Learning and convergence analysis of neural-type structured networks

A class of feedforward neural networks, structured networks, has recently been introduced as a method for solving matrix algebra problems in an inherently parallel formulation. A convergence analysis for the training of structured networks is presented. Since the learning techniques used in structured networks are also employed in the training of neural networks, the issue of convergence is discussed not only from a numerical algebra perspective but also as a means of deriving insight into connectionist learning. Bounds on the learning rate are developed under which exponential convergence of the weights to their correct values is proved for a class of matrix algebra problems that includes linear equation solving, matrix inversion, and Lyapunov equation solving. For a special class of problems, the orthogonalized back-propagation algorithm, an optimal recursive update law for minimizing a least-squares cost functional, is introduced. It guarantees exact convergence in one epoch. Several learning issues are investigated.

Construction of bacterial artificial chromosome (BAC/PAC) libraries

This unit describes the construction of BAC and PAC libraries. Two vectors, pCYPAC2 and pPAC4 have been used for preparing PAC libraries, and a new BAC vector pBACe3.6 has been developed for construction of BAC libraries. A support protocol describes preparation of PAC or BAC vector DNA for cloning by digestion with BamHI or EcoRI, simultaneous dephosphorylation with alkaline phosphatase, and subsequent purification through pulsed-field gel electrophoresis (PFGE). For the preparation of high-molecular weight DNA for cloning, support protocols provide procedures for embedding total genomic DNA from lymphocytes or animal tissue cells, respectively, in InCert agarose. Another support protocol details the next steps for the genomic DNA: partial digestion with MboI or with a combination of EcoRI endonuclease and EcoRI methylase, and subsequent size fractionation by preparative PFGE. The final support protocol covers the isolation of BAC and PAC plasmid DNA for analyzing clones.

Construction of bacterial artificial chromosome (BAC/PAC) libraries

Large-insert genomic libraries are necessary for physical mapping of large chromosomal regions, for isolation of complete genes, and for use as intermediates in DNA sequencing of entire genomes. Construction of BAC and PAC libraries is detailed in the unit, including preparation of PAC or BAC vector DNA for cloning by digestion with BamHI or EcoRI, dephosphorylation with alkaline phosphatase, and purification through pulsed-field gel electrophoresis (PFGE). For the preparation of high-molecular weight DNA for cloning, procedures for embedding total genomic DNA from lymphocytes or animal tissue cells are also provided. Other protocols detail partial digestion of genomic DNA with MboI or with a combination of EcoRI endonuclease and EcoRI methylase (including methods for optimizing the extent of digestion), and subsequent size fractionation by preparative PFGE. Finally, the isolation of BAC and PAC plasmid DNA for analyzing clones is also presented.

Identification of a haplosufficient 3.6-Mb region in human chromosome 11q14.3-->q21

Cytogenetic deletions are almost always associated with phenotypic abnormality and are very rarely transmitted. We have located a hitherto undescribed, familial deletion involving the region 11q14.3-->q21 in five individuals in a three-generation kindred. Four of the deletion carriers show no phenotypic abnormality; the other, who is the proband, was investigated for short stature and poor academic progress. In view of the apparent innocuous nature of this genetic imbalance, the deletion was investigated in detail to determine its size (3.6 Mb) and location with reference to molecular markers and genetic content. The deleted region is described by a contig of 37 BACS including the flanking regions, which we have assembled. Several possible contributory factors are considered, which might explain the lack of clinical significance of this large deletion. It is notable that there are few genes in this region and none have known functions. All most likely have copies elsewhere in the genome and a number of other hypothetical genes appear to be members of certain gene families, i.e. none is unique. Part of the region (1 Mb) is also duplicated at the pericentromeric region 11p11. Given the very low proportion of the genome occupied by single copy genes and their uneven distribution, regions such as this, which appear to be functionally haplosufficient, may be more common than hitherto recognised.

Gene therapy for inherited lung disorders: an insight into pulmonary defence

This review summarizes the latest developments in viral and nonviral gene delivery systems to the lung, and the problems that have to be overcome. Gene delivery has the potential to offer effective treatment to patients with life-threatening lung diseases such as cystic fibrosis and alpha(1)-antitrypsin deficiency, and could modify gene-environment relationships in asthma and other respiratory diseases. Phase I clinical trials conducted in the early 1990s showed that in principle gene transfer to the lung was safe. Although the preliminary results gave encouraging laboratory data, gene expression from viral or nonviral gene delivery systems was too inefficient or transient to offer clinical benefit. Initial optimism gave way to the realization that gene therapy to the lung was unlikely to be straightforward. The host innate and acquired immune system, which protects against infection from inhaled bacteria and viruses, represents a major barrier to successful gene transfer to the lung. A better understanding of the immunological barriers which exist in the lung may allow the development of pharmacological and/or immunological agents that modulate the host immune system to allow for a more continuous and regulated level of gene expression following gene transfer.

Targeted gene correction in the mdx mouse using short DNA fragments: towards application with bone marrow-derived cells for autologous remodeling of dystrophic muscle

In muscle, mutant genes can be targeted and corrected directly by intramuscular (i.m.) injection of corrective DNA, or by ex vivo delivery of DNA to myogenic cells, followed by cell transplantation. Short fragment homologous replacement (SFHR) has been used to repair the exon 23 nonsense transition at the Xp21.1 dys locus in cultured cells and also, directly in tibialis anterior from male mdx mice. Whilst mdx dys locus correction can be achieved in up to 20% of cells in culture, much lower efficiency is evident by i.m. injection. The major consideration for application of targeted gene correction to muscle is delivery throughout relevant tissues. Systemically injected bone marrow (BM)-derived cells from wt C57BL/10 ScSn mice are known to remodel mdx muscle when injected into the systemic route. Provided that non muscle-derived cell types most capable of muscle remodeling activity can be more specifically identified, isolated and expanded, cell therapy seems presently the most favorable vehicle by which to deliver gene correction throughout muscle tissues. Using wt bone marrow as a model, this study investigates systemic application of bone marrow-derived cells as potential vehicles to deliver corrected (ie wt) dys locus to dystrophic muscle. Intravenous (i.v.) and intraperitoneal (i.p.) injections of wt BM were given to lethally and sub-lethally irradiated mdx mice. Despite both i.v. and surviving i.p. groups containing wt dys loci in 100% and less than 1% of peripheral blood nuclei, respectively, both groups displayed equivalent levels of wt dys transcript in muscle RNA. These results suggest that the muscle remodeling activity observed in systemically injected BM cells is not likely to be found in the hemopoietic fraction.

Atomic force microscopy imaging of DNA-cationic liposome complexes optimised for gene transfection into neuronal cells

BACKGROUND

Cationic liposomes represent an important gene delivery system due to their low immunogenicity, but are relatively inefficient, with optimisation of DNA-liposome complexes (lipoplexes) for transfection necessary for each cell type of interest. There have been few studies examining optimisation in neuronal cell types or determining how the structure of lipoplexes affects transfection efficiency.

METHODS

Four commercially available cationic liposome formulations were used to optimise transfection efficiency in neuronal cells. The DNA to liposome ratio and the amount of DNA used in transfections were varied. Transfection efficiency was determined by the percentage of cells positive for the micro-galactosidase reporter gene product. The structure of lipoplexes was studied using atomic force microscopy. Lipoplexes were characterised further using dynamic light scattering to determine size and fluorescence techniques to show DNA compaction.

RESULTS

Optimal transfection conditions were found to differ between immortalised cell lines and primary cells. High transfection efficiencies in immortalised cell lines were achieved predominantly with multivalent cationic liposomes while primary neuronal cells showed optimal transfection efficiency with monovalent cationic liposomes. The structure of lipoplexes was observed with atomic force microscopy and showed globular complexes for multivalent cationic liposomes, while monovalent liposomes gave less compact structures. In support of this finding, high levels of DNA compaction with multivalent liposomes were observed using fluorescence quenching measurements for all DNA to liposome ratios tested. One monovalent liposome showed increasing levels of compaction with increasing liposome amount. Dynamic light scattering showed little change in complex size when the different lipoplexes were studied.

CONCLUSIONS

Optimisation of transfection efficiency was different for cell lines and primary neurons. Immortalised cells showed optimal transfection with multivalent liposomes while primary neurons showed optimal transfection with monovalent liposomes. The charge ratio of the monovalent liposome was below one, suggesting a different mechanism of lipoplex binding and uptake in primary neurons. The structure of lipoplexes, as

Engineering EGFP reporter constructs into a 200 kb human beta-globin BAC clone using GET Recombination

GET Recombination, a simple inducible homologous recombination system for Escherichia coli, was used to target insertion of an EGFP cassette between the start and termination codons of the beta-globin gene in a 200 kb BAC clone. The high degree of homology between the promoter regions of the beta- and delta-globin genes also allowed the simultaneous generation of a delta-globin reporter construct with the deletion of 8.8 kb of intervening sequences. Both constructs expressed EGFP after transient transfection of MEL cells. Similarly, targeting of the EGFP cassette between the promoter regions of the gamma-globin genes and the termination codon of the beta-globin gene enabled the generation of reporter constructs for both (A)gamma- and (G)gamma-globin genes, involving specific deletions of 24 and 29 kb of genomic sequence, respectively. Finally the EGFP cassette was also inserted between the epsilon- and beta-globin genes, with the simultaneous deletion of 44 kb of intervening sequence. The modified constructs were generated at high efficiency, illustrating the usefulness of GET Recombination to generate large deletions of specific sequences in BACs for functional studies. The establishment of stable erythropoietic cell lines with these globin constructs will facilitate the search for therapeutic agents that modify the expression of the individual globin genes in a physiologically relevant manner.

Insertion of disease-causing mutations in BACs by homologous recombination in Escherichia coli

We have used GET Recombination, an inducible homologous recombination system for Escherichia coli, to insert one of the most common thalassaemia mutations into the intact beta-globin locus in a second generation BAC vector. We first inserted a PCR fragment carrying the tetracycline resistance gene (TetR) into the beta-globin gene. All recombinant clones examined contained the TetR gene at the correct target site. Next, a PCR fragment with the IVS I-110 G-->A splicing mutation but no selectable marker was used to replace the TetR gene in a second round of GET Recombination. Recombinant clones were selected by plating on medium containing chlorotetracycline and fusaric acid. Although counterselection for the TetR gene is not very efficient, four recombinant colonies with the IVS I-110 mutation were identified among 480 clones screened. Analysis of the recombinant clones did not show any other modifications or rearrangements. Thus the TetR gene can be used in combination with GET Recombination to introduce point mutations and other modifications in BACs without leaving behind any operational sequences, in order to generate accurate cell and transgenic mouse models for various diseases.

Hb Bart's levels in cord blood and alpha-thalassemia mutations in Cyprus

The purpose of this study was to examine the frequency of alpha-thalassemia in the population of Cyprus using cord blood samples. The levels of Hb Bart's were compared with the hematological indices and the results correlated with the presence of alpha-thalassemia mutations. The protocols for the polymerase chain reaction detection of the six most common alpha-globin mutations encountered in Cyprus were optimized, and the frequency of each mutation was determined through the screening of 495 random cord blood samples. The total allele frequency for the mutations examined was 10.6%, of which 1% is due to the triplication of the alpha-globin genes. The -alpha(3.7 kb) deletion accounts for 72.8% of all detectable mutations, while the--MED-I and -(alpha)-20.5 kb mutations account for 7.8%. The level of Hb Bart's and the MCV and MCH values in cord blood samples were found to correlate closely with the severity of alpha-thalassemia, although the -alpha(3.7 kb) deletion and perhaps other mild alpha-thalassemia mutations may not give detectable Hb Bart's levels. A reasonably accurate estimate of the alpha-thalassemia carrier frequency may be obtained from cord blood studies if Hb Bart's estimates are combined with hematological indices. When molecular methods are added, these give the best way to use cord bloods to survey populations for alpha-thalassemia.

Sandhoff disease in Cyprus: population screening by biochemical and DNA analysis indicates a high frequency of carriers in the Maronite community

In the last 15 years, four patients with the infantile form of Sandhoff disease were diagnosed in four different families in Cyprus (population 703,000, birth rate 1.7%). Three of these cases came from the Christian Maronite community (less than 1% of the population) and one from the Greek community (84% of the population). This relatively large number of patients prompted us to initiate an epidemiological study in order to establish the frequency of the mutant allele in Cyprus. Carrier detection was initially based on the measurement of beta-hexosaminidase A and B in both leucocytes and serum. Using the enzyme test, 35 carriers were identified among 244 random Maronite samples and 15 among 28 Maronites with a family history of Sandhoff disease, but only one carrier was found out of 115 random samples from the Greek community. In parallel to the biochemical screening, DNA studies were undertaken in one of the three Maronite patients and in a Greek carrier related to the Greek patient. These studies resulted in the identification of two novel mutations, a deletion of A at nt76 and a G to C transversion at position 5 of the 5'-splice site of intron 8, which have been published. We subsequently screened the carriers detected in the biochemical study for these two mutations using PCR-based tests. Of 50 Maronite carriers examined, 42 were found to have the nt76 deletion. Eight Maronite samples, designated carriers from the biochemical results, were negative for both mutations. It is possible that these individuals were incorrectly classified as carriers since their enzyme values are equivocal, although the presence of another mutation has not been excluded. Two Greek Cypriot carriers and two obligate Lebanese carriers were negative for both mutations. We conclude that there is a high frequency of Sandhoff disease carriers in the Maronite community of Cyprus, approximately 1 in 7, and that a single mutation predominates in this population.

Chromosome localization and characterization of the mouse and human zinc finger protein 265 gene

The chromosome location and pattern of expression of the gene encoding the zinc finger protein 265 (alias "Zis") in human (ZNF265) and mouse (Zfp265) was determined. By interspecific backcross analysis, we mapped Zfp265 to mouse chromosome 3q. ZNF265 was localized to human chromosome 1p31 by fluorescence in situ hybridization. Since discovery of Zfp265 (in rat) came from studies of changes in renin expression in kidney cell lines, we examined the cell specificity of expression in kidney and also determined hybridization of cDNA with RNA in other tissues. We found that expression was not confined to renin mRNA-containing cells but was ubiquitous. Moreover, the fact that highly conserved homologs of ZNF265p exist in lower organisms (e.g., C4SR in Xenopus), suggests that this protein may have a generalized role in posttranscriptional mechanisms in various cell types and species.

Efficient and precise engineering of a 200 kb beta-globin human/bacterial artificial chromosome in E. coli DH10B using an inducible homologous recombination system

Gene therapy studies require techniques that allow alteration of human genomic DNA sequences. Bacterial artificial chromosome cloning systems (BACs/PACs) bridge the gap between vectors with small inserts and yeast artificial chromosomes (YACs). We report the use of a second generation BAC vector, pEBAC, containing eukaryotic selectable markers and combining some of the best features of the BAC, PAC and HAEC systems, into which a 185 kb sequence containing the human beta-globin gene cluster was retrofitted. To permit the introduction of mutations corresponding to those causing human pathology, we have adapted an inducible homologous recombination system for use in E. coli DH10B cells, the host strain for BACs and PACs. Using this system, we have introduced PCR fragments carrying a selectable marker and a reporter gene downstream of the IVS-110 splicing mutation into a specific site within the beta-globin gene sequence. The use of this inducible system minimises the risk of unwanted rearrangements by recombination between repetitive elements and allows the introduction of relevant modifications or reporters at any specific sequence within BACs/PACs in E. coli DH10B cells.

A genomic region encompassing a cluster of olfactory receptor genes and a myosin light chain kinase (MYLK) gene is duplicated on human chromosome regions 3q13-q21 and 3p13

The olfactory receptor (OR) multigene family is widely distributed in the human genome. We characterize here a new cluster of four OR genes (HGMW-approved symbols OR7E20P, OR7E6P, OR7E21P, and OR7E22P) on human chromosome 3p13 that is contained in an approximately 250-kb region. This region has been physically mapped, and a 106-kb portion containing the OR genes has been sequenced. All the OR sequences are disrupted by frameshifts and stop codons and appear to have arisen through local duplications. A myosin light chain kinase pseudogene (HGMW-approved symbol MYLKP) lies at one end of the OR gene cluster. Sequences spanning the entire region are also present at 3q13-q21, the site of the functional MYLK gene. This region duplicated locally before the divergence of primates, and the two paralogous copies were later separated to sites on either side of the centromere. This study increases our understanding of the evolution of the human genome. The 3p13 cluster is the first example of a tandem array of OR pseudogenes, and duplications of such clusters may account for the accumulation of a large number of pseudogenes in the human genome.

Muscle-specific locus control region activity associated with the human desmin gene

We describe the reproduction of the full pattern of expression of the muscle-specific desmin gene in transgenic mice using a 240-kb genomic clone spanning the human desmin locus. Analysis of RNA from adult tissues demonstrated that this fragment possesses all the necessary genetic regulatory elements required to provide reproducible, site-of-integration-independent, physiological levels of tissue-specific expression that is directly proportional to transgene copy number in all muscle cell types. In situ hybridization revealed that in marked contrast to murine desmin which is strongly expressed in the myotome of the somites, skeletal muscles, the heart, and smooth muscle of the vasculature by 9.5 days postcoitum, human desmin transgene expression was completely absent from smooth muscles, was very weak and restricted to the atrium and outflow tract within the heart, and was expressed at only 5% of murine desmin mRNA levels within the myotome of the somites. The spatial distribution and levels of human and mouse desmin expression were not coincident until 14.5 days postcoitum. Immunohistochemical analysis of human embryos at comparable stages of development showed that this transgene faithfully reproduces the human and not the mouse developmental expression pattern for this gene in transgenic mice. These results indicate that the 240-kb desmin genomic clone is capable of establishing an independent, chromatin domain in transgenic mice and provides the first definitive data for muscle-specific locus control region activity. In addition, our results demonstrate that the behavior of human transgenes in mice should, whenever possible, be compared to expression patterns for that gene in human embryonic as well as adult tissues.

Molecular bases of pseudo-homozygous APC resistance: the compound heterozygosity for FV R506Q and a FV null mutation results in the exclusive presence of FV Leiden molecules in plasma

Pseudo-homozygous APC resistance, the condition resulting from compound heterozygosity for FV R506Q (FV Leiden) and quantitative FV deficiency, provides a natural model to study the interaction between procoagulant and anticoagulant defects. This paper reports a complete FV characterization of a pseudo-homozygous APC resistant thrombotic patient. The expression of the patient's non-Leiden gene was found to be severely impaired both at the mRNA and protein levels. In particular, only FV Leiden molecules were detected in the patient's plasma by immunoblotting, which accounts for the observed marked APC resistance. Analysis of the FV cDNA obtained by reverse transcription of platelet RNA revealed that the mRNA of the non-Leiden gene was extremely reduced in amount. A PAC clone containing the whole FV gene was used to design primers for a complete FV exon scanning. A 2-bp insertion at nucleotide 3706 in the large exon 13 of the non-Leiden gene, predicting a frame-shift and premature termination of protein synthesis, was identified as responsible for the FV defect. Failure to find any case of pseudo-homozygous APC resistance in a large sample (6,804) of blood donors suggests that this condition is extremely rare among normal controls and that its detection is favoured by the thrombotic risk that it may confer.

Complete coding sequence, exon/intron arrangement and chromosome location of ZNF45, a KRAB-domain-containing gene

Zinc finger genes represent a large multigene family present in mammalian and other genomes. A subgroup of these genes contain a conserved motif, the KRAB domain, at the NH2-terminal region, which was recently shown to posses a potent transcriptional repression activity. For one such gene, ZNF45, we determined the complete coding sequence, along with the exon/intron arrangement and the splice junctions. Exon 2 codes exclusively for the KRAB-A element of 42 amino acids, and exon 4 contains the 19 zinc finger repeats. In addition, we generated an EST (expressed sequence tag) from the 3'-end of ZNF45 that we used for polymerase chain reaction screening of a P1-derived genomic library. We isolated a 65-kb clone that was used for localizing this gene on chromosome 19q13.2 by fluorescence in situ hybridization.

Genetic variation of recent Alu insertions in human populations

The Alu family of interspersed repeats is comprised of over 500,000 members which may be divided into discrete subfamilies based upon mutations held in common between members. Distinct subfamilies of Alu sequences have amplified within the human genome in recent evolutionary history. Several individual Alu family members have amplified so recently in human evolution that they are variable as to presence and absence at specific loci within different human populations. Here, we report on the distribution of six polymorphic Alu insertions in a survey of 563 individuals from 14 human population groups across several continents. Our results indicate that these polymorphic Alu insertions probably have an African origin and that there is a much smaller amount of genetic variation between European populations than that found between other population groups.

African origin of human-specific polymorphic Alu insertions

Alu elements are a family of interspersed repeats that have mobilized throughout primate genomes by retroposition from a few "master" genes. Among the 500,000 Alu elements in the human genome are members of the human-specific subfamily that are not fixed in the human species; that is, not all chromosomes carry an Alu element at a particular locus. Four such polymorphic human-specific Alu insertions were analyzed by a rapid, PCR-based assay that uses primers that flank the insertion point to determine genotypes based on the presence or absence of the Alu element. These four polymorphic Alu insertions were shown to be absent from the genomes of a number of nonhuman primates, consistent with their arising as human genetic polymorphisms sometime after the human/African ape divergence. Analysis of 664 unrelated individuals from 16 population groups from around the world revealed substantial levels of variation within population groups and significant genetic differentiation among groups. No significant associations were found among the four loci, consistent with their location on different chromosomes. A maximum-likelihood tree of population relationships showed four major groupings consisting of Africa, Europe, Asia/Americas, and Australia/New Guinea, which is concordant with similar trees based on other loci. A particularly useful feature of the polymorphic Alu insertions is that the ancestral state is known to be the absence of the Alu element, and the presence of the Alu element at a particular chromosomal site reflects a single, unique event in human evolution. A hypothetical ancestral group can then be included in the tree analysis, with the frequency of each insertion set to zero. The ancestral group connected to the maximum-likelihood tree within the African branch, which suggests an African origin of these polymorphic Alu insertions. These data are concordant with other diverse data sets, which lends further support to the recent African origin hypothesis for modern humans. Polymorphic Alu insertions represent a source of genetic variation for studying human population structure and evolution.

PCR probes for chromosome in situ hybridization of large-insert bacterial recombinants

We have developed a procedure for efficient in situ hybridization of bacterial recombinants created with various types of large-insert cloning vectors. Minimal quantities of crude DNA are amplified and labeled during the degenerate-oligonucleotide-primed polymerase chain reaction. The resulting probes generate high-intensity fluorescent hybridization signals on metaphase chromosomes and on interphase nuclei.

A new bacteriophage P1-derived vector for the propagation of large human DNA fragments

We have designed a P1 vector (pCYPAC-1) for the introduction of recombinant DNA into E. coli using electroporation procedures. The new cloning system, P1-derived artificial chromosomes (PACs), was used to establish an initial 15,000 clone library with an average insert size of 130-150 kilobase pairs (kb). No chimaerism has been observed in 34 clones, by fluorescence in situ hybridization. Similarly, no insert instability has been observed after extended culturing, for 20 clones. We conclude that the PAC cloning system will be useful in the mapping and detailed analysis of complex genomes.