkB domains to other AlkB sequences The results described above may indicate that the AlkB domains have been integrated into the replicase polyprotein relatively recently (see Discussion). In order to test for potential sources selected AlkB domains were compared to non-viral sequences. PSI-Blast was used to search the NCBI nr database, removing all viral hits in the final search report. Most of the remaining top-scoring hits were from bacteria. This included two different strains of Xanthomonas, X. axonopodis pv citri and X. campestris pv campestris . Xanthomonas attacks plants such as citrus, beans, grapevine, rice and cotton [ 13 ]. The search also returned high-scoring hits from another plant pathogen, Xylella fastidiosa . This bacterium infects a great variety of plants, including grapevine, citrus, periwinkle, almond, oleander and coffee [ 14 ]. Potential similarities in variable regions Pfam searches obviously will only identify known domain types in protein sequences. In order to identify potential similarities in regions that were not recognised by Pfam, systematic PSI-Blast searches were performed, using the polyprotein regions between the MT and HEL domains and searching against the NCBI database of reference sequences [ 15 ], excluding all viral entries. A maximum of 5 PSI-Blast iterations were allowed, with an inclusion threshold of 0.005. The expected homologues of the AlkB-domain were found with high confidence, as most of the E-values were < 1 × 10 -50 . Homologues of typical viral domains like the viral peptidases were obviously not found, as all viral database entries were excluded. Very few new similarities were found by these searches. Pepper ringspot virus ( Tobravirus , NCBI gi20178599) showed significant similarity to site-specific DNA-methyltransferase from Nostoc sp (E = 1 × 10 -74 ), as well as other cytosine 5C-specific DNA methylases. Bamboo mosaic virus ( Potexvirus , NCBI gi9627984) showed similarity to aggregation substance Asa1 from Enterococcus faecalis (E = 6 × 10 -34 ). A small number of additional similarities seemed to be caused by biased sequence properties (e.g. proline-rich regions), and were probably not significant. This included matches against mucin and cadherin-like proteins from Homo sapiens and multidomain presynaptic cytomatrix protein (piccolo) from Rattus norvegicus . In general the variable regions seemed to be truly variable, with very little similarity to other proteins, except for the Pfam domains already identified. Loss of domains in related polyprotein sequences As seen in Figures 2 and 5 , some closely related sequences are lacking specific domains in the sense that HMMER does not find a significant similarity to the Pfam entries for these domains. In order to understand the degree of sequence variation associated with this domain loss, as well as the general sequence variation in conserved vs. non-conserved regions of typical polyproteins, several dot plots were generated. The dot plot for two Carlavirus sequences, Potato virus M (NCBI gi9626090) and Aconitum latent virus (NCBI gi14251191), is shown in Figure 6 . The dot plot confirms that these two sequences are closely related in the MT, HEL and RdRp domains. However, there are significant differences in the region between MT and HEL. Potato virus M is lacking the AlkB domain whereas Aconitum latent virus is lacking the OTU domain. As seen from the dot plot, short regions of similarity close to the diagonal shows that both domains may have been present in an ancestral sequence. However, this region shows a high degree of sequence variation, and as indicated by the dot plot they are almost exclusively mutations. Non-essential or non-functional domains are probably rapidly lost. In this particular case, none of the typical AlkB motifs seem to be conserved in Potato virus M , indicating that this indeed is a non-functional AlkB domain. Figure 6 Dot plots for Potato virus M (NCBI gi9626090) and Aconitum latent virus (NCBI gi14251191). To the left the full sequences are shown, using the program default for similarity threshold, and to the right the region with AlkB, OTU and peptidase integration, using a slightly lower (more sensitive) threshold for sequence similarity. The Pfam regions corresponding to MT (magenta), AlkB (red), OTU (green), peptidase (blue), HEL (yellow) and RdRp (cyan) domains are indicated. Discussion The N-terminal domains of Flexiviridae and Tymoviridae are methyltransferases As described above the Pfam methyltransferase motif ( Vmethyltransf ) did not match any of the putative methyltransferase domains of Flexiviridae and Tymoviridae , despite the fact that they had been identified via PSI-Blast searches starting with known methyltransferases. Therefore an additional Pfam-type profile was generated. It is obviously a possibility that these domains in Flexiviridae and Tymoviridae are not methyltransferases, and that they are false positives from PSI-Blast. However, the essential residues of a typical viral methyltransferase motif are conserved in the alignment of these domains (data not shown) [ 16 ]. In Bamboo mosaic virus , which belongs to Flexiviridae , the residues H68, D122, R125 and Y213 have been identified as putative active site residues with similarity to the Sindbis virus-like methyltransferase [ 17 ], and it has been demonstrated that this region of the Bamboo mosaic virus has methyltransferase activity, as it catalyses the transfer of a methyl group from S-adenosylmethionine (AdoMet) to GTP or guanylylimidodiphosphate (GIDP). The corresponding sequence positions are almost completely conserved in the alignment of Flexiviridae and Tymoviridae N-terminal domains. This is most likely significant, as only 7 positions in total are completely conserved in this alignment, which means that the majority of the conserved positions are known to be essential for methyltransferase activity. Work e.g. by Hataya et al . seems to support the assumption that this sequence region is a methyltransferase domain [ 18 ]. It therefore seems likely that all the sequences with AlkB domains also contain functional MT, HEL and RdRp domains. The MT domains are probably involved in capping of genomic and subgenomic RNA [ 19 ]. The viral AlkB domains are most likely functional Based on the bioinformatic evidence generated here, it seems reasonable to assume that the viral AlkB domains identified by Pfam are functional. All the essential residues found in 2-oxoglutarate- and Fe(II)-dependent oxygenases are conserved, in particular the putative Fe 2+ coordinating H, D and H residues at alignment positions 19, 21 and 91 of Figure 3 , and the 2-oxoglutarate coordinating R at position 100. The conserved R at position 106 is also very characteristic of AlkB homologues [ 10 ]. The fact that all AlkB-like domains identified in these viral genomes are full-length, compared to the Pfam profile, also seems to support the hypothesis that these domains are functional. The AlkB domains are found in a subset of viral genomes The Pfam searches show that AlkB domains are found only in a subset of the viral genomes. This subset is phylogenetically consistent (see Figure 2 ), as it is mainly restricted to the Flexiviridae , and in particular to a subset of the Flexiviridae consisting of Viti , Capillo , Tricho , Fovea and Carlavirus . This subset is well separated from the remaining Flexiviridae in the phylogenetic analysis. The split seems to be robust from bootstrap analysis, therefore this family will be discussed here as two subfamilies, Flexiviridae 1 and 2. The same split was observed by Adams et al . in their recent analysis of the Flexiviridae family [ 20 ]. Most of the AlkB domains (15) are found in Flexiviridae 2. The remaining AlkB domains are found in Flexiviridae 1 (5) and Closteroviridae (2). In general, all the Flexiviridae 2 sequences have at least one extra domain in addition to MT, HEL and RdRp: either AlkB, OTU-like cysteine protease or a peptidase. Most other plant viruses that are included in this survey do not have additional domains, except for Tymoviridae where a peptidase domain seems to be common. For the remaining plant virus families included here (excluding Tymoviridae and Flexiviridae 2), only 14% seem to have additional domains. Introduction of AlkB domain in plant virus is probably a recent event The observed distribution of AlkB domains could most easily be explained by assuming that an ancestral AlkB domain was integrated into the genome of the last common ancestor of the Flexiviridae 2 subfamily. Subsequent virus generations derived from this common ancestor would then also contain an AlkB domain, except in those cases where the domain was lost again. This scenario could also include subsequent transfer to a small number of other virus families e.g. by recombination. If this scenario was correct, then one would expect the different domains of the polyprotein to have a similar evolutionary history. From the phylogenetic analysis (Table 2 ) this seems to be confirmed for the MT, HEL and RdRp domains, but not for the AlkB domain. This indicates that the AlkB domain may not have co-evolved with the other domains, at least until relatively recently. This seems to be confirmed by looking at the degree of co-evolution, which was analysed by computing pairwise distances between alignment regions representing the relevant domains (Figure 4 ). In the case of perfect co-evolution all points should fall on a diagonal. This seems to be the case for the MT, HEL and RdRp domains. However, the plot of the AlkB domain vs. these three domains for the same set of sequences does not show a similar correlation. Only some of the closely related sequence pairs in the upper right quadrant of the plot in Figure 4 show some degree of correlation for AlkB vs. RdRp. The most likely explanation seems to be that most of the AlkB domains have not co-evolved with the other domains for any significant period of time. This seems to rule out the possibility of ancient integration of the AlkB domain, except if we assume that an ancient viral AlkB domain has frequently recombined with other AlkB domains. However, it is difficult to distinguish a scenario with frequent recombination of AlkB domains from de novo integration, and the net effect on the properties observed here would be the same. As seen in Figure 4 , the range of score values is generally smaller for the AlkB domains than e.g. the RdRp domains, particularly if we exclude a couple of very high-scoring cases (see figure caption). On the other hand, the degree of sequence variation within the collection of AlkB domains is significant, average sequence identity for pairwise alignments is 38%, and only 10% of the positions are totally conserved. This can be consistent with a recent integration if we assume that several different AlkB-type vectors have been used for integration (see below for details). An increased mutation rate after integration could also have contributed to sequence diversity in this region. Moving the AlkB domain into a novel structural and functional context would have removed many of the original evolutionarily constraints, as well as introduced some new ones. This could have created a "punctuated equilibrium" type of situation, potentially leading to a very rapid evolution that could have introduced significant differences between the AlkB domains, independent of the evolution in the other domains. A high mutation rate seems to be the case for this region in general, as indicated in Figure 6 . Although the MT, HEL and RdRp domains seem to be well conserved from the dot plot, there are very large sequence variations in the intervening region. One sequence in Figure 6 has a well conserved AlkB domain, the other an OTU domain. The fact that there are very weak sequence similarities in these two domains in the dot plot indicates that both sequences originally had both domains. However, the fact that this similarity now is very weak and without any of the typical AlkB active site motifs also indicates a high mutation rate where non-essential domains are rapidly lost. Therefore the conservation of AlkB domains is a strong indication that they are functional, as already mentioned. The AlkB domains may represent several separate integrations If we assume that AlkB domains have been integrated relatively recently, then either de novo integration or recombination (horizontal gene transfer) may have been the main driving force for spreading the AlkB domain to new genomes. In the first case a large number of individual integrations could have lead to the present situation. If horizontal gene transfer was the main driving force, the initial number of integrations might have been quite small. It is not easy to differentiate between these two situations. The map of Pfam motifs in the variable region between the MT and HEL domains in Flexiviridae 2 polyproteins (Figure 5 ) shows that they have a very similar domain organisation, basically an AlkB domain followed by an OTU domain and a peptidase domain, located towards the C-terminal part of the sub-sequence. The relatively constant domain organisation seems to be consistent with a small number of initial integrations that were subsequently diffused to related genomes e.g. by homologous recombination. However, this is not fully consistent with the fact that the viruses with AlkB domains have been collected from hosts at very different locations, e.g. Canada, USA, Russia, Italy, Germany, France, India, Taiwan, China and Japan. Although import of virus-infected species or transmission by insects may transport viruses over significant distances, it is not obvious that this is enough to explain the observed distribution of AlkB-like domains. Therefore several independent integrations, mainly from closely related hosts, have to be considered as an alternative explanation. This explanation seems to be supported by the apparent lack of any consistent evolutionary relationships between the various AlkB domains, as seen in Table 2 . It is not easy to see how this model can be consistent with the observed similarities in domain organisation in Flexiviridae . Assuming that this region has a high degree of variability, one would expect the variability to affect localisation of integrated domains as well. However, it is possible that conserved regions e.g. in the polyprotein play a significant role in integration of novel domains. It may be relevant in this context that preliminary simulations indicate that e.g. the AlkB domains tend to form independent folding domains in the folded RNA structure of the polyprotein RNA (F. Drabløs, unpublished data). This property may possibly facilitate the insertion of such domains into the viral genome. The original AlkB integration may be of bacterial origin There are many groups of organisms that can act as vectors and spread viruses, including bacteria, fungi, nematodes, arthropods and arachnids. The plant viruses may have acquired the AlkB domain either from the vector or from the host itself. As already mentioned, searching with viral AlkB domains in protein sequence databases resulted mainly in bacterial sequences, including the plant pathogens X. fastidiosa and campestris . It is therefore a reasonable possibility that AlkB domains in plant viruses have originated from bacterial mRNA. It is also possible that the mRNA originated from other vectors or from the host itself, but at the present time this is not easily verified or disproved because of the limited number of insect and plant genomes that have been sequenced. The AlkB domain probably protects virus RNA against methylation It has previously been suggested that the viral AlkB domain may be involved in protecting the virus against the post-transcriptional gene silencing (PTGS) system of the host [ 1 ]. PTGS is known as one of a plant's intrinsic defence mechanisms against viruses [ 21 ]. Gene silencing can occur either through repression of transcription (transcriptional gene silencing – TGS) or through mRNA degradation, PTGS. The PTGS-mechanism in plants shows similarities to RNA interference (RNAi) in animals [ 22 ]. This mechanism results in the specific degradation of RNA. Degradation can be activated by introduction of transgenes, RNA viruses or DNA sequences homologous to expressed genes [ 23 ]. Many viruses have developed mechanisms to counteract PTGS in order to successfully infect plants [ 24 ]. Two of these suppressors of PTGS have been identified as Hc-Protease and the 2b protein of Cucumber mosaic virus [ 25 ]. Although both proteins suppress PTGS, it is likely that they do so via different mechanisms. Could the AlkB-like domain found in some of the plant viruses also be a suppressor of PTGS? Previously reported research indicates that methylation of transcribed sequences is somehow connected with PTGS, and the methylation can be mediated by a direct RNA-DNA interaction [ 26 ]. This RNA-directed DNA methylation has been described in plants, and leads to de novo methylation of nearly all cytosine residues within the region of sequence identity between RNA and DNA [ 27 ]. Both RNA methylation and methylation of host proteins that are essential for viral replication would be detrimental to the virus. It has already been mentioned that AlkB repairs 1-methyladenine and 3-methylcytosine by oxidative demethylation. It is therefore possible that AlkB demethylates the nucleotides methylated by the PTGS mechanism, helping the virus to overcome one of the major defence mechanisms of the plant. As shown here, only a subset of plant viruses have the AlkB domain. However, other viruses may be utilising naturally occurring AlkB proteins in the host. Viruses have to rely on a number of host proteins in order to replicate [ 28 ]. In some cases it is probably beneficial for the virus to integrate such genes into their own genome in order to ensure that they are accessible, although there will be a trade off between this advantage and the increased cost of maintaining a larger genome [ 29 ]. However, there is an alternative hypothesis with respect to the AlkB integration that also has to be considered. As discussed above, the AlkB domain seems to have been integrated relatively recently in viruses found at very different geographical locations, and the only obvious connection seems to be that most viruses belong to a subset of the Flexiviridae . However, the source of these viruses points at another common feature. As seen from the table given in Additional file 1 , AlkB domains are often found in viruses associated with grapevine, apple, cherry, citrus and blueberry – crops where the usage of pesticides is common. It is known that several common pesticides (e.g. methyl bromide and some organophosphorus compounds) may cause methylation of DNA and RNA [ 30 - 33 ]. An integrated repair domain for methylation damage as part of the viral replication complex would therefore give the virus a competitive advantage in a highly methylating environment. The application of such pesticides would probably also stimulate AlkB production e.g. in co-infecting bacteria, giving these viruses easy access to AlkB mRNA for integration into their RNA genome. It could be argued that a more active PTGS system in these plants would give a similar effect. However, in that case we would expect to see more ancient integrations of AlkB domains. It could also be argued that the presence of AlkB domains may be an artefact caused by promiscuous viral domains picking up available mRNA sequences during cultivation of viruses in the laboratory. However, given the large number of different laboratories involved, and the number of different hosts used (data not shown), this seems to be a very unlikely explanation. The hypothesis that environmental compounds, in particular pesticides, may have provoked the integration of AlkB domains into the viral genomes depends upon a high mutation rate and frequent integrations of non-viral domains. The integrations have to be recent, not only in relative terms, compared to other domains in the same genome, but also in absolute terms, compared to the progress of modern agriculture. The integrations also have to be frequent, in the sense that it is likely that integration could have happened several times, in different biotopes. It is difficult to estimate mutation rates in RNA viruses. They evolve very rapidly, and it is often difficult to assign reliable phylogenies. However, recent studies indicate that most ssRNA viruses have a mutation rate close to 10 -3 substitutions per site per year [ 34 ], e.g. the SARS virus has 1.16–3.30 × 10 -3 non-synonymous substitutions per site per year, which is considered to be a "moderate" ssRNA mutation rate [ 34 ]. If we assume that most ssRNA viruses have effective mutation rates within the same order of magnitude, a realistic mutation rate for the viruses included here might be something like 2.0 × 10 -3 . In that case, the MT, HEL and RdRp trees shown in Additional file 2 represent approximately between 325 and 750 years of evolution. In general the NJ trees estimate a slightly shorter evolutionary history (between 325 and 450 years) compared to the ML trees (between 550 and 750 years). In this estimate the Ampelovirus sequences have not been included, as they seem to have diverged from the remaining AlkB-containing viruses at a much earlier stage. If we believe that the AlkB integrations happened after the divergence of most sequence included here, as indicated by the lack of co-evolution in Figure 4 , it does not seem unrealistic to assume that most of these integrations happened within the last 50 – 100 years or so. This estimate is of course very approximate, in particular since we do not know the true mutation rate of these viruses. However, it shows that a likely time span for AlkB integration is compatible with the evolution of modern agriculture. Unfortunately, because of the lack of any robust phylogeny for the viral AlkB sequences it does not make sense to do a similar estimate for that domain. Although it is generally accepted that viruses frequently use recombination to acquire functionality [ 35 ], it is less well known how often this includes nonviral sequences. However, there are some well-documented examples, and in particular the properties of the ssRNA positive-strand Pestivirus may be relevant in this context. There are two biotopes of the pestiviruses, cytopathogenic (cp) and noncytopatogenic (noncp). The host is infected by the noncp form which is converted into the cp form by integration of a fragment of a cellular gene into the viral genome [ 36 ]. This introduces a protease cleavage site in the polyprotein. However, the important point here is that this happens as part of the normal infection process. It has been suggested that the integration is facilitated by the viral polymerase undergoing two subsequent template switches during minus-strand synthesis [ 37 ], although nonreplicative RNA recombination also may be a possibility [ 38 ]. Integration of cellular sequences have also been observed in other viruses, e.g. in influenza virus [ 39 ]. This shows that at least some viruses do have efficient mechanisms for recruitment of host genes into the viral genome. Therefore a recent and rapid integration of AlkB domains into selected plant virus genomes does not seem to be an unlikely scenario. This study has focused on the AlkB domain, mainly as an attempt to get a better understanding of potential functions associated with this domain. However, it is likely that additional information about integration patterns and the relative importance of de novo integration vs. recombination can be achieved by a closer investigation of the other variable domains, e.g. by looking at how they correlate with the evolution of the AlkB domains. Conclusions We believe that the viral AlkB-like domains are conventional repair domains targeted towards the viral RNA. The integration of AlkB domains into viral genomes may have been provoked by environmental methylating agents, e.g. the introduction of DNA/RNA-methylating pesticides in farming. The hypothesis [ 1 ] that the domain interferes with the PTGS system of plants can not be excluded, but seems to be less consistent with observed features of the AlkB integration. Methods The NCBI nr protein sequence database was searched with PSI-Blast [ 40 ], with the output limited to viral sequences. Multiple alignments were made with ClustalX version 1.8 [ 41 ]. The phylogenetic tree in Figure 2 was made from ClustalX alignments by MEGA2 [ 42 ], using the neighbour-joining (NJ) approach with complete deletion of gap positions, Poisson correction of distances and 500 bootstrap steps. Phylogenetic trees for sequence regions from sequences with AlkB domains were made with the NJ approach as described above, but with 10.000 bootstrap steps. Corresponding trees were also made by the maximum likelihood approach (ML) by Tree-Puzzle version 5.2 [ 43 ], using an exact likelihood function, the VT matrix [ 44 ] and 10.000 puzzling steps. The trees from Tree-Puzzle were visualised with TreeView version 1.6.6 [ 45 ], and the NJ and ML trees were compared with Component version 2.0 [ 46 ]. Significance of pairwise tree distances were estimated using the data of Day [ 12 ]. Pairwise distances between sequences, for comparing evolution of AlkB domains to other viral domains, were computed directly from ClustalX alignments with local tools, using the Blosum50 mutation matrix [ 47 ], but without any correction for multiple substitutions. Motifs in protein sequences were identified using HMMER version 2.3.2 [ 48 ] with the Pfam library version 11.0 [ 49 ]. A Pfam-type profile for the methyltransferase domains of Flexiviridae and Tymoviridae was generated from a ClustalX alignment, using hmmbuild and hmmcalibrate from the HMMER package. Visualisation of motif positions in viral sequences was generated directly from the HMMER output files using a local tool as an interface to the GNU [ 50 ] groff software. Systematic large scale searches with polyprotein subsequences were done locally with PSI-Blast and the NCBI reference sequence database [ 15 ]. Dot plots for comparison of viral protein sequences were generated with Dotter version 3.0 [ 51 ]. List of abbreviations used MT – Methyl transferase; HEL – Helicase; RdRp – RNA-dependent RNA polymerase; ssRNA – Single-stranded RNA; PTGS – Post-transcriptional gene silencing; 2OG – 2-oxoglutarate; (h)ABH – (human) AlkB homologue; OTU – Ovarian tumour-like protein; NJ – Neighbour-joining; ML – Maximum likelihood; SJA – Strict joint assertions. Authors' contributions MSB carried out all PSI-Blast searches, generated local (sub)sequence databases, and drafted the initial manuscript. FD conceived the study, carried out HMMER/Pfam searches, and estimated evolutionary distances. Both authors participated on sequence alignment, phylogenetic analysis and writing of the manuscript. Both authors have read and approved the final manuscript. Supplementary Material Additional File 1 Full listing (with GI numbers) of viral sequences and domains included in the analysis. Click here for file Additional File 2 Individual NJ and ML trees for relevant domains (MT, AlkB, HEL, RdRp). Click here for file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544882.xml |
544896 | FunnyBase: a systems level functional annotation of Fundulus ESTs for the analysis of gene expression | Background While studies of non-model organisms are critical for many research areas, such as evolution, development, and environmental biology, they present particular challenges for both experimental and computational genomic level research. Resources such as mass-produced microarrays and the computational tools linking these data to functional annotation at the system and pathway level are rarely available for non-model species. This type of "systems-level" analysis is critical to the understanding of patterns of gene expression that underlie biological processes. Results We describe a bioinformatics pipeline known as FunnyBase that has been used to store, annotate, and analyze 40,363 expressed sequence tags (ESTs) from the heart and liver of the fish, Fundulus heteroclitus . Primary annotations based on sequence similarity are linked to networks of systematic annotation in Gene Ontology (GO) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) and can be queried and computationally utilized in downstream analyses. Steps are taken to ensure that the annotation is self-consistent and that the structure of GO is used to identify higher level functions that may not be annotated directly. An integrated framework for cDNA library production, sequencing, quality control, expression data generation, and systems-level analysis is presented and utilized. In a case study, a set of genes, that had statistically significant regression between gene expression levels and environmental temperature along the Atlantic Coast, shows a statistically significant (P < 0.001) enrichment in genes associated with amine metabolism. Conclusion The methods described have application for functional genomics studies, particularly among non-model organisms. The web interface for FunnyBase can be accessed at . Data and source code are available by request at jpaschall@bioinfobase.umkc.edu . | Background Investigating patterns of gene expression using mouse and human microarrays has produced insights into cancer [ 1 , 2 ], cardiac diseases [ 3 - 6 ], and metabolic disorders [ 7 - 12 ]. These and many other functional genomics studies rely on full genomic sequence to establish well-annotated databases. Yet, microarrays based on EST collections are increasingly being used for diverse species, from honey bees to fish [ 13 - 20 ] and including simple diploblastic organisms [ 21 ]. These studies within a diversity of organisms provide insights not provided by 'model' species (species that are genetically well defined or with annotated genomes [ 22 ]). For example, 'non-model' organisms have provided insight into the natural variation in gene expression [ 23 ], social castes among bees [ 24 , 25 ], hypoxia [ 26 ], and physiological responses to variation in the thermal environment [ 27 , 28 ]. To investigate adaptive variation in gene expression we use the teleost Fundulus heteroclitus (killifish) [ 23 , 29 ]. The killifish Fundulus heteroclitus are distributed along the eastern coast of North America which has one of the steepest thermal clines in the world: northern populations have environmental temperatures more than 12°C below southern populations across 12 degrees of latitude. Migration among populations is sufficient to minimize random genetic drift [ 30 ] but not frequent enough to extinguish local adaptation [ 31 , 32 ]. Populations are large (>10,000) and affected by historical, demographic and selective constraints, providing a framework for the partitioning of variation in gene expression within and among populations. Additionally, the well-established phylogenetic relationship among Fundulus species can be used to discern adaptive changes [ 23 , 33 , 34 ]. These characteristics make F. heteroclitus an ideal species to investigate adaptive variation in gene expression. Microarrays from diverse EST collections offer opportunities to address many biological problems, but to effectively use this information often requires a locally generated bioinformatics approach. Tools like the TIGR Gene index [ 35 ] and Unigene [ 36 ] provide significant information on many species, yet these databases do not meet the needs of functional genomics projects for many non-model species. Currently, TIGR and NCBI provide gene indices for 28 and 23 animal species, respectively. Yet, there are 63 animal species with more than 10,000 ESTs [ 37 ]. The number of species with ESTs >10,000 has continued to grow, and there was approximately a 20% increase in the preceding three months. While annotation from these resources can be accessed through web-based homology searches, for many laboratory collections of ESTs it is difficult to use existing tools to achieve a systems-level view of gene functions and relationships. Rather than simply browsing functional information over the web for a different group's project, laboratories that produce novel EST collections and microarrays require customized databases providing access to integrated functional annotation as expression data are being analyzed. We have developed FunnyBase to meet these functional genomics needs. FunnyBase provides functional information for >40,000 ESTs from the teleost fish Fundulus heteroclitus , provides the means to quickly process, evaluate, and store annotation based on similarity searches of public resources, and integrates these data with species-specific clustering and microarray analysis. Perhaps ironically, the greatest challenge for functional annotation based on similarity searches is an overabundance of data. There are a number of databases to chose from, and often the single best hit from a given database search is not the most informative. FunnyBase implements a strategy to make maximum use of systems-level functional information from Gene Ontology (GO) [ 38 ] assignments and membership in metabolic pathways as defined by the Kyoto Encyclopedia of Genes and Genomes (KEGG) [ 39 ]. Specifically, several sequence databases are queried and results integrated to maximize the number of annotated sequences. Alignments and scores for all homology based associations are tracked, allowing further evaluation and statistical studies. Microarray data using genes annotated in FunnyBase can be systematically analyzed in the context of biological functions. We present a case study to illustrate how assessment of systems-level annotation can identify statistically significant functional differences among sets of genes. Results and discussion FunnyBase (Fig. 1 ) is divided into 3 modules: Sequence Pipeline, Hierarchical Annotation, and Microarray Production and Analysis. The Sequence Pipeline takes sequences and quality output files from the sequencer, applies vector screening, quality trimming, clone tracking, and clustering (described below) to produce a set of unique sequences that are deposited in the 'Sequence Data' and 'Cluster Data' tables. Notice "unique sequences" are a combination of singletons (single unique ESTs) and clusters of overlapping sequences. Figure 1 FunnyBase annotation scheme. The integration of the three FunnyBase modules: sequence pipeline, hierarchical annotations and microarray production and analysis. Database tables are shown in cylinders, arrows are data flow, and dashed lines indicate the integration of data from multiple sources. The Hierarchical Annotation module uses the consensus sequences from the clusters or singletons, and integrates primary annotation such as gene name and description with associated pathways and systems-level functional annotation. This may include gene function (e.g., enzyme catalyst), metabolic or signal pathway (e.g., oxidative phosphorylation), or biological function (e.g., protein translation). Sequence data from the first module and functional annotation from the second are matched using database similarity searches (BLASTX and BLASTN). E-values, bit scores and local alignments are stored in the 'Similarity Data' table for all significant matches. One of the strengths of FunnyBase is the use of different sequence databases (SwissProt [ 40 ], NCBI UniGene [ 36 ], and NCBI non-redundant NR [ 41 ]) to provide separate annotations. Although these databases are not completely independent, the three separate annotations provide verification of gene names. The third module, Microarray Production & Analysis provides a list of unique genes to be printed and integrates expression data from microarray experiments with the Hierarchical Annotation module. This provides functional annotation for expression data. FunnyBase annotation is accessible through the web or through local SQL queries and data-mining scripts. EST isolation and sequencing The overall strategy used to isolate and sequence thousands of Fundulus cDNAs was (1) generate a high quality unidirectional cDNA library, (2) normalize the library, (3) randomly pick colonies and amplify by PCR the cDNA within the vector, (4) sequence and identify PCR products, and (5) after approximately every 1,000 clones, subtract these from the normalized library and repeat steps 3–5. Details for all protocols are provided at and were used in the Comparative Functional Genomic course at Mount Desert Island, ME 2000. We sequenced 46,433 ESTs and 40,043 of these are available in the dbEST database at NCBI (dbEST identification numbers: 23,480,307 to 23,515,306; 23,520,047 to 23,525,409 and 24,320,128 to 24,320,184) as of June 26, 2004. Sequences in FunnyBase are identified by a number series: unique sequence number, array number, plate number and well identifier (example: 23434_125_001_H04). The remaining 6,966 un-submitted sequences failed to meet one of the sequence quality parameters. Two criteria are used for defining "good" sequences: 1) >100 bp of sequence with Phred score >20 or 2) form an overlapping cluster with other sequences. Of the 19,937 sequences processed with the current version of FunnyBase , 17,893 (90%) passed one of these quality parameters. In earlier iterations of FunnyBase , visual inspection and later a sequencer-specific quality measure equivalent to a Phred score of 15 were used as filters resulting in 3,603 of the first 5,760 sequences (63%) and 13,922 of the next 15,168 sequences (92%) meeting quality standards, respectively. Re-sequences account for 5,668 sequences, and 4,625 of these were submitted. Controls One of the most important steps for producing microarrays from cDNA libraries is being able to associate the bacteria containing the cDNAs of interest with the EST annotation. High-throughput procedures are highly prone to tracking errors including: loading plates into an automatic sequencer in the wrong order, orienting symmetric plates in the wrong direction, or mislabeling of plates. The ability to identify these types of mistakes requires controls for identifying plates and plate orientation. The FunnyBase system has a number of integrated quality control steps. First, a Ctenophore cDNA (NCBI: accession number: CN992733) that is unlike anything else in GenBank is used as a control. Controls are placed in 96-well plates in wells corresponding to the plate number and two orientation wells (A5 and F9). Sequences from 96-well plates are automatically scanned for these controls so that the identity and orientation are confirmed and a report is generated for manual review. Secondly, all clones used for microarray production are re-sequenced. This is necessary because individual cDNAs are cherry picked, re-grown and re-amplified, and each of these steps has the potential to introduce or magnify an error. For example, for a 6,000 gene array, a 5% error rate would result in 300 incorrect clones. Re-sequenced array plates are compared using pair-wise BLAST [ 42 ] against previous sequencing results so that the identity of printed microarray spots are verified. EST clustering EST projects generate a number of redundant sequences due to the random selection of cDNAs from tissue libraries (Table 1 ). Clustering redundant sequences is a critical first step of analysis in order to identify genes to target for subtraction. The program CAP3 by Xiaong Huang [ 43 ] was used to cluster EST sequences with a 30 bp overlap and 75 percent similarity. Table 1 The Ten Most Frequently Annotated ESTs. Clusters with the greatest number of annotated ESTs, the sequence id, number of ESTs that cluster together, and the e-value (probability of similarity) are listed. ID Number of ESTs Evalue Description 1616 977 0 Vitellogenin I precursor 2262 734 2e-88 Cytochrome c oxidase polypeptide II 1348 507 1e-102 Alpha-1-antitrypsin homolog precursor 1026 481 5e-61 Zona pellucida sperm-binding protein 3 precursor 1727 401 0 Serotransferrin precursor 555 397 0 Cytochrome c oxidase polypeptide I 640 369 8e-61 Zona pellucida sperm-binding protein B precursor 1549 351 2e-77 Apolipoprotein A-I precursor 570 331 1e-111 Cytochrome c oxidase polypeptide III 1178 304 2e-45 ATP synthase a chain FunnyBase contains a total of 40,043 EST sequences from F. heteroclitus heart and liver. Clustering with CAP3 yields 3,776 clusters that contain 30,688 ESTs (77%). The remaining 8,991 ESTs (23%) are singletons. By storing the results of clustering with annotation, FunnyBase easily identifies these genes and aids in the selection of genes to be used for library subtraction with the goal of picking less common transcripts. The 10 annotated clusters with the most sequences are listed in Table 1 . In microarray experiments these genes tend to be highly expressed and the fluorescent signal tends to saturate the photomultiplier tube. These genes also serve to verify microarray printing because the predicted spots for these genes have the strongest signal. The distribution of the number of clusters with two or more ESTs is depicted in Figure 2 . Although most clusters (2,581 or 68%) in FunnyBase have two or three-to-four sequences (Fig. 2 ), a small number of highly expressed genes form clusters with a large number of ESTs. For example, there are three clusters that contain 512 to 1,024 ESTs and ten clusters that contain 256 to 512 ESTs (bottom two bars for killifish in Fig. 2 ). This distribution is similar to other teleost fish EST collections (Fig 2 ). Notice, as more ESTs are added, clusters tend to get larger (more ESTs per cluster) rather than new small clusters growing in frequency. Of the 3,779 killifish clusters, 14% have more than eight ESTs, yet of the 14,714 Medaka clusters, 48% have eight or more ESTs. These distributions suggest that adding more EST sequences has diminishing returns. Figure 2 Frequency of cluster size class in teleosts. The frequency of the number of ESTs in each cluster is shown for Fundulus heteroclitus (Killifish: total number of clusters 3,779), Oryzias latipes (Medaka: total number of clusters 7,401), Oncorhynchus mykiss (Rainbow Trout: total number of clusters 11,405) and Danio rerio (Zebrafish: total number of clusters 14,714). For example, the first bar indicates that approximately 1,000 clusters contain 2 ESTs in all four teleost fish. Clusters for other species (not F. heteroclitus ) are based on NCBI UniGene. One of the objectives of EST projects is to isolate most, if not all, genes expressed in a tissue or organism. The increasing size of larger clusters with more sequencing efforts indicates that strategies to increase the probability of isolating new genes need to be employed. We used two strategies. First, we normalized the library to reduce the differences among expressed genes to less than 10-fold among rare and abundant mRNAs [ 44 , 45 ]. Using this technique we were able to reduce redundancy in annotated genes from 33% in the non-normalized library to 11% after normalization. Second, we targeted specific sequences for subtraction: annotated cDNAs with high frequencies were targeted in order to focus effort on picking new, rare sequences. Through subtraction we were able to increase the rate of discovery of new annotatable sequences from 24% to 36%. However, analysis of these results indicate that a set of highly expressed sequences, some of which were not subtracted because they were not in the set of annotated genes, still make up much of the EST library and should be the focus of future subtractions. Gene annotation Of the 12,776 unique ESTs (3,776 clusters and 8,991 singletons), 3,877 (30%) were annotated. The distribution of e-values for these annotations is shown in Figure 3 . Most (84%) of these ESTs have e-values less than 10 -5 . Among the clusters, 2,265 of 3, 779 (60%) were annotated as compared to 333 of 1,131 (30%) of confirmed high quality singletons. The lower percent of annotated singletons suggests that these are either rare fish-specific genes, or represent otherwise divergent, likely non-coding (5' or 3' UTR), regions. Figure 3 Distribution of e-values for annotations. Gene annotation is based on sequence similarity. The e-values, which describe the probability of random sequence similarity, are shown as the negative log value (e.g., 10 -5 = 5). Annotations for ESTs are based on similarity using BLASTX or BLASTN [ 42 ] to sequences in one of six-public databases: Swiss-Prot, Human UniGene, Danio rerio UniGene, Oncorhynchus mykiss Unigene, Oryzias latipes Unigene, or GenBank NR. FunnyBase includes locally parsed copies of these databases in a relational format. Thus, all searches are done locally and annotation features beyond the FASTA description can be queried. Consensus sequences from the Fundulus EST clusters as well as high quality single unique sequences (singletons) are used as query sequences for BLAST searches against these public databases. The use of consensus sequences, when available, allows sequences that do not contain regions of significant similarity with known protein (e.g., 5' or 3' noncoding regions) to be annotated if they are members of an annotated cluster. All hits with e-value less than 0.001 and their associated alignments are stored in the database and tracked with any associated functional annotation. Users can specify a custom level of significance when assessing the validity of homology based annotation. This record, which goes beyond storing a certain number of 'best hits', is critical because in many cases additional results may have a negligibly lower alignment score, but provide much more useful functional data. The use of multiple databases increases the total number of annotated ESTs (Fig. 4 ) as compared with any one source and provides opportunity to compare annotation between all three sources for 1,841 (47%) sequences. GenBank NR provided the most number of annotations, but these tend to be less informative (see systematic functional annotations below). Human Unigene provided an additional 311 (8% of total) annotations. SwissProt provided an additional 32 annotations (1% of total) with 743 fewer annotated sequences than the NR. However, SwissProt is uniformly well annotated as compared to NR where informative functional annotation can easily be buried by numerous uninformative hits at similar e-values. Besides increasing the number of annotations, comparing the annotations from multiple databases ensures that mistakes in the curation can be detected and information such as alternative gene names can be compiled from multiple sources. Figure 4 Venn diagram of annotations from three different databases. Number of unique ESTs annotated by three different databases: GenBank non-redundant (NR, total annotated = 3,534), Swiss-Prot (total annotated = 2,829), and human Unigene (total annotated = 2,390). Systematic functional annotation: KEGG and Gene Ontology In conjunction with performing similarity searches by BLAST, FunnyBase includes locally parsed representations of public databases such as SWISS-PROT in a relational database format. These databases provide additional information that cross-references other public resources such as GO, KEGG or OMIM [ 46 ] that are not available in the single FASTA description line returned by BLAST search. KEGG [ 39 ] is a unique tool that represents metabolic and signal-transduction pathways both visually and computationally. FunnyBase links annotated genes to enzymes in KEGG pathways based on enzyme commission (EC) numbers. These pathway associations are stored and queries can readily identify genes from a given pathway that show specific patterns of expression. For visual inspection of the pathway, the web interface links directly to the graphical KEGG pathways in which a gene occurs. Of the 3,877 annotated ESTs in FunnyBase , 588 (14%) participate in one or more pathways defined by KEGG. These 588 ESTs represent 105 different pathways. Table 2 provides a breakdown of the number of ESTs in FunnyBase for the 10 pathways associated with the largest number of distinct sequences (contigs or singletons). The extent that a given pathway is represented in FunnyBase can be used to identify metabolic differences among tissues [ 47 ] or in different species. Table 2 Number of distinct sequences in the Top 10 most common KEGG pathways. The KEGG pathway name and number of distinct sequences (clusters or singletons) from FunnyBase are presented. There are more distinct sequences than enzymes in a pathway because many enzymes have several protein subunits and many proteins have several different loci encoding the same subunit (e.g., NADH dehydrogenase, a protein complex of oxidative phosphorylation, has 26 protein subunits and 42 loci for these subunits). Sequence count for TOP 10 pathways Glycolysis/Gluconeogenesis 89 Oxidative phosphorylation 86 Fatty acid metabolism 70 Pyruvate metabolism 69 Tryptophan metabolism 66 Butanoate metabolism 48 Glycerolipid metabolism 47 Valine, leucine and isoleucine degradation 46 Glycine, serine and threonine metabolism 44 Propanoate metabolism 44 The Gene Ontology project (GO) has produced a structured vocabulary in the form of an acyclic directed graph that biologists can use to annotate genes in a systematic manner [ 38 ]. FunnyBase includes two non-trivial steps to make the best possible use of GO terms. First, many GO annotations are lost if only the single 'best hit' from a homology search is considered because GO annotation is applied most often to a few model species such as human that may not appear as the single 'best hit' in a list of BLAST results. FunnyBase identifies the gene name associated with the 'best hit' BLAST result and then uses all GO annotation associated with hits from the complete BLAST results that have the same gene name as the 'best hit' and an e-value of e < 10 -12 . The goal of this approach is to identify annotation associated with a single 'best hit' gene based on results that may come from multiple species (orthologous genes) and therefore may have varying degrees of sequence similarity due to phylogenetic distance, but to avoid the problem of selecting an inconsistent set of GO terms arising from gene families that share regions of sequence similarity but may have different functions. Secondly, GO annotation in public databases tends to annotate sequences with only the most specific GO term available, for example RNA polymerase II transcription factor activity , enhancer binding (GO:0003705) rather than the more general parent term transcription regulator activity (GO:0030528). However, in functional genomic analysis, significant patterns of expression may exist at the more general level of functional description. FunnyBase takes advantage of the connected parent-child relationship of GO terms provided by using the relational database version of GO available for download at to identify such relationships. These data are used to extract the tree of more general GO terms related to those provided by public databases. A FunnyBase script then re-annotates genes with this more complete set of GO terms. Of the 3,877 annotated genes, 1,912 (54%) are assigned one or more GO terms with a total 6,728 GO assignments being made directly based on information in public databases such as SwissProt. Using parent-child GO term relationship backtracking, an additional 34,112 GO term assignments were made, resulting in a final count of GO assignments of 36,024 excluding the most general terms that divide GO into three categories. Thus, on average, 19 GO terms are assigned to each of 1,912 annotated genes. Gene scaffolding: clustering of clusters Humans have approximately 30,000 expressed genes, yet there are over 1,000,000 human UniGenes (NCBI). Clearly, these clusters of cDNAs greatly overestimate the number of unique genes. Similarly, FunnyBase has multiple clusters for the same gene: 15 apolipoprotein I, 10 cytochrome oxidase I, and 53 vitellogenin clusters. To provide a more precise estimate of the number of unique genes, consensus sequences were queried against the 27,695 sequences from the Human RefSeq [ 48 ] database, then grouped by identical gene symbol. Of the 2,376 Fundulus clusters that were similar to a sequence in Human RefSeq (e-value < 10 -10 ), 1,818 (76%) had distinct gene annotations. This method of clustering clusters by similarity to well-annotated reference sequences provides a method to more accurately define the number of unique genes represented by an EST set. Case study: using functional annotation for microarray analysis As a case study in how functional annotation in FunnyBase can be integrated with microarray data in a rigorous manner, we used a data set based on a microarray of metabolic genes printed from ESTs annotated in FunnyBase [ 47 ]. Statistical analysis of this set of 363 metabolic genes identified a set of 62 genes that showed statistically significant regression between gene expression levels and temperature along the Atlantic coast. That is, among individuals collected from different locations along the thermocline and then acclimated to common physiological conditions for at least nine months before analysis, 17% of the metabolic genes had a linear relationship between the amount of mRNA and the environmental temperature these animals evolved in. Our hypothesis was that this set of 59 genes represents a functionally different set than those genes that do not show regression with temperature. To test this hypothesis we examined the frequency of genes annotated with a given GO term in the statistically significant gene set versus the non-significant genes. Figure 5 shows the relevant proportions in each set for GO terms that are represented by 5 or more ESTs in the significant set. For example, the GO term Amine Metabolism (GO:0009308) is assigned to 14% of the 62 statistically significant genes but only 3% of the non-significant genes (those that do not show significant regression with temperature). A Fisher-exact test indicates these frequencies (14% vs. 3%) represent different underlying distributions (p < 0.001). Specifically, genes involved in amine metabolism are overrepresented in the set of genes that show regression with temperature as compared with the remaining sequences. This significant increase is found for two other non-mutually exclusive GO terms: amino acid and derivative metabolism , and amino acid metabolism (p < 0.05). Other GO terms show a reverse trend although none were statistically significant. For example, ion transport (p = 0.08) and cell growth (p = 0.16) had few genes with a clinal variation in expression. These data suggest that the functions of genes influence whether they are affected by ecologically interesting patterns of expression (Fig 5 ). Figure 5 Distribution of significant and non-significant genes relative to GO terms. The relationship between gene expression from "common gardened" fish and the environment they evolved in was statistically analyzed and grouped by GO terms. Black bars represent genes whose levels of expression has a significant regression with the environmental thermal cline among populations (p < 0.05). Hatched bars represent the set of genes with no significant relationship to the thermal environment. Web interface The web interface provides public access to the FunnyBase system and dataset. Searches can query by keyword in annotation, gene name, GO term, metabolic pathway, clone or plate id, and BLAST homology search. All data including raw sequences, cluster memberships, cluster alignments, and alignments with homologous sequences are provided for the user to examine the source of annotations. Links associated with each annotation are made to external resources such as GO's AMIGO browser, KEGG pathways, SwissProt, and NCBI records. Other Fundulus sequences FunnyBase was constructed to annotate sequences for the analysis of gene expression. It provides identification and annotation for genes in the Crawford laboratory with a primary goal of identifying clones useful for the construction of microarrays. As such, other Fundulus sequences in Genbank are not included. However, FunnyBase forms the basis of the TIGR Killifish gene index that includes publicly available F. heteroclitus sequences. Conclusions Customized species specific EST databases are available for many species [ 17 , 18 , 21 , 49 - 56 ]. FunnyBase provides an integrated method to annotate ESTs with the most biologically relevant set of associations and provides several innovations for the production of ESTs for microarrays. Control sequences are identified in each 96-well plate so that mislabelled or inverted plates are automatically detected. Annotations are based upon several different public databases. The multiple annotations provide greater assurance about gene description and greater frequency of annotation than any one database. The most functionally informative innovation of FunnyBase is the process of culling through numerous primary similarity search results in order to identify links to systematic functional databases in GO and KEGG. These provide a discrete set of terms that can be analyzed statistically and that are organized into networks that represent biological knowledge of higher-level functional and pathway associations. The range of databases queried by similarity search and the tracking of homology beyond a single 'best' hit maximizes the opportunity to obtain this annotation. A richer set of GO terms is achieved by using all hits with e-values less than 10 -11 that represent the same gene as the 'best hit'. Additional GO terms that represent more general functions than those found in public annotation are derived through the parent-child relationship of the Gene Ontology. EC numbers provide links, via KEGG, to metabolic pathways and these stored terms can be used to investigate the relationship between gene expression in specific metabolic pathways including cardiac metabolism [ 57 ]. To provide a more accurate accounting of the number of unique genes, consensus sequence from clusters of ESTs were queried against the Human RefSeq database and those sequences sharing the same gene symbol are grouped based on this scaffolding information. These approaches use publicly available bioinformatics tools (BLAST, CAP3, Phred, Cross-Match, Perl, and the MySQL database management system). The application of theses tools in an appropriate framework as outlined in FunnyBase can be used to create a systems level functional genomics annotation system useful for EST databases to study biological processes among a rich diversity of organisms. Methods Organism The animal protocols used in the present study have been approved by the University of Miami Institute Animal Care and Use Committee. The teleost fish Fundulus heteroclitus used for ESTs were collected from two sites: Scorton Creek in Sandwich, MA, and Stone Harbor, NJ. These populations are in the central portion of the thermal cline and have relatively high levels of heterozygosity [ 32 ]. These fish were subjected to the following environmental regime before tissues were harvested for mRNA extraction: kept in controlled temperature and aeration conditions, and acclimated to common conditions (20°C, 15 ppt salinity) in re-circulating aquaria for at least nine months before experiments. Following this common acclimation a subset of fish were subjected to one of several stresses: 4°C, 34°C, hypoxia, or a complex mix of hydrocarbons. cDNA library To effectively isolate and sequence thousands of cDNAs for the production of microarrays, a unidirectional cDNA library with few non-recombinants was required. We created four cDNA libraries: heart libraries from non-stressed and stressed fish and liver libraries from non-stressed and stressed fish. The non-stressed F. heteroclitus cardiac and liver libraries were provided by Drs. S. Karchner and M. Hahn, WHOI [ 58 ] and were constructed using the UniZap λ cDNA Gigpack Gold cloning kit (Stratagene, La Jolla, CA, USA). The cardiac library was produced from 27 fish hearts (both sexes) sampled from Scorton Creek in Sandwich, MA. The cDNAs in these libraries are oriented such that the 5'end of each cDNA is ligated to EcoR1 and 3' poly A is ligated to XhoI. These libraries had less than 1% non-recombinants, i.e. 2 of 300 random clones from a non-normalized library had no inserts. The stressed libraries included 4 fish subjected to the four stressors (above) and 4 non-stressed individuals. Unidirectional heart and liver libraries were constructed such that the 5'end of each cDNA is ligated to EcoR1 and 3' poly A is ligated to XhoI of the plasmid vector pSmart (Lucigen, Middleton, WI, USA). The pSmart-cDNA vector was designed for EST work. The vector expresses kanamycin-resistance and has a terminator on both sides of the cDNA insertion site preventing expression of cDNA. These two attributes (non-expression and Kan-resistance) increase the stability of different genes in the library versus cDNA libraries in Amp libraries with Lac promoters (Crawford, unpublished). These libraries had less than 1% non-recombinants. Normalization of cDNA libraries reduces the differences among expressed genes to less than 10-fold among rare and abundant mRNAs [ 44 , 45 ]. Normalized libraries were produced by isolating cDNAs from approximately 10 12 plasmids. The cDNAs were isolated using PCR amplification with vector specific primers immediately 5' and 3' to the insertion site (EcoRI and XhoI sites). These PCR products (PCR-cDNAs) were denatured and hybridized to single stranded plasmids from the cardiac cDNA library. Taking advantage of Cot values, the most abundant cDNAs were annealed to the more abundant PCR products and were removed selectively by hydroxyapatite-column chromatography. The single-stranded plasmids in the flow-through were converted to double strands using the Sequenase DNA polymerase (Amersham, Piscataway, NJ, USA). DH10s E. coli (BRL) were transformed with these double-stranded plasmids by electroporation. The number of recovered plasmids and the resulting complexity of the normalized library depended on the duration of hybridization or Cot values. Two normalized libraries were made using either a 12 or 24 hour hybridization. The library from the 12-hour hybridization yielded 250,000 plasmids. The library from the 24 hour hybridization yielded 3,000 plasmids and had a greater representation of rare mRNAs and greater frequency of non-recombinants. Isolation and sequencing of cDNAs Characterization of cDNAs (growth of individual bacterial colonies containing plasmids, PCRs, purification of PCR products, sequencing reactions) used 96 well plates and octopipettes. To characterize cDNAs, 96 individual bacterial colonies from the normalized library were randomly chosen, and each was grown in 1.25 ml of Superbroth in 2 ml-96 well plates. After 18 hours of growth, two 250 ul bacterial glycerol stocks were made and stored in 96 well plates at -80°C. One microliter of these bacterial growths was used for PCR reactions using forward and reverse plasmid specific primers: (PucF = CGCCAGGGTTTTCCCAGTCACG, PucR = GAGCGGATAACAATTTCACACAGGAAA). PCR reactions had 0.2 mM dNTPs, 10 pmoles of each primer, 1 unit of Promega Taq (0.2 ul), and reaction buffer with detergents and DMSO (final concentrations: 50 mM Tris HCl, pH 9.2 (25°C), 16 mM (NH 4 )2SO 4 , 2.25 mM MgCl 2 , 2% (v/v) DMSO, 0.1% (v/v) Tween 20). Two-step thermal cycle conditions were used (94°C for 10 seconds; then 32 cycles of 94°C for 30 seconds followed by 70°C for 5 minutes; then 72°C for 15 minutes). PCR products were purified manually in 96 well format using Sephadex G-50 in a deep well plate with a 0.2 microfilter (Millipore, Billerica, USA) or robotically using AmPure (Agencourt, Beverly, MA, USA) and EvolP 3 96 pipetting liquid handling system (PerkinElmer Life Sciences Inc., Boston, MA, USA). PCR products were sequenced from the 5' end (relative to the mRNA) on an ABI 373 or ABI 3730 sequencer using ABI "Big Dye" reaction mix. We typically used 1/16 the amount of reaction mix, yielding 300 to 400 unambiguous bases. Sequences were purified using biotin primers and streptavidin coated magnetic beads (for the ABI 373) or Agencort CleanSeq (for the ABI 3730). Validation We used three procedures to verify that the correct sequence was associated with each cDNA. 1) Each 96-well plate had three wells with a "marker cDNA" ( Ctenophore cDNA #5, a random cDNA with no similarity to any sequence in GenBank). Two wells (#40 and #67) always contained the marker cDNA, and thus any misloading or mislabeling of sequencing lanes was identifiable. The third marker cDNA was placed in a well that corresponds to the plate number (e.g., plate 2 had the marker in well 2). 2) After the production of 12 plates, one row (8 wells) from each plate was re-sequenced. Thus, 8/96 or ~8% of all sequences and their locations were confirmed. 3) cDNAs used for microarrays were re-sequenced. These measures are important to ensure that the correct and known cDNAs are printed. Subtraction The complexity of the normalized library was reduced by subtracting the characterized cDNAs previously isolated from the normalized library. Subtraction greatly reduced the probability of isolating the same cDNA and thus improved the efficiency of screening the library for unique clones. Subtraction used a 100-fold molar excess of biotin-labeled antisense cDNAs produced by PCR using all the characterized cDNAs as substrates and vector-specific primers in which the 3' primer was labeled with biotin. These PCR products were hybridized to the cDNA libraries in the presence of oligo-dA and vector-specific oligos (that prevented non-specific hybridization to oligo-dT or vector sequences). After a 24 hour hybridization, genes in the library that bound to these biotin-labeled PCR products were removed with the use of magnetized, streptavidin coated beads. DH10s E. coli were transformed with the subtracted library by electroporation. Hardware and software Computational work was done on an Apple G5 dual 2 GHz processor system with 4 GB of RAM. Data are stored in a MySQL database, perl scripts were used extensively for parsing and loading data, and PHP was used on an APACHE web server to construct the user interface. Additional programs available from their authors are mentioned within context. Software and databases are described in Table 3 . Table 3 Software and Databases. Publicly available software and databases used for FunnyBase . The version and/or download date are listed. Resource Version and/or Download Date Software Stand-alone BLAST 2.2.8 Cross match 0.990329 CAP3 January, 2004 Public Sequence Similarity Databases Swiss-Prot NR 44 Human RefSeq June, 2004 Human Unigene June, 2004 Zebrafish Unigene June, 2004 Medaka Unigene June, 2004 Rainbow Trout Unigene June, 2004 June, 2004 Functional Annotation Gene Ontology 2004-06-04 KEGG June, 2004 Microarrays Microarrays were printed using a select 384 cDNAs from F. heteroclitus cardiac library encoding essential proteins for cellular metabolism isolated from over 40,000 expressed sequences . These 384 cDNAs were amplified with amine-linked primers and printed on 3-D Link Activated slides (Surmodics Inc., Eden Prairie, MN, USA) using GeneMachine OminGrider , and blocked following slide manufacturer protocols. The suite of 384 amplified cDNAs was printed as a group in four spatially separated replicates. Four hybridization zones of these four replicate arrays were printed per slide, with each zone set separated by a hydrophobic barrier. Samples were hybridized twice; once with Cy3 and once with Cy5 resulting in overall technical replication of 8-fold per sample. Sample preparation and hybridization RNA was extracted from tissue homogenate in a chaotropic buffer using phenol/cholorform/isoamyl alcohol and RNA quality was assessed using the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA). RNA for hybridization was prepared by amplification using a modified Eberwine protocol [ 59 ] using the Ambion Amino Allyl MessageAmp aRNA Kit. Cy3 and Cy5 were hybridized to slides and incubated 12–18 hours at 42°C. Following hybridization, slides were scanned using the Packard Bioscience ScanArray Express microarray scanner (PerkinElmer Life Sciences Inc., Boston, MA, USA) and images processed using ImaGene (Biodiscovery Inc., Marina del Rey, CA, USA). Authors' contributions JP designed, scripted and implemented FunnyBase and provided statistical analyses of database. MFO initiated, designed protocols and provided sequences for F. heteroclitus ' EST project. JDV optimized robotic interfaces for sequencing and sequenced ESTs. JLR and KJK sequenced ESTs. GJW collaborated on bioinformatics and database development. JAW provided microarray data and analyses. DLC initiated F. heteroclitus ' EST project and developed the database and annotation schemes for FunnyBase . All authors read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544896.xml |
551519 | Home visits by family physicians during the end-of-life: Does patient income or residence play a role? | Background With a growing trend for those with advanced cancer to die at home, there is a corresponding increase in need for primary medical care in that setting. Yet those with lower incomes and in rural regions are often challenged to have their health care needs met. This study examined the association between patient income and residence and the receipt of Family Physician (FP) home visits during the end-of-life among patients with cancer. Methods Data Sources/Study Setting . Secondary analysis of linked population-based data. Information pertaining to all patients who died due to lung, colorectal, breast or prostate cancer between 1992 and 1997 (N = 7,212) in the Canadian province of Nova Scotia (NS) was extracted from three administrative health databases and from Statistics Canada census records. Study Design . An ecological measure of income ('neighbourhood' median household income) was developed using census information. Multivariate logistic regression was then used to assess the association of income with the receipt of at least one home visit from a FP among all subjects and by region of residency during the end-of-life. Covariates in the initial multivariate model included patient demographics and alternative health services information such as total days spent as a hospital inpatient. Data Extraction Methods . Encrypted patient health card numbers were used to link all administrative health databases whereas the postal code was the link to Statistics Canada census information. Results Over 45% of all subjects received at least one home visit (n = 3265). Compared to those from low income areas, the log odds of receiving at least one home visit was significantly greater among subjects who reside in middle to high income neighbourhoods (for the highest income quintile, adjusted odds ratio [OR] = 1.37, 95% confidence interval [CI] = 1.15, 1.64; for upper-middle income, adjusted OR = 1.19, 95%CI = 1.02, 1.39; for middle income, adjusted OR = 1.33, 95%CI = 1.15, 1.54). This association was found to be primarily associated with residency outside of the largest metropolitan region of the province. Conclusion The likelihood of receiving a FP home visit during the end-of-life is associated with neighbourhood income particularly among patients living outside of a major metropolitan region. | Background In the last ten years, more and more of those dying of cancer in Canada are doing so out of hospital[ 1 ]. In Nova Scotia, a Canadian Maritime province with a total population of approximately 950,000 people, the proportion of cancer deaths occurring out of hospital has recently grown by fifty per cent[ 2 ]. This trend appears to be associated with a number of factors. More individuals with cancer are choosing to remain in the home setting, hospitals have down-sized thus reducing the number of beds available for end-of-life care [ 3 - 5 ], and there is growing availability of services in the community such as homecare and community based palliative care programs [ 6 - 8 ]. As this trend has developed, it has become even more important for patients and families to have access to medical care in the community. Such first line medical care, in Canada, is generally provided by a family physician, usually previously known to the patient. Initially those with terminal illness will obtain their medical care in the office or out patient setting. As they become sicker, however, there will come a time when getting from the home to the clinic office will be too difficult. At such a time, access to home visiting by a physician becomes very important [ 9 - 11 ]. Research has shown that access to a supportive family physician willing to make home visits is associated with a greater likelihood of a home death [ 12 - 14 ], as is access to a comprehensive palliative care program (PCP)[ 2 ]. There is evidence that those better off financially live longer and are in better health than poorer individuals[ 15 , 16 ]. Reasons for this have been postulated to include the fact that those with higher incomes have higher educational achievement, better living circumstances, and less risky health behaviours. Such better health may also be due, in part, to better access to services for those with fewer financial barriers. In Canada, such an association should not be the result of inadequacy of health services provided to those with lower incomes as our federal government has committed to the provision of a universal, accessible, comprehensive publicly administered health insurance system which aims to ensure that all residents have access to necessary hospital and physician services on a prepaid basis[ 17 ]. We wondered if access to terminal care home visiting by family physicians is better for those with higher incomes even in our publicly-funded health system. Therefore, the purpose of this study was to examine the association between income and the likelihood of receiving home visits by family physicians during the end-of-life among those with cancer. In addition, we examined the effect of regional residency, specifically residency in a major urban centre versus all other regions of NS (which are much smaller in size). Methods Data Data for this retrospective, population-based study were obtained through the linkage of individual-level information extracted from four administrative health databases: (1) the Nova Scotia Cancer Centre Oncology Information System (OPIS) which includes the Nova Scotia Cancer Registry (NSCR) and provincial vital statistics information, (2) the Nova Scotia Medical Services Insurance Physician Services (MSIPS), (3) the Nova Scotia Hospital Admissions / Separations (HAS) file, and (4) the Queen Elizabeth II Health Sciences Center Palliative Care Program (PCP). The MSIPS provides a record of all services provided by physicians to residents of Nova Scotia whereas the HAS contains information relating to all hospital inpatient and outpatient stays and procedures. Because individual-level income information is not available from these sources, the Postal Code Conversion File (PCCF) and 1996 Statistics Canada census data were used to develop an 'ecologic-level' proxy for household income, enumeration area median (EAM) income quintiles or 'neighbourhood income'. These aggregate measures are derived from census information grouped by provincial enumeration areas or 'neighbourhoods'. The resulting quintiles are based on the median income value of each enumeration area. Evidence suggests the use of such proxies in population-based studies is a valid alternative in situations where household level information in not available[ 18 ]. It is, however, important to recognize that ecologic measures represent conceptually distinct measures of SES even when individual measures are available [ 19 - 21 ]. Encrypted patient health care numbers were used to link all four administrative health databases whereas the postal code was the link to the PCCF and Statistics Canada census information. Ethics approval for this project was provided by the Queen Elizabeth II Health Sciences Centre research ethics committee. Subjects All adults identified on death certificates in the NSCR database as having died due to lung, colorectal, breast or prostate cancer death (International Classification of Diseases, 9 th revision [ICD9-CM]) from 1992 to 1997 were included as subjects. These four cancers represent the most common causes of cancer death in Nova Scotia. Measures Patient characteristics included sex, date of birth, region of residency (Halifax regional municipality [HRM], all regions outside of HRM), date of initial cancer diagnosis, year of death (1992–1997) cancer cause of death (lung, colorectal, breast, prostate), and neighbourhood income categorized as provincial quintiles (lower, lower middle, middle, upper middle, upper). Almost 40% (39.5%) of Nova Scotia's population resides in the HRM which spans a primarily urban geographical region. Within HRM's boundaries are all of Nova Scotia's major tertiary health care centres, several community hospitals and many specialized care programs. Although regions outside of HRM also encompass many towns with regional and community health care facilities, they span a much larger, diverse, geographical area and may be considered to be relatively more rural than HRM. Health services information was limited to each subject's 'survival time'. In our end-of-life research we have defined 'end-of-life' as the last 180 days of life, or if of shorter duration, from the date of initial cancer diagnosis to death. This six month time period is commonly used in end-of-life studies [ 22 - 24 ]. Total family physician home visits received during this survival time were counted and, due to the highly skewed distribution evidenced, also dichotomized to represent at least one home visit received or none. Additional health services of interest and potential covariates included the total number of ambulatory visits made to a family physicians by each subject, the total number of visits made by the subject to a specialist, the total number of days spent as a hospital inpatient, receipt of palliative radiotherapy, and whether or not the patient had been admitted to the PCP, a comprehensive palliative care program which has been operating since 1992. As an indirect measure of whether the patient was a resident of a long term care (LTC) facility during the end-of-life, a flag was created indicating whether a patient had received at least one family physician within a LTC centre. Analysis Following descriptive statistics and the application of nonparametric tests to assess median differences and cross-tabulations with chi-square analyses for association, regression techniques were employed to estimate the effect of neighbourhood income (EAM income quintile) on the receipt of family physician home visits. Our initial regression analysis retained the total number of home visits as a continuous dependent variable and involved negative binomial regression where differences in survival time were accounted for as an offset variable. This was followed by logistic regression techniques where the probability of receiving at least one FP home visit versus no home visits was assessed. For both forms of regression, unadjusted analyses were followed by multivariate where the initial model included neighbourhood income in addition to sex, year of death, age, cancer cause of death, region of residency, the number of visits made to a medical specialist, the receipt of palliative radiotherapy and admission to the PCP as covariates. To account for the possibility that the subject may not have been 'at home' during their 'survival time' and hence unable to receive a home visit, our LTC residency flag and the total days spent as a hospital inpatient were added to the model. To control for differences in 'survival time', the number of days from the initial cancer diagnosis date to death were categorized and added to the model. Subsequent modeling involved the sequential elimination of covariates and confounders found to no longer be significantly associated with the receipt of FP home visits in the multivariate model at the p = 0.05 level of significance. Since our administrative data do not provide the ability to make adjustments for regional differences such as the availability of alternative health services, physician density or community resources, as an alternative, we stratified each analysis by region of residency. All analyses were performed using SAS software[ 25 ]. Results In total, 7212 adults were identified from death certificate information as having died from lung, colorectal, prostate or breast cancer between 1992 and 1997 in Nova Scotia. Over 94% of these advanced cancer patients had seen a family physician at least once during the end-of-life. In total, home visits accounted for 29% of all ambulatory visits provided by family physicians with 3265 (45.3%) patients receiving at least one FP home visit. The total number of FP home visits received varied widely, from 0 to 89 with an average number of 2 (standard deviation [SD] 4.2) and median of 0. Table 1 records the number of home visits received within each neighbourhood income quintile and by region of residency. Although patients from upper and middle income neighbourhoods across all of Nova Scotia appear to receive a greater number of home visits than those from lower income neighbourhoods, examination by region of residency reveal that this gradient by income is primarily associated with residency outside of HRM. Furthermore, patients residing outside of HRM tend to receive fewer home visits in general (mean 1.75, SD 4.1; median 0, range 0–89) than those living within the metropolitan region (mean 2.53, SD 4.4; median 1, range 0–56). The differences between the mean and median number of visits by region of residency were significant at the p < 0.0001 level. Results were similar in the examination of home visits as a dichotomy. A greater proportion of patients residing in middle to upper income neighbourhoods received at least one home visit than those from lower income areas (Table 2 ). Again, after controlling for region of residency, this association was found only to apply to those residing in regions outside of HRM (p < 0.0001). Table 1 Family physician home visits by neighbourhood income quintiles and region of residency Family physician home visits Mean (standard deviation); Median (range) Region of residency Neighbourhood income quintile All adult Nova Scotians Halifax regional municipality All other regions Lower 1.67 (4.2); 0 (0–89) 2.30 (4.1); 0 (0–30) 1.53 (4.2); 0 (0–89) Lower middle 1.77 (4.1); 0 (0–69) 2.60 (4.6); 1 (0–31) 1.60 (4.0); 0 (0–69) Middle 2.25 (4.7); 0 (0–58) 2.96 (5.1); 1 (0–45) 2.02 (4.9); 0 (0–58) Upper middle 2.15 (4.0); 0 (0–56) 2.44 (4.4); 1 (0–56) 1.95 (3.6); 0 (0–24) Upper 2.42 (3.9); 1 (0–35) 2.36 (3.7); 1 (0–23) 2.65 (4.6); 0 (0–35) All 2.00 (4.2); 0 (0–89) 2.53 (4.4); 1 (0–56) 1.75 (4.1); 0 (0–89) Note: Mean and median visits differ significantly at the p < 0.0001 level Table 2 Characteristics of Nova Scotians by receipt of home visits during the end-of-life, 1992–1997 Characteristic Home visit receipt; No. (and %) of adult Nova Scotians * No home visit n = 3947 At least one home visit n = 3265 Neighbourhood income quintile † Lower 1029 (60.0) 685 (40.0) Lower middle 866 (58.4) 618 (41.6) Middle 747 (51.5) 704 (48.5) Upper middle 676 (52.3) 610 (47.4) Upper 399 (45.2) 483 (54.8) Sex † Male 2323 (56.9) 1763 (43.2) Female 1624 (52.0) 1502 (48.1) Year of death ‡ 1992 643 (56.3) 499 (43.7) 1993 637 (54.8) 525 (45.2) 1994 655 (52.7) 588 (47.3) 1995 648 (52.1) 597 (48.0) 1996 679 (54.7) 562 (45.3) 1997 685 (58.1) 494 (41.9) Age group § < 65 years 968 (55.6) 772 (44.4) 65–74 years 1165 (54.2) 986 (45.8) 75+ years 1814 (54.6) 1507 (45.4) Cancer case of death † Lung 2094 (57.0) 1580 (43.0) Colorectal 674 (55.1) 549 (44.9) Breast 629 (50.6) 614 (49.4) Prostate 550 (51.3) 522 (48.7) Survival time † <61 days 935 (78.4) 258 (21.6) 61–120 days 307 (49.8) 310 (50.2) 121–180+ days 2705 (50.1) 2697 (49.9) Region of residency † Halifax regional municipality 1082 (47.2) 1211 (52.8) All other regions of Nova Scotia 2855 (58.2) 2049 (41.8) Visit within a long term care center (LTC) † None 3421 (53.6) 2968 (46.5) At least one LTC visit 526 (63.9) 297 (36.1) Specialty visits ‡ 0–2 1066 (57.5) 789 (42.5) 3–6 1025 (54.4) 859 (45.6) 7–13 736 (54.8) 608 (45.2) 14+ 1120 (52.6) 1009 (47.4) Total days as hospital inpatient ‡ 0 568 (55.4) 458 (44.6) 1–12 1247 (55.3) 1007 (44.7) 13–31 1132 (52.0) 1044 (48.0) 32+ 1000 (57.0) 756 (43.1) Admission to palliative care program † No 3342 (59.2) 2301 (40.8) Yes 605 (38.6) 964 (61.4) Received palliative radiation † No 3085 (56.8) 2346 (43.2) Yes 862 (48.4) 919 (51.6) * Total number of patients by characteristic may vary due to missing values. Proportions are row percentages and may total more than 100 due to rounding. † Characteristic is associated with receipt of home visit (p < 0.001) ‡ Characteristic is associated with receipt of home visits (p < 0.05) § No significant association demonstrated Subject characteristics and health service utilization by receipt of at least one home visit are displayed in Table 2 . In addition to patients who received at least one FP home visit tending to reside in higher income neighbourhoods, they also were more likely to be female, have a breast cancer cause of death, survived at least 61 days from their initial cancer diagnosis date, did not receive a FP visit within a LTC facility, made more than 14 specialty visits, spent 13–31 days as a hospital inpatient during their survival time, received palliative radiotherapy, and were admitted into the PCP. Regression results incorporating the total number of home visits using negative binomial regression and those derived from logistic techniques assessing the log odds of receiving at least one home visit compared to none proved similar. Therefore, for ease of presentation, we present the logistic regression results only. Displayed in Table 3 are the adjusted multivariate logistic regression results examining the effect of 'neighbourhood' income and additional predictors on the receipt of at least one FP home visit during the end-of-life among all advanced cancer patients and by region of residency. Examination of the crude odds ratios (OR) and related confidence intervals (CI) indicate the log odds of receiving at least one home visit was significantly greater among subjects who reside in middle to high income neighbourhoods compared to those from low income. Following adjustments for all other significant predictors retained in the model, this significant association remained, although less strongly. Compared to advanced cancer patients from lower income neighbourhoods, those from upper income neighbourhoods were 37% more likely to receive at least one FP home visit (adjusted odds ratio [OR] 1.37, 95% confidence interval [CI] 1.15, 1.64). Cancer patients from the upper-middle and middle income neighbourhoods were also significantly more likely to have received at least one family physician home visit than those from the lowest income area (for upper-middle income adjusted OR 1.19, 95%CI 1.02, 1.39; for middle income adjusted OR 1.33, 95%CI 1.15, 1.54). However, this association is not experienced equally across the province. Although patients residing in the large metropolitan region of HRM tend to receive more FP home visits in general, the receipt of such visits are not associated with neighbourhood income. In contrast, among patients living outside the HRM, those from upper income neighbourhoods were more than twice as likely to receive a FP home visit than others residing in lower neighbourhood income areas (adjusted OR 2.23; 95%CI 1.63, 3.07). Table 3 Odds of receiving a home visit by income and other characteristics for Nova Scotia overall and by region Predictor Nova Scotia overall Halifax Regional Municipality [HRM] All regions outside of HRM Crude OR Adjusted* OR (and 95% CI) Adjusted* OR (and 95% CI) Adjusted* OR (and 95% CI) Neighbourhood income quintile Low 1.0 1.0 (-) 1.0 (-) 1.0 (-) Lower middle 1.07 1.04 (0.90, 1.21) 1.42 (0.99, 2.03) 0.98 (0.84, 1.16) Middle 1.42 1.33 (1.15, 1.54) 1.40 (1.00, 1.94) 1.30 (1.10, 1.53) Upper middle 1.36 1.19 (1.02, 1.39) 1.21 (0.90, 1.64) 1.20 (0.99, 1.44) Upper 1.82 1.37 (1.15, 1.64) 1.18 (0.88, 1.56) 2.23 (1.63, 3.07) Survival time <61 days 1.0 1.0 (-) 1.0 (-) 1.0 (-) 61–120 days 3.66 3.83 (3.08, 4.77) 3.91 (2.62, 5.82) 3.69 (2.81, 4.84) 121–180+ days 3.61 4.04 (3.45, 4.72) 3.64 (2.76, 4.81) 4.21 (3.47, 5.11) Admission to palliative care program No 1.0 1.0 (-) 1.0 (-) 1.0 (-) Yes 2.30 2.25 (1.97, 2.56) 3.05 (2.49, 3.74) 1.47 (1.15, 1.86) Visit within a long term care center (LTC) None 1.0 1.0 (-) 1.0 (-) 1.0 (-) At least one LTC visit 0.65 0.55 (0.46, 0.65) 0.47 (0.34, 0.64) 0.60 (0.48, 0.75) Age group < 65 years 1.0 1.0 (-) 1.0 (-) 1.0 (-) 65–74 years 1.06 1.25 (1.09, 1.44) 1.24 (0.97, 1.59) 1.28 (1.08, 1.51) 75+ years 1.04 1.41 (1.24, 1.61) 1.94 (1.52, 2.49) 1.29 (1.09, 1.51) Total days as a hospital inpatient 0 1.0 1.0 (-) 1.0 (-) 1.0 (-) 1–12 1.0 1.06 (0.90, 1.26) 1.02 (0.76, 1.37) 1.07 (0.87, 1.30) 13–31 1.14 1.07 (0.91, 1.27) 1.04 (0.77, 1.40) 1.08 (0.87, 1.32) 32+ 0.94 0.80 (0.67, 0.95) 0.87 (0.64, 1.20) 0.78 (0.63, 0.97) Year of death 1992 1.0 1.0 (-) 1.0 (-) 1.0 (-) 1993 1.06 1.05 (0.88, 1.25) 1.07 (0.78, 1.48) 1.02 (0.82, 1.27) 1994 1.16 1.17 (0.98, 1.40) 1.13 (0.81, 1.57) 1.16 (0.94, 1.43) 1995 1.19 1.10 (0.92, 1.31) 0.88 (0.63, 1.22) 1.18 (0.95, 1.45) 1996 1.07 0.94 (0.79, 1.13) 0.76 (0.55, 1.05) 1.02 (0.83, 1.27) 1997 0.93 0.82 (0.69, 0.98) 0.80 (0.57, 1.11) 0.82 (0.66, 1.01) Sex Male 1.0 1.0 1.0 (-) 1.0 (-) Female 1.22 1.15 (1.04, 1.28) 1.24 (1.03, 1.50) 1.11 (0.98, 1.25) Note: OR = odds ratio, CI = confidence interval *Adjusted for all other listed predictors Among all advanced cancer patients, additional factors predictive of receiving at least one FP home visit included a longer length of survival (for 121 to more than 180 days survival: adjusted OR 4.04; 95%CI 3.45, 4.72), admission to the QEII Palliative Care Program (PCP) (adjusted OR 2.25, 95%CI 1.97, 2.56), older age (for those 75 years and older: adjusted OR 1.41; 95%CI 1.24, 1.61) and being female (adjusted OR 1.15; 95%CI 1.04, 1.28). Patients who were in LTC at some point during their survival period (adjusted OR 0.55; 95%CI 0.46, 0.65) and those who spent 32 or more days in hospital compared to none (adjusted OR 0.80; 95%CI 0.67, 0.95) tended to be less likely to receive at least one FP home visit. Over time, the likelihood of receiving a home visit tended to decline. However, after accounting for all other predictors in the model, year of death was not a major factor. Cancer cause of death, receipt of physician specialty visits and undergoing palliative radiation were not associated with home visit receipt in the final multivariate model. Discussion The neighbourhood income of those dying of cancer is associated with the likelihood of receiving a home visit during the end-of-life by a family physician in Nova Scotia. The association found, however, appears to be modified by region of residence for those who died of cancer. It appears that income plays less of a role in predicting home visits by a family physician for those who live in the larger, urban centre of Halifax Regional Municipality. Given the finding that patients followed by the QEII Palliative Care Program are also more likely to receive family physician home visits, we speculate that the urban centre may provide a collaborative 'team-care' advantage to cancer patients. The publicly-funded PCP may act to equalize the opportunity to stay at home and facilitate family physician home involvement in ways rural locations may not be able to. In previous work, income was not associated with location of death but region of residency was[ 2 ]. We have also found that those who live in higher income areas tend to use the emergency department less[ 24 ]. In the United Kingdom, Aylin and colleagues[ 26 ] found, for the general population, those in social class 1 (highest income) received the fewest home visits. Their study also revealed a dose-response relationship in that as one moves to lower income class, the more likely one is to have received a home visit. Our study shows some gradient element, albeit in the opposite direction, but not as clearly. McNiece and Majeed found home visiting rates among patients with highest income to be half that of those with the lowest income[ 27 ]. Their study results and ours were adjusted for age and sex, such that the relationships between income, age and sex cannot be confounding the results. Aylin postulates that the reason for greater home visiting among those with lower income may be due to a number of factors including increased morbidity, poorer access to a car, and differing expectations of the services supplied by their general practitioners[ 26 ]. Such factors should also hold true for cancer patients. Nevertheless, our findings are opposite to those of Aylin. We hypothesize that when it comes to routine home visits for brief, episodic illnesses, the home visiting trend may be as Aylin suggests. However, for those who are at home and looking to stay at home with advanced cancer, there are more substantial financial issues driving whether this is likely to happen or not. In Nova Scotia, as in many Canadian settings, there is access to home visiting nurses and some other health professionals through the publicly-funded health system. However, as disease progresses, the ability of a family to support death at home depends on many other factors. These include the presence of a family member who can stay at home, the ability of a family member to manage the medications and symptom assessment along with health professionals, the cost of drugs which are paid for in hospital but not in the home (unless the patient has private health insurance or is 65 or more years old), the cost of equipment in the home, and possibly the cost of additional nursing or personal care workers in the home (variably covered by the public system, and only sometimes covered by private insurance)[ 17 ]. All of these factors point to the fact that those with greater income would be more likely to succeed at staying at home[ 7 ]. In addition, in more rural settings there is less access to specialized services such as palliative care programs. The home visits provided by family physicians may therefore be a direct response to the other capacities of patients and families to stay at home rather than being the critically enabling feature. Thus, the home visit is essential but not sufficient and without the rest of the support required, the patient will not be able to stay at home, thereby resulting in fewer home visits. Another interpretation is that those with lower incomes may actually make different choices about how they wish to receive health care and where they would like to spend their last days. Depending on the study cited, up to 80% of those with advanced cancer wish to spend their last days at home[ 7 , 28 - 30 ]. Grande and colleagues reported that those who lived in higher socioeconomic areas were more likely to die at home than their counterparts[ 31 ]. Sims found that those with cancer from social class IV and V (semi-skilled and unskilled occupations) were under-represented in deaths that occurred in hospices and homes when compared to those in social class I and II (professional occupations and managerial/technical occupations)[ 32 ]. All of this may be supported by any one or a combination of factors such as less desire to remain at home, less capacity (financial or otherwise) to remain at home or bias in the delivery of health service by professionals. Our study is the first we know of to show that the number of home visits made by family physicians to those at the end-of-life is also less for these individuals. Home visiting has long been an element of continuity of care across settings (office, hospital, home, nursing home) provided by family physicians. Some would argue that home visits may be influenced by the geography in which the physician operates daily. As a result, physician travel patterns to and from the office (when home visiting often occurs) may not take them through low income neighbourhoods, thus reducing the likelihood of a visit. The work of Aylin[ 26 ] and McNiece[ 27 ], however, does not support this. New initiatives are underway in Canada which may provide more opportunities for the enhanced presence of a range of health professionals in the homes of the dying. In response to the Romanow Commission[ 33 ], the federal government of Canada has initiated agreement with the provinces for them to provide coverage for enhanced home-based end-of-life care. In the future, we may see more nurses or nurse-practitioners making home visits as part of the community-based care team along with family physicians[ 34 ]. In rural or remote areas where there is a scarcity of family physicians, nurses and nurse practitioners with advanced assessment skills may play an even greater role. Our study has limitations. As we are using routinely collected data used for administrative and billing purposes there may be biases operating. The data reflect those family physicians who bill for the services they provide. It should also reflect the "shadow-billing" of those on alternate payment mechanisms (estimated to be less than five per cent of family physicians at the time of the study) but in reality, these physicians may have less incentive to capture these fee codes and so may under-report home visiting slightly. The data file used in this study was originally created for an alternate project looking at health service utilization among patients who died due to lung, colorectal, breast or prostate cancer. We were therefore limited to examining home visits provided to these patients only and are not able to report whether the use of family physician home services among those who died due to all other cancer causes is similar or different. We are unable to adjust for homecare utilization (data did not exist for the study years), family member caregiving status (no data available) or account for additional insurance coverage (above provincial) which may have covered additional costs associated with drugs, home nursing, home equipment, etc. Our attempt to account for service availability by region is crude. HRM is more homogeneous with respect to services than our combination all other regions outside of HRM; however, the effect evidenced may, therefore, be a conservative estimate. Conclusions And so, in conclusion, it appears that even in death those with fewer financial resources may be less likely to achieve the same access to health services as those better off. What does this mean for our care of the dying? We must examine carefully which elements are the cause of this inequality. If there is less desire among those with more financial barriers, we need to examine the origins of these desires. Is it fear of caring for those at home? Is it an established culture of caring to move loved ones to hospital at the end-of-life? These issues need identification if we are to support such families. If there truly are financial barriers to such things as drugs, equipment and personnel then we must redefine health policy to make these more accessible to those with fewer financial resources. And finally, we should attempt to understand whether or not there is any bias operating on the part of health professionals. All of these need further research if we are to ensure patients receive the care where they wish to as they approach death. Competing interests The author(s) declare that they have no competing interests. Authors' contributions FB and BL participated in the conceptualization and design of the project, the analysis and interpretation of the data, first-drafted the majority of the article, and incorporated co-author comments into the final draft. GJ participated in the interpretation of the data and the drafting and revising of the manuscript. All authors gave approval to this final version. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC551519.xml |
515367 | Population History and Natural Selection Shape Patterns of Genetic Variation in 132 Genes | Identifying regions of the human genome that have been targets of natural selection will provide important insights into human evolutionary history and may facilitate the identification of complex disease genes. Although the signature that natural selection imparts on DNA sequence variation is difficult to disentangle from the effects of neutral processes such as population demographic history, selective and demographic forces can be distinguished by analyzing multiple loci dispersed throughout the genome. We studied the molecular evolution of 132 genes by comprehensively resequencing them in 24 African-Americans and 23 European-Americans. We developed a rigorous computational approach for taking into account multiple hypothesis tests and demographic history and found that while many apparent selective events can instead be explained by demography, there is also strong evidence for positive or balancing selection at eight genes in the European-American population, but none in the African-American population. Our results suggest that the migration of modern humans out of Africa into new environments was accompanied by genetic adaptations to emergent selective forces. In addition, a region containing four contiguous genes on Chromosome 7 showed striking evidence of a recent selective sweep in European-Americans. More generally, our results have important implications for mapping genes underlying complex human diseases. | Introduction Despite intense study and interest, a detailed understanding of the evolutionary and demographic forces that have shaped extant patterns of human genomic variation remains elusive. An important goal in studies of DNA sequence variation is to identify loci that have been targets of natural selection and thus contribute to differences in fitness between individuals in a population. Identifying regions of the human genome that have been subject to natural selection will provide important insights into recent human history ( Sabeti et al. 2002 ; Tishkoff and Verrelli 2003 ), the function of genes ( Akey et al. 2002 ), and the mechanisms of evolutionary change ( Otto 2000 ), and it may also facilitate the identification of complex disease genes ( Jorde et al. 2001 ; Nielsen 2001 ). The neutral theory of molecular evolution ( Kimura 1968 ; King and Jukes 1969 ), which posits that the majority of polymorphisms have no appreciable effects on fitness, has been integral to recent studies of natural selection. Specifically, the neutral theory makes explicit and quantitative predictions about the amount, structure, and patterns of sequence variation expected under neutrality, and serves as a null hypothesis by which to evaluate the evidence for or against selection in empirical data ( Otto 2000 ; Nielsen 2001 ). Unfortunately, robust inferences of natural selection from DNA sequence data are difficult because of the confounding effects of population demographic history. For example, both positive selection and increases in population size lead to an excess of low-frequency alleles in a population relative to what is expected under a standard neutral model (i.e., a constant-size, randomly mating population at mutation-drift equilibrium). Therefore, rejection of the standard neutral model usually cannot be interpreted as unambiguous evidence for selection. One way out of this conundrum is to recognize that population demographic history affects patterns of variation at all loci in a genome in a similar manner, whereas natural selection acts upon specific loci ( Cavalli-Sforza 1966 ; Przeworski et al. 2000 ; Andolfatto 2001 ; Nielsen 2001 ). Thus, by sampling a large number of unlinked loci throughout the genome, it is in principle possible to distinguish between selection and demography. For instance, Akey et al. (2002) recently used this approach to infer the presence of selection in a genome-wide collection of single nucleotide polymorphisms (SNPs). However, studies based on SNPs that were initially identified in a small sample and subsequently genotyped in a larger sample are not ideally suited for detecting selection, because ascertainment bias (i.e., a systematic bias introduced into a dataset because of the way in which the data were collected) complicates downstream analyses ( Akey et al. 2003 ). However, DNA sequence data provides the opportunity to exhaustively catalog variation, which attenuates the problem of ascertainment bias and therefore is arguably the most powerful and direct approach for detecting selection. Here, we describe an extensive analysis of the molecular evolution of 132 genes that were comprehensively resequenced in 24 African-Americans and 23 European-Americans. In total, over 2.5 Mb of baseline reference DNA was sequenced, spanning 20 autosomal chromosomes and the X chromosome. The sampling of a large number of loci dispersed throughout the genome has allowed us to clarify the relative contributions of demography and selection to patterns of genetic variation at individual genes. Specifically, we developed a rigorous computational approach for taking into account multiple hypothesis tests and demographic history, and we found that while many apparent selective events can instead be explained by demography, there is also strong evidence for positive or balancing selection at eight genes in the European-derived population. In addition, we describe a striking example of a previously unreported recent selective sweep in European-Americans that spans four contiguous genes on Chromosome 7. More generally, our data provide insight into the demographic histories of African-American and European-American populations and have important implications for genetic association studies of complex diseases, as several of the genes showing evidence of selection have been implicated in susceptibility to complex human diseases. Results Statistical Tests Reveal Many Deviations from Neutrality We resequenced 132 genes primarily involved in inflammation, blood clotting, and blood pressure regulation and discovered a total of 12,890 SNPs ( Table S1 ). We first characterized patterns of genetic variation by calculating several common summary statistics of the within-population allele frequency distribution, including Tajima's D, Fu and Li's D*, Fu and Li's F*, and Fay and Wu's H. As is conventionally done, we initially determined whether these statistics were significantly different from what is expected under a standard neutral model by performing coalescent simulations under the simplifying assumption of no recombination. In total, 28 genes in the European-American sample and ten genes in the African-American sample were nominally significant (i.e., the observed test statistic differed from neutral expectations at p < 0.05) in one or more tests of the allele frequency distribution ( Figure 1 ). Thus, the European-American sample contained nearly three times as many significant genes as the African-American sample, and only three genes were significant in both samples (ABO, IL1RN, and TNFRSF1B) . Figure 1 Scatter Plot of Neutrality Test Statistics in European- and African-Americans Genes that are nominally significant ( p < 0.05) in European-Americans (EA), African-Americans (AA), or both populations are denoted by red, blue, and green circles, respectively. Genes that are not significant are shown as black dots. Two-sided tests were used for Tajima's D, Fu and Li's D*, and Fu and Li's F*, and a one-sided test was used for Fay and Wu's H. The direction of Tajima's D, Fu and Li's D*, and Fu and Li's F* is potentially informative about the evolutionary and demographic forces that a population has experienced. For example, negative values reflect an excess of rare polymorphisms in a population, which is consistent with either positive selection or an increase in population size. Positive values indicate an excess of intermediate-frequency alleles in a population and can result from either balancing selection or population bottlenecks. In the European-American sample, we observed eleven significantly positive and five significantly negative values for one or more of these three test statistics ( Figure 1 ). In the African-American sample, we observed two significantly positive and five significantly negative values for one or more of the test statistics ( Figure 1 ). The observations of both significantly positive and significantly negative values of Tajima's D, Fu and Li's D*, and Fu and Li's F*, combined with the largely nonoverlapping set of significant genes, could reflect selective pressures unique to one population (i.e., local adaptation), different demographic histories, spurious results, or most likely some complex combination of all of these factors. Although these results are intriguing, their interpretation is confounded by two issues: (1) We have not corrected for multiple hypothesis tests, and (2) rejection of the standard neutral model can result from either selective or demographic forces. In the subsequent sections, we develop approaches to address these issues with the dual goals of identifying genes that possess strong evidence of natural selection and of inferring population demographic history. Correcting for Multiple Hypothesis Tests In order to robustly correct for multiple hypothesis tests, the conventional practice of assuming no recombination when determining significance is not appropriate, because it results in conservative p values ( Wall 1999 ) and hence decreases the statistical power to detect deviations from neutrality. Although recombination can easily be incorporated into coalescent simulations, in practice it is difficult to accurately estimate recombination rates, which vary substantially across the genome ( Yu et al. 2001 ; McVean et al. 2004 ). To model the stochastic behavior and uncertainty in local rates of recombination, we reassessed the significance of Tajima's D, Fu and Li's D*, Fu and Li's F*, and Fay and Wu's H by coalescent simulations that incorporate recombination rates sampled from a Gamma(2, 0.5 × 10 –8 ) distribution (see Materials and Methods ). Finally, we corrected each statistic for multiple tests using the positive false discovery rate (FDR; Storey 2002 ) method, which determines the predicted proportion of “false positives” for the number of significant observations. In the European-American sample, we observed 22 genes that were significant at a FDR of 5% (i.e., we expect approximately one false positive in this set of genes) for one or more tests of the allele frequency distribution ( Tables 1 and S2 ). Thus, the number of significant genes in the European-American sample, after incorporating recombination and correcting for multiple tests, is very similar to the initial results where recombination was ignored and multiple tests were not corrected for. However, in the African-American sample there were no genes significant at a FDR of 5% for any of the tests of the allele frequency distribution (unpublished data). This result is consistent with the relatively small number of significant genes that were initially found before correcting for multiple tests ( Figure 1 ). Genes with the smallest FDR in African-Americans were ABO, F2RL1, and IL17B, which each had a FDR of 13.5% for Fu and Li's D*. Table 1 Significant Genes in European-Americans after Correcting for Multiple Tests D, D*, F*, and H denote Tajima's D, Fu and Li's D*, Fu and Li's F*, and Fay and Wu's H, respectively. Nominal p values determined from 10 4 coalescent simulations with recombination are shown in the column next to each statistic. The p values that are significant after correcting for multiple tests (FDR = 5%) are shown in bold Distinguishing between Selective and Demographic Forces Although neutrality tests of the allele frequency distribution reveal many significant deviations, it is impossible to unambiguously interpret these data as evidence for natural selection, because the null model used to assess significance makes unrealistic assumptions about population demographic history. In principle, it is possible to distinguish between demography and selection, because demography affects all loci in the genome, whereas selection acts upon specific loci. Thus, by sampling a large number of loci dispersed throughout the genome, we can begin to construct a more realistic null hypothesis by which to evaluate the evidence for or against selection ( Kreitman 2000 ). To this end, we used the empirical data to explore four different demographic models ( Figure 2 A), which we could then use to account for demographic influences on tests of natural selection. For each model, we used coalescent theory to simulate data over a broad range of parameters and identified the particular combination of parameters that most closely matched summary statistics (average Tajima's D, Fu and Li's D*, and Fu and Li's F*) of the observed data. Of the four demographic models, the European-American data are most consistent with a bottleneck occurring approximately 40,000 y ago, which is nearly identical to a previously reported estimate ( Sabeti et al. 2002 ). However, the confidence intervals for the observed summary statistics are broad, and various aspects of the data are also consistent with other models ( Figure 2 B). The African-American data are most consistent with either an exponential expansion or a relatively old and severe bottleneck ( Figure 2 ). Similarly, using DNA sequence variation from ten unlinked, noncoding loci, Pluzhnikov et al. (2002) found that an African Hausa sample was consistent with a recent population expansion (although they did not consider bottleneck models). Figure 2 Summary of the Four Demographic Models Considered in Each Population (A) Schematic diagram of each demographic model and its associated parameters (see Materials and Methods for details). Parameter values that match the observed data most closely for European-Americans (EA) and African-Americans (AA) are shown below the diagrams. (B) Average and 95% confidence intervals of Tajima's D (blue bars), Fu and Li's D* (red bars), and Fu and Li's F* (pale yellow bars) for the observed data and each demographic model (using the parameters that most closely match the empirical data). Results from the standard neutral model (Constant) are also shown. We reestimated the significance of Tajima's D, Fu and Li's D*, Fu and Li's F*, and Fay and Wu's H in each population for each of the four demographic models using the best-fit parameter values. All simulations included recombination and correction for multiple tests using the FDR method (with a FDR of 5%) as described above. Population history can clearly have a profound effect on tests of natural selection ( Figure 3 A and 3 B; see also Simonsen 1995 ; Przeworski 2002 ), and given the uncertainty in our knowledge of human demographic history, it is challenging to ascribe unusual patterns of genetic variation to either demography or selection. To address this problem, we identified genes whose statistical evidence for selection was robust to demographic history. We conservatively defined demographically robust selection genes as those that demonstrated significant evidence for selection in all five demographic models. We identified eight demographically robust selection genes in European-Americans, and zero in African-Americans ( Figure 3 C; Table 2 ). Thus, out of the 22 genes originally found to be significant (at a FDR of 5%) under a standard neutral model, our estimates suggest that demographic history can potentially account for approximately two-thirds of these observations. Figure 3 The Influence of Demographic History on Tests of Selection (A and B) The significance of observed values of Tajima's D (red), Fu and Li's D* (pale yellow), Fu and Li's F* (pale blue), and Fay and Wu's H (dark blue) were reassessed for each best-fit demographic model in European-Americans (A) and African-Americans (B). Results from the standard neutral model (Constant) are shown for comparison. The number of significant genes for each demographic model is noted above each category in (A) and (B). For example, there were a total of 19 significant test statistics across all four tests of neutrality assuming a bottleneck model for Europeans, which define ten unique genes. Therefore, each gene is supported by approximately two (19/10) tests of neutrality. (C) The distribution of the number of significant genes across the five demographic models in European-Americans and African-Americans. For example, in European-Americans, 40 genes were significant in at least one of the demographic models, and 27 genes were significant in at least two of the demographic models. Table 2 Demographically Robust Selection Genes in European-Americans Biological Process terms were assigned using the Panther classification scheme ( Thomas et al. 2003 ) Evidence for a Recent Selective Sweep on Chromosome 7q in European-Americans One particularly interesting region of the genome is located at 7q and contains four contiguous demographically robust selection genes ( EPHB6, TRPV6, TRPV5, and KEL; Figure 4 A). Collectively, the entire 115-kb region bears many of the hallmarks of a locus subject to a recent selective sweep: an excess of high-frequency-derived alleles ( Figure 4 B); an overall excess of rare polymorphisms, which results in an extreme skew of the site frequency spectrum reflected by sharply negative values of Tajima's D ( Figure 4 C); and a significant reduction in the amount of nucleotide diversity ( Figure 4 D). The signature of positive selection is seen only in European-Americans, suggesting that EPHB6, TRPV6, TRPV5, and/or KEL possess specific alleles that have conferred local adaptation to a unique environmental pressure in European-derived populations. Consistent with this hypothesis, we observed strong levels of population subdivision ( Figure 4 E) across the entire 115-kb region. The closest genes centromeric to EPHB6 and telomeric to KEL are approximately 42 kb and 64 kb away, respectively, suggesting that one or more of these four genes is the target of selection. However, we have not surveyed patterns of DNA sequence variation outside of the region delimited by EPHB6 and KEL, and therefore it is possible that the signature of selection extends even further. Based on the level of genetic variation on the putatively selected haplotype (see Materials and Methods ), we can provide a rough estimate of the time back to the selective sweep as approximately 10,000 y ago. Although this number should be interpreted cautiously, it suggests that selection operated recently. Figure 4 A Strong Signature of Positive Selection Spanning 115 kb on Chromosome 7q (A–D) Exons for EPHB6, TRPV6, TRPV5, and KEL are shown as gray vertical lines. A dashed black line indicates the boundary between EPHB6 and TRPV6 exons, which are approximately 1 kb apart. Transcriptional orientation is indicated by the arrows below exon positions. SNPs found in European-Americans and African-Americans are shown below. Noncoding, synonymous, and nonsynonymous SNPs are denoted as black, blue, and red vertical bars, respectively. The positions of three nonsynonymous SNPs in TRPV6 are shown with asterisks. For each of the resulting nonsynonymous amino acid changes, the most frequent amino acid in European-Americans is given first. The frequency of derived alleles, P D (B), sliding window plots of Tajima's D (C), and nucleotide diversity, π (D), are shown across the entire region. Gaps in the sliding window plots indicate positions where sequence data were not obtained. In (B–D), European- and African-American data are shown in red and black, respectively. (E) The distribution of F ST across the 115-kb region. The average F ST for all SNPs across the 132 genes is shown as a dashed red line. The dashed green line indicates the threshold for significantly ( p < 0.01) large values of F ST , determined by coalescent simulations. Discussion In summary, we have found that both population demographic history and natural selection shaped patterns of DNA sequence variation in the 132 genes studied here. By studying multiple unlinked loci dispersed throughout the genome, we were able to develop a rigorous computational approach to distinguish between the confounding effects of natural selection and demographic history on patterns of genetic variation. Using this strategy, we found that approximately two-thirds of the genes that were initially significant could be accounted for by population demographic history. Thus, our analyses clearly demonstrate the importance of considering both neutral and nonneutral forces when interpreting DNA sequence variation. An interesting feature of our data is that the majority of deviations from neutrality, and all of the demographically robust selection genes, are not shared between the two population samples, suggesting that local adaptation has played an important role in recent human evolutionary history. Consistent with this observation, several possible examples of local adaptation in humans have previously been reported ( Stephens et al. 1998 ; Rana et al. 1999 ; Hollox et al. 2001 ; Tishkoff et al. 2001 ; Currat et al. 2002 ; Fullerton et al. 2002 ; Gilad et al. 2002 ; Hamblin et al. 2002 ; Rockman et al. 2003 ). We hypothesize that the stronger signature of selection in the European-derived population may reflect the exposure of non-African populations to novel and evolutionarily recent selective pressures (e.g., unique dietary, climatic, and cultural environments) as modern humans migrated out of Africa and spread throughout the world. In contrast, the African-derived population may have experienced fewer evolutionarily recent selective forces. Theoretical studies have demonstrated that the power to detect a selective sweep is generally greatest if it occurred less than approximately 0.1 N e generations ago (i.e., approximately 20,000–25,000 y ago [ Kim and Stephan 2000 ; Przeworski 2002 ]), which is consistent with our hypothesis that signatures of selection in European-Americans reflect recent selective events. However, it is important to note that we have surveyed less than 1% of all human genes, and many of the genes that we did analyze are involved in mediating inflammatory and immune responses; thus our results may not be representative of the genome at large. Interestingly, Glinka et al. (2003) found that European-derived populations of Drosophila melanogaster demonstrated abundant evidence for recent selective sweeps, whereas African populations did not, which is strikingly similar to our results in humans. An alternative explanation for why we observed fewer significant results in African-Americans than in European-Americans is that African-Americans are an admixed population ( Parra et al. 1998 ), and the admixture process may mask the signature of selection. However, simulation studies in which we constructed an artificially admixed European-American sample with African-American chromosomes resulted in an increase in significant genes relative to the observed data (unpublished data). Therefore, to the extent that our simulations recapitulate the dynamics of the admixture process in African-Americans, admixture is unlikely to explain the discrepancies between the two samples. It is important to point out that some genes that do not meet our rigorous definition of a high-confidence selection gene may have nonetheless been targets of selection, such as ABO in African-Americans ( Table S2 ). In this initial survey we have elected to be conservative and identify genes that possess the strongest signatures of selection. Ultimately, it will be necessary to confirm our results in geographically diverse populations (a more comprehensive sampling of African populations is particularly needed), as well as in replicate samples of the populations we studied, and to functionally characterize the suspected targets of selection. Recently, Clark et al. (2003) presented an evolutionary analysis of 7,645 orthologous human-chimp-mouse gene trios by looking for accelerated rates of synonymous and nonsynonymous nucleotide substitution in either the human or the chimp lineages. In total, 50 genes overlap between our dataset and theirs ( Table S3 ), including three demographically robust selection genes ( TRPV6, EPHB6, and DCN; see Table 2 ). All three of the demographically robust selection genes also demonstrate statistically significant evidence ( p < 0.05) of accelerated evolution in either the human (TRPV6 and EPHB6) or chimp (DCN) lineage. In addition, Clark et al. (2003) found evidence for accelerated evolution in seven genes along the human lineage that did not demonstrate evidence for selection in our dataset ( Table S3 ). This observation may simply reflect either false negatives in our analysis or false positives in Clark et al. (2003) . However, it is important to note that the statistical methods and data used to detect selection in Clark et al. (2003) (divergence between species) are quite different from our methods (polymorphism within species), so completely overlapping results are not expected. More specifically, the analyses of Clark et al. (2003) will preferentially detect selective events between species, whereas our analyses will preferentially identify selection operating within species. In other words, these two methods are complimentary and may potentially detect selection operating over different time scales. In this respect, it is particularly interesting that the genes we identified as possessing the strongest evidence for recent selection in one human population also show evidence of selection in the human or chimp lineage following their divergence ( Clark et al. 2003 ). The strongest signature of selection that we observed occurs on Chromosome 7q in European-Americans. The signature of selection extends for at least 115 kb and spans the genes EPHB6, TRPV6, TRPV5, and KEL . To our knowledge, this is the largest footprint of selection that has been described in the human genome, and likely reflects the combination of strong and recent selective pressures and reduced recombination in this region (the average ratio of genetic to physical distance, cM/Mb, is approximately 0.68 according to the deCode map). Based on our current data it is impossible to identify which gene (or perhaps genes) has been the target (or targets) of selection. However, TRPV6 is a particularly interesting candidate, as it possesses three nonsynonymous amino acid substitutions (C157R, M378V, and M681T) that are each nearly fixed for the derived allele in European-Americans, show significant frequency differences between European-Americans and African-Americans, and are located in the most significant regions of both Tajima's D and reduced nucleotide diversity ( Figure 4 ). The program PolyPhen ( Ramensky et al. 2002 ) predicts that the C157R replacement may alter protein structure. Recently, TRPV6 was shown to be up-regulated in prostate cancer ( Wissenbach et al. 2001 ), and a susceptibility locus for aggressive prostate cancer was mapped to the TRPV6 region (7q31–33; Paiss et al. 2003 ). These observations, combined with the large difference in disease prevalence between Europeans and African-Americans ( Crawford 2003 ), make TRPV6 a strong candidate gene for prostate cancer susceptibility and/or aggressiveness. TRPV6, as well as TRPV5, constitute the rate-limiting step in kidney, intestine, and placenta calcium absorption ( Nijenhuis et al 2003 ; van de Graaf et al. 2003 ). Interestingly, Northern European populations have very high frequencies of the lactase persistence allele ( LCT*P; Hollox et al. 2001 ), which allows digestion of fresh milk throughout adulthood. It is widely accepted that strong selection has driven LCT*P to high frequency in Northern Europeans, beginning sometime after the domestication of animals approximately 9,000 y ago ( Feldman and Cavalli-Sforza 1989 ; Hollox et al. 2001 ; Bersaglieri et al. 2004 ). What has been debated, however, is the specific selective advantage conferred by lactase persistence ( Holden and Mace 1997 ). Our finding that TRPV6 and/or TRPV5 have been under strong selective pressure in Northern Europeans suggests that increased calcium absorption may have been the driving force behind selection for lactase persistence, which was originally hypothesized by Flatz and Rotthauwe (1973) . Although additional studies are clearly needed, our results provide additional insight into the molecular mechanisms of adaptation to a new dietary niche (i.e., high-lactose diets). More generally, our results have several implications for mapping genes underlying complex human diseases. Specifically, four of the high-confidence selection genes have been implicated in various complex diseases ( Table 3 ). If genes underlying complex diseases have experienced differential selective pressures, then this could in part explain the failure of many studies to replicate disease associations across populations ( Florez et al. 2003 ; Moore 2003 ). Finally, our data are consistent with the notion that variation in genes that was once beneficial may have become detrimental in the environmental and cultural milieu of contemporary human populations, akin to the “thrifty gene” hypothesis for type II diabetes ( Neel 1962 ). Table 3 Disease Associations with Demographically Robust Selection Genes Materials and Methods DNA samples and sequencing Human DNAs were obtained from the Coriell Institute (Camden, New Jersey, United States). We analyzed DNA from 24 African-Americans from the Human Variation Panel, African-American Panel of 50 (HD50AA) and DNA from 23 European-Americans derived from various CEPH pedigrees. We also sequenced each gene in a common chimpanzee ( Pan troglodytes ) to determine the derived allele for Fay and Wu's H test. These data were generated under the auspices of the SeattleSNPs Program for Genomic Applications, which resequences candidate genes involved in inflammatory processes in humans. In general, we resequenced the complete genomic region for each gene, including introns and approximately 2 kb 5′ of the gene and 1 kb 3′ of the gene using Big-Dye terminator chemistry on an ABI 3700 or ABI 3730XL (Applied Biosystems, Foster City, California, United States). For several exceptionally large genes, such as F13A1, less than complete coverage was obtained (see Table S1 ). All variants occurring once in the sample were confirmed with an additional sequencing run. Further experimental details and all of the raw data can be found at our website ( http://pga.gs.washington.edu/ ). Data analysis We calculated the following summary statistics of nucleotide variation for each gene: θ^= S/a n , where S is the number of segregating sites, and n is the sample size ( Watterson 1975 ); , where h i is an unbiased estimate of nucleotide diversity for the i th segregating site (see equation 12 in Tajima 1989 ) and η S , which is the number of singletons ( Fu and Li 1993 ). From these statistics we calculated several tests of the standard neutral model including Tajima's D ( Tajima 1989 ), Fu and Li's D* ( Fu and Li 1993 ), Fu and Li's F* ( Fu and Li 1993 ), and Fay and Wu's H statistic ( Fay and Wu 2000 ). In calculating Fu and Li's F*, we used the formulas provided in Simonsen et al. (1995) , which correct a typographical error in the original description of the method ( Fu and Li 1993 ). For a discussion of the similarities and differences of Tajima's D, Fu and Li's D*, Fu and Li's F*, and Fay and Wu's H, see Fu and Li (1993) , Simonsen et al. (1995) , and Przeworski (2002) . We initially assessed the significance of these statistics by comparing the observed values to 10 4 coalescent simulations ( Hudson 1983 ), conditional on the observed sample size and number of segregating sites, assuming a standard neutral model with no recombination. Coalescent simulations were performed using the program ms (obtained from R. Hudson's Web site [ http://home.uchicago.edu/~rhudson1/source.html ]). In order to correct for multiple tests, we repeated the coalescent simulations as described above, but included recombination. Following Pluzhnikov et al. (2002) , for each of the 10 4 coalescent realizations, we sampled the recombination rate from a Gamma(2, 0.5 × 10 –8 ) distribution whose expectation equals the average genome-wide recombination rate of 10 –8 /generation ( Hamblin et al. 2002 ). The positive FDR method was used to correct for multiple hypothesis tests using the software QVALUE ( Storey 2002 ; http://faculty.washington.edu/~jstorey/qvalue/ ). We quantified the allele frequency differences between the European- and African-American samples by the statistic F ST as described in Akey et al. (2002) . All of the analyses described above excluded insertion/deletion polymorphisms, but their inclusion does not affect any of our conclusions (unpublished data). We assigned PANTHER Biological Process terms ( Thomas et al. 2003 ) to each gene. We estimated the time since the selective sweep for the Chromosome 7q region in European-Americans by analyzing the amount of nucleotide diversity that has accumulated on the selected haplotype as described in Rozas et al. (2001) . We assumed that TRPV6 is the target of selection and the selected haplotype is defined by the C157R, M378V, and M681T polymorphisms. If mutations are Poisson-distributed, the expected number of segregating sites in a genealogy is E [ S ] = μE [ T ], where S, μ, and T denote segregating sites, neutral mutation rate of the locus, and total branch length of the genealogy, respectively. Assuming a star-shaped genealogy, E [ T ] = n × t , where n is the number of selected haplotypes. Thus, the time back to the selective sweep, t, can be estimated by S /( nμ ). For TRPV6 in European-Americans, n = 45 (i.e., 45 out of 46 haplotypes carry C157, M378, and M681), S = 11, and μ = 2.5 × 10 –5 . Demographic modeling We assessed the impact of demographic history on the robustness of the statistical tests of neutrality by using coalescent theory to simulate data under four different population histories, including a bottleneck, exponential expansion, population structure according to an island model that allows symmetric migration between demes, and population structure assuming population splitting with no subsequent migration. For each model we simulated data under a wide variety of parameters by conditioning on the observed sample size and θ^ W for each population. The bottleneck model is specified by the parameters F (the inbreeding coefficient) and t (the time in years measured from the present) at which the bottleneck occurred. Values of F and t considered were F = [0.05, 0.075, … , 0.40] and t = [10,000, 20,000, … , 100,000]. The exponential expansion model is determined by the parameters α (the growth rate/generation) and t (the time, in years measured from the present, at which the population began increasing in size). Values considered for α and t were: α = [0.0005, 0.001, … , 0.01] and t = [10,000, 20,000, … , 100,000]. The population structure under an island model is specified by the population migration rate between demes, M = 4 N o m, where N o and m are the effective subpopulation size and fraction of migrants in each subpopulation per generation, respectively. Values of M considered were M = [1, 2, … , 10]. The structure model assuming population splitting with no subsequent migration is determined by the parameter t (the time in years since the populations diverged). Values of t considered were t = [1,000, 2,000, …, 10,000]. In all simulations we assumed an effective population size of 10,000 and a generation time of 25 y in order to facilitate comparisons to a previous study ( Sabeti et al. 2002 ). The parameter space for each model included a full grid search, so we tested 160, 100, 10, and 10 parameter combinations for the bottleneck, expansion, structure (island), and structure (splitting) models, respectively. We performed 10 4 simulations for each parameter combination. For each demographic model, we calculated the average value of Tajima's D, Fu and Li's D*, and Fu and Li's F* and compared the results to the observed values of these statistics. For the bottleneck and exponential expansion models, we identified the parameter values that most closely matched the observed data by identifying the parameter combination that minimized the function , where T Oi and T Si denote the observed and simulated averages of Tajima's D, Fu and Li's D*, and Fu and Li's F*. For the demographic models of population structure we selected parameter values that matched the observed F ST . Finally, we reassessed the significance of the observed values of Tajima's D, Fu and Li's D*, Fu and Li's F*, and Fay and Wu's H by 10 4 coalescent simulations for each demographic model using the best-fit parameter values. Supporting Information Table S1 Summary Statistics of the 132 Genes (266 KB DOC). Click here for additional data file. Table S2 Neutrality Test Statistics (534 KB DOC). Click here for additional data file. Table S3 Overlap of Genes Analyzed by Clark et al. (2003) (87 KB DOC). Click here for additional data file. Accession Numbers LocusLink ID numbers ( http://www.ncbi.nlm.nih.gov/LocusLink/ ) for the genes discussed in this paper are ABO (28), ACE2 (59272), APOH (350), BDKRB2 (624), BF (629), C2 (717), CCR2 (1231), CD36 (948), CEBPB (1051), CRF (10882), CRP (1401), CSF2 (1437), CSF3 (1440), CSF3R (1441), CYP4A11 (1579), CYP4F2 (8529), DCN (1634), EPHB6 (2051), F10 (2159), F11 (2160), F12 (2161), F13A1 (2162), F2 (2147), F2R (2149), F2RL1 (2150), F2RL2 (2151), F2RL3 (9002), F3 (2152), F5 (2153), F7 (2155), F9 (2158), FGA (2243), FGB (2244), FGG (2266), FGL2 (10875), FSBP (10646), GP1BA (2811), ICAM1 (3383), IFNG (3458), IGF2 (3481), IGF2AS (51214), IL10 (3586), IL10RA (3587), IL10RB (3588), IL11 (3589), IL12A (3592), IL12B (3593), IL13 (3596), IL15RA (3601), IL17B (27190), IL19 (29949), IL1A (3552), IL1B (3553), IL1R1 (3554), IL1R2 (7850), IL1RN (3557), IL2 (3558), IL20 (50604), IL21R (50615), IL22 (50616), IL24 (11009), IL2RB (3560), IL3 (3562), IL4 (3565), IL4R (3566), IL5 (3567), IL6 (3569), IL8 (3576), IL9 (3578), IL9R (3581), IRAK4 (51135), ITGA2 (3673), ITGA8 (8516), JAK3 (3718), KEL (3792), KLK1 (3816), KLKB1 (3818), KNG (3827), LTA (4049), LTB (4050), MAP3K8 (1326), MC1R (4157), MMP3 (4314), MMP9 (4318), NOS3 (4846), PFC (5199), PLAT (5327), PLAU (5328), PLAUR (5329), PLG (5340), PON1 (5444), PON2 (5445), PPARA (5465), PPARG (5468), PROC (5624), PROCR (10544), PROS1 (5627), PROZ (8858), PTGS2 (5743), SCYA2 (6347), SELE (6401), SELL (6402), SELP (6403), SELPLG (6404), SERPINA5 (5104), SERPINC1 (462), SERPINE1 (5054), SFTPA1 (6435), SFTPA2 (6436), SFTPB (6439), SFTPC (6440), SFTPD (6441), SMP1 (23585), STAT4 (6775), STAT6 (6778), TF (7018), TFPI (7035), TGFB3 (7043), THBD (7056), TIRAP (114609), TNF (7124), TNFAIP1 (7126), TNFAIP2 (7127), TNFAIP3 (7126), TNFRSF1A (7132), TNFRSF1B (7133), TRAF6 (7189), TRPV5 (56302), TRPV6 (55503), VCAM1 (7412), VEGF (7422), and VTN (7448). Coriell ( http://coriell.undmj.edu/ ) repository numbers for human genomic DNAs sequenced for this study are as follows. DNAs from African-Americans were NA17101–NA17116 and NA17133–NA17140. DNAs from European-Americans were NA06990, NA07019, NA07348, NA07349, NA10830, NA10831, NA10842–NA10845, NA10848, NA10850–NA10854, NA10857, NA10858, NA10860, NA10861, NA12547, NA12548, and NA12560. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC515367.xml |
524179 | Targeting smoking cessation to high prevalence communities: outcomes from a pilot intervention for gay men | Background Cigarette smoking prevalence among gay men is twice that of population levels. A pilot community-level intervention was developed and evaluated aiming to meet UK Government cessation and cancer prevention targets. Methods Four 7-week withdrawal-oriented treatment groups combined nicotine replacement therapy with peer support. Self-report and carbon monoxide register data were collected at baseline and 7 weeks. N = 98 gay men were recruited through community newspapers and organisations in London UK. Results At 7 weeks, n = 44 (76%) were confirmed as quit using standard UK Government National Health Service monitoring forms. In multivariate analysis the single significant baseline variable associated with cessation was previous number of attempts at quitting (OR 1.48, p = 0.04). Conclusions This tailored community-level intervention successfully recruited a high-prevalence group, and the outcome data compares very favourably to national monitoring data (which reports an average of 53% success). Implications for national targeted services are considered. | Background Analysis of tobacco marketing has demonstrated lesbian and gay youth as an emerging target community [ 1 ], thereby reinforcing behaviour patterns that contribute to the adult gay smoking prevalence (39.7–47.8%) being up to twice that of adult heterosexuals [ 2 ]. Gay men have been disproportionately affected by HIV/AIDS disease in developed countries. HIV risk-taking behaviour is associated with cigarette smoking among HIV-negative gay men [ 3 ], and among gay men infected with HIV, studies of co-morbidity and survival have identified cigarette smoking as a significant risk factor for opportunistic infections [ 4 ] and rapid disease progression [ 5 ]. A review of the evidence on sexuality and cigarette smoking found the elevated rates of tobacco use to be consistent across international studies, and concluded with a strengthened call for targeted cessation interventions to lesbians, gays and bisexuals [ 6 ]. None have been published to date. The National Health Service (NHS) Cancer Plan, a UK Government health strategy, recommends that Primary Care Trusts (PCTs) take a commissioning lead in forming local alliances involving community groups, harnessing community efforts, and dissemination of effective interventions. [ 7 ]. In order to meet smoking cessation targets, PCTs are recommended to develop specialist smoking cessation services and develop links with local community groups [ 8 ]. This innovative pilot study aimed to design, recruit to, and deliver a series of pilot smoking cessation group interventions and to evaluate outcomes using standard UK Government assessment criteria. Methods Intervention design The intervention was developed and delivered by a community-based volunteer-led charity in London UK, with a remit to promote the health of gay men. Potential acceptability and effectiveness were maximised by providing an NHS-approved programme adapted for an appropriate service wholly facilitated and attended by gay men. Seven volunteers experienced in delivering group interventions within the organisation were trained in the 3 day course " Setting up and running specialist Smoking Cessation Clinics" , part of the Smoking Cessation Training and Research Programme (SCTRP) at St Bartholomew's and Royal London School of Medicine and St George's Hospital Medical School. The programme of withdrawal-oriented treatment combines groupwork, nicotine replacement therapy (obtained on prescription from general practitioners) and ongoing peer support throughout. An initial information session is followed by 6 closed group sessions, setting a quit data for week 3. This pilot consisted of 4 delivered groups, and each group consisted of 7 closed weekly meetings each of 2 hours. The service principle was for a non-judgemental environment where gay men could address socialising and gay social spaces, recreational drug use, sexuality and HIV and the impact of these on their motivations, and ability, to quit smoking. Several specific modifications were made to the taught model. Our intervention modified the SCTRP program's use of "Quit buddies" which promoted partnered support, instead creating "Quit cells" of 3 or 4 participants. This design modification was made in the light of other group interventions delivered by this community organisation in which reliance of a participant on more than one person for support has found to be more reliable. The information on Zyban was expanded to address contraindications with HIV antiretroviral combination therapies. Exercises from assertiveness training courses were imported to assist participants in clearly communicating the intention to remain a non-smoker. In general, group discussion and processes were focussed on culturally-specific contexts to gay men. A detailed intervention programme was written in order to promote consistency across the cycles of intervention delivery. Week 1: information on the course content as well as expectations of quit date are given along with information regarding potential side effects and how to deal with them. Week 2: what to expect when you quit and how to deal with reactions, information on the effects of carbon monoxide, preparation for quit date, personal action plan and the how to use a smoking diary. Week 3: information on how to use nicotine replacements, role play of assertive refusal of cigarettes, selection and formation of quit support cells, and personal statements of cessation. Week 4: group review of challenges of the first week of not smoking with reference to smoking diary and personal action plan, exploration of potential "alternative" support such as meditation and exercise, discussion of the challenges of drug use with respect to smoking cessation. Week 5: group review of previous week's experience, information on health benefits achieved to date and weight gain issues. Week 6: review of previous week's experience, identification of future sources of support. Week 7: review of previous week's experience, information of health benefits to date, elaboration on support sources, small celebration of the group's achievement. Recruitment Twenty-four recruitment advertisements were placed in free London-wide and national gay press, and accompanying editorial and articles were secured to support the recruitment process. Data collection and analysis Prior to the initial session, participants were sent the required UK Department of Health self-completion Smoking Cessation Service NHS Client Assessment Form. Carbon monoxide readings were taken at each session from week 2, using the "Smokealyser" calibrated carbon monoxide register, and readings were used in addition to self-report data to confirm smoking cessation at week 7. All intervention attendees were asked to give written permission for data collection purposes and were given guarantees of confidentiality. All data were entered into SPSS for windows V11. In line with NHS monitoring data requirements, the percentage of successful quitters was calculated as those who gave carbon monoxide readings and confirmed they had quit at week 7 as a percentage of those who set a quit date for week 3. Variables were entered individually into univariate binary logistic regressions, with cessation outcome as the dependent variable and participant baseline characteristics, attitudes and behaviour, and nicotine replacement methods as independent variables. Variables with p values below 0.25 were then entered stepwise into a multivariate logistic regression, with 95% confidence intervals (95% CI) reported. Results Participant characteristics Ninety-eight men registered to attend the intervention, and of these 76 attended at least the first session. Sixty-nine of men returned the assessment sheet, and the outcome analysis is of those 69 men. The mean age of participants was 37.1 years (range 23–63, SD = 7.2 years), and n = 63 (90%) reported their ethnicity as White. Forty-four men (64%) had been educated to degree level or higher, and n = 52 (75%) were in full time employment with a further 9 (13%) men medically retired, n = 5 (7%) unemployed, n = 2 (3%) in full time education and n = 1 (1%) retired. Seventeen men (25%) were entitled to free prescriptions (i.e. the welfare state pays for their prescribed medications). Sixty-five men (94%) reported that they drink alcohol, consuming a mean of 22.8 units per week (median = 20, SD = 19, range 1–120). Smoking behaviours at baseline The daily number of cigarettes smoked was as follows: 1–5 (n = 3, 4%); 5–10 (n = 5, 7%); 11–20 (n = 27, 39%); 21–30 (n = 21, 30%); 31–40 (n = 8, 12%); 41+ (n = 5, 7%). The first cigarette after waking was smoked during the following number of minutes after waking: 5 minutes (n = 19, 28%); 6–30 minutes (n = 31, 45%); 31–60 minutes (n = 7, 10%); 61+ minutes (n = 11, 16%). Smoking motivations are summarised in Table 1 . Table 1 Smoking behaviour at baseline (Strongly) agree Neither (Strongly) Disagree I enjoy smoking 44 (64) 11 (16) 13 (18) Smoking helps me cope with stress 38 (55) 15 (22) 15 (21) Smoking helps me to socialise 39 (57) 16 (23) 13 (19) Smoking helps me to cope with boredom 31 (45) 22 (32) 15 (21) I smoke to keep my weight down 5 (7) 7 (10) 55 (80) Health status and consultations Participants reported a mean 2.6 of consultations with their primary care General Practitioner (GP) in the previous year (median = 2, SD 3.5). Secondary/hospital consultations in the previous year were reported by n = 35 (52%) men, with a mean of 2.26 consultations for these men (median = 1, SD = 3.9). Thirty-four men (51%) had been recommended by their GP to give up cigarette smoking, and n = 26 (38%) men were currently on prescribed medication. Fourteen men (20%) were diagnosed HIV-positive, n = 25 (51%) HIV-negative, n = 16 (23%) untested and n = 4 (6%) refused to answer. The participants rated their health as follows: excellent n = 10 (14.5%); good n = 36 (52%); moderate n = 20 (29%); poor n = 2 (3%); very poor n = 1 (1%). Quitting motivations and history Sixty-one men (90%) had made a previous attempt to quit, and of those who had made an attempt the mean was 2.85 attempts (median 3, SD = 1.4). Previously employed nicotine replacement methods were gum n = 30 (49%), patches n = 30 (49%), nasal spray n = 3 (5%), inhalor n = 12 (20%), microtabs n = 3 (5%), nicotine lozenges n = 4 (7%), and Bupropion (Zyban) n = 12 (20%). Participants described the importance of this current attempt to quit as extremely important (n = 33, 48%); very important (n = 27, 39%); quite important (n = 9, 13%); not at all important (n = 0). Participants rated their chances of quitting for good on this attempt as extremely high (n = 10, 15%); very high (n = 27, 39%); quite high (n = 24, 35%); not very high (n = 7, 10%); very low (n = 1, 1%). Intervention attendance and outcomes Attendance at sessions was consistently high, of 532 person-sessions 13 sessions were missed. Non-attendance did not apparently cluster around a particular session. At week 3, of the 69 men who gave data, n = 58 men (84%) set a quit date. At week 7 (4 weeks after the quit date) n = 44 men (64%) were confirmed as having quit using the CO monitor, representing 58% of those who attended the first session, 76% of those who set a quit date and 64% of those who gave data at baseline and week 7. A further 3 men reported by telephone that they had quit smoking but did not attend the final session to give clinical data to verify. Nine men (13%) reported not having stopped smoking, n = 6 men (9%) set a quit date at week 3 and did not return to group, n = 7 men (10%) attended the first session only. For the purposes of this analysis, these 25 men were coded as not having quit in the following modelling. Variables associated with cessation outcomes in multivariate logistic regression This analysis considers those 44 men confirmed as having ceased compared to those 25 categorised as not having quit. Following univariate analysis (see Table 2 ), the 4 variables entered into the multivariate model were smoking for enjoyment, number of cigarettes per day, smoking to keep weight down, and number of previous attempts. Only the latter (continuous) variable was significantly associated with successful quitting at week 7 (OR = 1.48, 95% CI = 1.02, 2.14, p = 0.04). Data from the multivariate model are presented in Table 3 . Table 2 Univariate binary regression analysis of demographic and behavioural baseline data with respect to cessation outcomes Variable p Odds Ratio 95% CI Age 0.36 1.03 0.96, 1.11 "I enjoy smoking" 0.22* 1.38 0.82, 2.32 "Smoke helps me cope with stress" 0.39 0.82 0.53, 1.29 "Smoking helps me to socialise" 0.58 0.89 0.58, 1.36 "Smoking helps me to cope with boredom" 0.97 0.99 0.63, 1.55 "I Smoke to keep down weight" 0.18* 1.41 0.85, 2.33 No. of cigarettes per day 0.08* 0.67 0.42, 1.05 Time to 1 st daily cigarette 0.52 1.18 0.71, 1.96 No. of previous attempts 0.04* 1.44 1.02, 2.01 Importance of this attempt 0.80 0.91 0.45, 1.84 No. of GP visits 0.33 1.09 0.92, 1.29 No. of secondary care visits 0.43 1.06 0.92, 1.22 Expected chance of quitting success on this attempt 0.91 0.97 0.58, 1.62 Perceived health status 0.96 1.02 0.55, 1.90 * Included in multivariate analysis (see Table 3) Table 3 Multivariate analysis of variables identified as associated with cessation at week 7 (i.e. p < 0.25 in univariate analysis, Table 2). Variable p Odds Ratio 95% CI No of cigarettes per day 0.15 0.68 0.41, 1.14 Smoking to keep down weight 0.20 1.43 0.83, 2.45 No of previous attempts to give up 0.04* 1.48 1.02, 2.14 I enjoy smoking 0.38 1.30 0.73, 2.30 Discussion This pilot intervention has targeted a hitherto overlooked high smoking prevalence group, and has adapted a Government-approved intervention to meet the specific needs of gay men in an appropriate and acceptable setting. The success rate of 76% of men who had set a quit date being confirmed as having quit at week 7 compares extremely favourably to national monitoring data, which reports a success rate nationally 2001–2002 for smoking cessation services as 53% [ 9 ]. Public health targets must consider the needs of high prevalence communities, and this may be achieved through innovative development of existing effective services. However, this study has highlighted the lack of targeted interventions for gay men, and the evidence demonstrates further elevated health needs compared to the general population in the fields of alcohol and drug use [ 10 ] mental health [ 11 ] and cancer [ 12 , 13 ]. Further research may identify the factors which contributed to the effectiveness of this pilot complex participative intervention, including offering recruitment and delivery outside of community settings, measuring success rates for gay men in non-gay specific or tailored groups, and the usefulness of "quit cells". Longer-term follow-up data and increasing dosage to include a follow-up session would also provide further useful data. In order to refine the intervention for trial testing, qualitative data regarding the utility, acceptability and preferences for the content of specific sessions would be illuminating. Further, the non-randomised design without comparison group limits presents a limitation to the generalisability of findings, yet still offers cessation outcomes much better than standard national cessation data quoted above which were collected without quasi-experimental design using the same follow-up period. Data were not available on the 29 men who registered for the course but did not attend or complete baseline data, and so it is not possible to compare their demographics or smoking behaviours to those who took up the intervention. Certainly, replication of this first pilot would be necessary in other settings, e.g. non-metropolitan communities, where issues of feasibility and uptake should be addressed. Commissioners may consider the purchase of existing facilitators from cities to deliver in non-metropolitan areas where demand is likely to be lower, as smoking cessation service recommendations state that group leaders need to keep up to date with their skills and to use them on a regular basis [ 14 ]. Conclusions In order to meet the smoking cessation needs of this hitherto overlooked population, and to meet public health policy targets, a rigorous research agenda must be established. While the use of required standard outcome monitoring must be continued, rigorous experimental trials using longer term follow up and commonly reported measures are required. Complex participative interventions must be developed, as in this pilot, from evidence-based interventions with full programme description to ensure replication. The development of appropriate interventions must first pilot services to ensure that they are appropriately adapted to maximise acceptability and uptake among target communities. Lastly, provision of the service by skilled volunteer facilitators has ensured an acceptable, low-cost intervention with a rate of effectiveness in these four pilot groups that compares favourably to national non-targeted interventions outcomes calculated using standard assessment formula. Acceptability of the model appears high with respect to the low number of missed sessions. Voluntary sector provision and delivery should be considered as a low-cost and highly acceptable point of delivery for effective community-level smoking cessation interventions. Competing interests The authors declare that they have no competing interests. Authors' contributions JB and NH managed the pilot study. RH was responsible for data management and analysis and drafted the manuscript. All authors read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC524179.xml |
544841 | Early increases in plasminogen activator activity following partial hepatectomy in humans | Background Increases in urokinase-like plasminogen activator (uPA) activity are reported to be amongst the earliest events occurring in remnant liver following partial hepatectomy in rats, and have been proposed as a key component of the regenerative response. Remodelling of the extracellular matrix, conversion of single chain hepatocyte growth factor to the active two-chain form and a possible activation of a mitogenic signalling pathway have all been ascribed to the increased uPA activity. The present study aimed to determine whether similar early increases in uPA activity could be detected in the remnant liver following resection of metastatic tumours in surgical patients. Results Eighteen patients undergoing partial hepatectomy for the removal of hepatic metastases secondary to primary colonic tumours were studied. Increased plasminogen activator activity was found in the final liver samples for the group of patients in whom the resection size was at least 50%. For smaller resections, the increased activity was not observed. The increased activity did not correlate with the age of the patient or with the time between the start of resection and the end of the operation. There was, however, a negative correlation between plasminogen activator activity and the time for which blood supply to the liver was clamped. Conclusions Our findings are in accordance with those from experimental animal models and show, for the first time, that rapid increases in plasminogen activator activity can occur following similarly large liver resection in humans. Thus, increases in plasminogen activator activity are an early event in the remnant liver following major liver resection in man. Our observations provide support for the contention that increases in plasminogen activators play a key role in the initiation of hepatic regeneration in man. | Background Urokinase-like plasminogen activator (uPA), initially recognised by its ability to convert plasminogen to plasmin and to participate in the fibrinolytic cascade, is now considered to have a wider role, which encompasses metastatic invasion by tumour cells and liver regeneration. In regeneration of the liver following partial hepatectomy, uPA has a number of potential roles. These include initiating the remodelling of the extracellular matrix to allow cell division, activation of extra-cellular pro-metalloproteases and the release of the bound single-chain form of hepatocyte growth factor (HGF) from the extracellular matrix (ECM). In vitro uPA and tissue-like plasminogen activator (tPA) have been shown to convert single chain inactive HGF into the active two chain form [ 1 ] in cultures of hepatocytes. In normal rodent liver, both the inactive and active forms of HGF can be detected, with the predominance of the inactive form [ 2 ]. Following partial hepatectomy in the rat there is an early net decrease in the total amount of HGF in the liver, but the relative proportion of the single chain, inactive form, is decreased and the active two-chain form increased [ 2 ]. This implies an early proteolytic conversion, possibly mediated by the plasminogen activators. The importance of the uPA-plasminogen system to liver repair has been further demonstrated by the inability of plasminogen deficient animals to form regenerative nodules in response to acute liver injury [ 3 ]. As discussed by Mangnall et al. [ 4 ], uPA may also activate a signalling pathway leading to mitosis of the hepatocyte. Increases in uPA activity are amongst the earliest documented changes following partial hepatectomy in rats [ 5 ]. Raised uPA activity was detected in the remnant liver at one-minute post-hepatectomy and continued to increase for at least one hour, although there were no changes in the total amount of uPA protein detectable by Western blotting. The binding of uPA to the uPA receptor (uPAR) is also associated with an increase in uPA enzymatic activity [ 6 ]. In the rat partial hepatectomy model, the increase in uPA activity is thought to be due to an increase in the level of uPAR and subsequent binding and activation of uPA. In the remnant liver, increases in the amount of uPAR have been detected by Western blotting also as early as 1 min post hepatectomy and more clearly at 1 hour. This had decreased by 6 h and was back to basal levels by 24 h [ 5 ]. The mechanism underlying these changes remains unclear. Additional support for a role for uPA in the hepatic regenerative process comes from studies of uPA-deficient (uPA-/-) mice. In these animals, uptake of [ 3 H]-thymidine into DNA and mitotic index were reduced by almost half at 44 h post-hepatectomy (the peak time for control mice), suggesting a slower hepatocyte growth response [ 7 ]. In a separate study uPA-/-mice were treated with anti-Fas monoclonal antibody to induce extensive hepatocyte apoptosis. Fas (a member of the TNF-receptor superfamily) is present in the inactive state as a monomer, but on binding the appropriate ligand (in this case the antibody) the receptors aggregate and activate apoptosis leading to cell destruction. In these uPA-deficient animals, the regeneration response following anti-Fas treatment was delayed relative to normal control animals [ 8 ]. Generation of mature HGF and time of peak levels were delayed in the uPA-/-mice and peak levels of proliferating cell nuclear antigen at 96 h were also delayed relative to controls, which peaked at 48 h. Treatment of the uPA-/-mice with the uPA gene by lipofection reversed these effects. The results support a role for uPA in the generation of mature HGF and in the regeneration after Fas-mediated liver damage. More recently, studies with uPA or plasminogen deficient mice confirmed the requirement for plasminogen activation in liver regeneration and also showed a need for plasminogen in regeneration-associated hepatic angiogenesis [ 9 ]. Collectively, these studies strongly suggest that a very early increase in uPA activity is a key feature of the liver regenerative response in rodents. It is generally assumed that regeneration in the human liver follows a similar course but the relative paucity of studies in humans means that, at present, it is unclear whether a similar role for uPA exists in the regeneration of human liver. Though not necessarily identical, it is clear from the literature that regeneration in humans and rodents share similar mechanisms. Many of the cytokines and growth factors essential for regeneration in rodents [reviewed in [ 4 ]] are also found in increased amounts in the regenerating human liver, implying once more similar mechanisms. However, clear differences between species do exist; a notable example being the differences in the time at which DNA synthesis peaks in the remnant liver. In rats, this is at about 24 h; in mice, at about 40 h; and in man, at 180–200 h following hepatectomy. In the case of the human studies, this may partially reflect the relatively greater age of the patients since the rate of regeneration slows with age. Such age related effects are less likely in the rodent studies where the timing of hepatocyte entry into DNA synthesis following partial hepatectomy has been shown to be an intrinsic, cell-autonomous, feature [ 10 ]. Thus, although the basic mechanisms may be fundamentally similar, there are inherent differences between species (such as the timing of the cell cycle clock) which underscores the need not to assume that all aspects of regeneration operate identically in all mammals. The unique sensitivity of the human hepatocyte to TRAIL (tumour necrosis factor-related apoptosis-inducing ligand) [ 11 ] likewise emphasises the need for caution when extrapolating from rodent liver to human liver. The vast majority of the literature concerns regeneration in rats and mice and much less information is available from human studies since the opportunity to study liver regeneration in humans is generally limited to units specialising in liver surgery and is necessarily constrained by ethical considerations. Surgical removal of liver metastases affords the opportunity to obtain small samples of liver at the start, time of resection and time of wound closure approximating to the early sampling times in the animal studies. In this vein, the aim of the present study was to determine whether very early increases in uPA activity occur in the remnant liver following resection in man. Results The basal uPA activity associated with the membrane preparations showed a wide patient to patient variation ranging from 4 to 24 nmol/min/mg protein with a mean of 9.94 nmol/min/mg protein (n = 18, SD = 5.06). This variation in basal activity correlated neither with the age of the patient – linear regression analysis gave a slope of -0.03 and correlation analysis gave a Spearman coefficient of -0.097 ( p = 0.7), nor there was any difference between the values for male patients (mean = 9.14 nmol/min/mg, SD = 3.84, n = 7) and female patients (mean = 10.45, SD = 5.83, n = 11) with a non-significant unpaired Student's t test ( p = 0.61). The uPA activity associated with the membrane fractions prepared from samples taken during the operation is shown in Figure 1 , for all the patients studied. The activity of the final remnant fraction taken at the end of the operation was increased significantly above the activity of the other fractions. The increased activity of the final remnant fraction was, almost exclusively, confined to those patients who had undergone a resection estimated at 50% or greater (Figure 2A ) and there was no increase in those patients in whom the resection was less than 50% (Figure 2B ). The percentage change in uPA activity as a function of the resection size for the individual patients is shown in Figure 2C . There was no correlation between uPA activity and the size of the resection below about 50% resection, but a positive correlation was observed when the resection size was 50% or greater. Figure 1 Membrane associated plasminogen activator activity (nmol/min/mg protein) in the whole patient group. Samples were: Start (control sample taken at start of operation); 1 = Res 0; 2 = Rem 0; 3 = Res End; and 4 = Rem End, as described in the Methods. Values are means and the error bars are 95% confidence limits. Student's paired t test analysis showed the Start and Rem End samples to be statistically significantly different ( p = 0.01, n = 18). There were no statistical differences between start and any of the other samples. Figure 2 Membrane associated plasminogen activator activity (nmol/min/mg protein) in the group for whom the estimated resection size was 50% or greater and for the group where the resection size was less than 50%. Samples were: Start (control sample taken at start of operation); 1 = Res 0; 2 = Rem 0; 3 = Res End; and 4 = Rem End; as described in the Methods. Values are means and the error bars are 95% confidence limits. For the 50% and greater group (Figure 2A), Student's paired t test analysis showed the Start and Rem End samples to be statistically significantly different ( p = 0.002, n = 8). There were no statistical differences between start and any of the other samples. For the less than 50% group (Figure 2B), there were no statistical differences between any of the samples (n = 10). The relationship between increased uPA activity in the Rem End samples and extent of resection is shown in Figure 2C. There was no statistical correlation below 50% resection (Spearman correlation coefficient = -0.22, p = 0.268) but for 50% resection and higher a positive statistical correlation was observed (Spearman correlation coefficient = 0.67, p = 0.025, n = 9) Figure 3 shows examples of the zymography gels confirming the presence in the membrane fractions of plasminogen-dependent proteolytic activities. Minor bands in the plasminogen-free control gel (indicating proteolysis which was not plasminogen dependent) were occasionally seen. The activity in these bands was always very much less than the plasminogen dependent activities and plasminogen-free gels had to be more extensively destained in order to visualise these minor bands. Figures 3A and 3B show samples from a patient in whom the estimated resection size was 60% and Figures 3C and 3D show samples from a 15% resection. In both cases, major bands corresponding to high molecular weight uPA and tPA were clearly demonstrable in all the membrane fractions together with minor bands of higher molecular weight. The lanes in plasminogen-free control gel for the patient with the major resection all showed a single very faint band of approximately uniform intensity corresponding to a protein larger than tPA (Figure 3A and 3B ). In Figure 3C , a high molecular weight band present only in the lane corresponding to the remnant end sample appeared with approximately equal intensity in the plasminogen-free gel (Figure 3D ) indicating that this material was not a plasminogen activator. Although the final remnant sample in Fig 3A had increased uPA activity relative to the other samples, as determined by the fluorometric assay, the major bands of activity in the gel, corresponding to the uPA and tPA markers, showed little change. In our experience, this is a reflection of the qualitative nature of the plasminogen activator zymography. We find that with purified uPA and tPA proteins, an increase of about one order of magnitude is necessary before the bands produced in the gels are notably different. Figure 3 Zymography gels for patients with greater than 50% resection and less then 50% resection. Figures 3A and 3B are greater than 50% and Figures 3C and 3D are less than 50%. The gels in Figures 3A and 3C contained plasminogen and the gels in Figures 3B and 3D are the corresponding plasminogen-free control gels. The samples run were, Lane 1- uPA (low molecular weight standard 33 kD, 0.6 ng); Lane 2- tPA standard (65 kD, 1 ng); Lane 3- uPA (high molecular weight standard 54 kD, 0.75 ng); Lane 4–10 μl SeeBlue Plus2 pre-stained standard markers; Lanes 5–9 were washed membrane preparation (30 μg protein / lane); Lane 5 – start; Lane 6 – Res 0; Lane 7 – Rem 0; Lane 8 – Res End; and Lane 9 – Rem End. There were apparent increases in band size associated with the higher molecular weight minor bands in the remnant end sample. Although, at present, the precise nature of these high molecular weight bands is uncertain, similar high molecular weight plasminogen activators have been previously reported [ 12 , 13 ]. The most logical explanation would be an association with either uPAR or plasminogen activator inhibitor type 1 (PAI-1), and on the basis of the present observations such a complex is as likely to contain tPA as uPA. These were not observed in the samples from the patient with a smaller resection. At present, the biological significance of these changes in what appear to be relatively minor bands is uncertain, but there was an association with the large resection size and increased uPA activity, as measured by fluorimetry. Increases in uPA activity did not correlate with patient age (Figure 4A ) or the elapsed time interval between taking the remnant start and the remnant end samples (Figure 4B ). There was, however, a statistically significant negative correlation between the increase in activity and the total time for which the blood supply to the liver had been clamped during the operative procedure (Figure 4C ), suggesting a link between increased proteolytic activity and hepatic perfusion. Figure 4 Relationship between the increase in membrane associated uPA activity (nmol/min/mg protein) and (A) Age, (B) Elapsed time between the time of the resection and the time at which the final sample was obtained, and (C) the total time for which the portal vein was clamped during operation. Spearman correlation analysis showed no correlation for A ( p = 0.44, Spearman correlation coefficient = -0.071, n = 8), or B ( p = 0.27, Spearman correlation coefficient = 0.26, n = 8). In C however, a statistically significant correlation was observed ( p = 0.012, Spearman correlation coefficient = -0.89, n = 7). Discussion Increased uPA activity and increased levels of uPAR are among the earliest reported events in the remnant liver, following 70% partial hepatectomy in rodents [ 5 ]. Since then, several studies have emphasised the importance of the plasminogen system to hepatocyte proliferation and angiogenesis in the regenerating rodent liver [ 6 - 9 ]. We have shown here for the first time in humans that increases in plasminogen activator activity occur following hepatectomy. Increased activity was only seen in remnant liver at the end of the operative procedure when, at least, 15 min had elapsed between the time at which the resection was completed and the last remnant sample taken, and where the magnitude of the resection was estimated to be at least 50% of total liver volume. If, as proposed by Mars et al. [ 5 ], increased uPA activity is an essential feature at the start of regeneration, then these observations confirm the findings of animal studies that the magnitude of the regenerative response is dependent on the extent of the hepatectomy [ 14 , 15 ]. The present studies suggest that removal of, at least, half the liver mass is necessary to generate the biological signal that results in increased plasminogen activator activity. The lack of any increase in uPA in response to resections less than 50% compared to the positive correlation between increased uPA activity and increased resection above 50% suggests a threshold event around the 40 to 50% level. The plot shown in Figure 2C bears a striking resemblance to the data in the review by Bucher [ 14 ] showing incorporation of tritiated thymidine into DNA following hepatectomy in mature rats. A similar threshold point at about 40% resection, with no correlation below this level and a positive correlation above, was clearly demonstrated in those studies also. The mechanism by which resections greater than about 50% increasingly result in elevated uPA activity and increased DNA synthesis remains elusive. It is still unclear whether the same trigger is responsible for the increases in both systems. The possibility that the increased uPA activity seen here represents a response to injury rather than an early regenerative response cannot be totally discounted. However, in the rat partial hepatectomy model the anatomy allows removal of the major liver lobe without imposing surgical trauma on the remnant liver suggesting that increased uPA activity is not injury related. Zymography clearly showed several plasminogen activators to be associated with the membrane fractions. As expected, the major bands corresponded to the high molecular weight uPA and tPA markers. Although uPA and its receptor uPAR have been implicated in the initiation of the liver regeneration process [ 5 , 16 ], no similar role has been ascribed to tPA. The latter binds to both liver endothelial cells (via the mannose receptor) and hepatocytes (by the LDL receptor-related protein) as part of the process by which tPA is rapidly cleared from the circulation by the liver. To date, however, there is no evidence from rodent studies to suggest that binding of tPA to receptors is, in any way, involved in the response to hepatectomy. However, the present study clearly shows tPA activity associated with the liver membrane preparations, and given the ability of tPA to generate active HGF in vitro [ 1 ] the possibility of a role for tPA in the response of human liver to partial hepatectomy needs to be borne in mind. We also found several minor bands of higher molecular weight, the nature of which is uncertain. These could potentially represent larger forms of the plasminogen activator or the plasminogen activator tightly bound to some other protein. The most likely candidates for such a complex would be uPA associating with either uPAR or PAI-1. The final remnant sample obtained after major resection showed increased amounts of these higher molecular weight components. High molecular weight forms of uPA have been observed in the rat prostate following castration [ 12 ] and also in cultured Kaposi sarcoma cells [ 13 ]. In the latter, it was suggested that the high molecular weight form of uPA contributed to the characteristic hyperproliferative and invasive phenotype of the Kaposi sarcoma lesions. Increased uPA activity associated with increased metastatic activity seems well accepted and uPA and other members of the urokinase plasminogen activator system (including uPAR and PAI-1) have been selected as novel targets for potential tumour therapies [ 17 ]. Whether the high molecular weight forms of uPA are also characteristic of an increased proliferative activity in the liver remains to be fully established. Presently, the source of the increased uPA activity is uncertain. The very early increase in activity at 1 minute post-hepatectomy in the rat and the lack of any associated increase in mRNA for uPA, precludes any de novo protein synthesis [ 5 ]. In the present study, the increased activity at 15 minutes post-resection also seems too rapid for a mechanism requiring new protein synthesis. Mars et al. [ 5 ] suggest that the increased uPA activity seen in rats immediately following partial hepatectomy represents binding of uPA from the blood to the uPA receptor in the liver. The uPAR was undetectable on Western blots from rat liver prior to hepatectomy, but was present in the remnant liver as early as one minute post-hepatectomy and increased in amount during the next 60 minutes. It has been proposed that this increased uPAR binds uPA from the circulating blood resulting in the increased uPA activity within the liver. The suggestion that this is a key element in the initiation of the regeneration process highlights the need for adequate perfusion of the liver. Though the underlying molecular mechanisms remain unclear, interruption of hepatic perfusion generally has adverse effects on the regenerative response [ 4 ]. The present study supports the hypothesis that continued liver perfusion is important in the process by which increased uPA activity is generated. Firstly, increases in uPA activity did not occur in the liver that had been resected and removed from the circulation; secondly, for those patients in whom there was an increase in uPA activity, the magnitude of the increase was inversely related to the clamp time, i.e. , the longer the liver perfusion was interrupted the smaller the response. Thus, in the present study, increased uPA activity was negatively correlated with total clamp time suggesting that hypoxia, which has been shown to induce uPAR expression in cells in culture [ 18 - 20 ], is not a likely mediator of the uPA increase seen here. Finally, despite the proliferative capacity of hepatocytes and the ability of the liver to regenerate declining with age [ 14 , 15 ], we found no correlation between age and basal uPA activity and the increase in remnant liver uPA activity was also not age dependent. Conclusions In the present paper, we show early increases in uPA activity can be demonstrated in the remnant liver following resection of metastatic tumours in patients in whom the resection was estimated to be 50% or greater. To the best of our knowledge this is the first time this has been demonstrated. Such increases are amongst the earliest events following hepatectomy in rats, where they are considered to initiate changes in the extracellular matrix essential for subsequent hepatocyte division. Thus, our results support a similar role in the initiation of liver regeneration in man. Methods Patients The South Sheffield Research Ethics Committee approved the research protocol and fully informed consent was obtained. Eighteen patients undergoing partial hepatectomy for the removal of hepatic metastases secondary to primary colonic tumours were studied. There were 7 males and 11 females with an age range from 24 to 78 years (median 67.5). Operative Procedure Standard operative procedures were followed. The liver was mobilised and the resection delineated with diathermy. The portal inflow was clamped while resection with an ultrasonic dissector was carried out. Typically, the portal inflow was released every 15 minutes for 5 minutes intervals to prevent ischaemic damage and the total clamping time was recorded. Resection margins were sent separately for histopathology. The magnitude of the resection was estimated as percentage of the total liver volume, by the surgeon. The following samples were taken from tumour-free regions of the liver during the operation. A sample was obtained before the resection was started (this was labelled 'Start'). Immediately following resection samples were taken from the remnant liver (labelled 'Rem 0' for remnant liver at time 0) and from the resected liver as far away from the tumour as possible (labelled 'Res 0' for resected liver at time 0). The samples were placed in cryovials and immediately frozen in liquid nitrogen in the operating theatre. A second sample of the resected liver (labelled 'Res end') was kept at room temperature until the end of the operative procedure and was only transferred to liquid nitrogen when the final sample from the remnant liver was taken. The final sample ('Rem end' for remnant end) was taken from the remnant liver as late into the operation as possible and frozen immediately. The 'Res end' sample was also frozen at this time. The interval between the time of sampling 'Rem 0' and 'Rem end' ranged from 7 to 90 minutes. The median was 20.5 minutes and 15 of the 18 intervals were between 10 and 33 minutes. Samples were stored in liquid nitrogen in the laboratory and only thawed immediately prior to analysis. Materials Casein, plasminogen, uPA (high and low molecular weight forms) and tPA were purchased from Calbiochem (CN Biosciences Ltd., UK). The fluorometric substrates 7-amino-4-methylcoumarin (AMC), Z-Gly-Gly-Arg-AMC and EGR-CMK (Glu-Gly-Arg-Chloromethylketone) were from Bachem Ltd. (UK). Electrophoresis reagents were from BioRad and Geneflow Ltd. Other reagents were from Sigma-Aldrich Co Ltd. (Poole, UK). Sample preparation Liver samples were homogenised in a ten-fold volume of homogenisation buffer: 250 mM sucrose / 10 mM MOPS pH 7.4 containing the protease inhibitors E-64 (20 μM), Pepstatin A (20 μM), and EDTA (0.2 mM). Inhibitors against the serine proteases, which include uPA, were not included. Membrane preparation A membrane preparation was made by differential centrifugation of the homogenate in a TLS 55 swinging bucket rotor in a Beckman TL-100 bench top ultracentrifuge (Beckman Coulter Ltd., High Wycombe, UK). The homogenate was initially centrifuged at 40,000 g, for 20 minutes, to pellet large cell organelles such as nuclei and mitochondria. After centrifugation the fat at the top of each tube was removed with a piece of tissue and the supernatants transferred to clean tubes and recentrifuged at 105,000 g, for 1 hour. The membranous pellets were then washed twice by resuspending in homogenisation buffer and recentrifuged at 105,000 g, for 1 hour. All centrifugations were carried out at 4°C. The protein content of the homogenates and membrane preparations was determined by the BCA (bicinchoninic acid) method [ 21 ] using a kit from Sigma-Aldrich. uPA fluorometric assay uPA activity was determined by a fluorimetric continuous rate assay of Z-Gly-Gly-Arg-AMC hydrolysis using a Perkin Elmer LS50B fluorimeter linked to an IBM compatible computer running the FLUSYS software [ 22 ]. Cleavage at the Gly-Arg bond by uPA releases the AMC from its quenched state [ 23 ] and the rate at which fluorescence is produced taken as a measure of uPA activity. At the end of the assay, EGR-CMK (Glu-Gly-Arg-Chloromethylketone), a chemical inhibitor of uPA, was added to check that this compound inhibited the measured activity. Any activity still persisting was taken as not uPA-mediated and subtracted from the rate measured in the absence of EGR-CMK. Since the biologically relevant fraction of uPA is generally considered to be associated with its receptor uPAR and therefore localised to the cell membrane, uPA activity measurements were performed with washed membrane preparations rather than with total liver homogenates. Preliminary experiments demonstrated the necessary linear response between the measured activity and the volume of membrane preparation assayed (data not shown). All samples were assayed in triplicate at two sample volumes to ensure linearity of activity with amount of extract. The measured rates were then adjusted for the protein concentration of each sample to give a rate in nmoles/min/mg of protein. Zymography Zymography was carried out with 7.5% SDS PAGE two-substrate gels essentially as described by Bryson et al. [ 24 ]. The control, plasminogen-free, gels contained casein alone (final concentration of 6 mg/ml gel) and the test gels contained plasminogen at a final concentration of 9.3 μg (1.12 U) / ml gel in addition to the casein. Following electrophoresis gels were washed in 25% (v/v) Triton X-100, for 1 hour at room temperature, and then in 50 mM Tris (pH 7.6) for 16–20 hours and at 37°C, prior to staining with Coomassie blue. Purified uPA (high molecular weight and low molecular weight) and tPA (all from CalBiochem, CN Biosciences Ltd., Nottingham UK) and SeeBlue Plus2 Pre-stained standards (Invitrogen Life Technologies, Paisley, UK) were included on each gel as markers. Graph plotting and statistical analysis All Figures were generated and analysed with the GraphPad Prism package (version 3.0). Statistical analyses (Student's t tests, simple linear regression, Spearman correlations) were performed using the software in the cited package. Authors Contributions DM initiated the study, carried out the zymography experiments, prepared tissue extracts and drafted the manuscript. KS prepared tissue extracts and carried out the fluorometric assays. AWM and NCB participated in the design and coordination of the study. All authors have read and approved the final manuscript. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544841.xml |
555958 | Antiplatelet agents for prevention of pre-eclampsia and its consequences: a systematic review and individual patient data meta-analysis | Background There is now good evidence that antiplatelet agents (principally low dose aspirin) prevent pre-eclampsia, a leading cause of morbidity and mortality for pregnant women and their babies. A Cochrane Review identified moderate, but clinically important, reductions in the relative risks of pre-eclampsia (19%), preterm birth (7%) and perinatal mortality (16%) in women allocated antiplatelets, rather than placebo or no antiplatelet. Uncertainty remains, however, about whether some women (in terms of risk) benefit more than others, what dose of aspirin is best and when in pregnancy treatment should ideally start. Rather than undertake new trials, the best way to answer these questions is to utilise existing individual patient data from women enrolled in each trial. Methods/Design Systematic review with meta-analysis based on individual patient data. This involves the central collection, validation and re-analysis of thoroughly checked data from individual women in all the available randomised trials. The objective is to confirm that antiplatelet agents, given during pregnancy, will reduce the incidence of pre-eclampsia. The review will then determine the size of this effect, and whether antiplatelets delay the onset of pre-eclampsia or its impact on important outcomes for women and their babies. It will also explore whether the effect of antiplatelets differs by womens' risk profile; when commenced during pregnancy; and/or by dose. Discussion The PARIS (Perinatal Antiplatelet Review of International Studies) Collaboration has been formed to undertake the review. This will be the first individual patient data review in the perinatal field. Final results should be available by 2006–7. | Background Clinical significance of pre-eclampsia World-wide, over half a million women die each year of pregnancy related causes with 99% of these occurring in low resource countries [ 1 , 2 ]. An estimated 10–15% of these maternal deaths are associated with hypertensive disorders of pregnancy [ 3 ]. High blood pressure is common during pregnancy: approximately one in ten women will have their blood pressure recorded as above normal at some point before delivery [ 4 ]. For women who develop raised blood pressure but have no other complications, pregnancy outcome is similar to that for women with normal blood pressure. Pre-eclampsia, a multi-system disorder of pregnancy usually associated with high blood pressure (hypertension) and proteinuria, complicates 2–8% of pregnancies [ 5 ]. It can affect the mother's organs, leading to problems in liver, kidneys and brain, and to abnormalities of the clotting system. As the placenta is also involved, there are increased risks for the baby. The most common problems are poor fetal growth due to inadequate blood supply through the damaged placenta, and prematurity, related either to the spontaneous onset of pre-term labour or the need for an early, elective delivery. Although the outcome for most women is good, pre-eclampsia and eclampsia (the rare occurrence of seizures superimposed on the syndrome of pre-eclampsia) are major causes of maternal mortality. In low resource countries pre-eclampsia and eclampsia account for 10–15% of maternal deaths [ 3 ] whilst in high resource countries pre-eclampsia is consistently a leading cause of maternal mortality [ 6 , 7 ]. Perinatal mortality is also increased [ 8 , 9 ]. There is little good quality information about morbidity for either mother or baby, but it is likely that this too is high. For example, pre-eclampsia accounts for about one fifth of antenatal admissions [ 10 ], two thirds of referrals to day care assessment units [ 11 ] and a quarter of obstetric admissions to intensive care units [ 12 ]. Pre-eclampsia is an antecedent for up to 19% of pre-term births [ 13 ] and 12% of growth restricted babies [ 14 ]. Also of note is the high rate of intrauterine growth restriction (20–25%) in pregnancies complicated by pre-eclampsia and the possible lifelong health effects due to prenatal programming [ 15 ]. Pre-eclampsia is a multi-factorial condition. Although its aetiology remains unclear, there have been significant advances in the understanding of the pathophysiology of the disorder. The primary lesion is thought to be deficient trophoblast invasion of the maternal spiral arteries in the second trimester, leading to underperfusion of the uteroplacental circulation and placental ischaemia [ 16 ]. The resulting placental damage is thought to lead to release of factors into the maternal circulation, which are responsible for the maternal syndrome. Activation of platelets and the clotting system may occur early in the course of the disease, before clinical symptoms develop [ 17 , 18 ]. Deficient intravascular production of prostacyclin, a vasodilator, with excessive production of thromboxane, a platelet-derived vasoconstrictor and stimulant of platelet aggregation [ 19 , 20 ], have also been demonstrated to occur in pre-eclampsia. These observations have led to the hypothesis that antiplatelet agents, low dose aspirin (<300 mg/day) in particular, might prevent or delay the development of pre-eclampsia or reduce its severity and the risk of adverse events. It is further hypothesised that the effect of antiplatelets may be different if treatment is started before placental implantation is complete. [ 21 ]. If this hypothesis were correct, the greatest benefit should be seen in women who started treatment before 16 weeks gestation, with the effect attenuating with later onset of treatment. Similarly, it remains unclear as to the most appropriate dose of antiplatelet therapy for the prevention of pre-eclampsia in order to maximise benefits whilst minimising harms [ 22 ]. It has been suggested that low doses of aspirin may selectively inhibit the cyclo-oxygenase pathway in platelet production but not in vessel wall endothelium thereby diminishing the synthesis of thromboxane but not of prostacycline. A higher dose may inhibit both thromboxane and prostacycline thereby neutralising the effect of the intervention [ 20 ]. However, there is also limited evidence from randomised trials that a higher dose of aspirin may effect a greater reduction in the risk of pre-eclampsia [ 23 ]. Although it is known that pre-eclampsia is a multi-system disorder, the relationship between the placental pathology and maternal endothelial response is not fully understood. Numerous maternal factors can predispose to the disorder, such as previous pre-eclampsia, diabetes, renal disease, chronic hypertension or other risk factors [ 24 ]. The syndrome known as pre-eclampsia may also be more than one disease, each with distinct origins, pathologic characteristics and natural history, rather than one fundamental process with varying degrees of clinical severity [ 25 , 26 ]. Hence, the ability to assess the affect of antiplatelets on women with individual risk factors, or a series of risk factors, is of great relevance to clinicians and women. A meta-analysis based on data from individual women will enable the exploration of these hypotheses. Randomised trials of antiplatelet agents The effects of antiplatelet agents were first evaluated in small randomised trials, which reported striking reductions in the risk of hypertension and proteinuria [ 27 - 29 ]. These trials were too small to provide reliable information about other more substantive outcomes, such as perinatal mortality and preterm birth. Also, there was no information about the potential hazards of antiplatelet therapy, such as a possible increased risk of bleeding for both the woman and her baby, or possible effects on infant and child development. The promising results of these early trials led to several large studies around the world. Before these could be completed, however, the use of low dose aspirin had already become relatively widespread for women considered at increased risk of pre-eclampsia. Results of the larger trials were disappointing, as they failed to confirm any statistically significant reductions in substantive outcomes [ 30 ]. Nevertheless, the first Cochrane Review of these trials demonstrated that, when taken together, there are modest, but clinically important benefits [ 31 , 32 ]. Summary of systematic review of aggregate data in 2004 The updated Cochrane Review identified 51 trials with over 36,500 pregnant women evaluating antiplatelet agents, principally low dose aspirin, for the prevention of pre-eclampsia [ 23 ]. Nine of these trials included over 1000 women, and 15 involved less than 50 women. Fifty-one studies were excluded, mostly due to the non-availability of clinically relevant data. Aggregate data from the included trials demonstrated a 19% reduction (RR 0.81, 95% CI 0.75–0.88) in the risk of developing pre-eclampsia associated with the use of antiplatelet agents, rather than placebo or no antiplatelet. There was also a small (7%) reduction in the risk of pre-term birth (RR 0.93, 95% CI 0.89–0.98) and a 16% reduction in the risk of the baby dying (RR 0.84 95% CI 0.74–0.96). Based on the average risk of women included in these trials, about 70 women would have to be treated to prevent one case of pre-eclampsia, and 240 to prevent one baby death. These effects are much smaller than had initially been hoped for but, nevertheless, potentially they have considerable public health importance. The conclusion from this aggregate data review was that for most low-risk women large numbers of women would need to be treated with antiplatelets agents to prevent one episode of either pre-eclampsia or perinatal death. Whether there are specific high risk sub-groups of pregnant women for whom there might be greater benefits, remains unclear, as does the best time to initiate treatment, and at what dose. The aim is now to extend this review based on aggregate data, to utilise the available data for every individual woman in each trial to help address these remaining questions. The use of individual data for each woman will allow for more powerful and flexible analysis of both subgroups and outcomes [ 33 , 34 ]. This review will therefore provide more specific information to guide the care of women at risk of pre-eclampsia. Limitations of the review using published, aggregate data • Many studies were excluded because publications did not report sufficient information to allow them to be included in the review. • The published aggregate data is variable in the completeness of outcome reporting and in definitions used between trials. • The aggregate data meta-analyses were restricted to analysing outcomes for complete trials. As many trials included women with a wide range of risk profiles at trial entry, trials could only be defined by average values. Such aggregated outcomes such as 'proportion of women developing pre-eclampsia' conceal a range of severity of disorder and so it was not possible to explore fully whether the effectiveness of antiplatelet therapy differed according to risk. • It was also difficult to make precise recommendations about when to start treatment. While most trials could generally be classified as starting treatment at earlier or later in gestational ages, there was a wide within-study variation that could not be explored with aggregate data analyses. • There is a markedly skewed distribution of effect estimates for each trial around the summary effect on pre-eclampsia, with more small positive trials than small negative trials. This suggests the possibilities of publication bias or that the differences in the characteristics of the women enrolled in small and large studies has an important influence on the effects of antiplatelet agents. Ways of overcoming these limitations by using individual patient data • Obtaining individual patient data from previously excluded trials will allow assessment of whether these trials are eligible for inclusion, both in terms of available outcomes and methodological quality. This will potentially increase the power and scope of the analyses. • Trial level information obtained by direct discussion with the trialists enables clarification of the definitions and measurements used. Data for measured, but previously unreported, outcomes can also often be obtained and trialists, who are part of a Collaborative group, may be able to provide missing outcome data. Furthermore, we will be able to apply common definitions across trials based on each woman's baseline and clinical data. For example, where possible, we will define pre-eclampsia based on a set of uniform criteria. • Subgroup analyses will be performed for a number of different risk factors using individual patient's specific risk data, rather than using the aggregated risk profile of all women enrolled in a particular trial. • By obtaining gestational age at randomisation for each individual woman, more precise patient-based analyses can be performed exploring whether gestation at treatment commencement alters the effectiveness of antiplatelets. • Formation of a Collaborative group, is likely to lead to a more complete identification of all relevant trials, including those previously unpublished. This may help overcome the potential for publication bias. The patient-level data will also allow us to explore whether there were important differences in the characteristics of women enrolled in small and large trials. Methods/design Objectives The objective of this review is to confirm that antiplatelet agents, given during pregnancy, reduce the incidence of pre-eclampsia. The review will then determine the size of this effect, and whether antiplatelets reduce the severity of pre-eclampsia and/or its impact on important outcomes for women and their babies. It will also explore whether the effect of antiplatelets differs by womens' risk profile, when commenced during pregnancy, and/or by dose. The main questions to be addressed in this review are: • Do antiplatelet agents, primarily low dose aspirin, have clinically important benefits for women at risk of developing pre-eclampsia and their babies? Investigation of this hypothesis will also explore whether the treatment effect differs, in a clinically meaningful way, between women with different risk factors such as those with a history of early onset pre-eclampsia, renal disease, diabetes, chronic hypertension, or autoimmune disease. • Does the planned dose of aspirin affect outcome in terms of preventing or delaying the onset of pre-eclampsia or other adverse outcomes, such as preterm birth or perinatal death? • Do the effects of aspirin differ according to gestation at onset of treatment? Identifying studies The search strategy to identify potentially eligible studies will include a search of the register of trials developed and maintained by the Cochrane Collaboration Pregnancy and Childbirth Review Group. Details of how this register is maintained are available elsewhere [ 35 , 36 ], but it involves extensive searching of bibliographic databases such as MEDLINE, The Cochrane Controlled Trials Register and hand searching of relevant journals. Trialists will be asked if they know of any further studies. [See Additional file 1 ] for the list of trials potentially eligible for inclusion. In addition, all members of the Collaborative Group will be asked to notify any unpublished trials of which they are aware. Inclusion and exclusion criteria for studies The inclusion and exclusion criteria for the types of study designs, participants, interventions and data completeness to be included in the review are listed below. Each potentially eligible study will be assessed independently by two members of the Secretariat, unblinded to the trial's identity. Any differences of opinion regarding the assessment of the inclusion criteria will be resolved by discussion between the two assessors. If differences cannot be resolved, a third member of the Secretariat will be asked to assess the study. If individual patient data are unavailable from an eligible trial, the trial will remain included in the review and aggregate data will be used. a. Study design Studies will be included in the review if they were randomised trials. Quasi-random study designs, such as those using alternate allocation, will be excluded. The level of allocation concealment within each trial will be assessed according to the criteria outlined in the Cochrane Handbook [ 37 ], and classified as either adequate, unclear or clearly inadequate. These assessments will be made together with the outcomes of thorough data checking procedures. b. Participants Participants will be pregnant women at risk of developing pre-eclampsia. Women who started treatment postpartum will be excluded, as will those who already have a diagnosis of pre-eclampsia at trial entry (defined as hypertension with new onset proteinuria after 20 weeks gestation, not due to renal disease). c. Interventions The interventions will be any comparisons of an antiplatelet agent (such as low dose aspirin or dipyridamole), or any combination of antiplatelet agents, compared with placebo or no antiplatelet agent. This is regardless of dose, mode of administration and irrespective of whether the antiplatelet is in combination with another drug. Trials that assessed only physiological outcomes following a short duration of intended therapy will be excluded. d. Completeness of follow-up The main analyses will include all trials that fulfill the previous inclusion criteria, regardless of completeness of follow-up. Sensitivity analyses will be undertaken to assess the effect of the inclusion of data from trials where only small numbers of enrolled participants have available outcome data. The threshold for an acceptable level of data completeness may vary by outcome. For example, for long-term follow-up of women and children, data may be included if follow-up was less than 80% provided that substantive bias between the groups was unlikely. Other outcomes from each trial may only be included in the analysis if available for 80% or more of women. Data collection, data management and confidentiality The individual patient data provided by the Collaborators will be de-identified, re-coded as required and stored in a custom-designed Microsoft ACCESS database. It will not include any patient identifying information such as names or addresses. Electronic data will be located on a secure, password protected network server. Copies of hardcopy data will be stored in locked filing cabinets until converted into electronic format, and will then be securely destroyed. Only authorised personnel will have access to the data. All data will be securely stored and archived according to the policies of the major funder, the Australian National Health and Medical Research Council. The data will be checked with respect to range, internal consistency, consistency with published reports and missing items. Trial details such as randomisation methods, and dose and timing of the interventions will be cross-checked against any published reports, trial protocols and data collection sheets. Integrity of the randomisation process will be examined by reviewing the chronological randomisation sequence and pattern of assignment, as well as the balance of prognostic factors across treatment groups (taking into account stratification factors). Inconsistencies or missing data will be discussed with the individual trialists and attempts will be made to resolve any problems by consensus. Each trial will be analysed individually, and the resulting analyses and trial data will be sent to the trialists for verification. Data items requested from the trialists There has been extensive consultation with the PARIS Collaborative Group regarding what data items to collect for each woman in the analyses. The following section contains the list of data items requested from trialists, which has been compiled following this consultation. More detailed definitions for the data items listed below can be found in Tables 1 and 2 . Details of the suggested coding for each of the following variables can be found in [see Additional file 2 ]. A formal request for the provision of the individual patient data was sent in April 2004 [see Additional file 3 ]. Table 1 Key definitions for enrolment characteristics Variable Definition gestational hypertension de novo systolic BP ≥ 140 mmHg and/or diastolic BP ≥ 90 mmHg after 20 weeks' gestation, without proteinuria severe hypertension systolic BP ≥ 160 mmHg and/or diastolic BP ≥ 110 mmHg proteinuria ≥ 1+ on dipstick, or ≥ 300 mg/24 hours, or spot urine protein/creatinine ratio ≥ 30 mg/mmol pre-eclampsia (for women normotensive at trial entry) de novo systolic BP ≥ 140 mmHg and/or diastolic BP ≥ 90 mmHg after 20 weeks' gestation with new-onset proteinuria as described above pre-eclampsia (for women with chronic hypertension at trial entry) new-onset proteinuria as described above pre-eclampsia (for women with chronic hypertension and proteinuria at trial entry) signs and symptoms of superimposed pre-eclampsia after 20 weeks' gestation, for example worsening of hypertension or proteinuria early onset proteinuria proteinuria as defined above, occurring ≤ 33 weeks + 6 days of gestation early onset pre-eclampsia hypertension and early onset proteinuria as described above chronic hypertension Essential hypertension : BP ≥ 140/90 mmHg pre-conception or in first half of pregnancy without an underlying cause, or Secondary hypertension : hypertension associated with renal, renovascular, cardiac and endocrine disorders intrauterine growth restriction (IUGR) or small for gestational age (SGA) growth below the 3 rd centile, or as defined in the individual trial miscarriage / fetal death any death in utero perinatal death death in utero or within the first 7 days of life neonatal death live born and any reported death within the first 28 days Table 2 Key definitions for outcome measures Main outcomes Definition pre-eclampsia as defined in Table 1 pregnancy loss / neonatal death miscarriage, fetal death or death of a liveborn infant before hospital discharge pre-term birth pre-term birth : ≤ 37 weeks + 6 days of gestation moderately pre-term birth : ≤ 33 weeks + 6 days of gestation extremely pre-term birth : ≤ 27 weeks + 6 days of gestation small for gestational age (SGA) infant infant with birth-weight below the 3 rd centile, or as defined in the individual trial pregnancy with serious adverse outcome pregnancy with any of the above main outcomes for the woman or any fetus/baby, or the death of the woman. If sufficient data available, severe maternal morbidity will also be included in this definition. Other outcomes Definition early onset pre-eclampsia as defined in Table 1 maternal death death during pregnancy or up to 42 days after termination of the pregnancy antepartum haemorrhage any vaginal bleeding before the onset of labour placental abruption clear evidence of placental separation severe maternal morbidity including eclampsia, HELLP syndrome, DIC, pulmonary oedema, liver failure, renal failure or CVA/stroke infant death live born and any reported death from 29 days to 1 year of life or after hospital discharge neonatal bleeding abnormal bleeding in the neonatal period including periventricular haemorrhage, gastrointestinal, umbilical or other sites a. Characteristics of trials 1 informed consent 2 dates trial opened and closed to accrual 3 total number of women randomised 4 treatments used in each arm of the trial 5 intended duration of treatments 6 definitions of key outcomes used in the trial 7 method of random allocation 8 stratification factors used 9 methods of allocation concealment b. Characteristics of enrolled women at trial entry 1 unique identifier for the enrolled woman, coded for anonymity 2 date of randomisation 3 gestational age at randomisation, or best estimate of expected date of delivery or last menstrual period 4 woman's date of birth or age 5 any previous pregnancy 6 blood pressure (systolic, diastolic, whether diagnosed as raised) 7 presence of proteinuria 8 presence of oedema 9 risk factors – multiple pregnancy, autoimmune disease, renal disease, diabetes, chronic hypertension, previous gestational hypertension or pre-eclampsia/eclampsia (including early onset disease), family history of pregnancy-related hypertensive disorders, previous fetal growth restriction, previous perinatal death, abnormal uterine artery Doppler flow and abnormalities on other diagnostic tests c. Maternal data items 1 hypertension during pregnancy (highest blood pressure, diagnosis of severe hypertension) 2 proteinuria during pregnancy (including date and/or gestation at onset) 3 oedema during pregnancy 4 pre-eclampsia 5 drug treatment for pre-eclampsia or hypertension 6 severe maternal morbidity (including eclampsia, renal failure, disseminated intravascular coagulation, liver failure, HELLP syndrome, stroke) 7 onset of labour (spontaneous or induced or pre-labour caesarean section) 8 mode of delivery (vaginal, vaginal assisted, caesarean section) 9 antepartum haemorrhage (all, placental abruption) 10 postpartum haemorrhage and/or estimated blood loss at delivery 11 maternal mortality d. Fetal / neonatal / child data items (for each fetus) 1 gestational age at birth and/or date of birth 2 birthweight 3 gender 4 small for gestation age (as defined within each trial: including centile charts and cut-off point used) 5 miscarriage or stillbirth: date and/or gestational age at death/loss 6 neonatal or infant death: date and/or age at death 7 admission to special care baby unit or neonatal intensive care unit 8 use of assisted ventilation 9 number of days in hospital or date of hospital discharge 10 neonatal bleeding (for example, periventricular haemorrhage) 11 child growth and development (such as cerebral palsy, blindness, deafness, significant cognitive delay – as defined within each trial) Planned analyses This section contains a summary of the planned analyses. The full, detailed analysis plan will be discussed and agreed upon by the Collaborators before any data have been analysed. Analysis will aim to be of all women ever randomised and will be based on intention to treat. In the main analyses a two stage approach will be taken. Outcomes will be analysed in their original trial and then these separate results will be combined to give an overall measure of effect. A fixed effect model will be used and the assumption of homogeneity of treatment effects will be tested using the chi squared test. The I 2 statistic will also be used to assess consistency of results. 1. Outcomes to be analysed The main analyses comparing the effect of antiplatelet agents with placebo or no antiplatelet agents will be undertaken for all outcomes listed below, for the woman and any fetus/baby. The planned sub-group and sensitivity analyses will be restricted to the designated main outcomes listed as follows: a. Main outcomes • pre-eclampsia • pregnancy loss / neonatal death • pre-term birth • small for gestational age infant • pregnancy with serious adverse outcome b. Other outcomes • early onset pre-eclampsia • maternal death • severe maternal morbidity • antepartum haemorrhage • placental abruption • induction of labour • Caesarean section delivery • postpartum haemorrhage • gestation at delivery • infant admission to special care or neonatal intensive care unit • infant required assisted ventilation • neonatal bleeding • infant death 2. Planned sub-group analyses The planned sub-group analyses will be restricted to the designated main outcomes unless there are clear indications for expanding the analyses further. a. Trial-level characteristics The effect of antiplatelet therapy may vary across the trials in the meta-analysis because they have used different agents in different ways. To explore this further, analyses are planned whereby trials will be grouped according to the agent used and by dose. These analyses will focus on the main outcomes. Trials will be classified into subsets based on the following: (i) type of antiplatelet(s) Trials will be grouped by the type of antiplatelet agent (trials that used aspirin alone, trials that used other antiplatelet agents, trials that used both aspirin and other antiplatelets agents) given as the active treatment. (ii) daily dose of aspirin In trials that used aspirin alone, trials will be grouped by planned aspirin dose (<75 mg, 75–149 mg, ≥ 150 mg). b. Patient-level characteristics One of the strengths of individual patient data reviews is that it allows us to assess eligibility and outcome using individual women's characteristics. Subgroup analyses will explore whether any particular risk factors act as effect modifiers. That is, are there any particular types of women who benefit more or less from antiplatelet agents? These analyses will take account each individual woman's own characteristics, rather than relying on summary measures of the 'average' risk profile of all participants in an individual trial. Analyses will be undertaken to explore whether there are any particular types of women who benefit more or less from antiplatelet agents based on the following criteria: (i) risk factor profile for pre-eclampsia at trial entry Women normotensive at trial entry: - previous hypertensive disorders of pregnancy : previous early onset (≤33 weeks + 6 days gestation) pre-eclampsia or eclampsia / previous pre-eclampsia / previous gestational hypertension / no previous hypertensive disorders of pregnancy / no previous pregnancy but family history of hypertensive disorders of pregnancy / no previous pregnancy and no family history of hypertensive disorders of pregnancy - diabetes : pre-existing diabetes at enrolment / no pre-existing diabetes at enrolment - renal disease : pre-existing renal disease / no pre-existing renal disease - autoimmune disease : autoimmune disease / no autoimmune disease - multiple pregnancy : multiple pregnancy / singleton pregnancy - maternal age : <20 years / 20–35 years / >35 years. Maternal age may also be analysed as a continuous variable. - diagnostic test results : abnormal uterine artery Doppler scan / other diagnostic test abnormalities / no abnormal diagnostic test results - previous SGA : previous small for gestational age infant / previous infant not small for gestational age / no previous infant - primigravida : first pregnancy with no other risk factors / first pregnancy with one or more risk factors / second or subsequent pregnancy with one or more risk factors / second or subsequent pregnancy with no risk factors Women with hypertension at trial entry: - hypertension: gestational hypertension / chronic hypertension (ii) gestation at trial entry To determine whether antiplatelet agents are differentially effective if given earlier in pregnancy and to determine the magnitude of any difference, gestational age at randomisation will be primarily analysed as a continuous variable in regression analyses. However, a subgroup analysis with women classified according to the following categories may also be performed: <16 weeks, 16–19 completed weeks, 20–23 completed weeks, 24–27 completed weeks, ≥28 weeks gestation. If numbers are insufficient for any category, categories will be combined. 3. Planned sensitivity analyses a. To assess whether results are robust to the inclusion or exclusion of particular types of trials or patients, the following sensitivity analyses will be conducted: • exclusion of trials that did not use a placebo for the control group • exclusion of trials of small size • exclusion of poor quality trials (assessed by adequacy of allocation concealment, blinding, completeness of follow-up and other data checking procedures) b. To assess whether results are robust to different methods of analysis or different definitions of pre-eclampsia [ 38 , 39 ], the following sensitivity analyses will be conducted: • comparison of analyses using random effects and fixed effect models • comparison of analyses using individual patient data (IPD) only with analyses using individual patient data and aggregate data where IPD unavailable • comparison of analyses using different definitions of pre-eclampsia These sensitivity analyses will be carried out for the main outcomes. 4. Additional analyses Depending on what data are available, the level of heterogeneity encountered and available time a one-stage modeling approach may also be undertaken to further explore important key outcomes as appropriate. Ethical considerations Participants in the individual trials have previously given informed consent to participate in their respective trial. The data for this project are to be used for the purpose for which they were originally collected and are available through an agreement between all trialists of the PARIS Collaboration. These trialists remain the custodians of their original individual trial data at all times. Data are provided on the stipulation that all trials have received ethical clearances from their relevant bodies. Project management Membership of the PARIS Collaboration will be representative(s) from each of the trials contributing data to the review with an accompanying project coordination and data management structure as described in this section. The membership and responsibilities of each of these management groups is as follows: a. Steering Group The Steering Groupwill be responsible for project management decisions and will meet approximately 4–6 times per year, usually via teleconference. Membership: D Henderson-Smart 1 (co-chair), L Duley 2 (co-chair), L Askie 1 (project coordinator), M Showell 1 (project administrator), B Farrell 2 , L Stewart 3 , M Clarke 4 , J King 5 , C Roberts. 1 The first six members act as the Secretariat . 1 Centre for Perinatal Health Services Research, University of Sydney, Australia; 2 Resource Centre for Randomised Trials, University of Oxford, Oxford, UK; 3 Medical Research Council Clinical Trials Unit, London, UK; 4 UK Cochrane Centre, Oxford, UK; 5 Royal Women's Hospital, Melbourne, Australia. b. Advisory Group The aim of the Advisory Group is to facilitate representative input from the Collaborative Group to the Steering Group. Membership of the Advisory Group will include people who have contributed to the trials included in the project and other international experts. Each trial that recruited over 1000 women will be invited to have a representative on the Advisory Group. This group will not have regular meetings, but may be consulted from time to time by means of email or teleconference, and may have occasional ad hoc meetings. Co-chairs of the Advisory Group: C Redman, C Roberts. c. Collaborative Group All potentially eligible trialists will be contacted and invited to become members of the Collaborative Group. The corresponding author for each study will be contacted in the first instance. If there is no response, the associated statistician, data manager and/or other authors will be contacted. This process will be updated annually for the duration of the project, to ensure that new trialists are offered the opportunity to join the project and contribute their data. d. Project coordination centre The project will be coordinated from the Centre for Perinatal Health Services Research (CPHSR), University of Sydney, NSW, Australia. The coordination centre will be responsible for the daily management of the project including correspondence, newsletter production, maintaining current trialist contact information and meeting/teleconference organisation. e. Data management centre The data management centre, based at the UK Cochrane Centre, will be responsible for the receipt, storage, and analysis of project data as directed by the Collaborative Group via the Steering Group. f. Collaborators' meetings All members of the Collaboration, including the Steering Group, the Advisory Group, and representatives of each participating trial, will be invited to attend regular Collaborators' meetings. These meetings will be scheduled, where possible, to coincide with the biennial International Society for the Study of Hypertension Pregnancy (ISSHP) congresses. The meetings will be designed to allow maximum input from the participating trialists into the design, conduct, analysis and reporting of the project's results. The final Collaborators' meeting, at which the results will be presented for discussion, is scheduled for 2006 in Oxford, UK. The discussion at this meeting will provide the basis for the paper publication. Funding The National Health and Medical Research Council (NHMRC) of Australia have provided funding for the project through the Centre for Perinatal Health Services Research, University of Sydney. These funds are: a three year project grant (ID: 253636) for the overall project administration base in Sydney, and a Sidney Sax Public Health Postdoctoral Fellowship (ID: 245521), based in Oxford and Sydney. Additional support is being provided by the Resource Centre for Randomised Trials and the UK Cochrane Centre, located in Oxford, UK, and the Medical Research Council Clinical Trials Unit in London, UK. Publication policy The results of the project's analyses will be presented to, and discussed with, the Collaborative Group before publication. The aim of publication will be presentation of the results, rather than their interpretation. The main manuscript will be prepared by the Secretariat, and then circulated to the Steering and Advisory Groups for comment and revision. The revised draft paper then will be circulated to all members of the Collaborative Group for comment before publication. All publications using these data will be authored in the name of the PARIS Collaboration, as follows: Perinatal Antiplatelet Review of International Studies (PARIS) Collaboration. Discussion Despite good evidence that antiplatelet agents (principally low dose aspirin) reduce the incidence of pre-eclampsia and its consequences, such as preterm birth and perinatal mortality, uncertainty remains regarding whether some women (in terms of risk) benefit more than others, when in pregnancy treatment should ideally start, and whether treatment effectiveness is dependent on antiplatelet dose. The best way to answer these questions is to utilise existing individual patient data from all women enrolled in trials that have addressed this question. This approach has been described as the 'gold standard' of systematic review methodology as it allows for more powerful and flexible analysis of both subgroups and outcomes. The PARIS Collaboration has been formed to undertake a systematic review of all available trials, with meta-analysis based on individual patient data, to answer these important clinical questions. This will be the first individual patient data review in the perinatal field. Provision of data by the participating Collaborators commenced in 2004, and results will be ready for presentation in 2006. Following consultation and discussion with the Collaborative Group, the main publication is expected in early 2007. Competing interests The author(s) declare that they have no competing interests. Authors' contributions All authors, the named members of the PARIS Collaboration Steering Group, contributed to the development of the protocol, and read and approved the final manuscript. Pre-publication history The pre-publication history for this paper can be accessed here: Supplementary Material Additional File 1 List of eligible trials text document listing potentially eligible trials Click here for file Additional File 2 Suggested coding sheet table listing variables and suggested coding Click here for file Additional File 3 Data provision form form to collect trial level data and data provision procedures Click here for file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555958.xml |
551531 | Comparison of the oxidative phosphorylation (OXPHOS) nuclear genes in the genomes of Drosophila melanogaster, Drosophila pseudoobscura and Anopheles gambiae | An analysis of nuclear-encoded oxidative phosphorylation genes in Drosophila and Anopheles reveals that pairs of duplicated genes have strikingly different expression patterns. | Background The accessibility of whole-genome sequence data for several organisms, together with the development of efficient computer-based search tools, has revolutionized modern biology, allowing in-depth comparative analysis of genomes [ 1 - 4 ]. In many cases, comparisons among species at various levels of divergence have helped to define protein-coding genes, recognize nonfunctional genes, and find regulatory sequences and other functional elements in the genome. When applied to a set of genes correlated by function and/or subcellular localization of their products, intra- and interspecies comparative analyses can be especially efficient tools to obtain information on the functional constraints acting on the evolution of the gene set and on the mechanisms regulating its coordinate expression. A set of genes present in all eukaryotic genomes and expected to be subject to peculiar evolutionary constraints is represented by the genes involved in oxidative phosphorylation (OXPHOS), the primary energy-producing process in all aerobic organisms [ 5 ]. To generate cellular ATP, OXPHOS uses the products of both nuclear and mitochondrial genes, organized in five large complexes embedded in the lipid bilayer of the inner mitochondrial membrane. Except for complex II, which is formed by four proteins encoded by nuclear genes, the other respiratory complexes depend on both mitochondrial and nuclear genomes; so, assembling the OXPHOS complexes and fine tuning their activity to satisfy cell- and tissue-specific energy demands requires specialized regulatory mechanisms and evolutionary strategies to optimize the cross-talk between the two genomes and ensure the coordinated expression of their relevant products. Analysis of co-regulated mitochondrial and nuclear genes, and of the transcription factors regulating the functional network they constitute, might also be a useful approach to investigate the origin of mitochondrial dysfunction in humans. Disorders of mitochondrial oxidative phosphorylation are now recognized as the most common inborn errors of metabolism, affecting at least one in 5,000 newborn children [ 6 ]. In this context, the expanding spectrum of identified mitochondrial proteins provides an opportunity to test a whole new range of candidate genes whose mutations may be responsible for common human diseases. For example, a recent study by Mootha et al . [ 7 ] suggests a promising strategy for clarifying the molecular etiology of mitochondrial pathologies by profiling the tissue-specific expression pattern of candidate mitochondrial proteins. Despite the long evolutionary divergence time, many key pathways that control development and physiology are conserved between Drosophila and humans, and about 70% of the genes associated with human disease have direct counterparts in the Drosophila genome [ 8 , 9 ]. For example, the potential role of Drosophila as a model system for understanding the molecular mechanisms involved in human genetic disease is validated by the recent identification of a Drosophila mutation causing a necrotic phenotype that mimics in detail the diseases that arise from serpin mutations in humans [ 10 ]. It has been suggested that comparisons between D. melanogaster and other species of the genus Drosophila could provide a model system for developing and testing new algorithms and strategies for the functional annotation of complex genomes [ 3 ]. To obtain new information on the evolution of a set of genes that control a basic biological function by encoding products targeted to a specific cellular compartment, we have performed a comparative analysis of the OXPHOS genes of D. melanogaster and D. pseudoobscura ; the complete genome of the latter was recently made available by the Baylor Human Genome Sequencing Center. These two species are the only species of the Drosophila genus for which whole-genome sequence data exist at present [ 11 - 13 ]. We also took advantage of the complete sequence of the A. gambiae genome [ 14 ] to compare the Drosophila OXPHOS genes with those of this more distantly related dipteran (the divergence time between D. melanogaster and A. gambiae is thought to be approximately 250 million years, as compared to 46 million years between D. melanogaster and D. pseudoobscura [ 15 , 16 ]). Although extensive reshuffling within and between chromosomal regions is known to have occurred since the divergence of Anopheles from Drosophila [ 4 , 17 , 18 ], we show that in these organisms the conservation of the OXPHOS genes is still sufficient to permit their meaningful comparison. Here we report the identification of 78 D. pseudoobscura and 78 A. gambiae genes representing the counterparts of D. melanogaster OXPHOS genes which, in turn, were previously identified as putative orthologs of human OXPHOS genes [ 19 ]. We have annotated these genes, taking into account conservation in amino-acid sequence, intron-exon structure, intron length, and the presence of duplications in the genome. The conservation of genomic organization and evidence from evolutionary trees based on sequence similarity suggest that these genes are one-to-one orthologs in the three species, and that in many cases they originated (produced?) duplicates by transpositional and/or recombinational events during evolution. We have identified in the three dipteran genomes a total of 47 genes that probably originated by duplication of the above-mentioned genes, and we show that the duplicate gene has usually acquired a pattern of expression strikingly different from that of the gene from which it derived. Moreover, when the comparison is possible, the gene duplicate almost always shows a strongly testis-biased expression, in contrast to the soma-biased expression of its parent gene. Results and discussion Identification and comparative annotation of D. pseudoobscura and A. gambiae OXPHOS genes We have previously reported [ 19 ] the identification of 285 D. melanogaster nuclear genes encoding mitochondrial proteins that represent the counterparts of human peptides annotated in the Swiss-Prot database as mitochondrial [ 20 ]. On the basis of comparative evidence obtained by BLASTP analysis, 78 of these genes are involved in the OXPHOS system, encoding 66 proteins known to be components of the five large respiratory complexes and 12 proteins involved in oxidative phosphorylation as accessory proteins. To identify the putative counterparts of the D. melanogaster OXPHOS genes in D. pseudoobscura and A. gambiae we performed a TBLASTN search [ 13 , 21 ] on the whole genome sequences of these species using the amino-acid sequences of the 78 D. melanogaster peptides as queries. Sequences giving the best reciprocal BLAST hits were tentatively assumed to identify functional counterparts in two species if they could be aligned over at least 60% of the gene length and the BLAST E-score was less than 10 -30 . By these criteria, all the 78 D. melanogaster OXPHOS genes investigated have a counterpart both in D. pseudoobscura and in A. gambiae . To better compare the structure of the OXPHOS genes in the three dipteran species, we used the predicted coding sequences as queries for a search of expressed sequence tags (EST) [ 21 ], and used the retrieved sequences to annotate the transcribed noncoding sequences of the A. gambiae genes investigated. Although little EST information is available for D. pseudoobscura , it was still possible to predict unambiguously the exon-intron gene structure of the OXPHOS genes in this species, as well as the amino-acid sequence of their full-length products, by exploiting the high level of similarity with D. melanogaster . The results of BLAST analysis, together with the construction of phylogenetic trees that also include other genes that show lesser but still significant sequence similarity to the 78 genes assumed to be one-to-one orthologs in the three species investigated (see below), strongly suggest that the newly identified D. pseudoobscura and A. gambiae genes are the functional counterparts of the 78 D. melanogaster genes used as probes. Table 1 lists the 78 putative orthologous OXPHOS genes in the three dipteran genomes and their cytological location. For each gene, a record showing the gene map and reporting the annotated genomic sequences as well as the mRNA and protein sequences is available and can be queried at the MitoComp website [ 22 ] (see also Additional data files). MitoComp also compares the structure of the D. melanogaster , D. pseudoobscura and A. gambiae putative orthologous genes and their duplications when present (see below), and aligns the orthologous coding sequences (CDS), and also aligns their deduced amino-acid products with the corresponding human protein. Amino-acid sequence comparison For the products of the OXPHOS genes investigated, the D. melanogaster / D. pseudoobscura average amino-acid sequence identity is 88%, compared to 64% between D. melanogaster and A. gambiae . Figure 1 shows the frequency distribution of sequence identities, and Additional data file 1 lists all pairwise identity values between the products of the 78 OXPHOS genes when orthologous D. melanogaster / D. pseudoobscura , D. melanogaster / A. gambiae and D. melanogaster /human gene products are compared. A multiple alignment of each cluster of homologous proteins is shown at the MitoComp website [ 22 ]. It should be kept in mind that identity values reported in Figure 1 and in the table in Additional data file 1 were calculated on the whole sequence of the predicted unprocessed proteins; they are much higher if the putative amino-terminal pre-sequences are excluded, since such sequences, possessed by most mitochondrion-targeted products, show little amino-acid sequence conservation [ 23 , 24 ], although they do share specific physicochemical properties [ 25 , 26 ]. When only the predicted mature protein is considered, the average percentage identity increases to 90% between D. melanogaster and D. pseudoobscura , and to 70% between D. melanogaster and A. gambiae . A striking example of evolutionary conservation is provided by the genes encoding cytochrome c (an essential and ubiquitous protein found in all organisms) in the three dipteran species: the amino-acid sequences of the gene products are identical in D. melanogaster and D. pseudoobscura , whereas 96% identity is preserved between Drosophila and Anopheles . Coding sequences are also extremely conserved, suggesting that the nucleotide sequence itself is subject to strong evolutionary constraints, maybe due to codon usage bias. Only synonymous substitutions (21 out of 108 codons) were found on comparing D. melanogaster and D. pseudoobscura cytochrome c coding sequences, whereas 28 synonymous substitutions and only four nonsynonymous substitutions were observed between D. melanogaster and A. gambiae (see MitoComp website [ 22 ]). Gene structure comparisons It is well known that a given function may be supplied in different species by genes that are not directly derived from a common ancestor, that is, by paralogous, not orthologous, genes. Therefore, we thought it would be interesting to compare the structural organization of the OXPHOS genes in the three species investigated, on the principle that it should be possible to infer derivation from a common ancestor, that is, 'structural orthology', if an identical or very similar overall structure was preserved. As the introns of the putative orthologous OXPHOS genes in the three species are, as expected, too divergent in DNA sequence to be aligned, we used conservation of number of introns, conservation of their location in the coding sequence, and preservation of the reading frame with respect to the flanking exons as our primary criteria. With the only exception of Dpse \ CG5037 , putatively encoding protoheme IX farnesyltransferase, whose 5' genomic sequence was impossible to find in the relevant contig assembly, all other investigated D. pseudoobscura genes show a structural organization almost identical to that of their D. melanogaster counterparts. Of the 78 Anopheles genes studied, 39 maintain the structural organization observed in Drosophila , whereas gain or loss of introns occurred in 33, and in six the location of introns is not preserved at all. In agreement with a previous report [ 4 ], the intron-exon structure of the gene appears to be conserved in all three dipteran species when splicing of alternative coding exons occurs: the alternative splice forms of both the Drosophila NADH-ubiquinone oxidoreductase acyl carrier protein ( mtacp1 , CG9160 ) [ 27 ] and the Drosophila ATP synthase epsilon chain ( sun , CG9032 ) [ 19 ] have very similar counterparts in Anopheles , as shown by genomic structure comparison, alignment of splice variants and EST mapping (Figure 2 ). Genes encoding the acyl carrier protein ( mtacp1 ) in the three species are characterized by the mutually exclusive use of homologous exons that are repeated in tandem (Figure 2a ). The duplicate exons occur at the same location in the aligned amino-acid sequences, and are flanked on both sides by a phase 1 intron. When the sequences of the duplicated exons are compared, they show the expected divergence pattern (that is, the similarity between duplicate exons within a gene is less than the similarity of each exon to its equivalent in the orthologous gene). Evidence from genomic and transcribed sequences (GenBank accession numbers BI510891 and BI508135) shows that the duplicated mtacp1 exons are also preserved in the more distantly related insect Apis mellifera (honeybee) (Figure 2c,d ), indicating a specific adaptive benefit for this gene structure, as also suggested by the evolutionary convergence leading to the occurrence of alternative splicing in members of three different ion-channel gene families from Drosophila to humans [ 28 ]. However, there is no evidence from ESTs that duplicated mtacp1 exons undergo alternative splicing in vertebrates and nematodes. Analysis of intron length Interspecies comparison of the introns of putative orthologous genes indicates that there is little constraint on their nucleotide sequence, which undergoes nucleotide substitutions at a rate comparable to that of pseudogenes [ 29 ]. However, several observations suggest that intron size is subject to natural selection. For example, in D. melanogaster and several other organisms the distribution of intron length has been shown to be asymmetrical, with a large group of introns falling into a narrow distribution around a 'minimal' length and the remaining showing a much broader length distribution, ranging from hundreds to thousands of base-pairs [ 30 - 32 ]. Of the introns that interrupt the coding sequence in the 78 OXPHOS genes investigated in the present study, 88 (64.7%) of 136 in D. melanogaster , 96 (70.5%) of 136 in D. pseudoobscura and 87 (67.9%) of 128 in A. gambiae fall into the short-size class (Figure 3a ). However, in A. gambiae the length distribution of these introns appears slightly broader (62-150 bp, compared with 51-100 bp in both Drosophila species). The remaining introns show a broad length distribution, ranging from 151 to 4,702 bp with no clear boundary between classes. A comparison of the length of introns in corresponding positions in the putative D. melanogaster , D. pseudoobscura and A. gambiae orthologs suggests that changes from the short-size to the long-size (more than 300 bp) intron class, or the converse, have been rare in the evolutionary history of these species: only seven class changes were observed comparing D. melanogaster and D. pseudoobscura introns, and six between D. melanogaster and A. gambiae (Figure 3b ). On the whole, our data confirm the highly asymmetrical intron length distribution in D. melanogaster and extend this finding to the introns of the D. pseudoobscura and A. gambiae OXPHOS genes. OXPHOS gene duplications It is generally accepted that gene duplication is the basic process that underlies the diversification of genes and the origination of novel gene functions [ 33 ]; however, many features of this process are still elusive. To obtain more information on the molecular evolution of the genes involved in the OXPHOS system, we searched the genomes of D. melanogaster , D. pseudoobscura and A. gambiae for duplications of the 78 OXPHOS genes whose orthologs we have identified in the three species. Duplicate gene pairs were tentatively identified within each genome as best reciprocal hits with an E-value of less than 10 -20 in both directions in a TBLASTN search using the default parameters. Deciding whether two proteins may be considered homologous becomes difficult when their sequence identity is within the 20-30% range (the so-called 'twilight zone' [ 34 ]), and so the following additional criteria were used: first, the two sequences could be aligned over more than 60% of their length; second, the putative processed proteins encoded had to have more than 40% identity; and third, amino-acid percentage similarity had to be larger than percentage identity [ 35 ]. Even if meeting these criteria and reported as different genes in the ENSEMBL database [ 36 ], identical Anopheles nucleotide sequences were excluded from further analysis, as they are likely to reflect annotation artifacts. Duplications, or in some instances triplications, of 24 OXPHOS genes were found. Overall, we identified 47 genes (20 in D. melanogaster , 19 in D. pseudoobscura and eight in A. gambiae ) each of which shows significant similarity with one of the 78 OXPHOS genes reported above. When the structure of a member of a paralogous gene set indicates that it has been produced by retroposition, it seems reasonable to assume that it is derived from a pre-existing 'parent' gene. For duplicates not clearly originating by retroposition, we also assume, on the basis of the much higher level of conservation and expression, that the genes we find to be the structural orthologs in all three species are the parent ones, and in this case also we will henceforth refer to their paralogs as OXPHOS gene duplicates. The amino-acid percentage identity between the products of duplicate gene pairs ranges from 40% to 85%. For each of the OXPHOS gene duplicates, cytological localization, number of exons interrupting the coding sequence, and number of ESTs found in the D. melanogaster and A. gambiae EST databases are reported in Table 2 . Neighbor-joining trees derived from distance matrix analysis and showing the inferred evolutionary relationship between members of each gene cluster are available at the MitoComp website [ 22 ]. Duplications (or triplications) of 16 of the 78 OXPHOS genes investigated were found in both D. melanogaster and D. pseudoobscura . In such cases, to assign pairwise orthology, besides taking into account conservation of structural organization, given the general conservation of microsyntenic gene order in the two species, we used the products of D. melanogaster genes flanking the duplicate loci to search for homologous sequences also flanking the same genes in the D. pseudoobscura genome. The genomic organization of many OXPHOS duplicates shows that they were originated by retropositional events, because they are intronless, or have only very few introns that are likely to have been inserted into the coding sequence after the duplication event. In other cases, duplication apparently resulted from transposition of genomic DNA sequences or from recombinational events, as duplicate genes maintain an identical or very similar structural organization. On the basis of the presence of the duplication in both species, supported by evidence from evolutionary trees and conservation of microsyntenic gene order, it can be inferred that 15 of the duplications identified occurred before the D. melanogaster / D. pseudoobscura divergence (about 46 million years ago). On the other hand, five duplications were found only in D. melanogaster and four only in D. pseudoobscura ; in these instances, if the duplication occurred before the divergence of the two species, it has been followed by loss of one of the copies in the lineage leading to the species in which the gene is no longer duplicated. On the assumption that the rate of gene duplication is constant over time, this translates to approximately 0.0014 duplications per gene per million years (4 or 5 duplications per 78 genes per 46 million years) that achieved fixation and long-term preservation in the genome. This value is about twofold lower than the 0.0023 value calculated by Lynch and Conery [ 37 ] for the 13,601 genes of the whole genome of D. melanogaster . However, it can be argued that the rate of long-term preservation in the genome of OXPHOS gene duplicates cannot be meaningfully compared with the general rate of preservation of duplicates in the whole genome since, while recent data suggest that in eukaryotic genomes there is preferential duplication of conserved proteins [ 38 ], duplicates of genes that encode subunits of multiprotein complexes, as most of the genes we have investigated do, negatively influence the fitness of an organism [ 39 ], and are therefore unlikely to become fixed in the population. In summary, it appears reasonable to assume that the preservation in the genome of OXPHOS gene duplicates should occur very infrequently, unless special mechanisms allowing their fixation in the population are present (see the next section). In A. gambiae we found only four duplications and two triplications of the OXPHOS genes analyzed; of these, four involve genes also duplicated in one or both Drosophila species (Table 2 ). Pairwise orthology could not be assigned between Drosophila and Anopheles gene duplicates as neither microsynteny nor evolutionary trees provide sufficient evidence for the origin of the gene pairs from a single-copy gene before the Drosophila / Anopheles divergence. Expression pattern of OXPHOS gene duplicates The relative abundance of ESTs in a EST library may be assumed roughly to reflect the level of expression of each mRNA in the tissues from which the library was prepared. We therefore used the mRNA sequences predicted in silico to be transcribed from the OXPHOS duplicate genes investigated in this work as queries in a search of the public D. melanogaster and A. gambiae EST databases to infer the relative abundance of the mRNA copies from the hits scored. For each gene, the number of ESTs found in the databases is detailed in Table 2 . With the exception of one of the paralogs of the A. gambiae gene encoding ubiquinol-cytochrome c reductase core protein 1, in all cases the search found the number of ESTs originating from the duplicate gene was strikingly lower than that originating from the putative parent gene, in both D. melanogaster and A. gambiae (in total, 100 versus 1,747 in D. melanogaster and 60 versus 687 in A. gambiae ). A smaller number of ESTs originating from the OXPHOS gene duplicates was observed even in A. gambiae EST libraries that are normalized. Remarkably, and regardless of the mechanism of the duplication, in D. melanogaster , in which several organ-specific or developmental stage specific libraries are available, the search showed that the expression of the OXPHOS gene duplicates is strongly testis-biased, as 97 out of the 100 ESTs originating from them were found in testis-derived libraries, while only 27 out of the 1,769 ESTs originating from the parent genes were found in such libraries, the bulk of them being instead found in libraries derived from embryos or somatic tissues. Our finding that the expression of the OXPHOS gene originated by duplication is strongly testis-biased is validated by the data obtained by Parisi et al . [ 40 ] using the FlyGEM microarray to identify D. melanogaster genes showing ovary-, testis- or soma-biased expression. With the exception of CG7349 , CG30354 , CG30093 and CG12810 , for which no data were presented by Parisi et al . [ 40 ], all other genes reported in this work as OXPHOS gene duplicates were found in the genomic fraction showing testis-biased expression, whereas all the parent genes present in the dataset showed soma-biased expression. Additional data file 2 summarizes the relevant data extracted from Parisi et al . [ 40 ]. The pattern of strongly testis-biased expression of OXPHOS gene duplicates holds for a further sample of 40 duplications of genes annotated in the MitoDrome database [ 19 ] as encoding products that are mitochondrion-targeted but not involved in the OXPHOS system. For 15 of these no data are provided by Parisi et al . [ 40 ], but all the remaining 25 genes show a testes-biased expression (data not shown). Duplications of genes encoding OXPHOS subunits, for which stoichiometry is important, are likely to be strongly deleterious owing to the negative consequences of an imbalance in the concentration of the respiratory complex constituents, unless, as proposed by Lynch and Force [ 41 ], 'subfunctionalization' and/or a differential expression pattern of duplicate copies occurs. In this case, the duplicate OXPHOS genes would have a reduced or absent capacity to functionally complement mutations in their parent genes, in contrast to what is generally assumed to be the main short-term advantage of gene duplication. In D. melanogaster at least there is evidence for this, as FlyBase [ 42 ] and BDGP P-Element Gene Disruption Project [ 43 ] searches for P-insertion mutants in the D. melanogaster OXPHOS genes found that lethal alleles for 11 out of 19 D. melanogaster parent genes are known (see the MitoComp website [ 22 ]), indicating that loss-of-function of the parent gene cannot be compensated for by the presence of the gene duplicate. P-insertion mutants with an abnormal phenotype, indicating a functional divergence, are known for only one of the D. melanogaster OXPHOS gene duplicates - Cyt-c-d , encoding cytochrome c ). Interestingly, although Cyt-c-d is adjacent to its putative parent gene, Cyt-c-p , it shows a different pattern of expression, suggesting that the two genes must be regulated at individual gene level and not at chromatin domain level (see Table 2 ). A systematic investigation of the expression pattern of other D. melanogaster duplicate genes will be necessary to answer the question of whether the testis-biased expression pattern reported here is specific to the duplicates of genes encoding mitochondrial proteins, or is a more general phenomenon. According to the balance hypothesis, validated by experimental results obtained on yeast [ 39 ], single gene duplications involving genes encoding components of multiprotein complexes are expected to severely affect fitness. Therefore, the expression pattern we have observed could be a necessary condition to maintain some gene duplicates in the D. melanogaster genome, at least until they evolve a new useful function. Finally, as nothing is known about the tissue-specific pattern of expression of the genes investigated in D. pseudoobscura and Anopheles , it also remains unclear whether the testis-biased expression of gene copies originated by duplication is specific to D. melanogaster , or is also to be found in other dipterans, and possibly in other organisms. Codon usage in the OXPHOS genes Because of the preferential use of codons ending in C or G, the D. melanogaster coding sequences have an average GC content higher than the genomic average [ 44 , 45 ]. This is also true for the 78 D. melanogaster OXPHOS coding sequences reported in this work and for their D. pseudoobscura and A. gambiae counterparts (68% of the codons in the OXPHOS genes end in C or G in D. pseudoobscura and 77% in A. gambiae , compared to 74% in D. melanogaster ). In all three species, the coding sequences of OXPHOS gene duplicates show a lower percentage of codons ending in C or G, when compared to both the entire set of 78 orthologous OXPHOS genes and the gene subset including only their parent genes. In samples including all the OXPHOS gene duplicates annotated in this paper the aggregate percentage of C- or G-ending codons is 63%, 46% and 73% in D. melanogaster , D. pseudobscura and A. gambiae respectively, as compared with 70%, 64% and 88% in their parent genes. In D. pseudoobscura , the shift toward a higher percentage of A- or T-ending codons is also detected in the pattern of synonymous codon usage; for 12 of the 18 amino acids that are encoded by more than one codon, the most frequently used codon in the D. pseudoobscura gene duplicates is different from the one used in their parent genes (see Additional data file 3). Chromosomal arm location, interarm homology and microsynteny It has been reported that in many eukaryotes including yeast [ 46 ], C. elegans [ 47 ], D. melanogaster [ 48 , 49 ] and humans [ 50 ], genes with related functions and similar expression patterns tend to be clustered, suggesting that they share aspects of transcriptional regulation depending on their inclusion in the same chromatin domain. In particular, Boutanaev et al . [ 48 ] reported that in D. melanogaster clusters of three or more testis-specific genes are much more frequent than expected by chance. Therefore, we investigated the chromosomal distribution of the OXPHOS genes to determine whether clustering could be detected. In all three dipteran species considered, the 78 OXPHOS orthologous genes are randomly distributed on all chromosomal arms (Table 1 ). Two D. melanogaster genes ( Ucrh , encoding the 11 kDa subunit of ubiquinol-cytochrome c reductase, and CG40002 , encoding the AGGG subunit of NADH-ubiquinone oxidoreductase) have a heterochromatic location. No evidence of OXPHOS gene duplicate clustering was found either, despite the common testis-biased expression of such genes. Moreover, no evidence of clustering with other testis-specific genes was found when an EST database search for such genes was performed in the regions flanking the investigated gene duplicates. However, in accord with two studies reporting a significant deficit of genes with a male-biased expression on the D. melanogaster X chromosome [ 51 , 52 ], only one out of the 20 D. melanogaster OXPHOS gene duplicates, two out of 19 in D. pseudoobscura and none (out of eight) in A. gambiae were found to be X-linked (Table 2 ). It may be that duplications of X-linked genes encoding OXPHOS subunits would be especially deleterious because of the male X chromosome transcriptional hyperactivity, which allows dosage compensation. In all three dipteran species, a disproportionately high fraction of OXPHOS gene duplicates appears to be constituted of autosomal genes derived from parent genes located on the X chromosome (Table 2 ). As suggested by recent work on the generation and preservation of functional genes produced by retroposition both in Drosophila [ 53 ] and in the human and mouse genomes [ 54 ], this may be explained by a selective advantage for duplicates of X-linked genes that move to an autosomal location and so escape the X inactivation in early spermatogenesis that occurs both in Drosophila [ 55 ] and in mammals [ 56 ]. We would like to speculate that such selective advantage may be especially significant for duplicates of OXPHOS genes, given the heavy reliance of sperm on mitochondrial function. In fact, the excess of autosomal duplicates of X-linked genes is not observed for MitoDrome annotated genes not involved in the OXPHOS system (see above). However, as the general pattern of much lower, testis-biased expression holds even for OXPHOS and other mitochondrial gene duplicates that apparently derive from autosomal parental genes, and even for X-linked duplicates, this pattern (and the explanation of the evolutionary preservation of such genes) cannot only be due to the selective advantage of escaping X inactivation during spermatogenesis. With the exception of CG9603 , all euchromatic D. melanogaster orthologs maintain their localization on the homologuos D. pseudoobscura chromosomal arm (Table 3 ). CG9603 , encoding the VIIa polypeptide of cytochrome c oxidase, is located on the 3R chromosomal arm in D. melanogaster , whereas Dpse \ CG9603 , its counterpart in D. pseudoobscura , is located on XR; microsyntenic gene order with the flanking genes is conserved in both species, suggesting that a chromosomal rearrangement occurred after their divergence. OXPHOS gene duplicates also almost always maintain the same chromosomal location and microsyntenic gene order in D. melanogaster and in D. pseudoobscura . However, a more complex situation was observed with regard to the gene encoding subunit IV of cytochrome c oxidase, which is duplicated in D. melanogaster and triplicated in D. pseudoobscura (Table 2 ). On the basis of identical genomic organization, conserved chromosomal location and mycrosyntenic gene order Dpse \ CG10664 is inferred to be the ortholog of D. melanogaster CG10664 . Dm CG10396 , Dpse \ CG10396.1 and Dpse \ CG10396.2 are intronless, and neither interarm homology nor microsyntenic order offer any clue to their phylogenetic relationship. The dendrogram based on sequence divergence (see the MitoComp website [ 22 ], complex IV, subunit IV) suggests, however, that a duplication event occurred before the D. melanogaster/D. pseudobscura speciation, originating the CG10664 - CG10396 gene pair ( Dpse \CG10664- Dpse \CG10396 in D. pseudoobscura ). A further duplication event, occurring in the D. pseudoobscura lineage after the D. melanogaster/D. pseudoobscura divergence, probably created the Dpse \ CG10396.1 - Dpse \ CG10396.2 gene pair. In contrast to the maintained location of almost all investigated genes on homologous chromosomal arms in the two Drosophila species, when D. melanogaster and A. gambiae are compared the only meaningful correspondence found concerns the genes on the D. melanogaster 2L and the A. gambiae 3R chromosomal arms (Table 3 ). This result is consistent with previous reports that compared the location of homologous genes in D. melanogaster and A. gambiae , concluding that extensive reshuffling both within and between chromosomal regions has occurred since the divergence of the two species [ 4 , 17 ]. Conclusions We have catalogued 78 nuclear genes that control oxidative phosphorylation in three dipteran species and compiled a web-based dataset, MitoComp [ 22 ], that contains all the data on which this article is based and which is available with the online version of this article. We have conducted only some basic comparative analyses of the many which are possible using such a dataset, and it is our hope that it will provide a valuable resource for those looking for information about nuclear genes encoding mitochondrion-targeted products in the context of functional genomics and proteomics. Future studies based on this information, especially if the comparative analysis is extended to other species, will surely allow a better understanding of the evolutionary history of a set of genes that control a basic biological function, and also offer interesting insights into the mechanisms of their coordinated expression. In fact, a first in silico analysis of the D. melanogaster and D. pseudoobscura nuclear energy gene sequences suggests that a genetic regulatory circuit, based on a single regulatory element, coordinates the expression of the whole set of energy-producing genes in Drosophila [ 57 ]. The comparative analysis of the 78 OXPHOS genes in the three dipteran species shows a high level of amino-acid sequence identity, as well as a substantial conservation of intron-exon structure, indicating that these genes are under strong selective constraints. An unexpected and intriguing result of this study is that in D. melanogaster , duplication-originated OXPHOS genes are expressed at a much lower level (or possibly not expressed at all) in most or all the tissues where their parent genes are expressed, as judged by the abundance of ESTs derived from their transcripts in all libraries other than those derived from testis. On the other hand, OXPHOS gene duplicates have a strongly testis-biased pattern of expression, a finding validated by other authors with a different approach based on the use of microarrays [ 40 ]. In A. gambiae , although no testis-specific ESTs databases are available, a pattern of expression of almost all duplicate OXPHOS genes different from that of the gene from which they originated, and possibly limited to specific tissues, is suggested by the fact that in all EST libraries available the abundance of the sequences originated from the duplicate genes is very low when compared with that of the sequences derived from their respective parent genes. We suggest that, at least in D. melanogaster , the acquisition of a new, testis-biased pattern of expression may be required to maintain duplicates of certain genes in the genome. This may also allow rapid acquisition of new functions by the gene product(s), as it has recently been shown that proteins encoded by duplicated genes with a changed expression pattern often show accelerated evolution [ 58 , 59 ]. Subfunctionalization could then further favor the preservation of multiple paralogous genes. No data are at present available to support the possibility that our findings could be extrapolated to other gene sets or even to the whole genome. However, we propose that duplication of the genes encoding products that are part of multiprotein complexes may be especially deleterious, unless sequence divergence allowing only testis-specific expression of one of the duplicate copies occurs. In turn, this could facilitate the development of novel functions, which is usually assumed to be the main evolutionary advantage of gene duplication, providing a general mechanism for originating phenotypic changes that might also lead to species differentiation. Materials and methods To identify orthologous OXPHOS genes and their duplications in D. pseudoobscura and A. gambiae , contigs from BCM [ 13 ] and scaffolds from AnoBase [ 21 ] were searched using TBLASTN with the D. melanogaster OXPHOS peptides listed in the MitoDrome database [ 19 ] as queries. Amino-acid sequence identity and similarity values were obtained from pairwise alignments using the Needleman-Wunsch global alignment algorithm at the EMBL-EBI server [ 60 ]. Multiple sequence alignments of the OXPHOS amino-acid and coding sequences and visualization of the dendrograms were obtained using the MultAlin 5.4.1 software [ 61 ] from MultAlin server [ 62 ]. The genomic sequence of each gene was manually searched for intron-exon boundaries and the predicted mRNA sequence reconstructed in silico . A. gambiae mRNAs were assembled by overlapping ESTs extracted from AnoBase [ 21 ]. We have named each newly identified A. gambiae gene with the four-letter code 'agEG' followed by the last four or five digits of its Ensembl [ 36 ] gene number, excluding the multiple zeros of the prefix; the D. pseudoobscura genes were named with the code 'Dpse\CG' followed by the Celera number of their D. melanogaster counterparts. The D. pseudoobscura OXPHOS genes investigated here were assigned a chromosomal location where possible, using the putative chromosomal assignments available at BCM [ 13 ] for the majority of the large D. pseudoobscura contigs. We also utilized the Ensembl mosquito genome server [ 36 ] to identify and visualize the chromosomal location of the A. gambiae annotated OXPHOS DNA sequences. The D. melanogaster EST database, available from the National Center for Biotechnology Information (NCBI) contains ESTs from cDNA libraries obtained from different developmental stages and body parts. The relative abundance of the transcripts of duplicate or triplicate D. melanogaster OXPHOS genes was defined by counting their cognate ESTs in non-normalized cDNA libraries generated by the Berkeley Drosophila Genome Project (BDGP) [ 43 ] from embryos (LD), larvae/pupae (LP), and adult ovary (GM), head (GH) and testes (AT), and also the ESTs from adult testes generated at the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) [ 63 ]. ESTs from BDGP normalized EST libraries generated from head (RH) and embryos (RE) were also considered. The relative abundance of the transcripts of duplicate or triplicate A. gambiae OXPHOS genes was defined by counting their cognate ESTs in all libraries recovered from the Anobase server [ 21 ]. Since the number of sequences in the EST databases changes as new EST sequences are added, our values are calculated on the EST sequences present in the databases as of July 2004. The list of D. melanogaster P-insertion OXPHOS mutants is reported in the MitoComp website [ 22 ] and was mostly compiled using information from FlyBase [ 42 ] and from the BDGP P-Element Gene Disruption Project [ 43 ]. Additional data files A web-based dataset, MitoComp, contains all data on which this work is based and is available at [ 22 ]. It includes information on the cytological location of each gene, its genomic organization and the structure of its transcript(s). The genomic structures of the D. melanogaster , D. pseudoobscura and A. gambiae putative OXPHOS orthologs are shown and compared, and their deduced amino-acid products are aligned with the corresponding human protein. When paralogs of the gene exist, neighbor-joining trees derived from distance matrix analysis are also shown to visualize the evolutionary relationships between them. Additional data files available with the online version of this article are as follows. Additional data file 1 contains a table that reports pairwise amino-acid sequence conservation values between the D. melanogaster OXPHOS genes investigated and their D. pseudoobscura , A. gambiae and human counterparts. Additional data file 2 contains data extracted from the Parisi et al . dataset [ 40 ]. Additional data file 3 reports the codon usage in the orthologous and duplicate OXPHOS genes of D. melanogaster , D. pseudoobscura and A. gambiae . Supplementary Material Additional data file 1 A table that reports pairwise amino-acid sequence conservation values between the D. melanogaster OXPHOS genes investigated and their D. pseudoobscura , A. gambiae and human counterparts Click here for additional data file Additional data file 2 Data extracted from the Parisi et al . dataset Click here for additional data file Additional data file 3 The codon usage in the orthologous and duplicate OXPHOS genes of D. melanogaster , D. pseudoobscura and A. gambiae Click here for additional data file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC551531.xml |
529307 | Predicting gender differences as latent variables: summed scores, and individual item responses: a methods case study | Background Modeling latent variables such as physical disability is challenging since its measurement is performed through proxies. This poses significant methodological challenges. The objective of this article is to present three different methods to predict latent variables based on classical summed scores, individual item responses, and latent variable models. Methods This is a review of the literature and data analysis using "layers of information". Data was collected from the North Carolina Back Pain Project, using a modified version of the Roland Questionnaire. Results The three models are compared in relation to their goals and underlying concepts, previous clinical applications, data requirements, statistical theory, and practical applications. Initial linear regression models demonstrated a difference in disability between genders of 1.32 points (95% CI 0.65, 2.00) on a scale from 0–23. Subsequent item analysis found contradictory results across items, with no clear pattern. Finally, IRT models demonstrated three items were demonstrated to present differential item functioning. After these items were removed, the difference between genders was reduced to 0.78 points (95% CI, -0.99, 1.23). These results were shown to be robust with re-sampling methods. Conclusions Purported differences in the levels of a latent variable should be tested using different models to verify whether these differences are real or simply distorted by model assumptions. | Background Clinical researchers frequently use statistical models in an attempt to model outcomes that are not directly measured, also known as latent variables. Examples of such latent variables include mental health, quality of life, and physical disability. Although groups of items (questions) known as outcome scales can be assumed to measure latent variables, it is methodologically challenging to aggregate item responses into scores that accurately and reliably represent the latent variable. The aim of this study is to point that the choice of models with biased assumptions can lead to different conclusions regarding the associations between latent variables and predictors. Three alternative methods are presented: Prediction of latent variables measured as summed scores using linear regression models, prediction of individual item responses using logistic regression models and propensity scores to control for differences in item responses, and prediction of latent variables using Item Response Theory models with covariates. Since all three methods are statistically sophisticated, they will be described using the technique of "layers of information", and used to evaluate the purported association between gender and disability. Specifically, we will test whether this association can be explained by different reporting patterns. Methods Method of layers of information The method of "layers of information" was designed to explain complex statistical methods to audiences with a variety of previous quantitative backgrounds. Each layer is associated with a progressive level of complexity; thus, ensuring that readers with different needs can understand the technique to a level that will enable them to at least understand the statistical method of a clinical study (first layer) and ultimately to apply the statistical method to a new research study (last layer). In the current study, we have used five layers of information: (1) General description, (2) Examples of previous clinical applications, (3) Data requirements, (4) Statistical Theory, and (5) Analysis and Reporting. Layer 1 – General description A latent construct is a concept not directly measured, but that can be estimated through proxy measures. Physical disability is an example since its level is frequently inferred from responses given to a series of items in an outcomes scale measuring patients' ability to perform activities of daily living. Because latent variables cannot be directly measured and predicted, several statistical techniques were devised to approach this problem (Figure 1 ). Figure 1 Graphical description of three models to predict a latent construct 1a. Prediction based on summed scores 1b. Prediction based on regression on individual items 1c. Prediction based on latent variables 1. Outcome prediction based on summed scores The most common approach is to simply add patients' responses to each item; thus, creating a summed score. Summed scores are then used to determine significant predictors in a regression model (Figure 1a ). Two assumptions underlie this strategy. First, we assume that the contribution of each item to the latent variable is known. For example, in a disability scale where patients are questioned about their ability to "raise a glass of water" and to "raise a 40-pound bag", researchers assume that they know the exact amount of disability associated with each of the activities stated by these items. In a scale that does not discriminate between the level of disability associated with each item, the assumption would be that answers to each of these items would represent the same amount of disability, when, in fact, they may not. The second assumption when using summed scores is that each item measures the latent construct without any interference from extraneous factors. For example, it is assumed that two individuals with the same neck disability level, but different educational levels would have similar answer patterns for an item such as "I feel pain in my neck after reading for more than two hours". In this example this assumption might not be true since individuals with different educational levels may have different levels of exposure to a two-hour reading session and consequently have a different perception of the disability caused by such activity. Therefore, in spite of having the same disability level, they would probably provide different answers to the same item. This phenomenon is known as Differential Item Functioning, previously known as item bias. 2. Outcome prediction based on responses to individual items A second approach is to use answers from each item and then determine how each predictor is associated with individual item responses (Figure 1b ). Although apparently simple, this model no longer measures the association of each predictor with the latent variable of interest since individual items, and not the latent construct, is part of the model. In addition, if different items have contradictory levels and directions of association with each predictor, making inferences about the construct may be difficult or impossible. 3. Outcome prediction based on latent variables The last and most recent approach is to use statistical models that will concomitantly determine the latent construct level and its association with the predictor of interest (Figure 1c ). The main advantage of this method is that the assumptions made for summed scores are no longer necessary while, in contrast with the prediction based on individual items, a latent variable is still assumed. The main underlying assumptions of IRT models are that the association between item responses and the latent variable obeys a constant pattern across items, usually an S-shaped pattern, and that patterns of item-response are not influenced by any factor extraneous to the latent variable. Additional requirements include more powerful computers to execute the computations as well as larger sample sizes. Layer 2 – Examples of previous clinical applications Outcome prediction based on summed scores In a study designed to predict factors associated with post-treatment disability after lower-extremity soft tissue sarcoma, Davis [ 1 ] calculated summed scores from scales measuring impairment [ 2 , 3 ], physical disability [ 4 ], and quality of life [ 5 ]. Although the authors did not report whether the four scales complied with the assumptions of a linear regression model described in our first layer, they found that large tumor size, bone resection, motor nerve sacrifice, and complications were associated with poor outcomes. Outcome prediction based on responses to individual items In a study evaluating the prediction of visual disability based on individual objective measures of visual impairment, Bandeen-Roche [ 6 ] regressed individual items of Activities of Daily Vision scale [ 7 ] and then compared their results to the prediction based on summed scores. These authors found that whereas most vision covariates were similarly associated with different item responses, visual acuity was much more strongly associated with two activities ("difficulty reading signs at night and during the day", and "watching television") than with others ("descending steps in either type of light"). In addition, male gender and a greater number of comorbid conditions were also preferentially associated with difficulty watching television. Although these models bring new insights into the association between individual physical activities and their respective predictors, they cannot clarify whether these were true predictors or whether they simply presented different reporting patterns. Prediction based on latent variables To our knowledge, although multiple previous clinical research projects have used IRT for the determination of scale scores [ 8 ], no previous clinical articles have used IRT models with concomitant predictors. Potential clinical applications are any situations where the researcher is attempting to predict a latent construct based on a group of variables [ 8 ], but where a possibility of different reporting patterns or items with an association with different levels of the latent construct are present. Layer 3 – Data requirements Outcomes First, a latent construct has to be measured through a set of proxy variables. These indicators may have responses in various formats, including dichotomous (yes/no), ordinal (e.g., a little, moderate, a lot), or nominal (alternatives without a rank). IRT models assume that the latent construct is continuous and, in most cases, unidimensional, meaning that one single latent construct is assumed. Predictors Predictors can be continuous or categorical variables. Sample size Previous studies have estimated that, for logistic regression models, one should have at least 10 events per predicting variable [ 9 ], while for multiple linear regression models this number reduces to four (Freedman 1989). For IRT, some studies have estimated that models can be estimated with as few as 250 respondents, although 500 would be ideal in most scenarios [ 11 ]. This number may vary; however, depending on the response heterogeneity to the items in the original sample. As general rule, more heterogeneous responses usually require smaller sample sizes. Layer 4 – Statistical theory Outcome prediction based on summed scores Differences in summed scores according to a set of predictor or covariates can be described using linear regression. In these models, the summed score is represented by y using a linear combination of predictor variables x j , where j represents several predicting variables 1, 2, ..., p . It is assumed that no missing values are present for every observation. The fitted values, or predicted summed scores, are then the sum of coefficients β j multiplying each of the x j plus an intercept β 0 , although the later may be absent in some models. This model can be represented by: y - β 0 + β 1 x 1 +...+ β p x p Ordinary least-squares models estimate the coefficients to minimize the squared sum of residuals. If the response and predictors corresponding to the i th of n observations are y i , x i1 ,..., x ip , then the fitting criterion chooses the β j to minimize: The standard statistical theory of linear models makes the first formula more explicit by writing the model for the i th observation as: This model makes the following assumptions: The c i are independently and identically distributed; the c i have mean zero and finite variance σ 2 ; the c i have a normal distribution. Outcome prediction based on responses to individual items Individual responses to dichotomous items can be predicted by generalized linear models using a binomial distribution and, most commonly, a logit link function that will bound the probability of an answer to be between 0 (answer = no) and 1 (answer = yes). The logit link can be expressed by: where π is the probability of a positive answer and x is a vector with item responses . To linearize the function, the dichotomous response for each item can be algebraically transformed to: Notice that, in contrast to linear models, the logistic model does not have an error term since it models the probability of an event directly that will determine the variability of the binary outcome. Logistic models are estimated by maximum likelihood, which is a method to estimate regression coefficients that will maximize the likelihood of obtaining the data ( p (0| x ), where 0 is the latent construct. One of the problems with the prediction based on individual items is that items do not individually represent the latent construct. Therefore, if one is to predict individual answers, it would be interesting to at least account (adjust) for responses of the same patient to other items. This adjustment can be accomplished by propensity scores [ 12 ], which reduce all remaining items to a single composite variable that appropriately summarizes their responses. Compared to the multiple adjustment performed in logistic regression models, propensity scores have the advantage of making the adjustment more transparent. It is important to notice that although the covariates are used as predictors for the item-response, it is still impossible to infer whether this association was distorted by an association between item responses and extraneous variables rather than the association between item responses and the latent trait. Outcome prediction based on latent variables Although multiple models have been described for the regression of latent variables on predictors [ 6 ], we will concentrate on IRT. IRT assumes that the response of patients to individual items can be modeled with a two-level logistic regression where the log odds of patient i providing a positive answer to an item j is represented by: Where β j represents the difficulty of item j and u i represents the trait level associated with subject i . This equation holds true in the simplest IRT model known as Rasch or one-parameter logistic (1PL). Other models – two-parameter logistic, ordinal logistic – among others – are used according to the types of response alternatives presented by each item. Adding one additional parameter λ to represent the extent to which item j can discriminate between subjects of different trait levels, we obtain: Finally, if we add a predictor to this equation we will have where γ is the regression coefficient for predictor x . This model allows several advantages over the two models previously described in this layer, including the absence of assumptions from summed scores as well as the summarization of all items into a single latent variable. The most frequent assumptions in IRT models are that a single construct is measured and that observations are independent, conditional on the latent variable. Different IRT models will have different assumptions about the extent to which assumptions of summed scores can be relaxed. For example, 1-Parameter. Logistic Regression models assume that each item measures the latent trait with equivalent strength. One important practical aspect, when making use of IRT models with predictors, is to check quadrature point approximation used in the random-effects estimator. As a rule of thumb, if the coefficients do not change by more than a relative difference of 0.01%, then the choice of quadrature points does not significantly affect the outcome and the results may be confidently interpreted. Two aspects of random-effects models have the potential to make the quadrature approximation inaccurate: large group sizes and large correlations within groups [ 16 ]. Layer 5 – Analysis and reporting Data analysis To illustrate a practical application of the previously described models, we will use data from a cohort study of patients with low-back pain to evaluate the gender-disability association. Specifically, we will evaluate whether female patients either have more severe disability or simply whether they are more likely to give positive answers to some items while having equivalent physical disability levels. Several studies have found that, compared to men, women are usually associated with higher initial disability and pain scores after low-back pain episodes [ 14 , 15 ]. However, it is usually unnoticed that these studies do not directly measure disability, a latent construct, but rather measure patients' responses to items that are hypothesized to measure disability. In other words, the hypothesis is that the instrument accurately measures the construct, although the instrument is rarely re-evaluated by the time of measurement. In support of this important caveat is that previous studies have found that women have different responses to the stress caused by low-back pain when compared to men [ 16 ]. Therefore, the question of whether women really present with higher disability levels, simply have a different response to items measuring disability or both have higher disability and have a different response is open. A description of the cohort used for this analysis is presented in detail elsewhere [ 17 ]. Briefly, the cohort contains data on 1,633 patients with low-back pain answering 23 dichotomous items from the Roland Questionnaire modified by Patrick [ 18 , 19 ]. The item content for this scale is presented in Table 1 . The outcome of interest is physical disability represented by items of the modified Roland Questionnaire, and the main effect is gender. The association between these variables is adjusted for several potential confounders, including marital status (married, other), presence of workman's compensation (yes/no), and presence of private insurance (yes/no). All analyses were performed using Stata 8.0 for Linux (Stata Corporation, College Station, TX). Because Item Response Theory model with predictors are very computer intensive and individual models may take over 24 hours to run in personal computers, a special arrangement of the operating system was instituted to obtain maximal performance. These changes included establishing maximal priority (renice set to -20 to the Stata process, and running in a Linux "bigmem" kernel 4.20 with random allocation memory of 4 gigabytes). Additional measures to increase computational speed included data collapsing, frequency weights, and matrices with previous beta coefficients used as priors. Table 1 Item content for the modified Roland Questionnaire 1. I stay home most of the time because of my back problem or leg pain (sciatica) 2. I change position frequently to try and get my back or leg comfortable 3. I walk more slowly than usual because of my back problem or leg pain (sciatica) 4. Because of my back problem, I am not doing any of the jobs I usually do around the house 5. Because of my back problem, I use handrail to get upstairs 6. Because of my back problem, I have to hold onto something to get out of an easy chair (comfortable padded chair) 7. I get dressed more slowly than usual because of my back problem or leg pain (sciatica) 8. I only stand for short periods of time because of my back problem or leg pain (sciatica) 9. Because of my back problem, I try not to bend or kneel down 10. I find it difficult to get out of a chair because of my back problem or leg pain (sciatica) 11. I have trouble putting on my socks (or stockings) because of the pain in my back or leg 12. I find it difficult to turn over in bed because of my back problem or leg pain 13. I sleep less well because of my back problem 14. I avoid heavy jobs around the house because of my back problem 15. Because of my back problem, I am more irritable and bad tempered with people than usual 16. Because of my back problem, I go upstairs more slowly than usual 17. I stay in bed most of the time because of my back or leg pain (sciatica) 18. I keep rubbing or holding areas of my body that hurt or are uncomfortable 19. My back or leg is painful almost all the time 20. I only walk short distances because of my back problem 21. Because of my back problem, my sexual activity is decreased 22. Because of my back problem, I am doing less of the daily work around the house than I would usually do 23. I often express concern to other people what might be happening to my health Data preparation Briefly, our sample is composed by 1,633 individuals with a diagnosis of low-back pain. Most patients are females (52.3%), married (69.9%), white (83.0), and with medical insurance (68.3%). For linear and logistic regression models the data were placed in wide format, with individual variables representing patient responses to each item. For IRT models the data were presented in long compressed format (Figure 2 ). Figure 2 Sequence of Stata commands for the execution of the three sets of model Prediction based on summed scores When comparing the crude association between summed scores and gender, it was found that female patients had scores that were on average 1.46 (95% CI 0.73, 2.08) points higher than their male counterparts in a 0–23 scale. This association was further tested in a linear regression model (Figure 1a ) controlling for gender, insurance status (including workman's compensation), marital status, and income. The full model demonstrated that, adjusted to potential confounders, women report on average 1.32 (95% CI 0.65, 2.00) more points in the modified Roland scale than men. After backwards deletion, none of the previous potential confounders were proven to be substantial confounders using a cut point of 10% change the original point estimate. Since the distribution of summed scores of the modified Roland Questionnaire was not normal, we used regression diagnostics using plots to determine that the relationship between predicted and observed values did not display any violations of the regression assumptions. This was confirmed by a Ramsey regression specification error test (RESET) for omitted variables (p = 0.7371) although the Breusch-Pagan / Cook-Weisberg test demonstrated a trend towards heteroscedacity (p = 0.0777). In order to further verify the robustness of this association, an ordinal logistic regression model was used with cut-points at 0–7 (low summed score), 8–15 (medium summed score), and 16–23 (high summed score). This model was considered to adequately comply with the proportionality assumption (p = 0.776). Results for the ordinal regression model demonstrated that the predicted probability of a male having low, intermediate, and high scores were progressively decreasing: 0.38, 0.33, and 0.28, respectively. This pattern was in contrast with women, where the probabilities were ascending: 0.32, 0.33, and 0.35, respectively. In summary, all results from models using summed scores point to a significant association between female gender and high disability scores. It is unclear; however, whether this association can be explained by high disability levels or simply different report patterns between men and women. Prediction based on responses to individual items As a next step, the association between individual item responses and gender was evaluated using logistic regression models stratified by propensity scores adjusting for responses to other items (Figure 1b ). Propensity scores were determined by running logistic regression models that evaluated the probability of a positive response to an item adjusted for all remaining items and covariates except gender. These scores were then used to classify all observations into five different propensity score percentiles. The distribution of each of the covariates was found to be balanced among all four groups, indicating that the propensity scores were effective in "randomizing" the groups (Alcouffe 1999). The analysis across propensity strata demonstrated contradictory results, with male patients being significantly associated with positive responses to items 4 ("Because of my back problem, I am not doing any of the jobs I usually do around the house") and 8 ("I only stand for short periods of time because of my back problem or leg pain (sciatica)"), while female patients were significantly associated with positive responses on items 7 ("I get dressed more slowly than usual because of my back problem or leg pain (sciatica)"), 15 ("Because of my back problem, I am more irritable and bad tempered with people than usual"), 17 ("I stay in bed most of the time because of my back or leg pain (sciatica)"), and 19 ("My back or leg is painful almost all the time"). No single item was consistently associated with gender across all propensity score strata. A new model was then built adjusting for scores pooled across strata. The results demonstrated that most items were not associated with either gender, items 4 ("Because of my back problem, I am not doing any of the jobs I usually do around the house") and 8 ("I only stand for short periods of time because of my back problem or leg pain (sciatica)") being positively associated with male gender while items 7 ("I get dressed more slowly than usual because of my back problem or leg pain (sciatica)") and 15 ("Because of my back problem, I am more irritable and bad tempered with people than usual") being associated with female gender (Figure 3 ). Figure 3 Odds ratio of having a positive response to an item* *ORs above one represent a positive association between a positive item response with being a male Since logistic regression models do not control for the latent variable one cannot test whether the association between gender and individual item responses is related to an association with disability or simply caused by women being more likely to provide a positive response to a certain item in spite of having the same degree of disability. Prediction based on latent variables Finally, IRT models (Figure 1c ) were used to determine the association between gender and IRT scores. First, a crude association between male and IRT scores was calculated based on all 23 items. This model demonstrated that female gender continued to be significantly associated with higher disability (coefficient 0.34, log likelihood test p < 0.001). Notice that this value is presented in a new scale that can no longer be compared to the previous scores obtained from the modified Roland scale with a range from 0 to 23. To test the hypothesis that some items might present different reporting patterns, we tested for interaction terms between each item and gender. Our results demonstrated that items 7 ("I get dressed more slowly than usual because of my back problem or leg pain (sciatica)", Figure 4a ), 14 ("I avoid heavy jobs around the house because of my back problem", Figure 4b ), and 17 ("I stay in bed most of the time because of my back or leg pain (sciatica)", Figure 4c ) presented significant interactions with gender. An interaction with gender indicates that the item response is affected by gender; thus, demonstrating different reporting patterns. Although interpretations of item content are speculative, items 7 and 14 may indicate that male and female patients interpret these questions as a different type and level of activity, respectively, while item 17 may be associated with differential behaviors in relation to disability across genders. Figure 4 Item characteristic curves for items demonstrating differential item functioning 4a. Item 7 4b. Item 14 4c. Item 17 A new IRT model was then calculated, but now excluding all items with differential reporting patterns. The difference in disability reporting between men and women was reduced (coefficient 0.04, log likelihood p = 0.08), indicating that gender was no longer significantly associated with disability. In fact, when the same items were excluded from the summed score, a multiple linear regression model demonstrated that the difference between female and male patients had been reduced to 0.78 points (95% CI, -0.99, 1.23) on the original 0–23 scale (p = 0.06), a reduction of 53.4% compared to the original difference. Bootstrapping methods were used in the linear regression model to verify whether the association was robust after multiple sampling procedures had been applied to the models. The results demonstrated a variation of only 13.2%; thus, indicating that these results are robust provided that the sample is representative of the study population. In conclusion, one could infer that although women still have slightly more disability than men, much of the previously reported differences using the modified Roland were inflated by the presence of items with different reporting patterns in scales measuring disability. Conclusions We used three different regression models to investigate the association between gender and disability. Although summed models demonstrated a significant association between gender and disability, these models did not allow us to test whether this purported difference was related to the latent construct disability or to items presenting with differential item functioning. Analysis of the association within individual items demonstrated inconsistent associations with gender, with some items presenting a strong positive association with male gender while others had a positive association with female gender. Since these associations were made with the item response rather than the latent variable, it was impossible to verify whether these were valid representations of the construct of interest, associations with disability, or simply the effects of differential item functioning. Last, we examined the association between gender and disability measured as a latent variable. After removing items with differential item functioning, the association with gender was lessened and no longer significant. Therefore, we concluded that although a small difference between genders in relation to the disability associated with low back pain does exist, much of it is caused by differential item functioning than a true association with the disability construct. In summary, we advocate that the measurement of the association between latent variables and covariates be systematically performed using a combination of regression models to ensure that observed associations are not distorted by differential item functioning. Authors' contributions RP: design, analysis, manuscript writing; MT: design, analysis, manuscript revision; UG: design, analysis, manuscript revision; LDH: design, manuscript revision; DOJ: design, manuscript revision; TC: data collection, design, manuscript revision. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529307.xml |
529461 | Using GIS technology to identify areas of tuberculosis transmission and incidence | Background Currently in the U.S. it is recommended that tuberculosis screening and treatment programs be targeted at high-risk populations. While a strategy of targeted testing and treatment of persons most likely to develop tuberculosis is attractive, it is uncertain how best to accomplish this goal. In this study we seek to identify geographical areas where on-going tuberculosis transmission is occurring by linking Geographic Information Systems (GIS) technology with molecular surveillance. Methods This cross-sectional analysis was performed on data collected on persons newly diagnosed with culture positive tuberculosis at the Tarrant County Health Department (TCHD) between January 1, 1993 and December 31, 2000. Clinical isolates were molecularly characterized using IS6110-based RFLP analysis and spoligotyping methods to identify patients infected with the same strain. Residential addresses at the time of diagnosis of tuberculosis were geocoded and mapped according to strain characterization. Generalized estimating equations (GEE) analysis models were used to identify risk factors involved in clustering. Results Evaluation of the spatial distribution of cases within zip-code boundaries identified distinct areas of geographical distribution of same strain disease. We identified these geographical areas as having increased likelihood of on-going transmission. Based on this evidence we plan to perform geographically based screening and treatment programs. Conclusion Using GIS analysis combined with molecular epidemiological surveillance may be an effective method for identifying instances of local transmission. These methods can be used to enhance targeted screening and control efforts, with the goal of interruption of disease transmission and ultimately incidence reduction. | Background The application of molecular analysis to identify specific Mycobacterium tuberculosis strains (TB), in combination with traditional surveillance, has yielded insights into tuberculosis transmission [ 1 ]. These insights together with a downward trend in tuberculosis in the United States have resulted in the Center for Disease Control and Prevention re-evaluating the TB elimination strategy, and recommending that testing be targeted at specific high risk populations [ 2 , 3 ]. The Institute of Medicine (IOM) also recommended the development of more effective methods for identifying persons with recently acquired infections as an important component of new strategies to limit the spread of tuberculosis [ 4 ]. While a strategy of targeted testing and treatment of persons most likely to develop tuberculosis is attractive, it is uncertain how best to accomplish this goal. Persons with molecularly clustered tuberculosis isolates are assumed to be in the same chain of recent tuberculosis transmission [ 5 , 6 ]. Limited studies have been conducted to evaluate whether these clusters occur in predefined geographical areas [ 7 - 11 ]. If so, then geographically based screening and treatment could be an effective method for TB control programs to identify high risk populations. In this study we seek to determine if we can identify geographical areas with on-going tuberculosis transmission by linking Geographic Information Systems (GIS) technology with ongoing molecular surveillance. Methods This cross-sectional analysis was performed on data collected on all persons newly diagnosed with culture positive tuberculosis at the Tarrant County Health Department (TCHD) between January 1, 1993 and December 31, 2000. The TCHD serves the western portion of the Fort Worth-Dallas metropolitan area and includes a population of approximately 1.5 million [ 12 ]. The Fort Worth-Dallas metropolitan area is the ninth largest in the U.S. [ 13 ]. This study is part of the recent collaborative project sponsored by the Center for Disease Control and Prevention National Tuberculosis Genotyping and Surveillance Network for studying the molecular epidemiology of tuberculosis [ 14 ]. All data and materials; including isolates, isolate genotypes, demographic factors, and addresses; were collected prospectively. Moreover, one of the stated objectives of the National Tuberculosis Genotyping and Surveillance Network was to characterize places involved in potential TB transmission [ 15 ]. All positive isolates obtained from persons residing in Tarrant County were sent to the Texas Department of Health (TDH) for DNA fingerprinting. Only persons whose M. tuberculosis strains were typed by the Texas Department of Health Mycobacteriology Laboratory were analyzed. Clinical isolate IS6110-based RFLP analysis and spoligotyping analyses were utilized to identify patients infected with the same strain using published methods [ 16 , 17 ]. RFLP analysis using IS 6110 RFLP is a powerful tool for discerning one strain of M. tuberculosis from another when there are greater than 6 copies of IS 6110 however, a secondary typing method is needed to help differentiate strains with 6 or fewer IS6 110 copies [ 18 ]. For this project, isolates were considered to be clonally related (i.e., genotypically clustered) if they had identical IS 6110 patterns containing seven or more bands, or they had identical IS 6110 patterns containing six or fewer bands and identical spoligotypes. A geographic cluster was defined as two or more patients with molecularly related TB strains living in Tarrant County, TX. The proportion of cases due to ongoing transmission was estimated allowing one source case per cluster (i.e. n-1 method) [ 5 ]. Any patient who did not have both spoligotyping and RFLP analysis of IS 6110 performed on their M. tuberculosis isolate, and/or did not live within Tarrant County at the time of collection was excluded from the geographical analysis. Each eligible patient participated in a standard interview as part of their routine initial medical evaluation. Interview data collected included current and past employment, housing, alcohol and illicit drug use, incarceration history, sexual orientation, and psychiatric history. Persons paid daily for work were considered sporadically employed; others were employed or unemployed. Homelessness was defined as being without a permanent address for more than 3 days since 1991. If persons had a history of homelessness, paid rent by the day, or lived with a non-spousal roommate without paying rent, they were considered unstably housed. Alcoholism was defined by admission of daily consumption of three or more ounces of an alcoholic beverage; a history of alcohol-related conditions including cirrhosis, hepatitis, alcohol withdrawal seizures; or incarceration for alcohol use. Illicit drug use was defined by admission of use or documentation of being under the influence of an illicit drug. Persons were classified as having a history of incarceration if they had spent more than 24 hours in any criminal justice facility since 1991. Patients born in the U.S. or one of its territories were considered American-born; all others were considered foreign-born. All patients received HIV testing and counseling as part of standard clinical practice at the time of diagnosis. HIV status was determined from these tests. Residential address at the time of diagnosis of tuberculosis, including zip code, were geocoded using ArcView, 4.0, Geographic Information System Software, (ESRI, Redlands, CA). After geocoding, automatically and interactively, 94% of the cases were correctly matched. The numbers of cases were then aggregated by zip code and, for each zip code, an average of the total population reported for the US Census 2000 and US Census 1990 was used to calculate incidence. Population information was retrieved from the US Census Bureau [ 20 ] and the North Central Texas Council of Governments (NCTOG) [ 21 ]. The US Census Bureau website provides census data aggregated to certain boundaries (e.g.) block groups, blocks, census tracts, zip codes, counties, and states. The NCTOG is a collection of local governments in the Dallas Fort Worth area, provides demographic and GIS data for the region. The demographic data provided has been directly extracted from the US Census. Zip-code level boundaries were established for incidence comparison purpose using zip code tabulation areas (ZTCAs) [ 21 ]. The three-dimensional analysis was performed using Inverse Distance Weighting (IDW) [ 22 ]. Interpolation is the estimation of values for points in an area not actually sampled. There are many different types of interpolation, with IDW being the simplest interpolation method. A neighborhood about the interpolated point is identified and a weighted average is taken of the observation values within this neighborhood. The weights are a decreasing function of distance. The simplest weighting function is inverse power: w(d)= 1/d p with p > 0. For p = 1, the interpolated function is "cone-like" in the vicinity of the data points [ 22 ]. The resulting "cone" shows the clustering of data around the center point of a geographical area. Statistical analysis was performed utilizing SAS V.8 statistical software (SAS Institute, Cary, NC). Patients with genotypically clustered and unique strains (non-clustered) were compared regarding each categorical risk variable by using odds ratio as a measure of association. Risk categories included patient demographics, and tuberculosis risk factors such as homelessness, HIV-infection, incarceration, and foreign birth. Because members of the clustered cases are assumed to be related, generalized estimating equations (GEE) analysis [ 23 ] was performed to determine factors associated with infection of genotypically and geographically clustered strains of M. tuberculosis , and to derive maximum likelihood odds ratio and 95% confidence intervals for all variables. Age was the only continuous variable. Age statistics were analyzed by comparing the means between groups. A 95% confidence interval for the mean age difference was calculated by using the normal approximation, and an independent sample two-tailed student's t -test was used to assess the statistical significance of the mean age difference. The institutional review board of the University of North Texas Health Science Center at Fort Worth approved this investigation. Results From January 1, 1993 to December 31, 2000, there were 991 incident cases of tuberculosis in Tarrant County, Texas; M. tuberculosis was isolated from 828 (83.6%) cases. Of the 828 cases with a positive culture, 527 (63.6%) had viable clinical isolates for molecular analysis. Persons excluded because of no viable clinical isolation of M. tuberculosis did not differ statistically from those with those with viable isolates by age (p = 0.49), gender (p = 0.57), or location (p = 0.64). Two hundred and ninety-two (55.4%) patients met the criteria for molecular clustering. These patients were categorized into 48 clusters varying from 2 to 95 patients per group. In nine instances (1.7%), patients had an identical low copy RFLP pattern but did not have spoligotyping conducted, and were therefore classified as missing data (Figure 1 ). The proportion of cases attributable to ongoing transmission (allowing one source case per cluster i.e. n-1 method) is estimated to be (292 -48)/518 = 47%. Figure 1 Derivation of study population, Tarrant County, Texas 1993 – 2000 The mean age for the entire population was 44.7 ± 17.3 (SD), 44.1 ± 16.6 (SD) for those genotypically clustered, and 48.5 ± 17.6 (SD) for those with unique strains. Patients that were genotypically clustered differ significantly with age when compared to patients with unique strains, [p = 0.005]. One hundred and seventy-one (32.4%) patients were African-American, 165 (31.3%) were Caucasian, 109 (20.7%) were Hispanic, and 82 (15.6%) were Asian. African-Americans with tuberculosis were significantly more likely to have a clustered strain [OR = 2.7, 95% CI = 1.8, 4.0]. Alternatively, Asians [OR = 3.9, 95% CI = 2.3, 6.0], and Hispanics [OR = 1.9, 95% CI = 1.2, 2.9] were significantly more likely to have a unique strain. Three hundred and twenty-nine (67.4%) of the patients were males, and of these, 214 (65.0%) had clustered strains; 78 of 159 females (49.1%) had clustered strains. Males were more likely than females to have a strain that matched at least one other person in Tarrant County [OR = 1.9, 95% CI = 1.2, 2.8]. Persons with previous experience of homelessness were strongly associated with clustering suggesting a high rate of on-going transmission among this population. [OR = 12.4, 95% CI = 2.9, 52.1] (Table 1 ). Table 1 Selected factors associated with genotypic clustering *Within Clustering, CI = confidence interval; OR = Odds Ratio N (%)* OR 95% CI p -value Homelessness 33 (11.3) 12.4 2.9, 52.1 <0.001 Living in Zip Code 1 41 (14.0) 6.2 2.4, 16.1 <0.001 American born 235 (80.5) 5.3 3.5, 7.9 <0.001 African-American 123 (42.1) 2.7 1.8, 4.0 <0.001 Male gender 214 (73.3) 1.9 1.3, 2.8 0.001 Living in Zip Code 2 40 (13.7) 1.9 1.0, 3.6 0.038 Living in Zip Code 3 9 (34.6) 0.3 0.1, 0.7 0.03 Three hundred and twenty-one (65.7%) patients were born in the United States. Of those, 235 (73.2%) had clinical isolates that matched the isolate from at least one other person living in Tarrant County. One hundred and sixty-seven patients were born outside of the United States. Of those, 57 (34.1%) clinical isolates that matched the isolate from at least one other person living in Tarrant County. U.S. born individuals were significantly more likely to be genotypically clustered than foreign-born counterparts [OR = 5.3, 95% CI 3.5, 7.9]. The birth country of foreign-born patients varied. Of those born outside of the U.S, 77 (46.1%) were born in Latin America, 47 (28.1%) in Southeast Asia, 14 (8.4%) in Sub-Saharan Africa, 12 (7.2%) in Pacific Asia, 11 (6.6%) in South Asia, and 6 (3.6%) in Europe. Evaluation of the spatial distribution of number of cases within zip-code boundaries displayed a distinct geographical distribution of disease. The average incidence for the entire county during the study period was 5.9 cases per 100,000. Zip code 1 recorded the highest incidence of 94.3 cases per 100,000 populations, followed by zip code 2 with an average incidence of 55.2 cases per 100,000 population (Figure 2 ). These areas are characterized by low socioeconomic status, high unemployment rates, homelessness, drug use, and poor quality housing conditions. To examine how molecular clustering varies spatially by zip code, a map of percent molecular clustering at the zip-code level was produced (Figure 3 ). This map displayed the number of genotypically clustered cases divided by the total number of cases reported in that zip code. The map demonstrated that molecularly clustered disease is not homogenously distributed throughout the county. Figure 2 Average incidence of Tuberculosis by zip code Tarrant County, Texas (1993 – 2000). Specific zip codes of interest are labeled in green. Figure 3 Percent of patients genotypically cluster by zip code Tarrant County, Texas (1993 – 2000). Percent genotypically clustered cases = number of genotypically clustered cases/ total number of cases × 100 GIS analysis demonstrated that the areas with the highest incidence also have the highest proportion of persons with genotypically clustered isolates. A strong preponderance of clustering occurred in the urban center of Tarrant County. The highest proportion of persons with molecular clustered TB isolates (80.4% clustered) occurred in the same zip code with the highest incidence. Similarly, zip code 2 on the southeast border of zip code 1, recorded the second highest proportion of persons with molecular clustered TB isolates with 76.6% of all reported cases clustered. Cases reported in zip code 1 were more than six times as likely [OR = 6.2, 95% CI = 2.4, 16.1] than any other zip code to have isolates that match at least one other person living in Tarrant County. In zip code 3, we observed a morbidity that was more than triple the county average (22.3 cases per 100,000). Unlike other high morbidity areas, zip code 3 had a strong preponderance of unique strain distribution. In this zip code, 17 out of 26 (65.4%) patients had isolates that did not match any other patient in Tarrant County. Cases reported in this zip code were 70% less likely [OR = 0.3, 95% CI = 0.1, 0.7] to have a clustered strain, suggesting that the high rates of tuberculosis did not result from local on-going transmission. Discussion The number of cases in the United States is at its lowest point in history, with 15,075 cases reported in 2002 [ 24 ]. The role of treatment of LTBI in tuberculosis elimination is of increasing importance. The IOM recommended developing improved methods for identifying persons with recently acquired infections as an important component of strategic tuberculosis elimination in the United States [ 4 ]. This study uncovered geographical links to on-going tuberculosis transmission enhancing traditional public health surveillance. We found that by combining molecular strain characterization with GIS analysis that risk of on-going transmission was geographically focal (p = 0.003) with significant clustering of cases occurring in 3 of 59 zip codes. This demonstrated that the current methods of surveillance of contacts of persons with tuberculosis were not completely effective in interrupting disease transmission in these zip code areas. The use of molecular strain characterization methods in conjunction with traditional surveillance has led to the recognition of a number of risk factors associated with on-going transmission, and has identified numerous outbreaks of tuberculosis undetected by conventional approaches [ 25 - 27 ]. These studies have demonstrated the importance of non-household location based transmission, such as homeless shelters, and social settings such as bars and crack houses [ 26 - 29 ]. Urban centers have traditionally had higher rates of tuberculosis than rural areas [ 30 , 31 ]. Population density, poverty and overcrowding appear in most areas to be major factors for disease transmission [ 32 ]. We found similar risks in our population. Location factors, specifically where patients reside at the time of diagnosis, were found to be significantly associated for certain zip codes in Tarrant County. Cases in urban zip codes 1 [OR = 6.2; 95% CI = 2.4, 16.1] and 2 [OR = 1.9; 95% CI = 1.3, 2.8] were strongly associated to infection with a clustered strain as compared to the rest of the county. These zip codes are in the urban center of the county. Zip code 1 is also the site of the largest homeless shelter in the county. Inverse Distance Weighting of this area graphically (Figure 4 ) demonstrates the three-dimensional result of the interpolation, representing the burden of disease in this area. The resulting "cone" shows the geographic clustering of cases around a particular point of the zip code, the physical location of homeless shelter. Figure 4 Three-dimensional analysis using Inverse Distance Weighting Interpolation, Tarrant County, Texas (1993 – 2000). Inverse power: w(d)= 1/d p with p = 1. Zip code 1 consists of 264.89 acres. These finding are similar to those reported in Los Angeles where locations, specifically homeless shelters were identified as important sites of tuberculosis transmission [ 33 ]. Similarly, in Houston, locations, specifically bars, were as important as persons in uncovering epidemiological and genotypical links in outbreak investigations [ 34 ]. The authors of both of these studies suggested measures to reduce tuberculosis transmission should be based on locations as well as personal contacts [ 33 , 34 ]. We identified that 55% of our patients were clustered and 47% attributable to ongoing community transmission. This differs from a study conducted in a high incidence area of South Africa, where 72% of cases were clustered and 58% attributable to ongoing community transmission [ 11 ]. Our lower percentage of clustering and attributable on-going transmission may be related to a much lower reported overall morbidity or the effects of differences in programmatic interventions, such as contact investigation, targeted screening efforts or DOT completion rates. Although the majority of the tuberculosis morbidity within the developed world is strongly influenced by imported tuberculosis from high prevalence countries [ 35 , 36 ], the rates at which these individuals transmit disease to the general population remain low. We found that foreign-born cases were significantly more likely to have a unique strain [OR = 6.4, 95% CI = 4.1, 9.8] indicating that immigrants were less likely to be source of ongoing transmission of TB in Tarrant County. In a San Francisco based study, investigators identified only two instances of a foreign-born individual transmitting the disease to the native population [ 37 ]. Similarly, only 1.8% of transmission from infectious Somali immigrants was to the native population in the Netherlands over the period from 1992 to 1999 [ 38 ]. Historically, tracking these populations of foreign born to assess transmission has been difficult. GIS provides another approach for evaluating this issue. As this study illustrates, identifying geographical areas of increased incidence with a high percentage of unique strains may improve local surveillance methods to locate hard to reach foreign-born populations before transmission occurs. There are some limitations to this research approach. This is based on secondary data, which includes variables collected from a cross-sectional period of time. Although each case is an incident case at the time of diagnosis, under this cross-sectional design, exposure and disease outcomes are assessed simultaneously. In addition patients with tuberculosis may have moved shortly before their diagnosis. However, this should not cause systematic error (bias) or result in an association of clustering with specific locations, because these events would be expected to produce a random misclassification. Also persons exposed within certain zip codes may go on to reside elsewhere and later develop the disease, and result in an underestimate of the morbidity and that may be reflected in calculating associations. Finally, genotyping results were not available for a proportion of TB cases in this study. Some unique isolates might have clustered if some of the missing isolates had been available or if other cases with the same strain moved or are located outside the study area [ 39 ]. We therefore believe that estimates of the degree of clustering and the size of clusters are conservative. When using this approach TB control programs must select the appropriate geographical boundary to examine transmission in their area. For example, using zip codes may be too large a boundary in very populated metropolitan areas. Census block groups may provide greater resolution in determining localized transmission. Nor are the molecular techniques used without limitation. Patients are clustered according to their isolates having the same genotype. While IS 6110 RFLP is recognized as the most discriminatory method for genotyping M. tuberculosis isolates, the discriminatory ability of the technique decreases when there are fewer than 6 IS 6110 insertions in the genome. In this case, spoligotyping was used for further strain discrimination. However, it is still possible that some isolates classified as being the same strain based on identical genotypes may represent distantly related, but distinct, strains. Moreover, demonstration that particular patients have the same strain supports, but does not irrefutably prove, direct transmission between these patients as opposed to another source of infection. Conversely, strains continue to evolve, and the resulting genotypic differences over time can result in assigning isolates from cases of direct transmission to distinct strain lineages. Given that a small minority of the isolates had fewer than 6 IS 6110 bands (18.2%) or differed by the presence or absence of one band in an otherwise conserved pattern (3.7%), we believe that estimates of the degree of clustering and the size of clusters are conservative. Conclusion Using GIS analysis combined with molecular epidemiological surveillance can be an effective method for identifying tuberculosis transmission not identified during standard contact tracing methods. The application of these methods can be utilized in countries where contact tracing is routinely performed. These methods can enhance targeted screening and control efforts, with the goal of interruption of disease transmission and ultimately incidence reduction. This study demonstrates that using existing health data, GIS can identify previously undetected TB transmission. These results were used to design new targeted screening efforts [ 40 ]. Studies of these efforts are ongoing to demonstrate if identifying focal areas for targeted screening has utility in reducing TB transmission. List of abbreviations CI Confidence Interval GEE Generalized Estimating Equations GIS Geographic Information Systems HIV Human Immuno-deficiency Virus IDW Inverse Distance Weighting NCTCG North Central Texas Council of Governments OR Odds Ratio RFLP Restriction Fragment Length Polymorphism TB Tuberculosis TCHD Tarrant County Health Department TDH Texas Department of Health Authors' Contributions Study concept and design: PM, MB, SW Acquisition of data: PM, TQ, KJ, DD, GB, SW Analysis and interpretation of data: PM, MB, TQ, JO, SW Drafting of the manuscript: PM, SW Critical revision of the manuscript for important intellectual content: SW, PM, JO, TN Statistical expertise: KS, MB, PM Obtained funding: TQ, SW Administrative, technical or material support: GB Funding/Support This work was supported in part by the Centers for Disease Control and Prevention, National Tuberculosis Genotyping and Surveillance Network Cooperative Agreement U52/CCU600497-18, and Tuberculosis Epidemiologic Studies Consortium 200-2001-00084. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC529461.xml |
423158 | What's So Hot about Recombination Hotspots? | Recombination is a nearly ubituitous feature of genomes; where and when it occurs can provide insights about its evolution and can affect our ability to identify genes that cause disease | Consider a piece of text, either this one that you are now reading or any other. Surely they are all pretty much alike, in so far as they are all run-on strings of characters. In this same sense, we can envision that all DNA strands are alike because all are monotonous polymers with the same general chemical makeup. Indeed, this is how we think of DNA when considering its basic function of inheritance, in which all parts of all chromosomes must be duplicated and then passed from one cell generation to the next. The capacity for inheritance is fundamentally a consequence of DNA's general molecular structure, and not of its sequence per se, as Watson and Crick (1953) , and indeed Muller (1922) long before them, well appreciated. Muller did not know that genes are made of DNA, but he did realize that, whatever genes were made of, they must have a general capacity to replicate, regardless of the information they carry ( Muller 1922 ). But sequence does matter when DNA fulfills its other, more directly functional role. When the DNA that makes up a gene is exposed and expressed, when a gene is serving its individual function, then the detailed sequence means all. So where does recombination ( Box 1 ) fit in? Is recombination something that happens to DNA generally? Or does it happen to particular sequences? Bacteria have their chi (χ) sequence, which is a specific series of eight base pairs in the DNA of the bacterial chromosome that stimulate the action of proteins that bring about recombination ( Eggleston and West 1997 ). Similarly, the immunoglobulin genes of mammals have recombination signal sequences that are involved in V-J joining—a kind of somatic recombination involving the joining of a variable gene segment and a joining segment to form an immunoglobulin gene ( Krangel 2003 ). But does normal meiotic recombination depend on the local DNA sequence? In yeast, as well as mammals (mice and humans), the answer is partly yes, for it is clear that chromosomes have local recombination hotspots where crossing over is much more likely to occur than in other places on the chromosome. Recombination hotspots are local regions of chromosomes, on the order of one or two thousand base pairs of DNA (or less—their length is difficult to measure), in which recombination events tend to be concentrated. Often they are flanked by “coldspots,” regions of lower than average frequency of recombination ( Lichten and Goldman 1995 ). Diverse Implications of Recombination Hotspots: The Study of Meiosis and the Mapping of Human Disease Alleles Recombination hotspots are of strong interest to at least two quite different groups of biologists. For geneticists and cell biologists who study meiosis, the existence of recombination hotspots offers a way to learn what other processes are associated with recombination. This is partly how we know that homologous crossovers in yeast and other eukaryotes are initiated by the cleavage of single chromosomes, called “double-strand breaks” ( Box 1 ). It turns out that because of this causal linkage, the hotspots for doublestrand breaks and the hotspots for recombination are one and the same ( Game et al. 1989 ; Sun et al. 1989 ; Keeney et al. 1997 ; Lopes et al. 1999 ; Allers and Lichten 2001 ; Hunter 2003 ). For population geneticists, much of the interest in recombination hotspots comes from their possible effect on the patterns of DNA sequence variation along human chromosomes and from the possibility that these patterns could be used to map the position of alleles that cause disease. When multiple copies of the DNA sequence of a gene, or of a larger region of a chromosome, are aligned, they reveal the location and distribution of variation at individual nucleotide positions—single nucleotide polymorphisms (SNPs). Each particular sequence, or haplotype, will carry a configuration for the SNPs for that region ( Figure 1 ). Investigators have long known that SNPs that are adjacent or near each other tend to be highly correlated in their pattern and to exhibit strong linkage disequilibrium ( Box 1 ). It is this linkage disequilibrium that enables scientists to map the locations of mutations that cause heritable genetic diseases. If alleles that cause a disease have the same kind of linkage disequilibrium with nearby SNPs as SNPs generally have with each other, then one could search for genes with disease alleles by looking for a pattern of SNPs that is found only in people who have the disease. This general method for mapping disease alleles is called “association mapping,” and it is basically a search for linkage disequilibrium between disease alleles and other SNPs. Whether or not association mapping works depends on the actual patterns of linkage that occur among SNPs in human populations, and these patterns depend in turn on how much recombination has occurred in the past (as well as on other demographic and mutation processes). Figure 1 A Hypothetical Example of Eight Aligned Haplotypes for All the SNPs Found in a Region Base positions that are not variable are not shown. Blocks of adjacent SNPs that revealed no evidence of historical recombination are flanked by vertical red bars. Two longer haplotype blocks are indicated in green. The presence of historical recombination was discerned by the four-gamete criterion. In brief, if the haplotypes for two SNPs, each with two bases (e.g., A/G at one SNP and C/T at the second), reveal all four possible combinations (i.e., A-C, A-T, GC, and G-T) then this is evidence that there has been a recombination event between these SNPs in the history of the sample of haplotypes ( Hudson and Kaplan 1985 ). If haplotype blocks are long, then it is possible to represent much of the haplotype diversity using just a small sample of the SNPs found within that region. With the advent of larger human haplotype data sets, it has become clear that there are often fairly long regions with very high linkage disequilibrium ( Daly et al. 2001 ; Patil et al. 2001 ; Gabriel et al. 2002 ). This pattern of variation has been characterized as occurring in “haplotype blocks,” which are apparent regions of low recombination (or high linkage disequilibrium). Figure 1 shows a hypothetical example of haplotype blocks among eight haplotypes for a series of SNPs found over a region of a chromosome. Given diverse evidence of recombination hotspots in humans, a much discussed question is whether recombination hotspots play a large role in the formation of the pattern of haplotype blocks ( Wang et al. 2002 ; Innan et al. 2003 ; Phillips et al. 2003 ; Stumpf and Goldstein 2003 ). The occurrence of haplotype blocks has inspired the HAPMAP project ( http://www.hapmap.org/), which has the goal of identifying a subset of SNPs that capture most of the relevant linkage information in the human genome ( IHC 2003 ). If one had a subset of all common SNPs, with one or two per haplotype block, then this subset would contain much of the available information for association mapping of disease alleles. The Evolution of Recombination and (Possibly) Recombination Hotspots Recombination is a nearly ubiquitous feature of genomes, and a great many theories have been put forward to explain why it would be evolutionarily advantageous for genes to regularly break with one another to join new genes ( Barton and Charlesworth 1998 ). By and large these theories predict that recombination should occur more often where genes occur in higher concentration and that it should happen less often in areas of the genome where genes are spaced far apart. This expectation is roughly born out in the human genome, where recombination rates are higher in regions of the genome with higher gene density ( Fullerton et al. 2001 ; Kong et al. 2002 ). To consider the possible evolutionary advantages of individual recombination hotspots, we can draw from theory on the evolution of recombination modifiers. In particular, recent population genetic theory has brought to light some fairly general circumstances for which mutations that raise recombination rates would be favored by natural selection ( Barton 1995 ; Otto and Barton 1997 ; Otto and Barton 2001 ; Otto and Lenormand 2002 ). The basic idea is that linkage disequilibrium can easily occur (for many reasons) between two (or more) polymorphic sites that are under selection. When this occurs, an allele that raises the recombination rate (and decreases the linkage disequilibrium) can cause selection to act more efficiently. If an allele that is under positive or negative selection always occurs with an allele at another locus that is also under selection (i.e., the two loci are in strong linkage disequilibrium), then selection cannot act on one locus independently of the second locus. As new, multilocus configurations of beneficial alleles are generated (by recombination) and increase in frequency by selection, the modifiers of recombination that caused the production of those beneficial configurations increase in frequency with them. A key piece of evidence supporting this kind of theory of the evolution of recombination is directional selection, like that which occurs in artificial selection experiments, which often generates a correlated elevation in recombination rates ( Otto and Lenormand 2002 ). Connecting these ideas about the evolution of recombination modifiers to the question of recombination hotspots, we come to the possibility that individual hotspots may have arisen as a byproduct of linkage disequilibrium between genes on either side of the hotspot that were under selection. This situation would create a kind of selection pressure favoring recombinant haplotypes and thus also favoring those chromosomes that happen to have a high recombination rate between the selected genes. If true, then we might expect local recombination rates (i.e., hotspots and coldspots for recombination) to fluctuate in location and intensity, in ways that would be hard to precisely predict without knowing what genes have been under selection and what patterns of linkage disequilibria there may have been. In this light, the paper by Ptak et al. (2004) in this issue of PLoS Biology is especially interesting. They report that chimpanzees do not have a recombination hotspot in the TAP2 region where humans have a fairly well characterized recombination hotspot ( Jeffreys and Neumann 2002 ). Ptak et al.'s is a statistical study of linkage disequilibrium in the TAP2 region of chimpanzees and humans, and is less direct than the sperm-typing study of Jeffreys and Neumann (2002) . However the contrast in linkage patterns between humans and our closest relatives suggests that recombination hotspots can evolve fairly quickly. Functional Constraints on Recombination Hotspots As appealing as the recombination modifier theory of recombination hotspots may be, there is circumstantial evidence that argues against it and that suggests that recombination hotspots are not directly the byproduct of selection on alleles in linkage disequilibrium. Particularly important in this regard is that some wellstudied organisms (notably the worm Caenorhabditis elegans and the fruitfly Drosophila melanogaster ) have not shown evidence of recombination hotspots. If we compare these organisms with yeast and mammals, which do show hotspots, we gain some more insight into the factors affecting the evolution of hotspots. Recall that double-strand breaks are the sites where recombination is initiated during meiosis, and that this is true regardless of the presence of hotspots for both phenomena. Apparently it is the case in yeast and mammals that both recombination and double-strand breaks are also prerequisites for the proper formation of the synaptonemal complex (SC) ( Figure 2 ) and thus for proper orientation of the spindle apparatus and accurate segregation of chromosomes during meiosis ( Paques and Haber 1999 ; Lichten 2001 ; Hunter 2003 ; Page and Hawley 2003 ). In contrast, neither double-strand breaks nor recombination appear to be required for the formation of the SC in D. melanogaster or C. elegans ( Zickler and Kleckner 1999 ; MacQueen et al. 2002 ; McKim et al. 2002 ; Hunter 2003 ; Page and Hawley 2003 ). Double-strand breaks and recombination do indeed co-occur in these model organisms, and are required for proper chromosome segregation, but they occur after the formation of the SC. Both of these species have broad chromosomal regions where crossing over occurs at higher rates than others, but there have been no reports of local recombination hotspots. Figure 2 The Synaptonemal Complex (A) Model of the SC. Lateral elements (light blue rods) of homologous chromosomes align and synapse together via a meshwork of transverse filaments (black lines) and longitudinal filaments (dark blue rods). The longitudinal filaments are collectively referred to as the “central element” of the SC. Ellipsoidal structures called recombination nodules (gray ellipsoid) are constructed on the central region of the SC. As their name implies, recombination nodules are believed to be involved in facilitating meiotic recombination (crossing over). The chromatin (red loops) of each homologue is attached to its corresponding lateral element. Because there are two “sister chromatids” in each homologue, two loops are shown extending laterally from each point along a lateral element. (B) Top: Set of tomato SCs. Chromatin “sheaths” are visible around each SC. Bottom: Two tomato SCs. The chromatin has been stripped from the SCs, allowing the details of the SC to be observed. Each SC has a kinetochore (“ball-like” structure) at its centromere. Recombination nodules, ellipsoidal structures found on the central regions of SCs, mark the sites of crossover events (see inset). Images and legend courtesy of Daniel G. Peterson, Mississippi Genome Exploration Laboratory, Mississippi State University, Mississippi State, Mississippi, United States ( http://www.msstate.edu/research/mgel/index.htm) . Recombination during meiosis seems to be required for proper chromosome segregation; however, in those organisms where recombination and double-strand-break hotspots occur, these phenomena are also required for proper formation of the SC. It is as if the recombination machinery has been partly co-opted for chromosome alignment in some eukaryotes more so than in others. The implication of these findings is that recombination hotspots are byproducts of other functional constraints associated with the recombination process. This does not rule out the evolutionary theory of recombination modifiers, or that the location and intensity of recombination hotspots may evolve rapidly, but it does suggest that we may not need to invoke the evolutionary modifier theory to explain the existence of recombination hotspots. Conclusions Recombination hotspots co-occur with double-strand-break hotspots in some eukaryotes, and together these phenomena appear to play an important role in the formation of the SC in those organisms. Given the limited phylogenetic occurrence of recombination hotspots (i.e., their occurrence in some, but not all, species), general theories for the evolution of recombination may not be very helpful for understanding the existence of recombination hotspots. However, in those species where they do occur, it is quite possible that recombination hotspots do evolve in location and intensity. Furthermore, the presence of recombination hotspots in humans may have large effects on the length of local patterns of linkage disequilibrium (haplotype blocks) and thus on our ability to map disease alleles by their association with other markers. Box 1. Glossary Double-strand break: A break in both strands of a DNA molecule, as distinguished from a break in just one strand. Linkage disequilibrium: A pattern of association between two SNPs or two loci that each have multiple alleles, such that pairs of particular SNPs or alleles, one from each locus, tend to co-occur within individuals or genomes more often than would be expected if the loci are sorting independently of each other. Recombination: The process of one double-stranded DNA molecule joining with another; specifically in the context of meiosis, the process of two homologous chromosomes exchanging large portions of their DNA (this is also called “crossing over”). | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC423158.xml |
544855 | Predicting binding sites of hydrolase-inhibitor complexes by combining several methods | Background Protein-protein interactions play a critical role in protein function. Completion of many genomes is being followed rapidly by major efforts to identify interacting protein pairs experimentally in order to decipher the networks of interacting, coordinated-in-action proteins. Identification of protein-protein interaction sites and detection of specific amino acids that contribute to the specificity and the strength of protein interactions is an important problem with broad applications ranging from rational drug design to the analysis of metabolic and signal transduction networks. Results In order to increase the power of predictive methods for protein-protein interaction sites, we have developed a consensus methodology for combining four different methods. These approaches include: data mining using Support Vector Machines, threading through protein structures, prediction of conserved residues on the protein surface by analysis of phylogenetic trees, and the Conservatism of Conservatism method of Mirny and Shakhnovich. Results obtained on a dataset of hydrolase-inhibitor complexes demonstrate that the combination of all four methods yield improved predictions over the individual methods. Conclusions We developed a consensus method for predicting protein-protein interface residues by combining sequence and structure-based methods. The success of our consensus approach suggests that similar methodologies can be developed to improve prediction accuracies for other bioinformatic problems. | Background Protein-protein interactions play a critical role in protein function. Completion of many genomes is being followed rapidly by major efforts to identify experimentally interacting protein pairs in order to decipher the networks of interacting, coordinated-in-action proteins. Identification of protein-protein interaction sites and detection of specific residues that contribute to the specificity and strength of protein interactions is an important problem [ 1 - 3 ] with broad applications ranging from rational drug design to the analysis of metabolic and signal transduction networks. Experimental detection of residues on protein-protein interaction surfaces can come either from determination of the structure of protein-protein complexes or from various functional assays. The ability to predict interface residues at protein binding sites using computational methods can be used to guide the design of such functional experiments and to enhance gene annotations by identifying specific protein interaction domains within genes at a finer level of detail than is currently possible. Computational efforts to identify protein interaction surfaces [ 4 - 6 ] have been limited to date, and are needed because experimental determinations of protein structures and protein-protein complexes, lag behind the numbers of protein sequences. In particular, computational methods for identifying residues that participate in protein-protein interactions can be expected to assume an increasingly important role [ 4 , 5 ]. Based on the different characteristics of known protein-protein interaction sites [ 7 ], several methods have been proposed for predicting interface residues using a combination of sequence and structural information. These include methods based on the presence of "proline brackets"[ 8 ], patch analysis using a 6-parameter scoring function [ 9 , 10 ], analysis of the hydrophobicity distribution around a target residue [ 7 , 11 ], multiple sequence alignments [ 12 - 14 ], structure-based multimeric threading [ 15 ], and analysis of amino acid characteristics of spatial neighbors to a target residue using neural networks [ 16 , 17 ]. Our recent work has focused on prediction of interface residues by utilizing analyses of sequence neighbors to a target residue using SVM and Bayesian classifiers [ 2 , 3 ]. There is an acute need for multi-faceted approaches that utilize available databases of protein sequences, structures, protein complexes, phylogenies, as well as other sources of information for the data-driven discovery of sequence and structural correlates of protein-protein interactions [ 4 , 5 ]. By exploiting available databases of protein complexes, the data-driven discovery of sequence and structural correlates for protein-protein interactions offers a potentially powerful approach. Results and discussion Here we are using a dataset of 7 hydrolase complexes from the PDB, together with their sequence homologs. The application of our consensus method to other types of complexes, e.g . antibody-antigen complexes is currently under study and will be published later. It should be noted, however, that prediction of binding sites for other types of protein complexes, especially those involved in cell signaling, is likely to be more difficult than for the hydrolase-inhibitor complexes. Figure 1 shows an example of the consensus method prediction mapped on the structure of proteinase B from S. griseus in a complex with turkey ovomucoid inhibitor (PDB 3sgb [ 18 ]). The inhibitor (3sgb_I) is shown at the top in wire frame and the proteinase B chain (3sgb_E), is shown at bottom. Actual interface residues in the proteinase B chain, i.e., amino acids that form the binding site between proteinase B and the inhibitor, were extracted from the PDB structure (see Materials and Methods). Predicted interface and non-interface residues, identified by the consensus method, are shown as color coded atoms as follows: Red spheres = true positives (TP), actual interface residues that are predicted as such; Gray strands = true negatives (TN), non-interface residues that are predicted as such; Yellow spheres = false negatives (FN), interface residues that are misclassified as non-interface residues; Blue spheres = false positives (FP), non-interface residues that are misclassified as interface residues. Note that the binding site in proteinase B is strongly indicated, with 14 out of 15 interface residues correctly classified, along with 2 false positives. The primary amino acid sequence for proteinase B chain and the interface residue prediction results for the four individual methods and the consensus method are shown in Figure 2 . Actual interface residues are identified highlighted in red. The five lines below the amino acid sequence show the locations of interface residues predicted by the different methods (described in detail below): P = Phylogeny; C = Conservatism of Conservatism (CoC); S = Data mining by SVM; T = Threading; E = Consensus. Similar Figures for each protein studied in this work are provided in Supplementary Materials [see Additional files 1 , 2 , 3 , 4 , 5 , and 6 ]. Figure 1 Interface residues predictions mapped on the three dimensional structure of Proteinase B from Streptomyces griseus (3sgb). The target protein is shown in ribbons and atomic spheres; the inhibitor partner is shown at the top in faint wire frame. The residues are color coded as: red = true positives (TP), gray = true negatives (TN), yellow = false negatives (FN), and blue = false positives (FP). Red, yellow, and blue residues are shown in spacefill representation. Note that the actual interface residues extracted from the PDB structure include the red (TP) and yellow (FN) residues. Red and gray residues represent correct predictions of interface and non-interface residues (14 TP+ 210 TN = 224 correct predictions); yellow and blue residues represent incorrect predictions (1 FN + 2 FP= 3) Figure 2 Comparison of individual methods for interface residue prediction with the consensus method. Results are shown for Proteinase B from Streptomyces Griseus (3sgb_E), the same protein shown in Figure 1. Actual interfaces are highlighted in red. Interface residues predicted by each of five different methods are indicated as follows: P = Phylogeny (none predicted for this protein), C = Conservatism of Conservatism; S = Support Vector Machine; T = Threading; and E = Consensus. Amino acid residues present in the protein sequence, but not included in the PDB structure file, are indicated by "X"s in the sequence. The prediction results for all methods are shown in Table 1 and Table 2 . Table 1 shows a complete summary of the classification performance on the proteinase B chain for all 5 methods including the overall Sensitivity (Sen) and Specificity (Spec); Sensitivity (Sen+) and Specificity (Spec+) for interface residues (the "positive" class); and Correlation Coefficient (see Materials and Methods for definitions of these performance parameters). Table 2 shows the overall average performance results for all seven protein complexes studied in this work. Two kinds of averages are considered: the numerical average over each of 7 proteins in the dataset, i.e., the average on a "per protein" basis (<...> p ); and the average over the total number of residues, i.e., the average on a "per residue" basis (<...> r ). Table 1 Classification results for Proteinase B from S. griseus (3sgb_E). TP is the number of true positive; TN is the number of true negatives; FP is the number of false positives, and FN is the number of false negatives. Overall sensitivity, overall specificity, sensitivity+, specificity+, and correlation coefficient are defined in the text. 3SGBE TP # TN # FP # FN # Overall Sen Overall Spe Sen+ Spe+ CC Phylog. 0 212 0 15 0.94 0.91* 0 - 0* COC 15 194 18 0 0.92 0.96 1 0.45 0.64 SVM 3 205 7 12 0.92 0.90 0.20 0.30 0.20 Thread. 14 201 11 1 0.95 0.97 0.93 0.56 0.70 Cons. 14 210 2 1 0.99 0.99 0.94 0.88 0.90 Table 2 Overall Classification Performance Results Averaged over 7 Proteins. Average results for Sensitivity+, Specificity+, overall Sensitivity, overall Specificity, and Correlation Coefficient averaged over the 7 proteins in the dataset. <> p denotes averaging over the total number of proteins, <> r denotes averaging over the total number of residues. Method <Sen+> p <Spe+> p <Spe> p <Spe> r <Sen> p <Sen> r <Cor> p <Cor> r Phylog. 0.39 0.71 0.90 0.89 0.91 0.89 0.43 0.37 COC 0.71 0.31 0.89 0.88 0.81 0.80 0.38 0.37 SVM 0.51 0.41 0.89 0.88 0.88 0.88 0.39 0.37 Thread. 0.59 0.57 0.91 0.89 0.92 0.91 0.53 0.48 Cons. 0.70 0.56 0.92 0.91 0.90 0.89 0.56 0.55 Sequence and structure conservation Amino acid sequences are conserved for many different reasons related to the structure and function of proteins: for stability [ 19 , 20 ], enzyme active sites, subunit interfaces, facilitation of an essential motion (hinges), and binding sites. Developing methods to identify the reason for conservation of individual highly conserved residues is a difficult problem. This is one of the reasons that a combination of approaches may be more likely to permit identification of residues that participate in protein-protein interactions. Even identifying the conserved residues themselves is not completely straightforward, and as will be seen, different approaches will indicate the same residue being conserved to different extents. In this study, we take advantage of this by using several methods to identify sequence and structure conservation. Here we use two principal methods for this purpose, one based on phylogeny to identify sequence conservation and one based on Conservatism of Conservatism [ 21 ] to identify structure conservation. These two methods often identify different residues as being conserved. Phylogeny To identify protein residues that are conserved – perhaps due to their functional role in forming specific protein-protein interactions – we use ClustalX [ 22 ] multiple sequence alignments of protein sequences to generate phylogenetic trees (see Materials and Methods). Conserved residues are defined as those that are identical at a given position in more than 85% alignments, i.e., only 15% substitutions or gaps were allowed. This 85% cutoff value is found to give optimal results (data not shown). Because phylogenetic trees of closely related sequences result in many residues that satisfy this condition (due to the high conservation of sequences, apparently important for protein folding, located in the protein core) we filter the results to focus on surface residues by removing conserved residues residing inside the protein core, i.e., having low solvent accessibility (see Materials and Methods). As shown in Figure 2 , the phylogenetic method does not classify any of the amino acids in proteinase B chain (3sgb_E) as interface residues, i.e., TP = 0 and FP = 0. Thus, for the phylogenetic method prediction, the correlation coefficient (CC), which can range from -1 to +1, converges to zero, whereas overall specificity converges to 0.905. The latter misleading statistic is due to the large number of negative examples (non-interface residues), which are correctly classified. In cases such as this (with unbalanced numbers of positive and negative examples), sensitivity + and specificity + measures are especially useful because they more clearly reflect the ability of a method to detect "positive" interface residues. (See the Methods section for definition and further discussion of performance measures). Note that even though Figure 2 shows that the phylogenetic method does not identify any interface residues in this particular example, the results summarized in Table 1 for all seven proteins demonstrate that the ability of the phylogenetic method to correctly predict non-interface residues (reflected in the high overall sensitivity and specificity values), and in combination with other methods, to lead to significantly improved predictions. Conservatism of conservatism To detect structurally conserved residues that are possible binding sites we have used the Conservatism of Conservatism method (CoC) developed by Mirny and Shakhnovich[ 21 ] We use structural alignments generated by FSSP (fold classification based on structure-structure alignment of proteins) developed by Holm and Sander [ 23 ]) to identify protein families with folds similar to that of the each of the 7 proteins. For each family, HSSP [ 24 ] (homology-derived secondary structure of proteins) alignments are used to calculate the sequence entropy at each position of the alignment. The HSSP profile is based on the multiple alignment of a sequence and its potential structural homologues [ 25 ]. The structural alignment generated by FSSP is used to calculate the value of CoC (see Materials and Methods). Each residue in the protein chain was ranked according to its CoC value at a given position in the sequence. The top 75% of total residues ranked according to their CoC values are defined as conserved. We filter the results of the CoC ranking by removing all structurally conserved residues located inside the protein core by only choosing the residues that have a relative accessibility of at least 25 as calculated by DSSP [ 26 ] (dictionary of protein secondary structure). Interface residues in proteinase B predicted by this method are indicated by a "C" in Figure 2 . The overall performance of the CoC method is summarized in the second row of Tables 1 and 2 . Although the correlation coefficient of the COC method is in the same range of those obtained by phylogeny and support vector machines, 0.37, the sensitivity+ value, 0.71, is surpassed only by the consensus value. Therefore, a larger fraction of interface residues is predicted by CoC than the other three methods. However, the CoC method alone is not sufficient to successfully predict binding sites, and combining this method with other prediction techniques in the consensus method gives improved results (Tables 1 and 2 ). Data mining for binding residues We have generated a support vector machine (SVM) classifier to determine whether or not a surface residue is located in the interaction site using information about the sequence neighbors of a target residue. An 11-residue window consisting of the residue and its 10 sequence neighbors (5 on each side) is chosen empirically. Each amino acid in the 11 residue window is represented using 20 values obtained from the HSSP profile of the sequence. Each target residue is therefore associated with a 220 (11 × 20) element vector. The SVM learning algorithm is given a set of labeled examples of the form (X, Y) where X is the 220 element vector representing a target residue and Y is its corresponding class label, either interface or non-interface residue. The SVM algorithm generates a classifier which takes as input a 220 element vector that encodes a target residue to be classified and outputs a class label. Our previous study [ 2 ] reported results for classifiers constructed using a combined set of 115 proteins belonging to six different categories of complexes: antibody-antigen, protease-inhibitor, enzyme complexes, large protease complexes, G-proteins, cell cycle signaling proteins, signal transduction, and miscellaneous. In another study [ 3 ], we trained separate classifiers for each major category of complexes (e.g., protease-inhibitor complexes). In the case of protease-inhibitor complexes, leave-one-out experiments were performed on a set of 19 proteins. In each experiment, an SVM classifier was trained using a set of surface residues, labeled as interface or non-interface, from 18 of the 19 proteins. The resulting classifier was used to classify the surface residues of the remaining target protein into interface residue and non-interface residue categories. The interface residues obtained for 3sgb_E are reproduced in Figure 2 and marked by "S". The performance of the SVM classifier for the current test set of complexes is summarized in Tables 1 and 2 . The results show that SVM yields relatively high sensitivity+ (0.51) and specificity+ (0.41). Threading of sequences through structures of interface surfaces Structural threading was performed for the set of 7 protein complexes using a recently developed threading algorithm [ 27 ], which was first used in the CASP5 [ 28 ] competition. For each complex structure, we first extract the interfacial region, essentially as described earlier. Residue-residue contacts in the interfacial region are described with contact matrices. The total energy in this threading method is the sum of all pair-wise contact energies for the conformation. Detailed residue-level contact potentials were obtained from the Li, Tang and Wingreen [ 29 ] parameterization of the Miyazawa and Jernigan [ 30 ] matrix. We represent a protein sequence vector s by the hydrophobicity values of its amino acids h i obtained in this factorization and protein structure by the contact matrix Γ. The problem of finding the best alignment of a query sequence s with a structure having contact matrix Γ is to find the transformation from s to s' that optimizes the energy function. The optimum s' is the dominant eigenvector v 0 of the contact matrix Γ. There is a strong correlation between a protein sequence and the dominant eigenvector of its native structure's contact matrix. Here the transformation we seek is obtained by maximizing the correlation between s' and v 0 . This is an alignment problem, and a dynamic programming method from sequence alignment has been adapted to solve this problem [ 27 ]. For each sequence, threading is performed against structures in our template database and alignment results used only when the score exceeds a length-dependent threshold. From the alignments, residues involved in contacts at the interface are identified using a scale based on the number of times a particular residue is indicated and the strength of the threading score. The predicted binding sites for 3sgb_E by the threading method are marked in Figure 1 by "T" and the prediction results are summarized in Tables 1 and 2 . The threading-based approach is somewhat more successful than other methods based on its sensitivity+, selectivity+, and correlation coefficient values, but still not as good as the performance obtained by combining it with methods in the consensus approach. Consensus method for predicting protein binding sites Based on the results from the predictions with the four independent methods, we have developed a simple consensus method to obtain a better prediction. In the consensus method results presented here, an amino acid is considered to be an interface residue if any of the following conditions are met: i) at least three independent methods classify it as an "interface residue" ii) any two methods (except the Phylogeny-Threading pair) predict it For this set of proteins, the parameters for combining results in the consensus method have been empirically determined without a systematic comparison of the strengths and weaknesses of each method. We employ this simple approach because it provides demonstrable improvement in prediction performance over the individual methods. The consensus interface residue predictions are indicated by an "E" in Figure 1 , and performance results are summarized in the last rows of Tables 1 and 2 . The consensus method generally results in an enhanced correlation coefficient and sensitivity+, demonstrating the superior performance of the consensus method for identifying interface residues in this protein set. Predictions for each protein, provided in Supplementary Materials [see Additional files 1 , 2 , 3 , 4 , 5 , and 6 ], illustrate that the improvements can be even more pronounced when the individual predictions of all four methods are relatively weak. This suggests that combining diverse prediction methods may be an excellent approach for the prediction of the binding sites in protein complexes. Conclusions Each of the four prediction methods presented in this paper sheds a different light on the conservation and prediction of protein interaction sites, but none of the methods taken separately is as powerful as the combination of all four methods. The simple consensus approach presented here could perhaps be improved by generating an ensemble predictor with more detailed probabilities. Our current work is directed at this approach. It is clear that the present subject is an active field of research [ 31 - 38 ]. Methods Dataset of hydrolase-inhibitor complexes The dataset of 7 hydrolase-inhibitor complexes used in this work has been derived from a larger dataset of 70 protein heterocomplexes extracted from PDB by Chakrabarti and Janin [ 39 ] and used in our previous studies [ 2 , 3 ]. All are proteins from hydrolase-inhibitor complexes, with six being proteinases: 1acb_E [ 40 ] (chain E of PDB structure 1acb), 1fle_E[ 41 ], 1hia_A[ 42 ], 1avw_A[ 43 ], 2sic_E[ 44 ], 3sgb_E [ 18 ]; and one being a carboxypeptidase: 4cpa [ 45 ]. Definition of surface and interface residues Surface and interface residues for the proteins were identified based on information in the PDB coordinate files as previously described [ 2 , 3 ]. Briefly, solvent accessible surface areas (ASA) for each residue in the unbound protein and in the complex are calculated using DSSP [ 26 ]. A surface residue is defined as an interface residue if its calculated ASA in the complex is less than that in the monomer by at least 1 Å 2 [ 46 ]. In the extraction of interfacial region for threading, however, a distance-based definition of surface is used: a surface residue is defined as an interface residue if its side-chain center is within 6.5Å of the side-chain center of a residue belonging to another chain in the complex. Based on the ASA definitions, 41% of the residues in the set of 7 proteins were surface residues, corresponding to a total of 631 surface residues. Among these surface residues, 166 were defined as interface residues and 465 as non-interface residues (i.e. surface residues that are not in the interaction sites). Thus, on average, interface residues represent 26% of surface residues, or 11% of total residues for proteins in our dataset. Using phylogeny to identify conserved residues Many computational tools have been developed for identifying amino acids that are important for protein function/structure, but there is no consensus regarding the best measure for evolutionary conservation [ 47 ]. Evolutionary conservation can be decomposed into three components: i) the overall selective constraints – the number of changes observed at a site; ii) the pattern of amino acid substitutions – the number of amino acid types observed at a site; and iii) the effect of amino acid usage. We have established a reliable relationship between each measure and various aspects of structure. To explore the connection between sequence conservation and functional-structural importance, we proposed a new measure that can decompose the conservation into these three components [ 47 ]. This measure is based on phylogenetic analysis. The evolutionary rate at site k during lineage l from amino acids i to j ( i , j = 1,...20) can be expressed as λ kl ( i , j ) = c k × a lk × Q ( i , j | k ), where c k accounts for the rate variation among sites, a lk for site-specific lineage (or subtree) effect caused by functional divergence [ 48 ], and the 20 × 20 matrix Q ( i , j | k ) is the (site-specific) model for amino acid substitutions. The likelihood function for a given tree can be determined according to a Markov chain model [ 49 ]. We have developed an integrated computer program (DIVERGE [ 50 ]) that can map these predicted sites onto the protein surface to examine these relationships. We use the solvent accessibility data from DSSP [ 26 ] to restrict predicted conserved residues to those located on the protein surface. Conservatism of conservatism The phylogeny-based conservation of residues relies on sequence homology. It is well known, however, that many non-homologous proteins share similar folds [ 51 ]. It is therefore highly desirable to study the conservation of residues in proteins based on the structural superimposition of non-homologous proteins. In order to obtain insight into the evolutionary conservation of residues in proteins, we use the Conservatism of Conservatism method (CoC). The CoC method was developed by Mirny and Shakhnovich [ 21 ] for studying evolutionary conservation of residues in proteins with specific folds from the FSSP database [ 23 ]. With the FSSP database, Mirny and Shakhnovich performed an analysis of conserved residues in several common folds. The 20 naturally occurring amino acids were subdivided into 6 different classes, based on their physicochemical characteristics and frequencies of occurrence at different positions in multiple sequence alignments. The evolutionary conservatism within families of homologous proteins was measured through sequence entropy. Structural superimposition of different families of proteins with similar folds was used to calculate CoC for all positions of residues within a fold. Here we have applied a similar approach to identify structurally conserved residues involved in protein interactions. For each protein, we first calculate the sequence entropy at each position within a family of related sequences from the HSSP database [ 25 ] where is the frequency of the class i of residues (for each of the six classes) at position l in sequence in the multiple sequence alignment. Then we use the FSSP database to obtain the structural alignment. The structural superimposition of different families was used to calculate the conservatism of conservatism (CoC) where s m (l) is the intrafamily conservatism within the family m at position l , and M is the number of families. The CoC is the measure of the evolutionary conservation of the specific sites within the protein fold. Because the CoC method does not distinguish between residues at the protein surface evolutionarily conserved for functional reasons and residues inside the protein core that are conserved because of their importance to the folding process, we use solvent accessibility data for the unbound molecules to eliminate those conserved residues located inside the protein core. Data mining approaches to binding site identification Recent advances in machine learning [ 52 ] or data mining [ 53 ] offer a valuable approach to the data-driven discovery of complex relationships in computational biology [ 54 , 55 ]. In essence, a data mining approach uses a representative data training set to extract complex a priori unknown relationships, e.g., sequence correlates of protein-protein interactions. Examination of the resulting classifiers can help generate specific hypotheses that can be pursued using molecular and biophysical methods. For example, a classifier that is able to identify protein-protein interface residues on the basis of sequence or structural features can provide insights about sequence characteristics that contribute to important differences in function. The data mining approach for binding site identification consists of the following steps: • Identify the surface residues in each protein. • Label each residue in each protein as either an interface residue or a non-interface residue based on appropriate criteria for defining residues in interaction sites. • Use a machine learning algorithm to train and evaluate a classifier to categorize a target amino acid as either an interface or a non-interface residue. Different types of information about the target residue (e.g., the identity and physicochemical properties of its sequence neighbors, whether or not the target residue is a surface residue) can be supplied as input to the classifier. A variety of machine learning algorithms [ 52 , 54 ] can be used for this purpose. • Evaluate the classifier (typically using cross-validation or leave-one-out experiments) on independent test data (not used to train the classifier). • Apply the classifier to identify putative interface residues in a protein, given its sequence (and possibly its structure), but not the sequence or structure of its interaction partner. Here we have used a support vector machine (SVM) learning algorithm because SVMs are well-suited for the data-driven construction of high-dimensional patterns and are especially useful when the input is a real-valued pattern [ 56 ]. In addition, algorithms for constructing SVM classifiers effectively incorporate methods to avoid over-fitting the training data, thereby improving its generality, i.e., the performance of the resulting classifiers on test data. Support vector machine algorithms have proven effective in many applications, including text classification [ 57 ], gene expression analysis using microarray data [ 58 ], and predicting whether or not a pair of proteins is likely to interact [ 59 ]. Threading of sequences through structures of protein-protein interface surfaces In phylogenetic and data mining approaches, the properties of the protein-protein interface are deduced by concentrating on the sequence information contained in the protein pair under investigation. However, it is well accepted that the physical origin of the specificity of protein-protein interactions comes predominantly from their structures. Thus, in any thorough investigation of protein-protein interactions, it is essential to include information from structural studies. Here we have adapted methods employed in protein structure recognition [ 60 - 63 ] to the problem of predicting protein-protein interface residues. In the first stage, structural models for identifying protein-protein interfaces are generated from existing protein databank (PDB) structures by extracting portions of contacts between different protein chains. We found that if we define the interaction region by the criterion that backbone C α atoms on the two interacting chains are less than 15 Å apart, reasonably well connected fragments suitable for threading studies are obtained. In the second stage, after identifying a set of candidate template structures, threading is performed to examine the probability that a given model resembles the real interface. The threading algorithm is described in Cao et al. [ 27 ]. The threading alignments and scores obtained allowed us to predict which parts of each protein are in the interfacial region in the hydrolase-inhibitor complexes and to predict the most probable residue-residue contacts between the two proteins. Ensemble predictions for combining results from multiple methods Different approaches for identifying binding sites from amino acid sequence information yield different (sometimes contradictory, sometimes complementary) results. In such cases, approaches for combining results from multiple predictors have a potential importance. The key idea is that results obtained by using different approaches, which we will call classifiers henceforth, may be correlated (or, more generally, statistically dependent) due to a variety of reasons including the use of a common dataset for constructing or tuning classifiers, use of intermediate variables for encoding input to the classifiers, and similarities between methods (e.g., SVM, neural networks). Regardless of the source of statistical dependency, the goal is to develop methods for weighting the output of each classifier appropriately for the purpose of producing more accurate predictions. Our method takes as input the binary (True/False) output of each classifier (e.g., SVM, CoC) and produces as output a probability that the residue under consideration is an interface residue, using the outputs produced by each of the classifiers. Algorithms for learning Bayesian (or Markov networks) can be then used to learn the network of dependences and the relevant conditional probabilities. General evaluation measures for assessing the performance of classifiers Let TP denote the number of true positives – residues predicted to be interface residues that are actually interface residues; TN the number of true negatives – residues predicted not to be interface residues that are in fact not interface residues; FP the number false positives – residues predicted to be interface residues that are not interface residues; FN the number of false negatives – residues predicted not to be interface residues that actually are interface residues. Let N = TP + TN + FP + FN . Sensitivity (recall) and Specificity (precision) are defined for the positive (+) class as well as the negative (-) class. Sensitivity + = TP /( TP + FN ), Sensitivity - = TN /( TN + FP ), Specificity + = TP /( TP + FP ), Specificity - = TN /( TN + FN ). Overall sensitivity and overall specificity correspond to expected values of the corresponding measures averaged over both classes. The performance of the classifier is summarized by the correlation coefficient, which is given by The correlation coefficient ranges from -1 to 1 and is a measure of how predictions correlate with the actual data [ 64 ]. It is important to note, that when the number of negative instances is much larger than the number of positive instances – as is the case for prediction of interface residues – the Sensitivity+ and Specificity+ measures are more appropriate for assessing prediction performance than the overall Sensitivity and Specificity measures [ 64 ]. In the extreme case when a classifier predicts every example to be negative (due to a preponderance of negative training instances) these overall performance measures would still show a high success rate despite the obvious failure of the prediction method. In such cases, the Correlation Coefficient, as well as the Sensitivity+, which is a measure of the fraction of positive instances that are correctly predicted, and Specificity+, which is a measure of the fraction of the positive predictions that are actually positive instances, may provide better performance assessment. Of course, a meaningful comparison of the performance of different classification methods depends critically on the specific application and goal. Author's contributions CY, VH and DD performed data mining calculations. XG performed phylogenetic calculations. KMH, CZW, YI, DD, and HC worked on threading. TZS, AK, and RLJ worked on the implementation of CoC and the development of consensus methodology. Every author contributed to the final draft of the paper. Supplementary Material Additional File 1 Comparison of individual methods for interface residue prediction for bovine α-chymotrypsin (1acbe). Click here for file Additional File 2 Comparison of individual methods for interface residue prediction for porcine pancreatic trypsin (1avwa). Click here for file Additional File 3 Comparison of individual methods for interface residue prediction for porcine pancreatic elastase (1flee). Click here for file Additional File 4 Comparison of individual methods for interface residue prediction for kallikrein(1hiaa). Click here for file Additional File 5 Comparison of individual methods for interface residue prediction for subtilisin BPN' (2sice). Click here for file Additional File 6 Comparison of individual methods for interface residue prediction for carboxypeptidase A (4cpa). Click here for file | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC544855.xml |
555541 | Correction: Determination of the differentially expressed genes in microarray experiments using local FDR | null | Following publication of this work [ 1 ], the authors became aware that Local FDR has been originally defined by Efron & al. (2001) in a mixture model framework [ 2 ]. | /Users/keerthanasridhar/biomedlm/data/PMC000xxxxxx/PMC555541.xml |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.